Linear Forms in Logarithms and Applications

IRMA Lectures in Mathematics
and Theoretical Physics 28
Linear Forms in Logarithms and Applications

Yann Bugeaud
Yann Bugeaud
The aim of this book is to serve as an introductory text to the theory of linear
forms in the logarithms of algebraic numbers, with a special emphasis on a
Yann Bugeaud
large variety of its applications. We wish to help students and researchers to
learn what is hidden inside the blackbox ‘Baker’s theory of linear forms in
logarithms’ (in complex or in p-adic logarithms) and how this theory applies to
many Diophantine problems, including the effective resolution of Diophantine
Linear Forms in
equations, the abc-conjecture, and upper bounds for the irrationality measure
of some real numbers. Logarithms
Written for a broad audience, this accessible and self-contained book can be
used for graduate courses (some 30 exercises are supplied). Specialists will
appreciate the inclusion of over 30 open problems and the rich bibliography
and Applications
of over 450 references.
ISBN 978-3-03719-183-5
www.ems-ph.org
Bugeaud IRMA 28 | FONT: Rotis Sans | Farben: Pantone 287, Pantone 116 | 170 x 240 mm | RB: 15??? mm
IRMA Lectures in Mathematics and Theoretical Physics 28
Edited by Christian Kassel and Vladimir G. Turaev
Institut de Recherche Mathématique Avancée

CNRS et Université de Strasbourg
7 rue René Descartes
67084 Strasbourg Cedex
France
IRMA Lectures in Mathematics and Theoretical Physics
Edited by Christian Kassel and Vladimir G. Turaev
This series is devoted to the publication of research monographs, lecture notes, and other material
arising from programs of the Institut de Recherche Mathématique Avancée (Strasbourg, France).
The goal is to promote recent advances in mathematics and theoretical physics and to make them
accessible to wide circles of mathematicians, physicists, and students of these disciplines.
For a complete listing see our homepage at www.ems-ph.org.
8 AdS/CFT Correspondence: Einstein Metrics and Their Conformal Boundaries, Oliver Biquard (Ed.)
9 Differential Equations and Quantum Groups, D. Bertrand, B. Enriquez, C. Mitschi, C. Sabbah and
R. Schäfke (Eds.)
10 Physics and Number Theory, Louise Nyssen (Ed.)
11 Handbook of Teichmüller Theory, Volume I, Athanase Papadopoulos (Ed.)
12 Quantum Groups, Benjamin Enriquez (Ed.)
13 Handbook of Teichmüller Theory, Volume II, Athanase Papadopoulos (Ed.)
14 Michel Weber, Dynamical Systems and Processes
15 Renormalization and Galois Theory, Alain Connes, Frédéric Fauvet and Jean-Pierre Ramis (Eds.)
16 Handbook of Pseudo-Riemannian Geometry and Supersymmetry, Vicente Cortés (Ed.)
17 Handbook of Teichmüller Theory, Volume III, Athanase Papadopoulos (Ed.)
18 Strasbourg Master Class on Geometry, Athanase Papadopoulos (Ed.)
19 Handbook of Teichmüller Theory, Volume IV, Athanase Papadopoulos (Ed.)
20 Singularities in Geometry and Topology. Strasbourg 2009, Vincent Blanlœil and Toru Ohmoto
(Eds.)
21 Faà di Bruno Hopf Algebras, Dyson–Schwinger Equations, and Lie–Butcher Series,
Kurusch Ebrahimi-Fard and Frédéric Fauvet (Eds.)
22 Handbook of Hilbert Geometry, Athanase Papadopoulos and Marc Troyanov (Eds.)
23 Sophus Lie and Felix Klein: The Erlangen Program and Its Impact in Mathematics and Physics,
Lizhen Ji and Athanase Papadopoulos (Eds.)
24 Free Loop Spaces in Geometry and Topology, Janko Latschev and Alexandru Oancea (Eds.)
25 Takashi Shioya, Metric Measure Geometry. Gromov‘s Theory of Convergence and Concentration
of Metrics and Measures
26 Handbook of Teichmüller Theory, Volume V, Athanase Papadopoulos (Ed.)
27 Handbook of Teichmüller Theory, Volume VI, Athanase Papadopoulos (Ed.)
Yann Bugeaud
Linear Forms in
Logarithms
and Applications
Author:
Yann Bugeaud
Université de Strasbourg, CNRS
Institut de Recherche Mathématique Avancée, UMR 7501
7, rue René Descartes
67084 Strasbourg CEDEX
France
E-mail: bugeaud@math.unistra.fr
2010 Mathematics Subject Classification: Primary: 11-02, 11J86, 11D; Secondary: 11B37, 11D25, 11D41,
11D59, 11D61, 11D75, 11D88, 11J25, 11J81, 11J82
Key words: Baker’s theory, linear form in logarithms, Diophantine equation, Thue equation, abc-conjecture,
primitive divisor, irrationality measure, p-adic analysis
ISBN 978-3-03719-183-5
The Swiss National Library lists this publication in The Swiss Book, the Swiss national bibliography, and the
detailed bibliographic data are available on the Internet at http://www.helveticat.ch.
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting,
reproduction on microfilms or in other ways, and storage in data banks. For any kind of use permission of
the copyright owner must be obtained.
© 2018 European Mathematical Society

Contact address:
European Mathematical Society Publishing House

Seminar for Applied Mathematics
ETH-Zentrum SEW A21
CH-8092 Zürich
Switzerland
Email: info@ems-ph.org
Homepage: www.ems-ph.org
Typeset using the author’s TEX files: Nicole Bloye, Cardiff, UK

Printing and binding: Beltz Bad Langensalza GmbH, Bad Langensalza, Germany
∞ Printed on acid free paper
987654321
Preface
Hay que saber buscar aunque uno no sepa qué es lo que busca.
Roberto Bolaño
— Les maths, m’sieur, ça fonctionne toujours à côté de ses grolles.

Jean Amila
In 1748 Leonhard Euler published Introductio in analysin infinitorum where, among

several fundamental results, he established the relationship ei D 1 and gave explicitly
the continued fraction expansions of e and e2 . He also made a conjecture concerning the
nature of quotients of logarithms of rational numbers, which can be formulated as follows:
For any two positive rational numbers r, s with r different from 1, the number log s=log r
is either rational (in which case there are non-zero integers a, b such that r a D s b )
or transcendental.
Recall that a complex number is called algebraic if it is a root of a non-zero polynomial
with integer coefficients and a complex number whichpis not algebraic is called transcen-
dental. Euler’s conjecture implies, for example, that 2 2 is irrational (if it were rational,
p p
then log 2 2 divided by log 2, which is equal to 2, would be rational or transcendental).
It can be reformulated as follows:
If a is a positive rational number different from 1 and ˇ an irrational real algebraic
number, then aˇ is irrational.
In 1900, David Hilbert proposed a list of twenty-three open problems and presented
ten of them in Paris at the second International Congress of Mathematicians. His seventh
problem expands the arithmetical nature of the numbers under consideration in Euler’s
conjecture and asks whether (observe that e D . 1/ i )
the expression ˛ ˇ for an algebraic

p
base ˛ different from 0 and 1 and an irrational algebraic
exponent ˇ, e.g. the number 2 or e , always represents a transcendental or at least an
2
irrational number.
Here and below, unless otherwise specified, by algebraic number we mean complex
algebraic number. Hilbert believed that the Riemann Hypothesis would be settled long
before his seventh problem. This was not the case: the seventh problem was eventually
vi Preface
solved in 1934, independently and simultaneously, by Aleksandr Gelfond and Theodor

Schneider, by different methods. They established that, for any non-zero algebraic numbers
˛1 ; ˛2 ; ˇ1 ; ˇ2 with log ˛1 and log ˛2 linearly independent over the rationals (here and
below, log denotes the principal determination of the logarithm function), we have
ƒ2 WD ˇ1 log ˛1 C ˇ2 log ˛2 6D 0:
Since the formulation is different, let us add some explanation. Under the hypotheses of
Hilbert’s seventh problem, the complex numbers log ˛ and log ˛ ˇ are linearly independent
over the rationals and, assuming furthermore that ˛ ˇ is algebraic, we derive from the
Gelfond–Schneider theorem that ˇ, equal to the quotient of the logarithm of ˛ ˇ by the
logarithm of ˛, cannot be algebraic, a contradiction.
Subsequently, Gelfond derived a lower bound for jƒ2 j and, a few years later, he realized
that an extension of his result to linear forms in an arbitrarily large number of logarithms of
algebraic numbers would enable one to solve many challenging problems in Diophantine
approximation and in the theory of Diophantine equations.
This program was realized by Alan Baker in a series of four papers published between
1966 and 1968 in the journal Mathematika. He made the long awaited breakthrough, by
showing that, if ˛1 ; : : : ; ˛n are non-zero algebraic numbers such that log ˛1 ; : : : ; log ˛n are
linearly independent over the rationals, and if ˇ1 ; : : : ; ˇn are non-zero algebraic numbers,
then
ƒn WD ˇ1 log ˛1 C C ˇn log ˛n 6D 0:
In addition, he derived a lower bound for jƒn j, thereby giving the expected extension of the
Gelfond–Schneider theorem. In his work, Baker generated a large class of transcendental
numbers not previously identified and showed how the underlying theory can be used
to answer a wide range of Diophantine problems, including the effective resolution of
many classical Diophantine equations. He was awarded a Fields Medal in 1970 at the
International Congress of Mathematicians in Nice.
It then became clear that further progress, refinements, and extensions of the theory
would have important consequences. This area was at that time flourishing and developing
very rapidly, both from a theoretical point of view (with improvements obtained by Baker
and Feldman, among others, on the lower bounds for jƒn j) and regarding its applications.
A spectacular achievement was the proof by Robert Tijdeman in 1976 that the Catalan
equation x m y n D 1, in the integer unknowns x; y; m; n all greater than 1, has only
finitely many solutions (Preda Mihăilescu established in 2002 that 32 23 D 1 is the only
solution to this equation).
The aim of the present monograph is to serve as an introductory text to Baker’s theory
of linear forms in the logarithms of algebraic numbers, with a special emphasis on a large
variety of its applications, mainly to Diophantine questions. We wish to help students and
researchers to learn what is hidden inside the blackbox “Baker’s theory of linear forms
in logarithms” (in complex or in p-adic logarithms) and how this theory applies to many
Diophantine problems.
Chapter 1 gives the reader a concise historical introduction to the theory. In Chapter 2,
we gather several explicit lower bounds for jƒn j and its p-adic analogue, which were es-
tablished by Waldschmidt, Matveev, Laurent, Mignotte and Nesterenko, Yu, and Bugeaud
Preface vii
and Laurent, and which will be used in the subsequent chapters. In all but one of these
estimates, ˇ1 ; : : : ; ˇn are integers, a special case sufficient for most of the applications.
The lower bounds are then expressed in terms of the maximum B of their absolute values
and take the form
log jƒn j > c.n; D/ .log 2A1 / : : : .log 2An / .log 2B/;
where c.n; D/ is an explicit real number depending only on n and the degree D of the
algebraic number field generated by ˛1 ; : : : ; ˛n and Aj is the maximum of the absolute
values of the coefficients of the minimal defining polynomial of ˛j over the rational
integers, for j D 1; : : : ; n. The crucial achievements of Baker are the logarithmic
dependence on B and the fact that an admissible value for c.n; D/ can be explicitly
computed.
We consider in Chapter 3 Diophantine problems for which the reduction to linear
forms in complex logarithms is almost straightforward. These problems include explicit
lower bounds for the distance between powers of 2 and powers of 3, effective irrationality
measures for n-th roots of rational numbers, lower bounds for the greatest prime factor
of n.n C 1/, where n is a positive integer, perfect powers in linear recurrence sequences
of integers, etc.
Chapter 4 is devoted to applications to classical families of Diophantine equations. In
the works of Thue and Siegel, it was established that unit equations, Thue equations, and
super- and hyperelliptic equations have only finitely many integer solutions, but the proofs
were ineffective, in the sense that they did not yield upper bounds for the absolute values
of the solutions and, consequently, were of very little help for the complete resolution
of the equations. The theory of linear forms in logarithms induced dramatic changes in
the field of Diophantine equations and we explain how it can be applied to establish, in
an effective way, that unit equations, Thue equations, super- and hyperelliptic equations,
the Catalan equation, etc., have only finitely many integer solutions. This chapter also
contains a complete proof, following Bilu and Bugeaud [72], of an effective improvement
of Liouville’s inequality (which states that an algebraic number of degree d cannot be
approximated by rational numbers at an order greater than d ) derived ultimately from an
estimate for linear forms in two complex logarithms proved in Chapter 11.
When the algebraic numbers ˛1 ; : : : ; ˛n occurring in the linear form ƒn are all rational
numbers very close to 1, the lower bounds for jƒn j can be considerably improved. Several
applications of this refinement are listed in Chapter 5. They include effective irrationality
measures for n-th roots of rational numbers close to 1 and striking results on the Thue
equation ax n by n D c.
Chapter 6 presents various applications of the theory of linear forms in p-adic log-
arithms, in particular towards Waring’s problem and, again, to perfect powers in linear
recurrence sequences of integers. It also includes extensions of results established in
Chapter 4: unit equations, Thue equations, super- and hyperelliptic equations have only
finitely many solutions in the rational numbers, whose denominators are divisible by
prime numbers from a given, finite set, and, moreover, the size of these solutions can be
effectively bounded.
Primitive divisors of terms of binary recurrence sequences are discussed in Chapter 7.
We partially prove a deep result of Bilu, Hanrot, and Voutier [77] on the primitive
viii Preface
divisors of Lucas and Lehmer numbers and discuss some of its applications to Diophantine
equations. Then, following Stewart [400], we confirm a conjecture of Erdős and show
n
p n 3, the greatest prime factor of 2
that, for every integer 1 exceeds some positive
real number times n log n= log log n.
In Chapter 8, we follow Stewart and Yu [405] to establish partial results towards the
abc-conjecture, which claims that, for every positive real number ", there exists a positive
real number ."/, depending only on ", such that, for all coprime, positive integers a; b;
and c with a C b D c, we have
Y 1C"
c < ."/ p ;
pjabc
the product being taken over the distinct prime factors of abc. Specifically, we show
how to combine complex and p-adic estimates to prove the existence of an effectively
computable positive real number such that, for all positive coprime integers a; b; and c
with a C b D c, we have
Y 1=3 Y 3
log c < p log p :
pjabc pjabc
There are only a few known applications of the theory of simultaneous linear forms in
logarithms, developed by Loxton in 1986. Two of them are presented in Chapter 9. A pfirst
gives us an upper bound for the number of perfect powers in the interval ŒN; N C N ,
for every sufficiently large integer N . A second shows that, under a suitable assumption,
a system of two Pellian equations has at most one solution.
Given a finite set of multiplicatively dependent algebraic numbers, we establish in
Chapter 10 that these numbers satisfy a multiplicative dependence relation with small
exponents. A key ingredient for the proof is a lower bound for the Weil height of a
non-zero algebraic number which is not a root of unity.
Full proofs of estimates for linear forms in two complex logarithms, which, in partic-
ular, imply lower estimates for the difference between integral powers of real algebraic
numbers, are given in Chapter 11. Analogous estimates for linear forms in two p-adic
logarithms, that is, upper estimates for the p-adic valuation of the difference between in-
tegral powers of algebraic numbers are given in Chapter 12. An estimate for linear forms
in an arbitrary number of complex logarithms is derived in Chapter 4 from the estimate
for linear forms in two complex logarithms established in Chapter 11. While the former
estimate is not as strong and general as the estimates stated in Chapter 2, it is sufficiently
precise for many applications.
We collect open problems in Chapter 13. The thirteen chapters are complemented by
six appendices, which, mostly without proofs, gather classical results on approximation
by rational numbers, the theory of heights, algebraic number theory, and p-adic analysis.
We have tried, admittedly without too much success, to curb our taste for extensive
bibliographies. No effort has been made towards exhaustivity, including in the list of
bibliographic references, and the topics covered in this textbook reflect somehow the
personal taste of the author.
Preface ix
Inevitably, there is some overlap between this monograph and the monograph [376]
of Shorey and Tijdeman, which, although over thirty years old, remains an invaluable
reference for anyone interested in Diophantine equations. In particular, the content of
Chapter 4 (except Section 4.1) is treated in [376] in much greater generality. There is also
some overlap with Sprindžuk’s book [386] and the monograph of Evertse and Győry [182].
Regarding the theory of linear forms in logarithms, Chapters 2 and 11 can be seen as an
introduction to the book of Waldschmidt [432]. As far as we are aware, the content of
Chapters 5, 7, 8, 9, and 12 and several other parts of the present monograph have never
appeared in books.
To keep this book reasonably short and accessible to graduate and post-graduate
students, the results are not proved in their greatest generality and proofs of the best
known lower bounds for linear forms in an arbitrary number of complex (resp., p-adic)
logarithms are not given.
Many colleagues sent me comments, remarks, and suggestions. I am grateful to all of
them. Special thanks are due to Samuel Le Fourn, who very carefully read the manuscript
and sent me many insightful suggestions.
This book was written while I was director of the ‘Institut de Recherche Mathématique
Avancée’.
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Frequently used notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
1. Brief introduction to linear forms in logarithms . . . . . . . . . . . . . . . . . 1

1.1. Linear forms in complex logarithms . . . . . . . . . . . . . . . . . . . . 1
1.2. Linear forms in p-adic logarithms . . . . . . . . . . . . . . . . . . . . . 6
1.3. Linear forms in elliptic logarithms . . . . . . . . . . . . . . . . . . . . . 6
2. Lower bounds for linear forms in complex and p-adic logarithms . . . . . . . . 9
2.1. Lower bounds for linear forms in complex logarithms . . . . . . . . . . . 9
2.2. Multiplicative dependence relations between algebraic numbers . . . . . . 17
2.3. Lower bounds for linear forms in p-adic logarithms . . . . . . . . . . . . 17
2.4. Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3. First applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1. On the distance between powers of 2 and powers of 3 . . . . . . . . . . . 23
3.2. Effective irrationality measures for quotients of logarithms of integers . . 25
3.3. On the distance between two integral S -units . . . . . . . . . . . . . . . 25
3.4. Effective irrationality measures for n-th roots of algebraic numbers . . . . 26
3.5. On the greatest prime factor of values of integer polynomials . . . . . . . 27
3.6. On the greatest prime factor of terms of linear recurrence sequences . . . 30
3.7. Perfect powers in linear recurrence sequences . . . . . . . . . . . . . . . 32
3.8. Simultaneous Pellian equations and Diophantine quadruples . . . . . . . 35
3.9. On the representation of integers in distinct bases . . . . . . . . . . . . . 39
3.10. Further applications (without proofs) . . . . . . . . . . . . . . . . . . . . 43
3.11. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.12. Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4. Classical families of Diophantine equations . . . . . . . . . . . . . . . . . . . 47
4.1. Proof of Baker’s Theorem 1.10 . . . . . . . . . . . . . . . . . . . . . . . 47
4.2. The unit equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3. The Thue equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.4. Effective improvement of Liouville’s inequality . . . . . . . . . . . . . . 55
4.5. The superelliptic and hyperelliptic equations . . . . . . . . . . . . . . . . 55
4.6. The Diophantine equation x 2 C C D y n . . . . . . . . . . . . . . . . . . 57
4.7. Perfect powers at integer values of polynomials . . . . . . . . . . . . . . 60
xii Contents
4.8. The Catalan equation and the Pillai conjecture . . . . . . . . . . . . . . . 62

4.9. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.10. Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5. Further applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.1. Effective irrationality measures for quotients of logarithms of rational
numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2. Effective irrationality measures for n-th roots of rational numbers . . . . . 68
5.3. The Thue equation ax n by n D c . . . . . . . . . . . . . . . . . . . . . 69
5.4. A generalization of Diophantine quadruples . . . . . . . . . . . . . . . . 71
5.5. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.6. Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6. Applications of linear forms in p-adic logarithms . . . . . . . . . . . . . . . . 75
6.1. On the p-adic distance between two integral S -units . . . . . . . . . . . . 75
6.2. Waring’s problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.3. On the b-ary expansion of an algebraic number . . . . . . . . . . . . . . 77
6.4. Repunits and perfect powers . . . . . . . . . . . . . . . . . . . . . . . . 80
6.5. Perfect powers with few binary digits . . . . . . . . . . . . . . . . . . . . 81
6.6. The S -unit equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.7. The Thue–Mahler equation and other classical equations . . . . . . . . . 85
6.8. Perfect powers in binary recurrence sequences . . . . . . . . . . . . . . . 86
6.9. Perfect powers as sum of two integral S -units . . . . . . . . . . . . . . . 88
6.10. On the digital representation of integral S -units . . . . . . . . . . . . . . 89
6.11. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.12. Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
7. Primitive divisors and the greatest prime factor of 2n 1 . . . . . . . . . . . . 95
7.1. Primitive divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.2. Primitive divisors of Lucas–Lehmer sequences . . . . . . . . . . . . . . 98
7.3. The Diophantine equation x 2 C C D y n , continued . . . . . . . . . . . . 101
7.4. On the number of solutions to the Diophantine equation x 2 C D D p n . . 102
7.5. On the greatest prime factor of 2n 1 . . . . . . . . . . . . . . . . . . . 103
7.6. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.7. Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
8. The abc-conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
8.1. Effective results towards the abc-conjecture . . . . . . . . . . . . . . . . 109
8.2. Proofs of Theorems 8.2 and 8.3 . . . . . . . . . . . . . . . . . . . . . . . 110
8.3. Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.4. Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
9. Simultaneous linear forms in logarithms and applications . . . . . . . . . . . . 117
9.1. A theorem of Loxton . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
9.2. Perfect powers in short intervals . . . . . . . . . . . . . . . . . . . . . . 118
9.3. Simultaneous Pellian equations with at most one solution . . . . . . . . . 121
9.4. Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
9.5. Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Contents xiii
10. Multiplicative dependence relations between algebraic numbers . . . . . . . . . 125

10.1. Lower bound for the height of an algebraic number . . . . . . . . . . . . 125
10.2. Existence of small multiplicative dependence relations . . . . . . . . . . 127
11. Lower bounds for linear forms in two complex logarithms: proofs . . . . . . . . 131
11.1. Three estimates for linear forms in two complex logarithms . . . . . . . . 131
11.2. An auxiliary inequality involving several parameters . . . . . . . . . . . . 132
11.3. Proof of Theorem 11.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
11.4. Deduction of Theorem 11.1 from Theorem 11.4 . . . . . . . . . . . . . . 138
12. Lower bounds for linear forms in two p-adic logarithms: proofs . . . . . . . . . 143
12.1. Three estimates for linear forms in two p-adic logarithms . . . . . . . . . 143
12.2. Two auxiliary inequalities involving several parameters . . . . . . . . . . 145
12.3. Proofs of Theorems 12.4 and 12.5 . . . . . . . . . . . . . . . . . . . . . 147
13. Open problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
13.1. Classical conjectures in transcendence theory . . . . . . . . . . . . . . . 157
13.2. Diophantine equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
13.3. Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
A. Approximation by rational numbers . . . . . . . . . . . . . . . . . . . . . . . 167
B. Heights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
B.1. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
B.2. The Liouville inequality . . . . . . . . . . . . . . . . . . . . . . . . . . 176
B.3. Linear forms in one p-adic logarithm . . . . . . . . . . . . . . . . . . . . 177
C. Auxiliary results on algebraic number fields . . . . . . . . . . . . . . . . . . . 179
C.1. Real quadratic fields and Pellian equations . . . . . . . . . . . . . . . . . 179
C.2. Algebraic number fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
D. Classical results on prime numbers . . . . . . . . . . . . . . . . . . . . . . . . 183
E. A zero lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
F. Tools from complex and p-adic analysis . . . . . . . . . . . . . . . . . . . . . 191
F.1. The Schwarz lemma in complex analysis . . . . . . . . . . . . . . . . . . 191
F.2. Auxiliary results on p-adic fields . . . . . . . . . . . . . . . . . . . . . . 191
F.3. The Schwarz lemma in p-adic analysis . . . . . . . . . . . . . . . . . . . 193
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Frequently used notation
log : unless otherwise specified, log z denotes the principal determination

of the logarithm of the non-zero complex number z, that is, writing
z D rei with r positive and in . ; , we have log z D log r Ci .
e: base of the natural logarithm.
deg : degree (of a polynomial, of an algebraic number).
det : determinant.
positive : strictly positive.
bxc : largest integer x.
dxe : smallest integer x.
fg : fractional part.
kk: distance to the nearest integer.
Card : cardinality (of a finite set).
perfect power : integer of the form ab , with a 1 and b 2 integers.
p1 < p2 < : the set of all prime numbers ranged in increasing order.
q1 < < qs : a collection of s distinct prime numbers, not necessarily consecutive.
P Œn : greatest prime factor of the integer n, with P Œ0 D P Œ˙1 D 1.
!.n/ : number of distinct prime factors of the positive integer n, with
!.1/ D 0.
': Euler totient function (Definition D.3).
: Möbius function (Definition D.3).
ˆd .X; Y / : homogeneous d -th cyclotomic polynomial.
h.˛/ : logarithmic Weil height of the algebraic number ˛ (Definition B.4).
L.P / : length of the polynomial P .X1 ; : : : ; Xn / (Definition B.9).
1
j jp : p-adic absolute value, normalized such that jpjp D p .
vp : p-adic valuation, normalized such that vp .p/ D 1.
S: finite, non-empty set of prime numbers (or, sometimes, of places of
an algebraic number field).
xvi Frequently used notation
jnjp 1 .
Q
ŒnS : S -part of the non-zero integer n, defined by ŒnS D p2S
S -unit : rational number whose numerator and denominator are only com-
posed of prime numbers in S .
integral S -unit : rational integer being an S -unit.
ab: the quantity a is less than b times an absolute, positive, effectively
computable real number.
a c1 ;:::;cn b : the quantity a is less than b times an effectively computable, positive,
real number, which depends at most on c1 ; : : : ; cn .
a ineff b : the quantity a is less than b times an absolute, positive real number.
logp ; expp : p-adic logarithm and p-adic exponential functions.
SN : set of permutations of f1; : : : ; N g.
eK ; fK : ramification index and residue degree of a p-adic field K.
a; b; c; p : ideals of an algebraic number field.
ep ; fp : ramification index and residue degree of an ideal p.

: Galois conjugacy.

K, OK , OK , MK , an algebraic number field, its ring of integers, group of units, set of
1
MK , DK , hK : places, set of infinite places, discriminant, and class number.
Chapter 1
Brief introduction to linear forms in logarithms
We start this textbook with a concise introduction to the theory of linear forms in
logarithms. For much more comprehensive historical surveys and many bibliographic
references, the reader is directed to the introductory text [441] and the monographs
[28, 38, 141, 243, 419, 432]. Throughout, unless otherwise specified, log z denotes the
principal determination of the logarithm of the non-zero complex number z, that is, writ-
ing z D rei with r positive and in . ; , we have log z D log r C i. In particular,
log. 1/ is equal to i.
1.1. Linear forms in complex logarithms

A complex number is called algebraic if it is a root of a non-zero polynomial with integer
coefficients. A complex number which is not algebraic is called transcendental. Rational
numbers and their integer roots are obvious examples of algebraic numbers. The first
examples of transcendental numbers were given, and even in a totally explicit form, by
Liouville [264] in 1844, in a note where he proved that a real algebraic irrational number
cannot be too close, in a suitable sense, to rational numbers; see Theorem A.5 for a precise
formulation.
It is one thing to construct transcendental numbers; it is another, much more difficult
one, to prove the transcendence of some explicitly given complex numbers. The first
result in this direction is the proof of the transcendence of e, obtained by Hermite in 1873.
Shortly thereafter, in 1882, Lindemann established that is transcendental. The now
famous Hermite–Lindemann theorem reads as follows.
Theorem 1.1. For any non-zero complex number ˇ, at least one of the two numbers ˇ
and eˇ is transcendental.
Theorem 1.1 includes the transcendence of e (take ˇ D 1) and that of (take ˇ D i).
It also shows that log 2 is transcendental and, more generally, that any determination of
the logarithm of an algebraic number different from 0 and 1 is transcendental.
In 1885 Weierstrass extended Theorem 1.1 as follows.
Theorem 1.2. Let n 2 be an integer. Let ˛1 ; : : : ; ˛n be distinct algebraic numbers.
Then, e˛1 ; : : : ; e˛n are linearly independent over the field of algebraic numbers.
Since monomials in e˛1; : : : ;e˛n are exponentials of integral combinations of ˛1 ; : : : ; ˛n ,
it is easy to show that Theorem 1.2 is equivalent to the following statement: If n 2
2 1. Brief introduction to linear forms in logarithms
and ˛1 ; : : : ; ˛n are algebraic numbers that are linearly independent over the field of ratio-
nal numbers, then e˛1 ; : : : ; e˛n are algebraically independent over the field of algebraic
numbers (this means that no non-zero polynomials with algebraic coefficients vanish at
.e˛1 ; : : : ; e˛n /).
In 1900, David Hilbert proposed a list of twenty-three open problems and presented ten
of them in Paris at the second conference of the International Congress of Mathematicians.
His seventh problem is the following (observe that e D . 1/ i ):
The expression ˛ ˇ for an algebraic base

p
˛ different from 0 and 1 and an irrational
algebraic exponent ˇ, e.g. the number 2 or e , always represents a transcendental or
2
at least an irrational number.

This problem was solved in 1934 independently and simultaneously by Gelfond [197]
and Schneider [359], by different methods.
Theorem 1.3. For any non-zero algebraic numbers ˛1 ; ˛2 ; ˇ1 ; ˇ2 with log ˛1 and log ˛2
linearly independent over the rationals, we have
ˇ1 log ˛1 C ˇ2 log ˛2 6D 0:
Theorem 1.3 was generalized to linear combinations of n logarithms of algebraic

numbers by Baker [16, 17] in 1966 and 1967.
Theorem 1.4. Let n 2 be an integer. Let ˛1 ; : : : ; ˛n be non-zero algebraic numbers
and log any fixed determination of the logarithm function. If log ˛1 ; : : : ; log ˛n are
linearly independent over the rationals, then they are linearly independent over the field
of algebraic numbers.
Shortly thereafter, Baker [18] extended his previous result as follows.
Theorem 1.5. Let n be a positive integer. Let ˛1 ; : : : ; ˛n be non-zero algebraic numbers
and log ˛1 ; : : : ; log ˛n any determinations of their logarithms. If log ˛1 ; : : : ; log ˛n are
linearly independent over the rationals, then 1; log ˛1 ; : : : ; log ˛n are linearly independent
over the field of algebraic numbers.
It readily follows from Theorem 1.5 that the complex number eˇ0 ˛1ˇ1 ˛nˇn is tran-
scendental for all non-zero algebraic numbers ˛1 ; : : : ; ˛n , ˇ0 ; : : : ; ˇn . Furthermore,
Theorem 1.5 includes Theorem 1.1.
Theorems 1.3 to 1.5 show that any expression of the form
ˇ0 C ˇ1 log ˛1 C C ˇn log ˛n ; (1.1)
where ˛1 ; : : : ; ˛n , ˇ1 ; : : : ; ˇn are non-zero algebraic numbers and ˇ0 is algebraic, vanishes

only in trivial cases. A natural question is then to bound from below its absolute value
(when non-zero).
For the sake of simplification, we assume in the discussion below that the algebraic
numbers involved are all rational numbers. Let n 2 be an integer. For j D 1; : : : ; n,
let yxjj be a non-zero rational number, bj a non-zero integer, and set
B WD maxf3; jb1 j; : : : ; jbn jg and Aj WD maxf3; jxj j; jyj jg: (1.2)

1.1. Linear forms in complex logarithms 3
We consider the rational number

x b1 x bn
1 n
ƒ WD 1: (1.3)
y1 yn
Since we wish to bound jƒj from below, we may assume that jƒj 12 . Then, the linear
form in logarithms of rational numbers , defined by
x1 xn
WD log.1 C ƒ/ D b1 log C C bn log ;
y1 yn
satisfies
jƒj
jj 2jƒj:
2
A trivial estimate of the denominator of (1.3) gives that ƒ D 0 or
n
X n
X
log jƒj jbj j log maxfjxj j; jyj jg B log Aj : (1.4)
j D1 j D1
The dependence on the Aj ’s in (1.4) is very satisfactory, unlike the dependence on B. For
applications to Diophantine problems, we require a better estimate in terms of B than the
one given in (1.4), even if it comes with a weaker one in terms of the Aj ’s. For example,
replacing B by o.B/ is sufficient in many cases (see, for example, Theorems 3.10, 3.13,
and 4.9), but not in all cases (see, for example, Theorems 3.3, 3.4, and 5.1). The next
lemma shows that B cannot be replaced by o.log B/.
Lemma 1.6. Let n; a1 ; : : : ; an be integers, all of which are greater than or equal to 2.
Set A D maxfa1 ; : : : ; an g. Then, for every integer B greater than 2n log A, there exist
rational integers b1 ; : : : ; bn with
0 < maxfjb1 j; : : : ; jbn jg B

and
2n log A
ja1b1 anbn 1j :
Bn 1
Lemma 1.6 is a direct consequence of the Dirichlet Schubfachprinzip applied to the
points b1 log a1 C C bn log an with 0 b1 ; : : : ; bn B, which all lie in the interval
Œ0; nB log A.
The first effective improvement of (1.4) was obtained by Gelfond [198] in 1935 in the
case n D 2. He proved that, for multiplicatively independent positive rational numbers
x1 x2
; , for an arbitrary positive real number ", and for all integers b1 ; b2 , not both 0, we
y1 y2
have ˇ x b1 x b2 ˇ
ˇ 1 2
1ˇ x1 ; x2 ; " exp .log B/5C" ;
ˇ
y1 y2
ˇ
y1 y2
where B D maxf3; jb1 j; jb2 jg. Throughout this book, the notation a c1 ;:::;cn b
(resp., a ineff
c1 ;:::;cn b) means that the quantity a is greater than b times an effectively
computable positive real number (resp., a positive real number), which depends at most
on c1 ; : : : ; cn . Subsequently, Gelfond improved his own result and totally explicit estimates
were provided by Schinzel [354] in 1967.
Gelfond [200] also gave an estimate valid for a linear form in an arbitrary number of
logarithms.
Theorem 1.7. Let n 2 be an integer and a1 ; : : : ; an positive rational numbers which are
multiplicatively independent. Let ı be a positive real number. Let b1 ; : : : ; bn be rational
integers, not all zero, and set B D maxf3; jb1 j; : : : ; jbn jg. Then, we have
ja1b1 anbn 1j ineff

n;a1 ;:::;an ; ı exp. ıB/:
The proof of Theorem 1.7, which rests on a theorem of Siegel, does not enable us to
compute effectively the implicit numerical constant. In his book Gelfond [200] pointed
out the importance of getting an effective version of Theorem 1.7.
For n 3, the first non-trivial effective lower bound for the quantity ƒ defined in (1.3)
was given by Baker [16] in 1966. With B as in (1.2), he obtained that either ƒ D 0 or, for
every positive real number ", we have
ˇ x b1 x bn ˇ
ˇ 1 n
.log B/nC1C" :

1ˇ n; x1 ;:::; xn ; " exp
ˇ
y1 yn
ˇ
y1 yn
His result is much more general and applies to expressions of the form (1.1).
Shortly thereafter, Feldman [185, 186] established the following refinement of Baker’s
lower bound.
Theorem 1.8. Let n 2 be an integer and a1 ; : : : ; an positive rational numbers which
are multiplicatively independent. Let b1 ; : : : ; bn be rational integers, not all zero, and
set B D maxf3; jb1 j; : : : ; jbn jg. Then, there exists a positive, effectively computable real
number C , depending only on n; a1 ; : : : ; an , such that
ja1b1 anbn 1j exp. C log B/ D B C

: (1.5)
Lemma 1.6 shows that the dependence on B in the estimate (1.5) is essentially best
possible, but, for applications, it is much desirable to determine precisely how C depends
on the rational numbers a1 ; : : : ; an .
A first explicit result in this direction was given by Baker [19] in 1968. Throughout
this chapter, the height of an algebraic number is the naïve height, that is, the maximum of
the absolute values of the coefficients of its minimal defining polynomial over the integers.
We reproduce below a consequence of the main theorem of [19].
Theorem 1.9. Assume that n 2 and that ˛1 ; : : : ; ˛n are non-zero algebraic numbers,
whose degrees do not exceed D and whose heights do not exceed A, where D 4 and
A 4. Let ı be a real number with 0 < ı 1. Let b1 ; : : : ; bn be rational integers, not
all zero, and set B D maxf3; jb1 j; : : : ; jbn jg. If
ınB
e
j˛1b1 ˛nbn 1j < ;
2
then
2 2
B < .4.nC1/ ı 1
D 2.nC1/ log A/.2nC3/ :
1.1. Linear forms in complex logarithms 5
In many applications, especially to classical families of Diophantine equations, we are

led to bound from below expressions of the form j˛1b1 ˛nbn ˛nC1 1j, where ˛nC1 has a
large height; see Chapter 4. In this respect, Theorem 1.9 is not plainly satisfactory, since
˛1 ; : : : ; ˛n play the same rôle, irrespective of their height and their exponent. Its following
refinement, established by Baker [26], appears to be very useful.
Theorem 1.10. Assume that n 1 and let ˛1 ; : : : ; ˛nC1 be non-zero algebraic numbers
of degree at most D. Let the heights of ˛1 ; : : : ; ˛n and ˛nC1 be at most A and AnC1
respectively, where A 2 and AnC1 2. Let ı be a positive real number. Let b1 ; : : : ; bn
be integers, not all zero, and set B D maxf3; jb1 j; : : : ; jbn jg. If
0 < j˛1b1 ˛nbn ˛nC1 1j < e ıB

;
then
B n;D;A;ı log AnC1 :
In the present textbook, we give a complete proof of Theorem 1.10. We derive it from
a lower bound for linear forms in two logarithms; see Section 4.1.
In 1975, Baker [27] established a subsequent refinement of Theorem 1.8.
Theorem 1.11. Assume that n 2 and that ˛1 ; : : : ; ˛n are non-zero algebraic numbers,
whose degrees do not exceed D. Assume that, for j D 1; : : : ; n, the height of ˛j does
not exceed Aj , where Aj 2. Let b1 ; : : : ; bn be rational integers, not all zero, and set
B D maxf3; jb1 j; : : : ; jbn jg. If ˛1b1 ˛nbn is not equal to 1, then
j˛1b1 ˛nbn 1j > B C … log …

; (1.6)
where
… D log A1 log An
and C is an effectively computable real number depending only on n and D.

As was pointed out by Baker below the statement of his theorem, it would be of much
interest to eliminate log … in (1.6). Also, Theorem 1.11 does not include Theorems 1.9
and 1.10, nor the main result of Baker’s paper [25].
The deletion of the log … factor (which is mainly interesting from a theoretical point of
view, but also has several applications, see, for example, Theorem 3.6) has been achieved
independently by Wüstholz [440] and by Philippon and Waldschmidt [329]. Thus, at the
end of the ’80s, it was established that, with ƒ, A1 ; : : : ; An , and B as in (1.2) and (1.3),
there exists an effectively computable real number c.n/, depending only on the number n
of rational numbers involved in (1.3), such that the lower estimate
ˇ x b1 x bn ˇ
ˇ 1 n
1ˇ c.n/ log A1 log An log B
ˇ
logˇ (1.7)
y1 yn
holds, when ƒ is non-zero. Since then, several authors have managed to considerably
reduce the value of the real number c.n/. However, it remains an open problem to replace
the product log A1 log An by the sum log A1 C C log An ; see Chapter 13.
1.2. Linear forms in p-adic logarithms

Let p be a prime number. In parallel with the development of the theory of linear forms
in complex logarithms, progress has been regularly made towards its p-adic analogue.
For a non-zero rational number x, let vp .x/ denote the exponent of p in the decompo-
sition of x as a product of prime powers. With ƒ; A1 ; : : : ; An , and B as in (1.2) and (1.3),
a trivial estimate shows that
x b1 x bn log.2Ab1 Abn / n
1 n 1 n B X
vp 1 1C log Aj : (1.8)
y1 yn log p log p
j D1
The theory of p-adic linear forms in logarithms gives a much better result in terms of the
dependence on B, namely, that there exists an effectively computable real number c 0 .n/,
depending only on the number n of rational numbers involved, such that the upper estimate
x b1 x bn
1 n
vp 1 c 0 .n/ p log A1 log An log B (1.9)
y1 yn
holds, when ƒ is non-zero. In terms of B, this is much better than (1.8) and analogous
to (1.7). One of the major open problems in the theory is to remove (or at least to improve)
the dependence on p in (1.9), which remains very unsatisfactory and is ultimately a
consequence of the fact that the radius of convergence of the p-adic exponential function
is finite.
A good reference for an historical introduction is [448]. Mahler [278] established
in 1932 the p-adic analogue of the Hermite–Lindemann theorem and three years later [280]
the p-adic analogue of the Gelfond–Schneider theorem. Gelfond [199] proved a quan-
titative estimate for linear forms in two p-adic logarithms, which was later refined by
Schinzel [354].
Estimates for linear forms in an arbitrary number of p-adic logarithms were obtained by
Brumer [99], Sprindžuk [384, 385], Coates [149], Kaufman [238], Baker and Coates [34],
van der Poorten [335], Dong [168, 169], and Yu [443, 445–447], among others.
1.3. Linear forms in elliptic logarithms

A few years after the birth of the theory of linear forms in logarithms, it was realized
that the techniques used in the proofs can be applied to any commutative algebraic group.
Initial steps towards the derivation of elliptic analogues were made by Baker [23], who
considerably extended earlier results of Schneider [360]. Elliptic curves with complex mul-
tiplication were studied by Masser [284] in 1975 and, a year later, Coates and Lang [151]
established the first lower bounds in the case of Abelian varieties with complex multi-
plication. Subsequently, Philippon and Waldschmidt [330] and Hirata-Kohno [229, 230]
obtained rather general statements.
In the case of elliptic logarithms, the first totally explicit estimates were given by
David [162]. These have applications to the complete determination of the integral points
on an elliptic curve, as is very well explained by Stroeker and Tzanakis [408], Gebel,
Pethő, and Zimmer [196], and Tzanakis [420, 421].
1.3. Linear forms in elliptic logarithms 7
Let n be a positive integer and E1 ; : : : ; En elliptic curves over an algebraic number

field K. For j D 1; : : : ; n, consider a Weierstrass model of Ej and }j the associated
Weierstrass function. Let uj be a complex number such that }j .uj / is in K [ f1g.
Such a complex number uj is an elliptic logarithm of an algebraic point of Ej . Also,
let ˇ0 ; ˇ1 ; : : : ; ˇn be elements of K. Let B 3 denote an upper bound for the heights
of ˇ0 ; ˇ1 ; : : : ; ˇn .
Set
ƒe WD ˇ0 C ˇ1 u1 C C ˇn un :
David and Hirata-Kohno [163] established that there exists an effectively computable
positive real number c, which depends only on K; n; u1 ; : : : ; un and the curves E1 ; : : : ; En ,
such that we have jƒe j B c if ƒe is non-zero.
For the state-of-the-art and generalisations to Abelian varieties, the reader is directed
to [163] and to Gaudron’s papers [192–194].
Lower bounds for linear forms in p-adic elliptic logarithms were given by Rémond
and Urfels [345] and Hirata-Kohno and Takada [232]; see also Fuchs and Pham [189].
Chapter 2
Lower bounds for linear forms
in complex and p-adic logarithms
We list below several estimates for linear forms in complex and p-adic logarithms, which
will be used throughout the book. We give only few bibliographic references and direct
the reader to the monograph of Waldschmidt [432] for further information, in particular
to its Section 10.4. Most of the results quoted are corollaries of more precise estimates, so
the reader wishing to apply the theory of linear forms in logarithms should better consult
the original papers to find the sharpest bounds available to date.
In this chapter we write completely explicit estimates. This will, however, not be
the case for most of the results presented in the next chapters, where we will often make
no effort to give explicit values for the numerical constants. Throughout, h denotes the
(logarithmic) Weil height; see Definition B.4.
2.1. Lower bounds for linear forms in complex logarithms

We start with a general theorem of Waldschmidt [430,432] on inhomogeneous linear forms
in logarithms of algebraic numbers with algebraic coefficients.
Theorem 2.1. Let n 1 be an integer. Let ˛1 ; : : : ; ˛n be non-zero algebraic numbers. Let
log ˛1 ; : : : ; log ˛n be determinations of their logarithm and assume that log ˛1 ; : : : ; log ˛n
are linearly independent over the rationals. Let ˇ0 ; : : : ; ˇn be algebraic numbers, not
all zero. Let D be the degree over Q of the number field Q.˛1 ; : : : ; ˛n ; ˇ0 ; : : : ; ˇn /.
Let E; E , and A1 ; : : : ; An be real numbers with
D
E E 1=D e1=D ; E e; E ;
log E
and
n E log E o
log Aj max h.˛j /; j log ˛j j; ; 1 j n:
D D
Let B be a real number with
D log Aj
B E ; B max ; log B max h.ˇj /:
1j n log E 0j n
10 2. Lower bounds for linear forms in complex and p-adic logarithms
Then, we have
log jˇ0 C ˇ1 log ˛1 C C ˇn log ˛n j

2nC25 n3nC9 D nC2 log A1 : : : log An log B log E .log E/ n 1
:
Assume that we are in the homogeneous rational case, that is, assume that ˇ0 D 0
and ˇ1 ; : : : ; ˇn are rational integers b1 ; : : : ; bn with bn ¤ 0. Let B 0 be a real number
satisfying
n jb j jbj j o
n
B 0 E ; B 0 max C :
1j n 1 log Aj log An
Then, we have
log jb1 log ˛1 C C bn log ˛n j

2nC26 n3nC9 D nC2 log A1 : : : log An log B 0 log E .log E/ n 1
:
In particular, choosing E D e and E D 3D, we get
log jb1 log ˛1 C C bn log ˛n j

2nC26 n3nC9 D nC2 log.3D/ log A1 : : : log An log B 0 :
Proof. This follows from Theorem 9.1 of [432], taking into account Remark 3 on page 303
and Proposition 9.18 of [432].
By Proposition 9.21 of [432], the assumption in Theorem 2.1 that log ˛1 ; : : : ; log ˛n are
linearly independent over the rationals can be relaxed to the assumptions ˇ0 C ˇ1 log ˛1 C
C ˇn log ˛n ¤ 0 and
D 3 .log B0 /.log Aj /.log E / .log D/.log E/2 ; 1 j n;
where B0 D B in the general case and B0 D B 0 in the homogeneous rational case.

Theorem 2.1 plainly includes Theorem 1.5. It also contains Theorem 1.11 and the
lower bound (1.7). Taking n 2 and bn D 1 in the homogeneous rational case of
Theorem 2.1 and setting B WD maxf3; jb1 j; : : : ; jbn 1 jg, we get
B
log jb1 log ˛1 C C bn 1 log ˛n 1 C log ˛n j n;D log A1 : : : log An log :
log An
This has many important applications, in particular to Diophantine equations; see Theo-
rems 4.1, 4.3, and 4.5. Assuming that there exists a real number ı such that 0 < ı 12
and jb1 log ˛1 C C bn 1 log ˛n 1 C log ˛n j < e ıB , we deduce that
B n;D ı 1 log A1 : : : log An 1 log ı 1 log.D n 1 A1 : : : An 1 / log An :

This (and the discussion below explaining how the linear form (2.1) and the quantity (2.2)
are related) shows that Theorem 2.1 also contains Theorem 1.10.
In the homogeneous rational case, a result similar to Theorem 2.1 was proved in-
dependently by Baker and Wüstholz [37, 38]. Since their estimate (which has a better
2.1. Lower bounds for linear forms in complex logarithms 11
dependence on n than in Theorem 2.1, namely the factor n3n is replaced by n2n ) does not
include the useful parameters E and B 0 , and is superseded by Theorem 2.2 below, we do
not quote it.
The parameter E originates in papers by Shorey [367, 368] and is of interest when
˛1 ; : : : ; ˛n are real and very close to 1, in which case it can be chosen to be very large.
By assumption, log E cannot exceed D min1j n log Aj . However, in some cases, it can
be taken close to this quantity. To see this in the homogeneous rational case, assume that,
for j D 1; : : : ; n, we have ˛j D 1 C x1j , for an integer xj 3. Then, setting
E D E D min xj and Aj D xj C 1; 1 j n;
1j n
we see that, since B 0 E, Theorem 2.1 implies the lower bound
log A1 : : : log An
log jb1 log ˛1 C Cbn log ˛n j n log maxf3; jb1 j; : : : ; jbn jg:
.min1j n log Aj /n 1
In the most favourable cases, for example when there exists a real number M such that
max1j n xj .min1j n xj /M , we get

log jb1 log ˛1 C C bn log ˛n j n Mn 2
max log Aj log maxf3; jb1 j; : : : ; jbn jg;
1j n
thus replacing the product of the log Aj , as it occurs in the statement of Theorem 2.1, with
their maximum. Further explanations are given below Theorem 2.5, in Chapter 5, and in
Section 10.4.3 of [432].
In the course of this textbook, we mention in passing a few applications of the first
assertion of Theorem 2.1 (at the beginning of Section 3.3 and in Section 3.10), but we
only apply estimates for homogeneous linear forms in logarithms with integer coefficients.
Therefore, we focus our attention on the second assertion of Theorem 2.1 and its subsequent
improvements and refinements.
Let n 2 be an integer. Let ˛1 ; : : : ; ˛n be non-zero algebraic numbers. Let b1 ; : : : ; bn
be integers. Let log ˛1 ; : : : ; log ˛n be any determination of the logarithms of ˛1 ; : : : ; ˛n .
The theory of linear forms in logarithms provides us with lower bounds for the absolute
value of the linear form
b1 log ˛1 C C bn log ˛n ; (2.1)
when it is non-zero. This yields lower bounds for the quantity
j˛1b1 : : : ˛nbn 1j; (2.2)
which occurs frequently in Diophantine questions. We choose below to mainly consider

quantities of the form (2.2) and not (2.1). When ˛1 ; : : : ; ˛n are all real numbers, both
forms are essentially equivalent since log.1 C x/ D x C O.x 2 / in the neighborhood of
the origin. This is however not the case when complex non-real numbers are among
˛1 ; : : : ; ˛n . Indeed, denoting by log the principal determination of the logarithm, to say
that (2.2) is small does not imply that b1 log ˛1 C C bn log ˛n is close to 0, but merely
that b1 log ˛1 C C bn log ˛n is close to an integer multiple of 2i. Precisely, we derive

the existence of an integer b0 with jb0 j jb1 j C C jbn j and such that the linear form
b0 log. 1/ C b1 log ˛1 C C bn log ˛n
is close to 0.
The next statement is a corollary of the, at present time, best known general estimate,
due to Matveev [290, 291]. The crucial improvement on the earlier results [37, 430]
concerns the dependence on the number n of logarithms: it is exponential in n, and not of
the form ncn as in the earlier estimates.
Theorem 2.2. Let n 1 be an integer. Let ˛1 ; : : : ; ˛n be non-zero algebraic numbers.
Let D be the degree over Q of a number field containing ˛1 ; : : : ; ˛n . Let A1 ; : : : ; An be
real numbers with
n j log ˛j j 0:16 o
log Aj max h.˛j /; ; ; 1 j n:
D D
Let b1 ; : : : ; bn be integers and set
B D maxfjb1 j; : : : ; jbn jg
and
n n log Aj oo
B 00 D max 1; max jbj j W1j n :
log An
Then, we have
log j˛1b1 : : : ˛nbn 330nC4 .nC1/5:5 D nC2 log.eD/ log A1 : : : log An log.enB/
1j >
(2.3)
and, if n 2 and ˛1 ; : : : ; ˛n are all real numbers, we get the better lower bound
log j˛1b1 : : : ˛nbn 2 30nC3 n4:5 D nC2 log.eD/ log A1 : : : log An log.eB/:
1j >
(2.4)
The same statements hold with B replaced by maxfB 00 ; nB=.D log An /g in (2.3) and
with B replaced by B 00 in (2.4).
Proof that Theorem 2.2 follows from Matveev’s results. Inequality (2.4) with B replaced
or not by B 00 is an immediate consequence of Corollary 2.3 of Matveev [291]. Denote
by log the principal determination of the logarithm. If j˛1b1 : : : ˛nbn 1j < 21 , then there
exists an integer b0 , with jb0 j n B, such that
WD jb0 log. 1/ C b1 log ˛1 C C bn log ˛n j
satisfies j˛1b1 : : : ˛nbn 1j

2
. Noticing that j log. 1/j D and h. 1/ D 0, we set

log A0 D D and deduce (2.3) from Corollary 2.3 of Matveev [291].
Since
log A0 nB
jb0 j ;
log An D log An
we see that B can be replaced by maxfB 00 ; nB=.D log An /g in (2.3).
We stress that the definition of B 0 in Theorem 2.1 is slightly different from that of B 00
in Theorem 2.2. The lower bound (2.4) with B 00 in place of B is crucial for the proof of
Theorem 3.8.
Many Diophantine problems can be reduced to lower bounds for linear forms in two or
three logarithms. While, in the case of three logarithms, we do not have very satisfactory
estimates (but see [298]), Laurent, Mignotte, and Nesterenko [254] obtained in 1995 rather
sharp lower bounds for linear forms in two logarithms. The quality of their result is an
illustration of the method of interpolation determinants, which was introduced in this
context by Laurent [252]. The current best known estimates have been established by
Laurent in 2008 in [253]. We display below an estimate obtained in Corollary 1 of [253]
and three consequences of the main result of [254].
Theorem 2.3. Let ˛1 and ˛2 be multiplicatively independent algebraic numbers. Set
D 0 D ŒQ.˛1 ; ˛2 / W Q=ŒR.˛1 ; ˛2 / W R. Let A1 and A2 be real numbers such that
n 1 j log ˛j j o
log Aj max h.˛j /; 0 ; ; j D 1; 2:
D D0
Let b1 and b2 be integers, not both zero, and set
n jb1 j jb2 j 20 o
log B 0 D max log C C 0:21; ;1 :
D 0 log A2 D 0 log A1 D0
Then, we have the lower bound
log jb1 log ˛1 C b2 log ˛2 j 25:2 D 04 .log A1 /.log A2 /.log B 0 /2 : (2.5)
For n D 2, the numerical constant in (2.4) exceeds 109 . It has been substantially
reduced in Theorem 2.3. This is crucial for applications to the complete resolution of
Diophantine equations. Very roughly speaking, when a Diophantine problem can be
reduced to linear forms in only two logarithms, then it can (often, this is not always true!
See Problem 13.12) be completely solved.
A weaker version (weaker only in terms of the numerical constants) of Theorem 2.3
is proved in this book; see Theorem 11.1. Note also that, unlike in Theorem 2.3, the
algebraic numbers ˛1 and ˛2 are not assumed to be multiplicatively independent in
Theorem 11.1. This allows us to deduce directly from Theorem 2.3 a slightly weaker
version of Theorem 2.6, see Theorem 11.3.
Observe that the dependence on B 0 in Theorem 2.3 is not best possible and worse
than in Theorems 2.1 and 2.2, since log B 0 occurs squared. Gouillon [202] established
a lower bound for jb1 log ˛1 C b2 log ˛2 j, where the dependence on B 0 occurs through a
factor .log B 0 / only. His numerical constants being not so small, it is better for most of
the applications to use the bounds established in [253, 254]. However, to have the best
possible dependence in B 0 is essential for the proofs of Theorems 3.3 and 5.1.
As in Theorem 2.1, an extra parameter E was introduced in [254]. We reproduce
below Corollaire 3 of [254].
Theorem 2.4. Let ˛1 and ˛2 be multiplicatively independent positive real algebraic
numbers. Let D be the degree of the number field Q.˛1 ; ˛2 /. Let A1 and A2 be real
numbers such that

n 1 j log ˛j j o
log Aj max h.˛j /; ; ; j D 1; 2:
D D
Let b1 and b2 be integers, not both zero. Let E be a real number with
n D log A D log A2 o
1
E 1 C min ; (2.6)
j log ˛1 j j log ˛2 j
and set
n jb1 j jb2 j 10 log E 1 o
log B 0 D max log C C log log E C 0:47; ; :
D log A2 D log A1 D 2
Assume furthermore that 2 E minfA3D=2

1 ; A3D=2
2 g. Then,
log jb1 log ˛1 C b2 log ˛2 j 35:1D 4 .log A1 /.log A2 /.log B 0 /2 .log E/ 3

:
In [254], the authors defined the parameter E to be equal to the right hand side of (2.6).
However, it easily follows from the proof that Theorem 2.4 as stated is correct.
A weaker version (weaker only in terms of the numerical constants) of Theorem 2.4 is
proved in this book; see Theorem 11.2.
The following consequence of Theorem 2.4 emphasizes the rôle of the parameter E in
a particular case which occurs frequently in applications.
Theorem 2.5. Let x1 ; x2 ; y1 ; y2 be positive integers with x1 2, x1 ¤ y1 , and
y2 < x2 56 y2 . Let b be a positive integer and assume that . yx11 /b ¤ yx22 . Define
the parameter by
x2
D 1 C x2 :
y2
Then, we have log x2 > 1 and
ˇ y b x ˇ 35:2 n log b o2
ˇ 1 2
1ˇ .log x1 / max 1 C ; 10 :
ˇ
logˇ (2.7)
x1 y2 log x2
Since yx22 > 1, we always have < 1. The estimate is stronger when is very close
to 1, that is, when yx22 is very close to 1. Under the assumption of Theorem 2.5, we get
instead of (2.5) a lower bound of the form
.log A1 /.log A2 /
.log B 0 /2 ;
log. yx22 1/
with A1 D maxfx1 ; 3g and A2 D maxfx2 ; 3g. This crucial improvement upon the
“classical” estimate (2.5) turns out to have many spectacular applications, some of which
being given in Chapter 5; see Theorems 5.2, 5.4, and 5.6.
Proof of Theorem 2.5 assuming Theorem 2.4. If y1 > x1 , then the left hand side of (2.7)
exceeds log. yx11 1/, thus it exceeds log x1 and (2.7) holds.
If y1 < x1 and x1 5, then . yx11 /b 45 and (2.7) holds since yx22 56 .
If x1 > y1 and yx11 and yx22 are multiplicatively dependent, then there exist coprime
positive integers c; d and positive integers u; v such that yx11 D . dc /u and yx22 D . dc /v . We
then get
ˇ y b x ˇ ˇ d ub v ˇ 1 1
ˇ 1 2
1ˇ D ˇ 1ˇ ;
ˇ ˇ ˇ
x1 y2 c maxfc; d g x1
ˇ
and the theorem holds.

Thus, we assume that yx11 and yx22 are multiplicatively independent, x1 > y1 , and
x1 6. Observe that (2.7) holds if x1 < x2 =2. To see this, it suffices to observe that,
since y1 < x1 , we then get
y b x 1 1 1
1 2
0< 1 1C <1 ;
x1 y2 x1 x2 2x1
which implies that
ˇ y b x ˇ 1
ˇ 1 2
1ˇ :
ˇ
x1 y2 2x1
ˇ
Thus, in the sequel, we assume that x2 2x1 . We apply Theorem 2.4 with ˛1 D yx11
and ˛2 D yx22 . Then, D D 1, A1 D x1 and A2 D x2 (note that the assumption y2 < x2 65 y2
implies that x2 6). Furthermore, we set
E D x2 :
Observe that E 5, since 5x2 6y2 . This proves the first assertion of the theorem.
Furthermore, we have E minfx13=2 ; x23=2 g, since x1 6 and < 1.
We deduce from log.1 C x2 / x2 that
log x2 log x2
1C x2 D 1 C 1 C x2 log x2 E:
log y2 log.1 C x2 /
Furthermore, we may assume that

ˇ y b x ˇ
ˇ 1 2
1ˇ x1 10 ;
ˇ
x1 y2
ˇ
since otherwise the theorem clearly holds. Using this inequality and 5 x2 2x1 , we
deduce that
x1 x2
b log log C 2x1 10 x2 C 211 x2 10 2x2 ;
y1 y2
thus
log x1 b
1C x1 1 C x2 log x1 E;
log y1 2
since x1 6. Consequently, we have checked that

n log x log x o
1 2
E 1 C min ; :
log yx11 log yx22
The assumptions of Theorem 2.4 are then satisfied and we derive that
ˇ x x ˇ log x log B 0 2
2 1 ˇ 2
b log ˇ 35:1 .log x1 /
ˇ
logˇlog
y2 y1 log E log E
35:1 log B 0 2
.log x1 / ;
log E
where
log logbx2 C log1x1
( )
log B 0 log log E C 0:47
D max C ; 10
log E log E log E
n log b o
max 1 C ; 10 :
log x2
1
We conclude by using that every real number z with jzj 2
satisfies j log.1 C z/j 2jzj.
This completes the proof of the theorem.
For some applications, we need a lower bound for quantities of the shape j˛ b 1j,
where ˛ is a complex algebraic number of modulus 1. Such an estimate does not follow
from Theorem 2.3, since in its statement ˛1 and ˛2 are assumed to be multiplicatively
independent. We display a consequence of Théorème 3 of [254].
Theorem 2.6. Let ˛ be a complex algebraic number of modulus 1 which is not a root of
unity. Let b be a positive integer. Set
ŒQ.˛/ W Q n 20 j log ˛j o
D0 D ; log A D max 0
; 11 0
C h.˛/ ;
2 D D
and
n 17 1 b 5:1 o
log B 0 D max 0
; p ; log C 2:35 C 0 :
D 10 D 0 25 D
Then, we have
log j˛ b 1j 9.D 0 /3 .log A/.log B 0 /2 :
Proof that Theorem 2.6 follows from Théorème 3 of [254]. Recall that log denotes the
principal determination of the logarithm. If j˛ b 1j < 31 , then there exists an integer b0
with jb0 j b such that
jb0 log. 1/ C b log ˛j
j˛ b 1j :
2
We then use the inequalities
1 1 1 1 1
C C
2D 0 log A 68:9 40 68:9 25
to get from Théorème 3 of [254] that
log jb0 log. 1/ C b log ˛j 8:87.D 0 /3 .log A/.log B 0 /2 :
This yields the desired estimate.
2.2. Multiplicative dependence relations between algebraic numbers 17
2.2. Multiplicative dependence relations between algebraic numbers

The results displayed in the previous section provide us with lower bounds for the quantity
jb1 log ˛1 C C bn log ˛n j;
when it is non-zero. However, in some situations, we derive linear forms in logarithms

that are equal to zero. It is then often useful to find rational integers b10 ; : : : ; bn0 , not all
zero, with small absolute values and such that
b10 log ˛1 C C bn0 log ˛n D 0:
A result of this type was given by Loxton and van der Poorten [271]. We quote below a
version of Waldschmidt [432].
Theorem 2.7. Let m 2 be an integer and ˛1 ; : : : ; ˛m multiplicatively dependent non-
zero algebraic numbers. Let log ˛1 ; : : : ; log ˛m be any determination of their logarithms.
Let D be the degree of the number field generated by ˛1 ; : : : ; ˛m over Q. For j D 1; : : : ; m,
let Aj be a real number satisfying
n j log ˛j j o
log Aj max h.˛j /; ;1 :
D
Then there exist rational integers n1 ; : : : ; nm , not all zero, such that
n1 log ˛1 C C nm log ˛m D 0
and
m 1 .log A1 / .log Am /
jnj j 11.m 1/D 3 ; for j D 1; : : : ; m.
log Aj
A full proof of Theorem 2.7, together with additional references, is given in Chapter 10.
Theorem 2.7 is used in the proof of Theorem 3.16.
2.3. Lower bounds for linear forms in p-adic logarithms

Let p be a prime number and K an algebraic number field. Let OK denote the ring
of integers of K. Let p be a prime ideal in K, lying above the prime number p, and
denote by ep its ramification index, that is, the exponent of p in the decomposition of
the ideal pOK in a product of prime ideals; see Section B.1. For a non-zero algebraic
number ˛ in K, let vp .˛/ denote the exponent of p in the decomposition of the fractional
ideal ˛OK in a product of prime ideals and set
vp .˛/
vp .˛/ D :
ep
This defines a valuation vp on K which extends the p-adic valuation vp on Q normalized

in such a way that vp .p/ D 1.
Let ˛1 ; : : : ; ˛n be elements of K and b1 ; : : : ; bn non-zero rational integers. We look

for an upper bound for the quantity
vp .˛1b1 : : : ˛nbn 1/;
where b1 ; : : : ; bn denote rational integers, not all zero, when ˛1b1 : : : ˛nbn is not equal to 1.
We begin with an easy estimate.
Theorem 2.8. Let p be a prime number, b 2 an integer, and ˛ an algebraic number of
degree D. If ˛ b is not equal to 1, then
log b pD 1 log 2
vp .˛ b 1/ < C 2D h.˛/ C 2D :
log p log p log p
The proof of Theorem 2.8 is given in Section B.3. Theorem 2.8 is trivial unless
vp .˛/ D 0, in which case the factor .p D 1/ corresponds to the upper bound for the
smallest positive integer s such that vp .˛ s 1/ is positive. The dependence on p in
Theorem 2.8 was slightly improved by Yamada [442] by means of a subtle variation
of the proof of the main estimate of [129] for linear forms in two p-adic logarithms.
Yamada’s result, reproduced as Theorem 12.3 and established in Chapter 12, has a striking
application to an old conjecture of Erdős; see Theorem 7.11.
The next result is a slight simplification of the estimate given on page 190 of Yu’s
paper [449].
Theorem 2.9. Let p be a prime number and ˛1 ; : : : ; ˛n algebraic numbers in an algebraic
number field of degree D. Let b1 ; : : : ; bn denote rational integers such that ˛1b1 : : : ˛nbn is
not equal to 1. Let A1 ; : : : ; An ; B be real numbers with
n 1 o
log Aj max h.˛j /; ; 1 j n;
16e2 D 2
and
B maxfjb1 j; : : : ; jbn j; 3g:
If n 2, then we have
pD 1
vp .˛1b1 : : : ˛nbn 1/ < .16eD/2.nC1/ n5=2 .log.2nD//2 D n log A1 : : : log An log B:
.log p/2
Theorem 2.9 allows us to improve slightly the dependence on p in Theorem 2.8; see
also Theorem 12.3. Note that (as in Theorems 2.10 and 2.11, but unlike in Theorem 2.12
and in [443]) we do not assume that vp .˛j / D 0 for j D 1; : : : ; n.
The factor .p D 1/ in Theorem 2.9 and in Theorems 2.10 to 2.12 is ultimately due to
the fact that the radius of convergence of the p-adic exponential function is finite, equal
to p 1=.p 1/ . Removing it is a major open problem in the theory. This can be done under
the rather restrictive assumption that vp .˛j 1/ > 0 for j D 1; : : : ; n; see [168, 169].
Theorem 2.9 should be compared with Theorem 2.2. We have exactly the same de-
pendence on the parameters n, log A1 ; : : : ; log An , and B. A crucial point in Theorem 2.9
is the dependence on n, which is essential in the proofs of Theorems 8.2 and 8.3 towards
2.3. Lower bounds for linear forms in p-adic logarithms 19
the abc-conjecture. These proofs also require an estimate for vp .˛1b1 : : : ˛nbn 1/ with a
(slightly) better dependence on p. For this reason, we quote the result stated on page 30
of [446].
Theorem 2.10. We keep the notation of Theorem 2.9. If n 1, then we have
pD 1 2.nC1/
vp .˛1b1 : : : ˛nbn 1/ < 12 6.n C 1/D log.e5 nD/
log p
n h.˛ / o n h.˛ / o
1 n
max ; 1 max ; 1 log B:
log p log p
Yu [449] obtained an improvement of Theorem 2.10, with a better dependence on n,

namely with nn in place of .n C 1/2n . We refer to [449] for a precise statement.
The next statement, extracted from page 191 of Yu’s paper [449], is the p-adic analogue
of the main result of [25].
Theorem 2.11. We keep the notation of Theorem 2.9. Let Bn be a real number such that
B Bn jbn j:
Assume that
vp .bn / vp .bj /; j D 1; : : : ; n:
1
Let ı be a real number with 0 < ı 2
. Then, we have
pD 1
vp .˛1b1 : : : ˛nbn 1/ < .16eD/2.nC1/ n3=2 .log.2nD//2 D n
.log p/2
n ıB o
max .log A1 / .log An /.log T /; ; (2.8)
Bn c0 .n; D/
where
Bn
T D c1 .n; D/p .nC1/D .log A1 / .log An 1/
ı
and
c0 .n; D/ D .2D/2nC1 log.2D/ log3 .3D/; c1 .n; D/ D 2e.nC1/.6nC5/ D 3n log.2D/:
In particular, we get either
B 2Bn .log A1 / .log An / c0 .n; D/ (2.9)
or
pD
vp .˛1b1 : : : ˛nbn 1/ < .16eD/2.nC1/ n3=2 .log.2nD//2 D n
.log p/2
n c .n; D/p .nC1/D B o
1
.log A1 / .log An / max 1; log : (2.10)
c0 .n; D/ log An
The formulation of Theorem 2.11 looks complicated, but it essentially corresponds to

the introduction of the parameter B 0 in Theorem 2.1 above. To get the second statement,
we have selected
.log A1 / .log An /Bn
ı0 D c0 .n; D/
B
and distinguished the two cases ı0 12 , which gives (2.9), and ı0 < 12 , which, by (2.8)
with ı D ı0 , gives (2.10).
In the special case bn D 1, the statements of Theorems 2.2 and 2.11 can be merged;
see Theorem 3.2.8 of [183].
As for linear forms in complex logarithms, the case n D 2 is very important for
applications. The next theorem reproduces one of the corollaries of the main result
of [129].
Theorem 2.12. Let p be a prime number. Let ˛1 and ˛2 be multiplicatively independent
algebraic numbers with vp .˛1 / D vp .˛2 / D 0. Set D D ŒQ.˛1 ; ˛2 / W Q. Let A1 and A2
be real numbers with
n log p o
log Aj max h.˛j /; ; j D 1; 2:
D
Let b1 and b2 be positive integers and set
n b1 b2 10 log p o
log B 0 D max log C C log log p C 0:4; ; 10 :
D log A2 D log A1 D
Then, we have the upper bound
24p.p D 1/ 4
vp .˛1b1 ˛2b2 / D .log A1 /.log A2 / .log B 0 /2 : (2.11)
.p 1/.log p/4
The assumption that ˛1 and ˛2 are multiplicatively independent is not really restrictive
for the applications and is useful to get small numerical values in (2.11). The assumption
that ˛1 and ˛2 are p-adic units is not restrictive. Indeed, if only one among vp .˛1 /
and vp .˛2 / is positive, then vp .˛1b1 ˛2b2 / D 0, while, if both of them are positive,
then vp .˛1b1 ˛2b2 / is bounded from below by a positive real number times the minimum
of b1 and b2 .
A version of Theorem 2.12 (which is weaker only in terms of the numerical constants)
is proved in this book; see Theorem 12.1.
Unlike Theorem 2.9, Theorem 2.12 cannot be applied to bound vp .˛ b 1/ from above
because of the assumption of multiplicative independence, but see Theorem 12.3.
Theorem 2.12 should be compared with Theorem 2.3. In particular, the numerical
constant is very small and the quantity log B 0 also occurs squared.
We end this section with an improvement of Theorem 2.12 when ˛1 and ˛2 are rational
numbers which are p-adically close to 1.
Let yx11 and yx22 be non-zero rational numbers. Theorem 2.13 below, originated in [109],
provides an explicit upper bound for the p-adic valuation of the difference between integer
powers of yx11 and yx22 .
2.3. Lower bounds for linear forms in p-adic logarithms 21
We suppose that there exist a positive integer g and a real number E with E > 1C p 1 1
and
x g
1
vp 1 E; (2.12)
y1
and at least one of the two following conditions

x g
2
vp 1 E (2.13)
y2
or
x g
2
vp 1 1 and vp .b2 / vp .b1 / (2.14)
y2
is satisfied. For instance, if p is an odd prime number, then (2.12) and (2.13) hold with
g D 1 if yx11 and yx22 are both congruent to 1 modulo p 2 .
We stress that the assumption “(2.13) or (2.14) holds” is more restrictive than the
assumption “vp .. yx22 /g 1/ > 0” made in [109]. However, it seems that the latter
assumption is not sufficient to derive Theorem 2 from Theorem 1 in [109], a fact which
has been overlooked in [109].
Theorem 2.13. Let yx11 and yx22 be multiplicatively independent rational numbers. Let
b1 ; b2 be positive integers. Assume that there exist a positive integer g and a real
number E greater than 1 C p 1 1 such that (2.12) holds, as well as (2.13) or (2.14). Let
A1 ; A2 be real numbers with
log Aj maxflog jxj j; log jyj j; E log pg; j D 1; 2;

and put
n b1 b2 o
log B 0 D max log C C log.E log p/ C 0:4; 6 E log p; 5 :
D log A2 D log A1

x b1 x b2 36:1 g
1 2
vp .log A1 /.log A2 /.log B 0 /2 ; (2.15)
y1 y2 E 3 .log p/4
if p is odd or if p D 2 and v2 . yx22 1/ 2.

x1 x2
Observe that, if and are both congruent to 1 modulo p 3 , then we can take E D 3
y1 y2
in Theorem 2.13 and the right hand side of (2.15) can be bounded independently of p.
A weaker version of Theorem 2.13 is proved in this book; see Theorem 12.2.
The parameter E in Theorem 2.13 should be compared with the parameter E in
Theorems 2.1 and 2.4. Let us illustrate its importance with the following example taken
from [117] (see also Exercise 6.11). Apply Theorem 2.13 with the prime number p to
bound from above the p-adic valuation v of . yx /b .1 C cp k /, for positive integers k, b,
c, x, y, M with k 3, 1 c < p M k , x > p k , and gcd.p; b/ D 1. The latter assumption
implies that, if v 1, then the p-adic valuations of . yx /b 1 and yx 1 are positive
and equal. Consequently, p k divides yx 1 and we apply Theorem 2.13 with g D 1,

A1 D p .M C1/k , A2 D maxfx; yg, and E D k to get
36:1 log A1
v .log A2 / .6 C log b/2 ;
log p E.log p/
thus
36:1.M C 1/
v .log A2 / .6 C log b/2 :
log p
This upper bound is independent of k. It considerably improves the upper bound given
by (2.11) and can be viewed as an analogue of Theorem 2.5 (with 1 replaced with M C 1).
2.4. Notes
F Linear forms in one logarithm have been studied by Mignotte and Waldschmidt [300].
Let ˛ be an algebraic number of degree D 2 and of small height. In their lower bound
for j˛ 1j (that is, for j log ˛j), the dependence on D is much better than the one given
by Liouville’s inequality Theorem B.10; see also [134]. Amoroso [7] proved that these
results are essentially sharp; see also [171]. A p-adic analogue has been worked out
in [105].
F In the homogeneous rational case, a complete proof of Theorem 2.1 up to the dependence
on n of the numerical constant has been given by Waldschmidt in [433]. Nesterenko [316]
has written a detailed proof of Theorem 2.2 in the special case where ˛1 ; : : : ; ˛n are rational
numbers. Aleksentsev [5] established an estimate of a similar strength as Theorem 2.2.
F There is a large gap between the size of the numerical constants in the best known lower
bounds for linear forms in two and in three logarithms. For applications to Diophantine
problems, it would be of greatest interest to improve the lower estimates for linear forms
in three logarithms; see [298] for a step in this direction.
F Bugeaud [111] obtained lower bounds for linear forms in two logarithms and simul-
taneously for several p-adic valuations; see [52, 54, 111] for applications to Diophantine
equations.
F Simultaneous linear forms in logarithms are discussed in Chapter 9.
Chapter 3
First applications
This chapter is principally devoted to Diophantine problems from which one can derive a
linear form in complex logarithms in an almost obvious way, or, at least, in an easy way.
We have listed them in increasing order of difficulty.
We postpone to Chapter 4 applications of the theory of linear forms in complex
logarithms to classical families of Diophantine equations, including unit equations, Thue
equations, elliptic equations, etc.
3.1. On the distance between powers of 2 and powers of 3

One of the simplest applications of the theory of linear forms in complex logarithms shows
that the distance between a power of 2 and the power of 3 closest to it tends to infinity
when the power of 2 tends to infinity. In addition, it provides us with an explicit lower
bound for this distance. Of course, and this will be clear in Section 3.3, the integers 2
and 3 can be replaced by any pair of multiplicatively independent positive integers.
Theorem 3.1. For all positive integers m and n, we have
8:4108
j2m 3n j > 2m .em/ :
Proof. Let n 2 be an integer and define m and m0 by the conditions

0 0 0 0
2m < 3n < 2m C1 and j3n 2m j D minf3n 2m ; 2m C1 3n g:
Then, m is equal to m0 or to m0 C 1 and
j2m 3n j < 2m ; .m 1/ log 2 < n log 3 < .m C 1/ log 2: (3.1)
The problem of finding a lower bound for j2m 3n j clearly reduces to this case. Also, we
have m n. To show that 2m cannot be too close to 3n , it is sufficient to prove that the
quantity
ƒ WD 3n 2 m 1
is not too small in absolute value. Such a result is a direct consequence of Theorem 2.2,
which, applied with ˛1 D 3; A1 D 3; ˛2 D 2; A2 D 2, gives that
log jƒj > 305 211=2 .log 2/ .log 3/ .log em/:

24 3. First applications
We conclude that
8:4108
j3n 2 m
1j > .em/ ;
and the theorem is established.
As an application of Theorem 3.1, we list all the powers of 3 which differ from a power
of 2 by at most 20. Our main auxiliary tool is the theory of continued fractions, briefly
explained in Appendix A.
Theorem 3.2. The only solutions in positive integers m; n with n 4 to the inequality
j2m 3n j 20 (3.2)
are given by j26 34 j D 17 and j28 35 j D 13.
Proof. Let n 6 and m be integers satisfying (3.2). Applying Theorem 3.1 to Inequal-
ity (3.2), we get
8:4108
20 > 2m .em/ ;
which implies
log 20 > m log 2 8:4 108 .1 C log m/;
giving m < 4 1010 and
.m C 1/.log 2/
n< < 3 1010 ;
log 3
by (3.1). Since j log.1 C x/j 2jxj for any real number x with jxj < 21 , the inequality
j2m 3n j 20 implies
ˇ ˇ
ˇ log 3 m ˇ
ˇ ˇ 40 3 n : (3.3)
ˇ log 2 n ˇ n log 2
Observe that the right-hand side of (3.3) is less than 2n1 2 , since n 6. By Theorem A.2,
log 3
the rational number mn
is a convergent of the continued fraction expansion of WD log 2
.
Furthermore, for n < N WD 3 1010 , the smallest value of jm nj is obtained for the
last convergent of the continued fraction expansion of with denominator less than N .
The computation of this expansion shows that
ˇ ˇ
ˇ log 3 m ˇˇ 18
ˇ
ˇ log 2 > 9:6 10 ; for 0 < n < 3 1010 :
nˇ
Comparing (3.3) with this estimate, we deduce that n 36. A rapid verification in the
range 6 n 36 completes the proof that (3.2) has no solution with n 6.
3.2. Effective irrationality measures for quotients of logarithms of integers 25
3.2. Effective irrationality measures for quotients

of logarithms of integers
Let a and b be multiplicatively independent positive rational numbers. It follows straight-
a
forwardly from the first assertion of Theorem 2.1 that the real number log
log b
is transcendental
and, by arguing as in the proof of Theorem 3.1, we show that it is not a Liouville number
(see Definition A.6) and give an effective upper bound for its irrationality exponent (see
Definition A.3).
a1
Theorem 3.3. Let a1 ; a2 ; b1 ; b2 be positive integers such that the rational numbers a2
b1
and b2
are multiplicatively independent. Then, the effective irrationality exponent of the
real number .log aa12 /=.log bb21 / satisfies
log aa12

eff .log maxfa1 ; a2 g/ .log maxfb1 ; b2 g/:
log bb12
a1 b1
A strengthening of Theorem 3.3 in the special case where a2
and b2
are rational
numbers close to 1 is given in Section 5.1.
Proof. Let pq be a convergent to .log aa12 /=.log bb21 / with q 2. Since an irrational
real number and its inverse have the same irrationality exponent, we can assume that
j log aa12 j < j log bb12 j. It then follows from Theorem 2.1 (or from Theorem 2.2) that
ˇ ˇ
ˇ a1 b1 ˇˇ
logˇˇq log p log .log maxfa1 ; a2 g/ .log maxfb1 ; b2 g/ .log q/:
a2 b2 ˇ
Thus, there exists an effectively computable, absolute real number C such that
ˇ log aa12 jq log aa12 p log bb12 j

ˇ ˇ
p ˇˇ C.log maxfa1 ;a2 g/ .log maxfb1 ;b2 g/
ˇ D q :
log bb12 qˇ q log b1
ˇ
b2
This proves the theorem.

The fact that the dependence on B in Theorem 2.2 occurs through the factor .log B/ and
not through the factor .log B/2 , as in Theorem 2.3, is crucial for the proof of Theorem 3.3.
3.3. On the distance between two integral S -units

For a finite set S of prime numbers, an integral S -unit is, by definition, an integer all of
whose prime factors are in S . The following result extends Theorem 3.1. It was proved
by Tijdeman [414] in 1973.
Theorem 3.4. Let S denote a finite, non-empty set of prime numbers and .xj /j 1 the
increasing sequence of all positive integers whose prime factors belong to S . There exists
an effectively computable real number C , depending only on S , such that
C
xj C1 xj xj .log 2xj / ; j 1:
Proof. Write S D fq1 ; : : : ; qs g, where q1 ; : : : ; qs are prime numbers and q1 < < qs .
Let u; v with 3 u < v be two consecutive elements in the sequence .xj /j 1 and write
s s
qiui ; qivi :
Y Y
uD vD
i D1 i D1
Without any loss of generality, we can assume that s 2, v 2u, and that u and v are
coprime. Thus, for every i D 1; : : : ; s, at least one of ui ; vi is zero. Set
s
v
qivi ui
Y
ƒ WD 1D 1:
u
i D1
Put B D maxf3; jv1 u1 j; : : : ; jvs us jg and observe that
2B=3 q1B=3 v 2u: (3.4)

It follows from Theorem 2.2 that
s
Y
sC4 5
log ƒ 30 s log qi .log eB/:
i D1
Combined with (3.4), this gives

v s
Y 3e log.2u/
log 1 30sC4 s 5 log qi log :
u log 2
iD1
Taking the exponential of both sides and multiplying by u, we get the assertion of the
theorem.
In view of the next result, established by Tijdeman [415] in 1974, Theorem 3.4 is
essentially best possible.
Theorem 3.5. Let P be an integer and S a finite set of at least two distinct prime numbers
all being at most equal to P . Let .xj /j 1 be the increasing sequence of all positive
integers whose prime divisors belong to S . There exists an effectively computable real
number C , depending only on P , such that
C
xj C1 xj xj .log 2xj / ; j 1:
It is an open problem of Erdős to establish that, for S being the set of all prime numbers
less than P , Theorem 3.5 holds with a real number C D C.P / which tends to infinity
with P .
3.4. Effective irrationality measures for n-th roots

of algebraic numbers
We have seen in Theorem 3.3 how the theory of linear forms in logarithms can be applied
to give an effective upper bound for the irrationality exponent of certain transcendental
real numbers. The next result addresses a special class of real algebraic numbers.
3.5. On the greatest prime factor of values of integer polynomials 27
Theorem 3.6. Let n 3 be an integerp and a positive, real algebraic number. Then, the
effective irrationality exponent of n satisfies
p
eff . n / log n:
Furthermore, if a; b are integers with 1 b < a, then we have

p
eff . n a=b / .log a/.log n/;
and, if a 2n , then we get

p
eff . n a=b / 11000 log a: (3.5)
p n is sufficiently large in terms of , Theorem 3.6 improves the upper bound

When
eff . n / n deg./, given by Liouville’s Theoremp A.5. It remains, however, very
far frompRoth’s Theorem A.7, which asserts that . n / D 2 but gives no information
on eff . n /.
In Section 5.2 we considerably strengthen Theorem 3.6 when is a rational number
very close to 1.
p
Proof. Let n 3 be an integer and a positive, real algebraic number. Let q
be a
p
convergent of n with q 2. Observe that
ˇp
ˇn p ˇˇ 1 ˇ p n ˇ
ˇ ˇ n
ˇ ˇ
q 2 maxf1; jjg q
ˇ ˇ:
It directly follows from Theorem 2.2 that

ˇ p n ˇ
ˇ> C./.log q/.log n/;
ˇ ˇ
logˇ
q
where C./ is an effectively computable real number which can be expressed in terms of
the height and the degree of . If is the rational number ab with 1 b < a, then C./ can
be taken to be an absolute real number times log a. This proves the first two statements of
the theorem.
For large values of n, we obtain a stronger result (not only numerically) by applying
Theorem 2.3. Indeed, if a; b are integers with 1 b < a, it directly gives the lower bound
ˇ p a ˇ n 2n o2
ˇ 25:2.log p/.log a/ max log C 0:21; 20 ;
ˇ ˇ
logˇn log log
q b log a
if p is sufficiently large. From this and under the assumption a 2n , we deduce (3.5).
3.5. On the greatest prime factor of values of integer polynomials

Let f .X / be a non-zero integer polynomial with at least two distinct roots. Let n be an
integer. Using the theory of linear forms in logarithms, Shorey and Tijdeman [374] were
the first to give an effective lower bound for the greatest prime factor of f .n/ as n tends
to infinity. In 1998 Tijdeman [418] noticed that, for the polynomial X.X C 1/, the use of
Theorem 2.2 allows us to improve the results of [374]. In his doctoral thesis, Haristoy [221]
extended Tijdeman’s result to an arbitrary polynomial f .X /. For an integer n, let denote
by P Œn its greatest prime factor with the convention that P Œ0 D P Œ˙1 D 1.
Theorem 3.7. Let f .X / be an integer polynomial with at least two distinct roots. Then,
we have
log log log n
P Œf .n/ f log log n ; for n 107 : (3.6)
log log log log n
A result similar to Theorem 3.7 has been established in [217]; see also Chapter 8.
If we apply Theorem 2.1 (with a weaker dependence on the number of logarithms in
the linear form than in Theorem 2.2) instead of Theorem 2.2, then we would get the weaker
lower bound
P Œf .n/ f log log n; for n 100;
which was obtained in [374].
Proof. To avoid technical complications, we treat only the case of the polynomial X.X C1/.
For an integer n at least equal to 100, write
n.n C 1/ D p1u1 : : : pkuk ;
where p1 < p2 < denotes the sequence of prime numbers in increasing order and the
integers u1 ; : : : ; uk are non-negative. There exist disjoint non-empty subsets I and J of
f1; : : : ; kg such that I [ J D f1; : : : ; kg and
Y u Y u
nD pi i ; n C 1 D pj j :
i2I j 2J
Then, there exist "1 ; : : : ; "k in f˙1g such that

nC1 1
1 D jp1"1 u1 : : : pk"k uk 1j D : (3.7)
n n
Observe that
n C 1 2uj ; j D 1; : : : ; k;
thus
max uj 2 log n: (3.8)
1j k
Since pj j log j , for j D 2; : : : ; k, by Theorem D.2, we deduce from (3.8) and

Theorems 2.2 and D.2 that there exist effectively computable, absolute real numbers
c1 ; c2 ; c3 such that
log jp1"1 u1 : : : pk"k uk 1j c1k .log p1 / .log pk / .log log n/

k log log k
c2 .log log n/;
thus, by (3.7),
k log log k
log n c3 log log n :
3.5. On the greatest prime factor of values of integer polynomials 29
By Theorem D.2, this implies that

log log log n
pk k log k log log n ;
log log log log n
as asserted.
The general case of an arbitrary monic, integer polynomial with distinct roots ˛ and ˇ
rests on the same idea, but is technically more complicated. Let OK denote the ring
of integers of the algebraic number field K WD Q.˛; ˇ/ and h its class number. Let n
be an integer greater than j˛j and jˇj. Write the integer ideal .n ˛/.n ˇ/OK as a
product of powers of prime ideals and raise this equality to the power h. Then, the quantity
.n ˛/h .n ˇ/ h , which is equal to 1 C O. n1 /, can, with the help of Proposition C.5, be
expressed as a product of powers of algebraic numbers in K whose heights are controlled.
It then follows from Theorem 2.2 that this product cannot be too close to 1, hence the
result. We leave the details to the reader.
Let S be a finite non-empty set of prime numbers. For a non-zero integer n, write
Y
ŒnS WD jnjp 1 ;
p2S
where j jp is the p-adic absolute value normalized such that jpjp 1 D p. Said differently,
ŒnS is the greatest divisor of n, all of whose prime factors belong to S .
In 1984, Stewart [396] applied the theory of linear forms in logarithms to prove non-
trivial effective upper bounds for Œn.n C 1/ : : : .n C k/S , for any positive integer k. His
result was later extended by Gross and Vincent [205] as follows; see also [122].
Theorem 3.8. Let f .X / be an integer polynomial with at least two distinct roots and S
a finite, non-empty set of prime numbers. There exist effectively computable positive real
numbers c1 and c2 , depending only on f .X / and S , such that, for every non-zero integer n
which is not a root of f .X /, we have
Œf .n/S < c1 jf .n/j1 c2
:
Proof. To avoid technical complications, we treat only the case of the polynomial X.X C1/.
We leave to the reader the details of the general case. Let n be an integer with jnj 2.
Write S D fq1 ; : : : ; qs g, where s is the cardinality of S and q1 < < qs . Let u1 ; : : : ; us
be non-negative integers and a an integer coprime with q1 qs such that
n.n C 1/ D q1u1 : : : qsus a:
There exist disjoint non-empty subsets I and J of f1; : : : ; sg such that I [ J D f1; : : : ; sg
and integers a0 ; a00 such that a D a0 a00 and
Y u Y u
n D a0 qi i ; n C 1 D a00 qj j :
i 2I j 2J
Then, there exist "1 ; : : : ; "s in f˙1g such that

1 nC1 a00
D 1 D q1"1 u1 : : : qs"s us 1 DW ƒ:
n n a0
Noticing that uj 2 log jnj for j D 1; : : : ; s, we deduce from Theorem 2.2 applied to ƒ
with the quantity B 00 that there exists an effectively computable real number c3 such that
2.log jnj/.log q /
s
log jnj c3s .log q1 / .log qs /.log maxfja0 j; ja00 j; 2g/ log 0 00
:
log maxfja j; ja j; 2g
This shows that there exists an effectively computable positive real number c4 , depending
only on the set S , such that
maxfja0 j; ja00 j; 2g jnjc4 :
If jnj > 21=c4 , then we get a maxfja0 j; ja00 j; 2g jnjc4 and
n.n C 1/ n.n C 1/
Œn.n C 1/S D q1u1 : : : qsus D .n.n C 1//1 c4 =3
;
a jnjc4
which gives the desired result.

We now explain how a suitable version of Theorem 3.8 with explicit values for c1 and c2
implies Theorem 3.7. Observe that, setting C WD c3s .log q1 / .log qs /, an admissible
value for c4 is given by
c4 D c5 C 1 .log.C log qs // 1 ;
for some suitable absolute real number c5 . If S is the set composed of the first s prime
s log log s
numbers, then it follows from Theorem D.2 that c4 exceeds c6 for some absolute
real number c6 . If P Œn.n C 1/ D ps , then maxfja j; ja j; 2g D 2 and jnjc4 2, thus,
0 00
s log log s
log jnj 2c6 :
Combined with Theorem D.2, this gives
log log log n

P Œn.n C 1/ log log n ;
log log log log n
and we recover Theorem 3.7.
3.6. On the greatest prime factor of terms

of linear recurrence sequences
Let k be a positive integer, a1 ; : : : ; ak and u0 ; : : : ; uk 1 be integers such that ak is non-zero
and u0 ; : : : ; uk 1 are not all zero. Put
un D a1 un 1 C C ak un k; for n k:
The sequence .un /n0 is a linear recurrence sequence of integers of order k. Its companion
(or, characteristic) polynomial
G.z/ WD z k a1 z k 1
ak
3.6. On the greatest prime factor of terms of linear recurrence sequences 31
factors as
t
Y
G.z/ D .z ˛i /ì ;
i D1
where ˛1 ; : : : ; ˛t are distinct algebraic numbers, ordered in such a way that j˛1 j
j˛t j, and `1 ; : : : ; `t are positive integers. Then, (see e.g. [179] or Chapter C of [376])
there exist polynomials f1 .X /; : : : ; ft .X / of degrees less than `1 ; : : : ; `t , respectively,
and with coefficients in Q.˛1 ; : : : ; ˛t /, such that
un D f1 .n/˛1n C C ft .n/˛tn ; for n 0:
The recurrence sequence .un /n0 is said to be degenerate if there are integers i; j with
1 i < j t such that ˛˛ji is a root of unity. We say that .un /n0 has a dominant root
if j˛1 j > j˛2 j and f1 .X / is not the zero polynomial.
The next result was established in [399]; see also [395].
Theorem 3.9. Let u WD .un /n0 be a recurrence sequence of integers given by
un D f1 .n/˛1n C C ft .n/˛tn ; for n 0;
and having a dominant root ˛1 . For any integer n greater than 100 and such that un is
not equal to f1 .n/˛1n , the greatest prime factor of un satisfies
log log n
P Œun u log n : (3.9)
log log log n
Proof. We establish a slightly more general result. We consider a sequence of non-zero
integers v WD .vn /n0 with the property that there are in .0; 1/ and a positive real
number C such that
jvn f .n/˛ n j C j˛j n ; n 0;
where f .X / is a non-zero polynomial whose coefficients are algebraic numbers and ˛
is an algebraic number with j˛j > 1. Clearly, a recurrence sequence having a dominant
root ˛ has the above property.
Let p1 ; p2 ; : : : be the sequence of prime numbers in increasing order. Let n be a
positive integer such that f .n/ is non-zero and vn ¤ f .n/˛ n . Assume that the greatest
prime factor of vn is equal to pk , that is, assume that
vn D p1r1 pkrk ;
where r1 ; : : : ; rk are non-negative integers and rk 1. Then, we have
ƒ WD jp1r1 pkrk f .n/ 1

˛ n
1j C jf .n/j 1
j˛j. 1/n
;
and, since < 1, we get

log ƒ v . n/:
n
As vn ¤ f .n/˛ , the quantity ƒ is non-zero and Theorem 2.2 implies that
log ƒ v c1k .log p1 / .log pk / .log n/2 ;

where c1 is an effectively computable, positive, absolute real number. Comparing both

estimates for log ƒ, we obtain that
n v .c1 log pk /k .log n/2 ;
thus, by Theorem D.2,

pk
log n v log log pk :
log pk
This implies that
log log n
P Œvn D pk v log n ;
log log log n
and the theorem is established.
The lower bound for P Œun in (3.9) is of similar strength as the one for P Œf .n/ in
Theorem 3.7, since jun j grows exponentially fast in n. Both proofs show that f in (3.6)
and u in (3.9) can be replaced by .1 "/, for any positive " and any sufficiently large
integer n.
3.7. Perfect powers in linear recurrence sequences

We continue the study of arithmetical properties of linear recurrence sequences of integers
having a dominant root and show that no term of such a sequence can be a very large
power of an integer greater than one.
Theorem 3.10. Let u WD .un /n0 be a recurrence sequence of integers given by
un D f1 .n/˛1n C C ft .n/˛tn ; for n 0;
and having a dominant root ˛1 . Assume that ˛1 is a simple root, that is, the polynomial
f1 .X / is equal to a non-zero algebraic number f1 . Then, the equation
un D y q ;
in integers n; y; q with q a prime number, jyj 2, and un ¤ f1 ˛1n , implies that
q u 1:
Theorem 3.10 was proved in [372]; see also [190, 325]. More can be said in the case
of binary recurrence sequences, see Section 6.9.
Proof. We establish a slightly more general result. We consider a sequence of non-zero
integers v WD .vn /n0 with the property that there are in .0; 1/ and a positive real
number C such that
jvn f ˛ n j C j˛j n ; n 0;
where f is a non-zero algebraic number and ˛ is an algebraic number with j˛j > 1.
Let n; y; q with q a prime number be such that vn D y q , jyj 2, and vn ¤ f ˛ n .
Then, we have
C
jf 1 ˛ n y q 1j j˛j. 1/n : (3.10)
jf j
3.7. Perfect powers in linear recurrence sequences 33
There exist integers k and r such that n D kq C r with jrj q2 . Rewriting (3.10) as
ˇ y q ˇ C
ƒ WD ˇf 1 k ˛ r 1ˇ j˛j. 1/n ;
ˇ ˇ
˛ jf j
we observe that
log ƒ v . n/: (3.11)
k
˛
By Theorem B.5, the height of y
is bounded from above by the sum of log jyj and k
times the height of ˛. Since
kq
q log jyj v n .k C 1/q and n v q log jyj; (3.12)
2
we get that h.˛ k =y/ v log jyj. It then follows from Theorem 2.2 that
log ƒ v .log jyj/.log q/;
which, combined with (3.11) and (3.12), gives
q log jyj v n v .log jyj/.log q/;
and we get q v 1. This proves the theorem.
We point out a very special case of Theorem 3.10.
Corollary 3.11. Let ˛ and ˇ be algebraic integers such that ˛ C ˇ, ˛ˇ are rational
integers and ˛ > jˇj. Then, the equation
˛n ˇn
D yq .resp., ˛ n C ˇ n D y q /; (3.13)
˛ ˇ
in integers n; y; q with jyj 2 and q a prime number, implies that
q ˛;ˇ 1:
Let us discuss a celebrated example of resolution of equations of the form (3.13),
namely the determination
p
of all the perfect powerspamong the Fibonacci and Lucas num-
bers. Set D 1C2 5 and observe that 1 D 1 2 5 . Let the integer sequences .Fn /n0
and .Ln /n0 be defined respectively by
n . 1 n
/
Fn D 1/
; Ln D n C . 1 n
/ ;
.
so that the first elements of the Fibonacci sequence .Fn /n0 and of the Lucas sequence
.Ln /n0 are given by
0; 1; 1; 2; 3; 5; 8; 13; 21; 34; 55; 89; 144; 233; : : :
and
2; 1; 3; 4; 7; 11; 18; 29; 47; 76; : : : ;
respectively. The Fibonacci sequence is an emblematic example of a Lucas–Lehmer
sequence (see Section 7.2). The following result has been established in [137].
Theorem 3.12. The only perfect powers in the sequence .Fn /n0 are 0, 1, 8, and 144.
The only perfect powers in the sequence .Ln /n0 are 1 and 4.
Outline of the proof. The complete proof is much too involved to be included in the
present textbook, however, it is appropriate to discuss the main steps of the proof of the
first assertion. The lists of squares and cubes in the Fibonacci sequence were obtained
by Ljunggren [265, 266] and London and Finkelstein [268], respectively. Their proofs are
elementary, in the sense that they do not rest on the theory of linear forms in logarithms.
The equation Fn D y 5 was solved by Pethő [324], by using a linear form in logarithms
and congruence conditions; see [137] for additional references.
Let n 13 be an integer, q 7 a prime number,
p and suppose that Fn D y q for some
1
integer y 2. Then, noticing that C D 5 and proceeding exactly as in the proof
of Theorem 3.10, we derive that there exist integers k and r such that n D kq C r with
jrj q2 and
ˇp y q ˇ
ƒ WD ˇ 5 k r 1ˇ y 2q :
ˇ ˇ

The quantity ƒ is a linear form in three logarithms. Since h.y= k / log y it follows
from Theorem 2.2 that
log ƒ .log y/ .log q/;
and, comparing both estimates, we deduce an upper bound for the exponent q. Actually,
a special estimate for linear forms in three logarithms proved in [137] (see also [298])
implies the sharper upper bound q < 2 108 , a range which is suitable for computer
calculations. Then, a sieve performed by using the “modular method” allows us to prove
that r D ˙1, for any prime number q between 7 and 2 108 . This shows that our linear
form in three logarithms can be written as a linear form in only two logarithms, namely
ˇ y q p ˇ
ƒ D ˇ k . ˙1 5/ 1ˇ:
ˇ ˇ

An application of Theorem 2.3 (actually, of the main result of [254]) now leads to the
much better upper bound q 733. It then remains to treat a reasonably small number of
equations. A second sieve, again performed by using the “modular method”, allows us to
prove that, for any prime number q between 7 and 733, the index n must be very large. To
get a contradiction, we have to bound n from above. To do this, observe that the Fibonacci
and Lucas numbers are linked by the relation
5Fm2 L2m D 4; for m 1 odd:
Since we have assumed that Fn D y q , this implies that the equation
X 2 C 4 D 5Y 2q
has a solution in positive integers X; Y with Y > 1, given by .X; Y / D .Ln ; y/. This
equation is similar to the equations considered in Section 4.6 below, with, however, an
important difference. Namely, since
. 1/2 C 4 D 1 C 4 D 5 D 5 12q ;
3.8. Simultaneous Pellian equations and Diophantine quadruples 35
it has always solutions, regardless of the prime number q. At present, we are not able to
solve it completely, but we can explicitly bound the absolute values of its solutions. The
authors of [137] have proceed as follows.
Let X; Y be positive integers with Y 2 and X 2 C 4 D 5Y 2q . By factoring over ZŒi
the left-hand side of this equality, we deduce the existence of integers a and b with
a2 C b 2 D Y 2 and
˙4i D .2 C i/.a C ib/q .2 i/.a ib/q :
Dividing both members of the inequality by 2i, we get
Œq=2
!
X q
˙2D2 a2k . 1/.q 2k 1/=2 b q 2k
2k
kD0
Œq=2
!
X q
C a2kC1 . 1/.q 2k 1/=2
bq 2k 1
:
2k C 1
kD0
We infer that a is even. Consequently, .b; a2 / is an integer solution of the Diophantine

equation
q
!
b.q k/=2c q
X
. 4/ Z k T q k D ˙1;
k
kD0
in integer unknowns Z and T . We recognize a Thue equation of degree q. As explained in
Section 4.3 below, the theory of linear form in logarithms allows us to bound from above
the absolute values of its solutions. Thus, we get an explicit upper bound for jaj and jbj,
hence for Y . This in turn implies an explicit upper bound for the index n. Since this bound
is smaller than the bound obtained via the second sieve, we have reached a contradiction.
Consequently, there are no q-th powers greater than 1 in the Fibonacci sequence, where
q 7 is a prime number. This completes the outline of the proof of the theorem.
3.8. Simultaneous Pellian equations and Diophantine quadruples

Let a 2 be an integer which is not a perfect square and u a non-zero integer. By
Theorem C.2, the Pellian equation
x2 ay 2 D u;
in integers x; y, has either no solutions or infinitely many. This motivates the study of
systems of two Pellian equations having an unknown in common. Let b 2 be an integer
which is not a perfect square, distinct from a, and v a non-zero integer. We consider the
systems
x2 ay 2 D u; x2 bz 2 D v; in integers x; y; z; (3.14)
2 2 2 2
x ay D u; y bz D v; in integers x; y; z; (3.15)
2 2 2 2
and x ay D u; z by D v; in integers x; y; z:
Solving any of these systems amounts to determine the common values of two binary
recurrence sequences of integers; see [44] and the references quoted therein.
Theorem 3.13. Let a; b be positive integers which are not perfect squares. Let u; v be
non-zero integers such that av ¤ bu. Then, all the solutions in positive integers x; y; z to
the system of simultaneous Pellian equations
x2 ay 2 D u; z2 by 2 D v (3.16)
satisfy
maxfx; y; zg a;b;u;v 1:
A similar result holds for the systems (3.14) and (3.15) of simultaneous Pellian equa-
tions.
Let x; y; z be integers satisfying (3.16). Setting w D xz, we derive the equation
w 2 D .ay 2 C u/.by 2 C v/;
which belongs to a general family of Diophantine equations treated in Theorem 4.7.

Observe that the polynomial .aX C u/.bX C v/ is not a square when av ¤ bu.
Proof. Let " andp be the fundamental
p units (see Definition C.3) of the rings of integers
of the fields Q. a/ and Q. b/, respectively. Let p x, y, and z be p positive integers
satisfying (3.16). Since the norm over Q of x C y a (resp., z C y b) is u (resp., v),
it follows from Proposition
p p there exist nonnegative integers m; n and algebraic
C.5 that
numbers ˛ in Q. a/ and ˇ in Q. b/, with h.˛/ a;u 1 and h.ˇ/ b;v 1, such that
p p
x C y a D ˛"m and z C y b D ˇn :
Consequently, we have
p
2y a D ˛"m ˛ ." /m
p
and 2y b D ˇn ˇ . /n ;
where the superscript denotes the Galois conjugacy. Since j" j D " 1
and j j D 1
,
we get
ƒ WD j˛ˇ 1 "m n 1j a;b;u;v ." m C n /: (3.17)
As pointed out at the end of Section C.1, the units " and are at least equal to the Golden
Ratio, thus it follows from (3.17) and Theorem 2.2 that
log maxfm; ng a;b;u;v log ƒ a;b;u;v . minfm; ng/:
Since minfm; ng a;b;u;v maxfm; ng, this shows that m and n are bounded by an
effectively computable real number depending on a; b; u, and v. Consequently, we have
proved that maxfx; y; zg a;b;u;v 1.
1 33 17
The Greek mathematician Diophantus observed that the rational numbers 16 , 16 , 4 ,
105
and 16 have the following property: the product of any two of them increased by 1 is
a square of a rational number. Later, Fermat noted that the set f1; 3; 8; 120g shares the
same property. For m 3, we call a Diophantine m-tuple any set of m positive integers
a1 ; : : : ; am such that ai aj C 1 is a perfect square whenever 1 i < j m. It was
3.8. Simultaneous Pellian equations and Diophantine quadruples 37
known already to Euler that there are infinitely many Diophantine quadruples. Baker and
Davenport [35] established that 120 is the unique positive integer t such that f1; 3; 8; tg
is a Diophantine quadruple. Among the broad literature on that topic, let us mention
that Dujella [172] proved that there exist no Diophantine sextuple and that every element
26
of a Diophantine quintuple is less than 1010 . The question of the non-existence of a
Diophantine quintuple remains a very challenging open problem.
The proof of the result of Baker and Davenport [35] reduces to the complete resolution
of a system of simultaneous Pellian equations.
Theorem 3.14. The only positive integer t such that f1; 3; 8; tg is a Diophantine quadruple
is 120.
A key ingredient in the proof of Theorem 3.14 is Lemma 3.15, usually called the
Baker–Davenport lemma. Throughout, k k denotes the distance to the nearest integer.
Lemma 3.15. Let and ˇ be real numbers. Let A > 0 and C > 1 be real numbers. Let
K 6; M 1; p; q be integers satisfying
2
1 q KM; jq pj < : (3.18)
KM
If
3
kqˇk ; (3.19)
K
then the inequality
m
jm n C ˇj < AC (3.20)
has no solutions in integers m; n, with
log AK 2 M
< m < M: (3.21)
log C
Proof. Let m; n be integers and assume that m satisfies (3.21). Setting WD q p and
multiplying (3.20) by q, we get
m
j.pm nq/ C m C ˇqj D j.q /m nq C ˇqj < AqC : (3.22)
2
Since m < M , we have jmj < K
, while m > .log AK 2 M /=.log C / implies that
m Aq 1
AqC < :
AK 2 M K
By (3.22), we deduce that
m 1 2 3
kqˇk < jmj C AqC < C ;
K K K
a contradiction with (3.19).
Outline of the proof of Theorem 3.14. Let t be a positive integer such that f1; 3; 8; tg is a
Diophantine quadruple. Then, there exist positive integers x; y; z such that
t C 1 D x2; 3t C 1 D y 2 ; 8t C 1 D z 2 :
In particular, we get
3x 2 y2 D 2 and 8x 2 z 2 D 7:
Classical results (see Section C.1) tell us that there exist non-negative integers m; n and "
in f˙1g such that the positive integers x; y; z are given by
p p p
y C x 3 D .1 C 3/.2 C 3/m
and
p p p
z C x 8 D ." C 8/.3 C 8/n :
Consequently, we arrive at
p p
1C 3 p m 3 1 p
2x D p .2 C 3/ C p .2 C 3/ m
3 3
p p
"C 8 p n 8 " p
D p .3 C 8/ C p .3 C 8/ n :
8 8
Thus, assuming m 3, we get n < m and, after some calculation,
p p
p p .1 C 3/ 8 2
0 < m log.2 C 3/ n log.3 C 8/ C log p p < p : (3.23)
." C 8/ 3 .2 C 3/2m
Combined with Theorem 2.2, we deduce an upper bound for m. Using Theorem 1.9,
Baker and Davenport obtained the upper bound
m < 10487 : (3.24)

p
Furthermore, dividing (3.23) by log.3 C 8/, we get the inequality
jm0 n C ˇ0 j < A0 C0 m ; (3.25)
where p p p
log.2 C 3/ 1 3/ 8 .1 C
0 WD p ; ˇ0 WD p log p p ;
log.3 C 8/ log.3 C 8/ ." C 8/ 3
and
2 p
A0 WD p ; C0 WD .2 C 3/2 :
log.3 C 8/
Classical results from metrical number theory (see e.g. Theorem II in Chapter VII of [144])
assert that, for almost all (with respect to the Lebesgue measure) pairs . 0 ; ˇ 0 / of real
numbers and for every positive real number ", the inequality
jm 0 n C ˇ0j < m 1 "
has only finitely many solutions in integers m; n. Thus, it is rather unlikely that there exist
large integers m; n satisfying (3.25).
Since ˇ0 is non-zero, we cannot use continued fractions as in the proof of Theorem 3.2.
However, Lemma 3.15 gives us a numerical method to reduce the bound (3.24). We
3.9. On the representation of integers in distinct bases 39
proceed as follows. Using the theory of continued fractions (see Appendix A), we find
suitable integers p; q; K; M with M > M0 which fulfill the assumptions of Lemma 3.15
applied with 0 ; ˇ0 ; A0 , and C0 . For instance, one can select for pq a convergent to , the
factor 2 in the numerator in (3.18) being there to allow some more flexibility.
Baker and Davenport applied Lemma 3.15 with M D 10487 and K D 1033 . They
took for pq the last convergent with q 10520 to the rational number 1 , whose 1040 first
decimals coincide with those of 0 and whose other decimals are 0. It then follows that
there are no solutions with m 500. These small values of m can then be treated directly,
or we can again apply Lemma 3.15 to refine the estimate m 499, before treating directly
the remaining values of m.
3.9. On the representation of integers in distinct bases

Let a and b be integers at least equal to 2. In 1973, Senge and Straus [363] proved that
the number of integers, the sum of whose digits in each of the bases a and b lies below a
fixed bound, is finite if and only if a and b are multiplicatively independent. Their proof
rests on the Roth Theorem A.7 and, hence, is ineffective. Using the theory of linear forms
in logarithms, Stewart [394] succeeded in establishing an effective version of Senge and
Straus’ theorem. He showed that if a and b are multiplicatively independent, then, for
every c 1, each integer m > 25 whose sum of digits in base a as well as in base b is
bounded by c satisfies
log log m
< 2c C 1; (3.26)
log log log m C c1
where c1 is a positive real number which is effectively computable in terms of a and b
only; see also [272, 295].
Besides the number of non-zero digits in the representation of an integer m, we may
also estimate the number of blocks composed of the same digit in this representation.
This was first considered by Blecksmith, Filaseta, and Nicol [79], who proved that, for
multiplicatively independent positive integers a and b, we have
lim BC.an ; b/ D C1;

n!C1
where BC.m; b/ stands for the number of times that a digit different from the previous one
is read in the b-ary representation of the positive integer m. Their result was subsequently
quantified by Barat, Tichy, and Tijdeman [40], who, under the same assumption, showed
that there are effectively computable real numbers c2 ; n0 , depending only on a and b,
such that
log n
BC.an ; b/ c2 ; for n > n0 :
log log n
Their proof is similar to that of Stewart [394]. We note that if m has at most k non-
zero digits in its b-ary representation, then BC.m; b/ does not exceed 2k. However, the
converse is not true, since a number m for which BC.m; b/ is small may have many non-
zero digits in its b-ary representation. The next statement, whose proof follows Stewart’s
arguments, extends (3.26).
Theorem 3.16. Let a and b be multiplicatively independent integers. Then we have
log log m
BC.m; a/ C BC.m; b/ ; (3.27)
log log log m C C
for m > 25, where C is a positive number which is effectively computable in terms of a
and b only.
Proof. Let m > a4 be an integer and let its a-ary representation be given by
m D ah a h C C a1 a C a0 ;
with ah ¤ 0. Define the integers n1 < n2 < < nr by nr D h, a0 D D an1 ,

an1 ¤ an1 C1 and anj C1 D D anj C1 , anj C1 ¤ anj C1 C1 for j 1, until we reach ah .
Then, observe that
an1 C1 1 an2 n1 1 anr nr 1 1

m D an 1 C an2 an1 C1 C C anr anr 1 C1
a 1 a 1 a 1
an1 an1 an2 n1 C1 an2 an3 n2 C1 anr nr C1
D C a C a C C a :
a 1 a 1 a 1 a 1
Likewise, if
m D bk b k C C b1 b C b0 ;
where bk ¤ 0, denotes the b-ary representation of m, we define the integers `1 <
`2 < < `t by `t D k, b0 D D b`1 , b`1 ¤ b`1 C1 and b`j C1 D D b`j C1 ,
b`j C1 ¤ b`j C1 C1 for j 1, until we reach bk . We have
b`1 b` b`2 `1 C1 b`2 b`3 `2 C1 b`t `t C1

mD C 1 b C b C C b :
b 1 b 1 b 1 b 1
Put D c1 log log m for a real number c1 3 which depends on a and b and will be
fixed later. Since m exceeds ee , we have 3. Define k to be the integer satisfying
log m
k < kC1 ; (3.28)
4 log a
and put
I1 D .1; ; I2 D .; 2 ; : : : ; Ik D . k 1
; k :
Assume that m is large enough to satisfy
log m 4c1 .log a/.log log m/: (3.29)
This ensures that k 1.

If each of the intervals I1 ; : : : ; Ik contains at least one term of the form nr nu with
1 u r 1 or of the form `t `v with 1 v t 1, then
BC.m; a/ C BC.m; b/ r C t 2 k: (3.30)

3.9. On the representation of integers in distinct bases 41
However, it follows from (3.28) that
.k C 1/ log > log log m log.4 log a/:
Combined with (3.30) and our choice of , this gives

log log m log.4 log a/
BC.m; a/ C BC.m; b/ 1;
log log log m C log c1
which implies (3.27) for a suitable C .
Therefore, we may assume that there is an integer s, with 1 s k, for which the
interval Is contains no term of the form nr nu with 1 u r 1 or of the form `t `v
with 1 v t 1. Define the integers p and q by the inequalities
nr np s 1
and nr np 1 > s;
and
`t `q s 1
and `t `q 1 > s;
with the convention that n0 D `0 D 0. Then, setting
A1 D anr anr C1 np
C .anr 1
anr /anr 1 C1 np
C C .anp anpC1 /a
and
A2 D .anp 1
anp /anp 1 C1
C .anp 2
anp 1 /anp 2 C1
C C .an1 an2 /an1 C1 an1 ;
we have
.a 1/m D A1 anp C A2 ;
with
0 < A1 < anr np C2
; 0 < jA2 j < anp 1 C2
:
Analogously, defining B1 and B2 in a similar way, we have
.b 1/m D B1 b `q C B2 ;
with
0 < B1 < b `t `q C2
; 0 < jB2 j < b `q 1 C2
:
We observe that
.b 1/.A1 anp C A2 /
1D ;
.a 1/.B1 b `q C B2 /
thus,
1/A1 anp 1 ˇˇ A1 anp B2 A2 B1 b `q ˇˇ
ˇ ˇ ˇ ˇ
ˇ .b ˇ b
ƒ WD ˇˇ 1ˇˇ
.a 1/B1 b `q a 1 ˇ B1 b `q .B1 b `q C B2 / ˇ
jA2 j A1 anp B2 (3.31)
C
.a 1/m .a 1/mB1 b `q
nr Cnp `t C`q
a;b .a 1
Cb 1
/:
Assume that ƒ D 0. Then, we have

.b 1/A1
log C np log a `q log b D 0: (3.32)
.a 1/B1
By Theorem 2.7, there exist rational integers x1 ; x2 ; x3 , not all zero, such that
.b 1/A1
x1 log C x2 log a C x3 log b D 0 (3.33)
.a 1/B1
and
maxfjx1 j; jx2 j; jx3 jg a;b log.A1 C B1 /:
Since s k and
log A1 a s 1
; log B1 b s 1
;
k 1
we deduce that jx2 j a;b and
log m
jx2 j < ;
4 log a
if m is large enough, by (3.28) and the definition of .
Also, using that np nr s 1 , we get that
log m
np > ;
4 log a
if m is large enough in terms of a and b, thus, np > jx2 j. Since a and b are multiplicatively
independent, x1 is non-zero. We multiply (3.32) by x1 and subtract the left-hand side
of (3.33) to get
.x1 np x2 / log a C .x3 x1 `q / log b D 0:
Noticing that x1 np x2 is non-zero, this contradicts our assumption that a and b are
multiplicatively independent.
Consequently, ƒ is non-zero. Since np nr 2 log m and `q `t 2 log m, it
follows from Theorem 2.2 that
ˇ .b 1/A1 anp
ˇ ˇ
ˇ
log ˇ
ˇ
`
1ˇˇ a;b log.A1 C B1 / log log m
.a 1/B1 b q
(3.34)
a;b .nr np C `t `q / log log m
a;b s 1
log log m:
On the other hand, we derive from (3.31) that there exist effectively computable positive
numbers c2 and c3 , depending only on a and b, such that
log ƒ > c2 s c3 : (3.35)
The combination of (3.34) and (3.35) shows that there exists an effectively computable
positive real number c4 , depending only on a and b, such that
c4 log log m:
Selecting now c1 D 2c4 , we have reached a contradiction. This shows that (3.29) does not
hold, thus, m is bounded from above by an effectively computable number which depends
only on a and b. This finishes the proof of the theorem.
3.10. Further applications (without proofs) 43
3.10. Further applications (without proofs)

We list below several applications of the theory of linear forms in complex logarithms,
which are not related to Diophantine equations. We do not claim to be exhaustive. Our
goal is merely to show that this theory applies to a large variety of problems.
– A notorious application of the theory of linear forms in logarithms concerns the Gauss
conjecture according to which there are only nine imaginary quadratic fields with class
number 1. This was observed by Baker [16] (see also Chapter 5 of [28] and Section 3.1
of [38]), while the conjecture was independently proved by Stark [387] with a different
method. Subsequently, Baker [24] and Stark [388] used the theory of linear forms in
logarithms of algebraic numbers to prove that all imaginary quadratic fields with class
number 2 can be effectively determined; see also [389] and [305,391] for the complete list
of these fields. It should be pointed out that this work on class numbers used linear forms
whose coefficients are quadratic (and not rational) numbers.
– Chowla raised the question whether there exists Pa rational valued function n 7! f .n/,
periodic with period a prime number, such that 1 nD1 f .n/=n D 0. Baker, Birch, and
Wirsing [32] used the theory of linear forms in logarithms with algebraic coefficients (an
earlier version of Theorem 2.1) to give a negative answer to Chowla question; see also
Chapter 23 of [308].
– Let q 2 be an integer, a non-trivial Dirichlet character modulo q, and set L.1; / WD
.n/
e2i=q /; : : :
P
n1 n . Since L.1; / can be expressed as a linear form in log.1
2i.q 1/=q
: : : ; log.1 e / with algebraic coefficients, it follows from Theorem 2.1 that
L.1; / is transcendental; see Chapters 20 and 25 of [308].
– Adhikari, Saradha, Shorey, and Tijdeman [3] investigated the transcendence of series
S D C1
P
nD0 f .n/=Q.n/, where Q.X / is a polynomial with only simple rational roots
and f W Z ! Q S is periodic, or f .X / is in QŒX
S , or f is an exponential polynomial.
Roughly speaking, they proved that either S is a computable algebraic number (that is, an
algebraic number that can be explicitly determined as a function of its defining parameters),
or S is transcendental. The proofs rest mainly on a lower bound for linear forms in
logarithms of algebraic numbers with algebraic coefficients, as given in Theorem 2.1.
– Let a; b be multiplicatively independent positive integers. Let be an irrational real
number, which is not a Liouville number. By means of Theorem 2.2 (the case n D 2),
Bourgain, Lindenstrauss, Michel, and Venkatesh [91] proved that the set fam b n W
0 m; n N g, viewed modulo 1, has no gaps of length greater than .log log N / c ,
where the real number c depends on a; b, and the irrationality exponent of . Harrap
and Haynes [222] applied Theorem 2.2 to a problem closely related to the Littlewood
conjecture in Diophantine approximation.
– Gorodnik and Spatzier [201] used Theorem 2.1 to investigate 3-mixing properties of
Z` -actions. The parameter B 0 is crucial in their proof.
– Kühne [247] applied the theory of linear forms in logarithms to give an effective
treatment of the André–Oort conjecture for a product of modular curves.
3.11. Exercises
Exercise 3.1. Discuss how Theorem 3.3 extends to multiplicatively independent positive
algebraic numbers a and b.
Exercise 3.2 (see [115, 370]). Let .pn =qn /n1 be the sequence of convergents to an
irrational algebraic number. Establish an effective lower bound for P Œpn qn .
Exercise 3.3. What can be said if the dominant root is not assumed to be simple in
Theorem 3.10? And if y is assumed to be fixed?
Exercise 3.4 (see [122]). Let f .X / be an irreducible integer polynomial of degree at
least two and n a positive integer which is not a root of f .X /. Establish an effective lower
bound for the greatest square free factor of f .n/.
Exercise 3.5 (see [294]). Let u D .un /n0 and v D .vm /m0 be linear recurrence
sequences of integers with a dominant root. Find a condition under which the equation
un D vm has only finitely many solutions in integers m; n.
Exercise 3.6 (see [410]). Let a 2 be an integer and q an odd prime number. Use
Theorem 2.3 to prove that if the integers x; y satisfy x 2 C a2 D 2y q , then q log a.
[Hint. Factor x 2 C a2 over ZŒi.]
Exercise 3.7 (see [403]). Let a; b; c be integers with a b > c > 0. Apply Theorem 2.2
to prove that
log a
P Œ.ab C 1/.bc C 1/.ca C 1/ log : (3.36)
log.c C 1/
b acC1 .acC1/.bcC1/
[Hint. Introduce the quantities ƒ1 D c
abC1
1 and ƒ2 D c 2 .abC1/
1.]
Exercise 3.8 (see [121]). Let u WD .un /n0 be a recurrence sequence of integers having a
dominant root. Let S be a finite, non-empty set of prime numbers. Prove that there exists
an effectively computable positive constant c.u; S /, depending only on u and S , such that
for every positive integer n we have
Œun S u;S jun j1 c.u;S /

:
Exercise 3.9. Let u WD .un /n0 be a recurrence sequence of integers having a dominant
root. Assume that there exist positive integers n; y; q and a non-zero rational number t
such that jun j 2 and un D ty q . Prove that we have q u maxfh.t /; 1g.
3.12. Notes
F Kim and Stewart [241] considered the case where the set S in Theorem 3.5 is an infinite
set of prime numbers.
F Using the hypergeometric method, Bennett, Filaseta, and Trifonov [56, 57] obtained
upper bounds for Œn.n C 1/S and Œn2 C 7S , wherep S D f2; 3g and S D f2g, respectively.
For instance, they established that Œn2 C 7f2g < n for every positive integer n not in
f1; 3; 5; 11; 181g (observe that 12 C 1, 32 C 1, 52 C 1, 112 C 1, and 1812 C 1 are all powers
of 2). When applicable, their method gives much stronger numerical results than those
obtained by following the proof of Theorem 3.8.
3.12. Notes 45
F The proof of Theorem 3.9 allows us to get a lower bound for the greatest prime factor
of the integer part of .3=2/n ; see [277] for related results.
F A large variety of results on linear recurrence sequences can be found in the mono-
graph [179]. In many cases, they are established under the dominant root assumption.
Sometimes, but not often, this extra assumption can be removed. This is for instance the
case for the growth of linear recurrence sequences [180, 337] and for the finiteness of
integral values for the ratio of two linear recurrence sequences [156].
F Consider the Tribonacci sequence defined by T0 D T1 D 0, T2 D 1 and the recurrence
relation
TnC3 D TnC2 C TnC1 C Tn ; n 0:
Observe that the polynomial P .X / D X 3 X 2 X 1 has a real root in the interval .1; 2/
and two complex conjugate roots in the open unit disc. Theorem 3.10 applies to show
that, for every prime number q large enough, the equation Tn D y q has no solution
with n 4. More generally, Pethő [325] established that if .un /n0 is a third order
recurrence sequence such that its companion polynomial is irreducible and has a dominant
root, then there are only finitely many perfect powers in .un /n0 . This result is, however,
not effective and does not enable us to get an upper bound for the largest perfect power
in .un /n0 . As pointed out in [326], we do not even know whether T2 D T3 D 1, T5 D 4,
T10 D 81, T16 D 3136 D 562 , and T18 D 10609 D 1032 are the only positive squares in
the sequence .Tn /n0 .
We can as well define Tn for negative integers n using the same recurrence relation.
Then, Theorem 3.10 does not apply anymore to the equation T n D y q in positive
integers n; y; q, since the companion polynomial X 3 P .X 1 / has two roots with the same
modulus, namely the inverses of the two complex conjugate roots of P .X /. This is an
open problem [326] to prove that, for every prime number q large enough, the equation
T n D y q has no solution with n 4.
F Bugeaud, Cipu, and Mignotte [119] have determined the Fibonacci and Lucas numbers
with at most four binary digits. The largest such number is 2584, the eighteenth Fibonacci
number, which can be expressed as 211 C 29 C 24 C 23 . For further results on integers
having at most three digits in two distinct bases, see [147, 276] and the references quoted
therein.
Chapter 4
Classical families of Diophantine equations
In this chapter, we show how the theory of linear forms in logarithms can be applied
to prove effective upper bounds for the size of the integer solutions to several classes
of Diophantine equations, including unit equations, Thue equations, and superelliptic
equations.
We begin by establishing how a weaker version of Theorem 2.2 can be derived from the
estimate of linear forms in two logarithms given in Theorem 2.3. Such a weaker version of
Theorem 2.2 was obtained by Baker as early as in 1973 (see Theorem 1.10). It is sufficient
for proving many results gathered in the present chapter, including Theorems 4.3 to 4.7,
but, apparently, not for establishing Theorems 4.9 and 4.11.
4.1. Proof of Baker’s Theorem 1.10

Let ˛1 ; : : : ; ˛n ; ˛nC1 be complex algebraic numbers. Let b1 ; : : : ; bn be non-zero integers
and assume that
ƒ WD ˛1b1 ˛nbn ˛nC1 1 ¤ 0:
By Liouville’s inequality Theorem B.10, there exists a positive real number c, depending
only on ˛1 ; : : : ; ˛nC1 , such that
cB
jƒj > e ; where B WD maxfjb1 j; : : : ; jbn jg:
As mentioned in Chapter 1, Baker established that, if there exists a positive real

number ı such that jƒj < e ıB , then B is bounded by some quantity depending only on
n; ı; ˛1 ; : : : ; ˛n times the maximum of 1 and the height of ˛nC1 .
Following an idea of Bombieri and Cohen [86], Bilu and Bugeaud [72] showed how
Baker’s Theorem 1.10 can be derived from the lower bound for linear forms in two complex
logarithms established in [254] and reproduced as Theorem 2.3; see also Theorem 11.1.
The next statement was proved in [72]; see also [432].
Theorem 4.1. Let ˛1 ; : : : ; ˛n ; ˛nC1 be complex algebraic numbers in an algebraic number
field of degree D. Let b1 ; : : : ; bn be non-zero integers and set
B WD maxfjb1 j; : : : ; jbn jg:
If there exists a positive real number ı such that
0 < j˛1b1 ˛nbn ˛nC1 1j < e ıB

; (4.1)
48 4. Classical families of Diophantine equations
then
B ˛1 ;:::;˛n ;D;ı;n maxfh.˛nC1 /; 1g:
Clearly, Theorem 4.1 is considerably weaker than Theorem 2.2. However, its proof is
much simpler and it is sufficient for many interesting applications, as will be shown in the
next sections. A key ingredient is the quantity B 0 (instead of B) occurring in Theorem 11.1.
The proof of Theorem 4.1 actually gives us a more precise upper bound for B, where the
dependence of the implied constant is made explicit in terms of ˛1 ; : : : ; ˛n ; D; ı; n. We
stress that the present monograph includes a full proof of Theorem 11.1, thus a full proof
of Theorem 4.1.
For the proof of Theorem 4.1 we need a special case of a lemma of Bombieri and
Cohen [86].
Lemma 4.2. Let b0 ; b1 ; : : : ; bn ; N; Q be integers with N Q 2. There exist a
positive integer r and integers t0 ; t1 ; : : : ; tn , not all zero, such that b N
Q
c r N and
1=.nC1/ jbi j
jbi rti j rQ C ; i D 0; : : : ; n:
r
Proof. By Minkowski’s theorem (see e.g. Chapter II in [357] or Appendix B of [113])
there exist rational integers t; t0 ; t1 ; : : : ; tn , not all zero, such that jtj Q and
ˇ b ˇ
ˇ i 1=.nC1/
ti ˇ Q ; i D 0; : : : ; n:
ˇ
(4.2)
N
ˇt
We have t ¤ 0 since, otherwise, we would get t0 D t1 D D tn D 0 by (4.2), which is

excluded. Replacing t by t if necessary, we may assume that 1 t Q. Let r be the
integer part of Nt . Since N Q t 1, we get N r b N Q
c 1 and, consequently,
for i D 0; : : : ; n, we have
ˇ tb
i
N b t ˇ
i ˇ jbi j
jbi rti j D ˇr ti C r ˇ rQ 1=.nC1/ C ;
ˇ
N t N r
as asserted.
Proof of Theorem 4.1. Keep the assumption of the theorem. Let N and Q be integers
which will be chosen later and satisfy
B N Q 2: (4.3)
Recall that log denotes the principal determination of the logarithm. If j˛1b1 ˛nbn ˛nC1 1j
< 13 , then there exists an integer b0 , with jb0 j nB C 1, such that
WD b0 log. 1/ C b1 log ˛1 C C bn log ˛n C log ˛nC1
satisfies j˛1b1 ˛nbn ˛nC1 1j jj 2

.
Let r; t0 ; t1 ; : : : ; tn be the integers given by Lemma 4.2 applied to b0 ; b1 ; : : : ; bn ; N;
and Q. Set ˛0 D 1, A0 WD e=D ,
˛ D ˛0t0 ˛1t1 ˛ntn ; and D ˛0b0 rt0

˛1b1 rt1
˛nbn rtn
˛nC1 :
4.1. Proof of Baker’s Theorem 1.10 49
Define `.˛/ and `. / by

n
X n
X
`.˛/ D ti log ˛i ; `. / D .bi rti / log ˛i C log ˛nC1 :
iD0 iD0
We point out that `.˛/ and `. / satisfy exp.`.˛// D ˛ and exp.`. // D , respectively,
but they have no reason to be the principal determinations of the logarithms of ˛ and .
We have expressed the linear form as
D r`.˛/ C `. /;
which is a linear form in two logarithms. We will use Theorem 11.1 to bound jj from
below.
For i D 0; : : : ; n, observe that
1=.nC1/ jbi j 1=.nC1/ nB C 1

jbi rti j rQ C rQ C ;
r r
jbi j C jbi rti j nB C 1 nB C 1 1=.nC1/ 4nB
jti j C CQ ;
r r r2 r
1=.nC1/ B
since Q 1 r
by (4.3). For i D 1; : : : ; n C 1, define Ai by
n j log ˛i j 1 o
log Ai WD max h.˛i /; ; :
D D
We deduce the estimates
n n
n j`.˛/j 1 o X 4nB X
max h.˛/; ; jti j log Ai log Ai ; (4.4)
D D r
i D0 i D0
n
n j`. /j 1 o X
max h. /; ; jbi rti j log Ai C log AnC1
D D
i D0
n

1=.nC1/ 2nB X
rQ C log Ai C log AnC1 : (4.5)
r
i D0
Assume that Q satisfies the inequality
B max 4Q2C1=.nC1/ ; 2.log AnC1 /Q1C1=.nC1/

˚
(4.6)
and set
N D Q max B 1=2 Q1=.2nC2/ ; .log AnC1 /Q1=.nC1/ :
˙ ˚
Then, (4.3) holds and we get the upper bounds (4.4) and (4.5). Furthermore, r b NQ
cD N
Q
B 1=.nC1/ 1=.nC1/
implies r rQ and log AnC1 rQ . The former inequality, combined
with (4.5), gives
n j`. /j 1 o
max h. /; ; 3nrQ 1=.nC1/ .1 C log A0 C log A1 C C log An / DW G:
D D
Since is non-zero, it follows from Theorem 11.1 that

n
nB X r 2
log jj D4 log Ai G log C2
r G
i D0
and there exists an effectively computable positive real number c1 such that
n
jj X 2
log > c1 n2 BD 4 Q 1=.nC1/
log Ai .log Q/2 :
2
iD0
Setting
n
X 2
ı WD c1 n2 D 4 Q 1=.nC1/
log Ai .log Q/2 ; (4.7)
i D0
we have established that
log j˛1b1 ˛nbn ˛nC1 1j ıB:
This contradicts our assumption (4.1), thus (4.6) does not hold and we get
B < max 4Q2C1=.nC1/ ; 2.log AnC1 /Q1C1=.nC1/

˚
(4.8)
Setting ız WD minfı; 1g, we deduce from (4.7) that

n
X 2.nC1/ n
X 2.nC1/
Q .c2 n/4n D 4.nC1/ ız n 1
log Ai log nD ız 1 log Ai ;
iD0 i D0
where c2 is an effectively computable positive real number. The theorem then follows
from (4.8) and the fact that
n j log ˛nC1 j 1 o
log AnC1 D max h.˛nC1 /; ; h.˛nC1 / C ;
D D D
by the definition of the height.
4.2. The unit equation

Let K be an algebraic number field. Many Diophantine problems reduce to equations of
the form
˛x C ˇy D 1;
where ˛; ˇ are given elements of K and the unknowns x; y are elements of its unit

group OK . These equations are called unit equations. Explicit upper bound for the solu-
tions of unit equations were given for the first time by Győry [210]; see also Bugeaud and
Győry [124], Győry and Yu [217], and the monographs [182, 183] by Evertse and Győry,
where the reader can find many bibliographic references and numerous applications.
4.2. The unit equation 51
Theorem 4.3. Let K be an algebraic number field of degree d and discriminant DK .

Let ˛, ˇ be non-zero elements of K. The unit equation
˛x C ˇy D 1 (4.9)

has only finitely many solutions in algebraic units x and y in OK , and all of them satisfy
maxfh.x/; h.y/g d;DK maxfh.˛/; h.ˇ/; 1g: (4.10)

Proof. Let x, y be in OK such that (4.9) holds. We assume that h.x/ h.y/. Let r
be the unit rank of K and f1 ; : : : ; r g a fundamental system of units in K satisfying the
inequalities stated in Lemma C.4. Then, we can write
y D b11 : : : br r ; (4.11)
where is a root of˚ unity in K and b1 ; : : : ; br are rational integers.

Put B WD max jb1 j; : : : ; jbr j; 3 and let v1 ; : : : ; vrC1 denote the non-equivalent com-
plex places on K. Then (4.11) implies that
r
X
log jyjvj D bi log ji jvj ; j D 1; : : : ; r:
iD1
Using Cramer’s rule and the definition of the height, we deduce from (4.11) and Lemma C.4
that
B d;DK h.y/ d;DK B: (4.12)
Let v be a complex place for which jxjv is minimal. It then follows from Definition B.4
that 1
h.x/ D h log jxjv :
x
Assume, without any loss of generality, that h.x/ 2dh.˛/. Then, we deduce from (4.12)
that
h.x/
log j˛xjv log j˛jv C log jxjv dh.˛/ h.x/
2 (4.13)
h.y/
d;DK . B/:
2
Let be a complex embedding such that jxjv D j .x/j. By setting ˛rC1 D ˇ we deduce
from (4.9) that
j˛xjv D jb11 : : : br r ˛rC1 1jv D j .1 /b1 : : : .r /br .˛rC1 / 1j: (4.14)
It then follows from (4.13) and Theorem 4.1 that
B d;DK maxfh.˛rC1 /; 1g d;DK maxfh.ˇ/; 1g;
thus, by (4.11),
h.y/ d;DK B d;DK maxfh.ˇ/; 1g:
We deduce from Theorem B.5 that
h.x/ h.˛x/ C h.˛/ C log 2

h.y/ C h.ˇ/ C h.˛/ C log 4 d;DK maxfh.˛/; h.ˇ/; 1g:
By Northcott’s Theorem B.7, this implies that (4.9) has only finitely many solutions in

x; y in OK and completes the proof of the theorem.
If we are interested in the size of the constant implicit in d;DK , a better result follows
by applying Theorem 2.2 directly to (4.14) rather than using Theorem 4.1; see Section 6.6.
Observe that the upper bound in (4.10) is linear in maxfh.˛/; h.ˇ/; 1g. This is
ultimately a consequence of the quantity B 0 occurring in the estimates of linear forms in
logarithms.
4.3. The Thue equation

Let F .X; Y / be an homogeneous, binary, integer polynomial of degree at least three and
such that F .X; 1/ has at least three distinct roots. Let m be a non-zero integer. In 1909,
Axel Thue [412] established that the equation
F .x; y/ D m; in integers x; y,
has only finitely many solutions. This equation is called the Thue equation. Thue’s
method, however, does not yield upper bounds for the absolute values of its solutions. The
first general effective result on Thue’s equations was established by Baker [20]. We also
refer to [33, 125, 216] and the monograph [182] for general totally explicit statements, and
to [125, 182, 216, 376] for an extensive list of bibliographic references and applications.
Theorem 4.4. Let F .X; Y / be an homogeneous, irreducible polynomial with integer
coefficients and of degree at least three. Let m be a non-zero integer. Then, there exists a
positive real number C , depending only on F .X; Y /, such that all the integer solutions x; y
to the Diophantine equation
F .x; y/ D m
satisfy maxfjxj; jyjg j2mjC .
We establish a more general statement, which will be needed in Section 4.5.
Theorem 4.5. Let K be an algebraic number field of degree d and discriminant DK . Let
n 3 be an integer and ˛1 ; : : : ; ˛n distinct algebraic numbers in K. Let a and m be
non-zero algebraic integers in K. Then, the equation
a.x ˛1 y/ .x ˛n y/ D m (4.15)
has only finitely many solutions in algebraic integers x and y in K, and all of them satisfy
maxfh.x/; h.y/g n;d;DK ;a;˛ maxfh.m/; 1g; (4.16)
where ˛ denotes the n-tuple .˛1 ; : : : ; ˛n /.

4.3. The Thue equation 53
The assumption n 3 cannot be replaced by n 2 in view of the Pellian equation.

Theorem 4.5 not only shows that the heights of x and y are effectively bounded, but
also that the bound is linear in maxfh.m/; 1g. Again, this is ultimately a consequence of
the introduction of the quantity B 0 in the estimates for linear forms in logarithms.
Proof. Let x and y be algebraic integers in K satisfying (4.15). Multiplying both sides
of (4.15) by an 1 times a suitable positive integer depending on ˛1 ; : : : ; ˛n , we see that we
may assume in the sequel that a D 1 and that ˛1 ; : : : ; ˛n are algebraic integers. Suppose
also that the heights of x and y are very large in terms of d; DK , and ˛. By applying if
needed a complex embedding of K, we may assume that
log maxfjxj; jyjg d maxfh.x/; h.y/; 1g: (4.17)
There exists at most one index i such that
jyj
jx ˛i yj < min j˛j ˛k j:
2 1j <kn
Since the heights of x and y are supposed to be sufficiently large, it follows from (4.15)
and (4.17) that this index exists and, without any loss of generality, we take i D 1. We
then get jxj ˛ jyj and, successively, the inequalities
jx ˛j yj ˛ maxfjxj; jyjg; j D 2; : : : ; n;
nC1
(4.18)
jx ˛1 yj n;˛ jmj maxfjxj; jyjg :
We also assume that maxfjxj; jyjg jmj (otherwise, (4.16) clearly holds) and get
nC2
jx ˛1 yj n;˛ maxfjxj; jyjg : (4.19)
Set
ˇj D x ˛j y; j D 1; 2; 3;
and observe that
.˛1 ˛2 /ˇ3 C .˛2 ˛3 /ˇ1 C .˛3 ˛1 /ˇ2 D 0:
Consider the quantity

˛2 ˛3 x ˛1 y x ˛3 y ˛2 ˛1
ƒ WD D 1:
˛3 ˛1 x ˛2 y x ˛2 y ˛3 ˛1
Let r be the rank of the group of units of the field L WD K.˛1 ; ˛2 ; ˛3 / and 1 ; : : : ; r a
system of fundamental units of L satisfying the inequalities stated in Lemma C.4. Note
that the degree of L is at most equal to d 3 and the absolute value of its discriminant is
bounded from above in terms of ˛1 ; ˛2 ; ˛3 ; d; and DK , by Lemma C.7.
By Proposition C.5, there exist algebraic integers 2 ; 3 in L and integers b1 ; : : : ; br ,

c1 ; : : : ; cr such that
x ˛2 y D 2 b11 : : : br r ; x ˛3 y D 3 c11 : : : crr ;
h. 2 /; h. 3 / d;DK ;˛ log j2 NormL=Q mj; (4.20)
and
B WD maxfjb1 j; : : : ; jbr j; jc1 j; : : : ; jcr jg

d;DK ;˛ 1 C h.x ˛2 y/ C h.x ˛3 y/ d;DK ;˛ X; (4.21)
where X WD maxfh.x/; h.y/; 1g. It follows from (4.17), (4.18), (4.19), and (4.21) that
log jƒj n;d;DK ;˛ . log maxfjxj; jyjg/ n;d;DK ;˛ . X / n;d;DK ;˛ . B/: (4.22)
Observe that
3 .˛2 ˛1 /
ƒ D c11 b1
: : : crr
1: br
2 .˛3 ˛1 /
Write WD 3 .˛2 ˛1 /=. 2 .˛3 ˛1 // and note that h./ d;DK ;˛ log M , where we
have set M WD j2 NormL=Q mj. We immediately deduce from (4.22) and Theorem 4.1
that
B n;d;DK ;˛ maxfh./; 1g n;d;DK ;˛ log M: (4.23)
Using (4.20) to bound the heights of
.˛3 ˛2 /x D ˛3 .x ˛2 y/ ˛2 .x ˛3 y/
and .˛3 ˛2 /y, we get
h.x/; h.y/ d;DK ;˛ .B C log M /: (4.24)
Thus, by (4.23), we obtain
X n;d;DK ;˛ log M:
Since log M d maxfh.m/; 1g, this proves (4.16) and we deduce from Northcott’s
Theorem B.7 that (4.15) has only finitely many solutions x; y in K. This completes the
proof of the theorem.
Instead of using Theorem 4.1, we can as well apply Theorem 2.2, which gives
B
log jƒj n;d;DK ;˛ h./ log 1 C ;
h./
so that, by (4.21), we obtain
B B
n;d;DK ;˛ log 1 C :
h./ h./
From this, we deduce that
B n;d;DK ;˛ maxfh./; 1g n;d;DK ;˛ log M;
and we conclude by using (4.24). This more direct proof yields the best known explicit
bounds.
4.4. Effective improvement of Liouville’s inequality 55
4.4. Effective improvement of Liouville’s inequality

An outstanding open problem in Diophantine approximation is to establish an effective
version of Roth’s Theorem A.7. Results in this direction were given in Section 3.4 for
specific classes of algebraic numbers. In 1968 Feldman [185] was the first to prove that
the effective irrationality exponent of an arbitrary real algebraic number of degree greater
than two is strictly less than its degree.
The present textbook contains a full proof of Theorem 4.6, which is derived from an
estimate for linear forms in only two complex logarithms; see [72].
Theorem 4.6. Let ˛ be an algebraic number of degree n 3. Then, there exists an
effectively computable positive real number .˛/, depending only on ˛, such that
eff .˛/ n .˛/:
Proof. Set ˛1 D ˛ and let a.X ˛1 / .X ˛n / with a 1 denote the minimal

defining polynomial of ˛ over the integers. Let x and y be non-zero integers and set X WD
maxfjxj; jyjg. It follows from Theorem 4.4 that there exists an effectively computable
positive number C , depending only on ˛, such that
X 1=C .n 1/C1=C
jx ˛yj ˛ X :
2ajx ˛2 yj : : : jx ˛n yj
It follows from Theorem A.4 that Theorem 4.6 cannot hold for quadratic numbers; see,
however, Theorem 6.3 for a result of a similar flavour valid for quadratic numbers.
4.5. The superelliptic and hyperelliptic equations

In 1926 Siegel [377] established that, for any integers m, n with m 3 and n 2, for any
non-zero integers a, b, and k, the equation
ax m by n D k; in integers x, y;
has only finitely many solutions. His method does not enable us to compute upper bounds
for jxj and jyj. However, this can be achieved since the end of the ’60s, as one of the
many applications of the theory of linear forms in logarithms; see [21, 22] and [92, 101]
for more recent and more general results.
Theorem 4.7. Let f .X / be an integer polynomial of degree m without multiple roots,
and let n be an integer. Assume that m 2, n 2, and mn 6. Let t be a non-zero
integer. Then, every solution in integers x; y to
f .x/ D ty n (4.25)
satisfies maxfjxj; jyjg f;t;n 1.

Proof. Let K be the splitting field of f .X /. By multiplying both sides of (4.25) by the
.m 1/-th power of the leading coefficient of f .X /, we can assume that f .X / is monic.
Assume first that n 3. It follows from Proposition C.6 and Theorem B.5 that there
exist distinct roots ˛1 ; ˛2 of f .X /, algebraic integers 1 , 2 in K, and ˇ1 , ˇ2 in K such
that
x ˛1 D ˇ1 1n ; x ˛2 D ˇ2 2n ;
and
h.ˇ1 /; h.ˇ2 / n;t;f 1:
By Northcott’s Theorem B.7, this shows that the pair .ˇ1 ; ˇ2 / belong to a finite set of pairs
of algebraic numbers and that . 1 ; 2 / is solution to the Thue equation (over the number
field K)
ˇ1 X n ˇ2 Y n D ˛2 ˛1 : (4.26)
Since the polynomial ˇ1 X n ˇ2 has n 3 distinct roots, it follows from Theorem 4.5
that (4.26) has only finitely many solutions in algebraic integers X, Y in K, thus the set
of quadruples .ˇ1 ; ˇ2 ; 1 ; 2 / is finite. Furthermore, we get an effective upper bound for
the heights of 1 and 2 , hence for jxj and jyj, in terms of f .X /, t , and n.
The case n D 2 is slightly more involved. It follows from Proposition C.6 and
Theorem B.5 that there exist distinct roots ˛1 ; ˛2 ; ˛3 of f .X /, algebraic integers 1 ; 2 ; 3
in K, and ˇ1 ; ˇ2 ; ˇ3 in K such that
x ˛1 D ˇ1 12 ; x ˛2 D ˇ2 22 ; x ˛3 D ˇ3 32 :
and
h.ˇ1 /; h.ˇ2 /; h.ˇ3 / t;f 1:
By Northcott’s Theorem B.7, this shows that the triple .ˇ1 ; ˇ2 ; ˇ3 / belongs to a finite
set of triples of algebraic numbers. For i D 1; 2; 3, fix a square root ˇi1=2 of ˇi . Set
L D K.ˇ11=2 ; ˇ21=2 ; ˇ31=2 /. Since
˛i ˛j D ˇj j2 ˇi i2 ; for 1 i < j 3,
the norms over Q of ˇj1=2 j ˇi1=2 i , where 1 i < j 3, are bounded in terms
of f .X / only. Put
ı1 D ˇ21=2 2 ˇ31=2 3 ; ı2 D ˇ31=2 3 ˇ11=2 1 ; ı3 D ˇ11=2 1 ˇ21=2 2 ;

ı1C D ˇ21=2 2 C ˇ31=2 3 ; ı2C D ˇ31=2 3 C ˇ11=2 1 ; ı3C D ˇ11=2 1 C ˇ21=2 2 :
For i D 1; 2; 3, we can write
ıi D i "i ; ıiC D iC "C

i ;
where, by Proposition C.5, i ; iC are in L and have their heights bounded in terms
of f .X / and the degree and discriminant of L, and "i ; "C
i are unknown units in L.
4.6. The Diophantine equation x 2 C C D y n 57
We check that
1C "C
1 2C "C
2 C 3 "3 D 0
and 1C "C
1 2 "2 3C "C
3 D 0:
It then follows from Theorem 4.3 on unit equations that h."C C C

1 ="3 / and h."1 ="3 / are
bounded in terms of f .X / and the degree and discriminant of L. Since
˛2 ˛1 D ı3 ı3C D 3 3C "3 "C
3;
we get
"C "C ˛2 ˛1
."C 2
1/ D
1

C "
1
:
"3 3 3 3C
This implies that the height of "C C
1 and, consequently, the height of ı1 are bounded in
terms of f .X / and the degree and discriminant of L. The same conclusion holds for the
heights of "1 and ı1 . Since
.ı1 C ı1C /2 D 4ˇ2 22 D 4.x ˛2 /;
it then follows that jxj is bounded in terms of f .X / and the degree and discriminant of L.
Note that the degree of L is at most equal to 8 times the degree of K. Furthermore, by
Lemma C.7, the discriminant of L is effectively bounded in terms of the discriminant
of K, the degree of L, and the heights of ˇ1 ; ˇ2 ; ˇ3 , which are all t;f 1. Since K is the
splitting field of f .X /, its discriminant and height are f 1. Putting all this together, we
have established that maxfjxj; jyjg f;t 1. This completes the proof of the theorem.
4.6. The Diophantine equation x 2 C C D y n

The Diophantine equation x 2 C 1 D y n was solved in 1850 by V. A. Lebesgue [256]
by means of an elegant 2-adic argument; see Section 2.1 of [73]. Since then, many
mathematicians have investigated the equation
x2 C C D yn; (4.27)
where C is a fixed non-zero integer and x; y; n are unknown integers with n 3. We do
not consider the case n D 2, since it reduces to .x C y/.x y/ D C , which is easy to
solve. For references on the case C positive, the reader is directed to [138, 155] and the
articles quoted therein.
A major difference between the cases C negative and C positive is the existence of
values of C for which x 2 C C D y n has a solution for every (odd) value of n. This
happens precisely when C is negative and equal to minus a square plus or minus one. In
that case, the equation x 2 C C D ˙1 has a solution and this prevents the direct use of
congruences to solve completely x 2 C C D y n for n 3 odd, since y D 1 or y D 1
gives then always a solution. Obviously, this situation cannot occur for positive values
of C .
Ramanujan [343] noticed that the equation x 2 C 7 D 2n has at least 5 solutions in
positive integers x; n, given by x D 1; 3; 5; 11; and 181. Subsequently, Nagell [310, 311]
(see also his collected papers [313]) established in 1948 that there are no more solutions
and, nowadays, this equation is called the Ramanujan–Nagell equation. This is the most
famous special case of (4.27).
All the solutions of (4.27) with 1 C 100 are listed at the end of [138]. The
hundred corresponding equations are not all equally difficult to solve. Below we focus
our attention on three specific values of C , namely 2; 7, and 25, and explain how one
derives a linear form in logarithms from each of the corresponding equations. For a
further discussion on the equation x 2 C C D y n , see the short Section 7.3. Section 7.4 is
devoted to the number of solutions to the equation x 2 C D D p n .
Theorem 4.8. The only solutions in positive integers x, y, n to the generalised Ramanujan–
Nagell equation
x 2 C 7 D y n ; n 3;
satisfy x D 1, 3, 5, 11, 181. The only solutions in positive integers x; y; n to the equations
x 2 C 2 D y n and x 2 C 25 D y n with n 3 satisfy x D 5 and x D 10, respectively.
Outline of the proof. In the sequel, we assume that x 20 and n is an odd prime number q.
Let x; y; q be such that x 2 C 2 D y q . Arguing modulo 4, we see that, since q 3,
the integer x must be odd. Write
p p
x 2 C 2 D .x C i 2/.x i 2/:
p p p
Since every common divisor of x C i 2 and x pi 2 also divides p their difference 2i 2,
we pdeduce from the fact that x is odd that x C i 2 and x i 2 are coprime. The ring
ZŒi 2 is a unique factorization domain, thus there exist rational integers a; b such that
p p
x C i 2 D .a C ib 2/q : (4.28)
The quantity
ˇ a C ib p2 q ˇ ˇ x C ip2 ˇ
ƒ2 WD ˇ p 1ˇ D ˇ p 1ˇ
ˇ ˇ ˇ ˇ
a ib 2 x i 2
p
is very small, bounded from above by x4 . Since the Galois conjugate of WD a C ib 2 is
equal to its inverse and
2.a2 2b 2 /
C 1 D ;
a2 C 2b 2
we get
log x
h./ log.a2 C b 2 / :
q
It then follows from Theorem 2.6 that
log x
log 4 log x log ƒ2 .log q/2 ;
q
and we derive an absolute upper bound on q. This bound, obtained by means of an estimate
of linear forms in two logarithms, is rather small. Then, for each value of q between 5
4.6. The Diophantine equation x 2 C C D y n 59
and this bound, sieving by several prime numbers congruent to 1 modulo q, we can rather
quickly establish that x 2 C 2 D y q has no solutions. If x 2 C 2 D y 3 , then we get
p p
x C i 2 D a.a2 6b 2 / C ib.3a2 2b 2 / 2;
thus b D 1 and a D ˙1. This shows that 52 C 2 D 33 is the only solution of x 2 C 2 D y 3

in positive integers.
Instead of Theorem 2.6, we could of course use Theorem 2.2 and get q log q,
but the numerical upper bound for q will then be larger, because of the larger size of the
numerical constant in Theorem 2.2.
p
Let x; y; q be such that x 2 C 7 D y q . We work in the
p ring ZŒ.1
p C i 7/=2, which
2
is a unique factorization
p p Write x C 7 D .x C i 7/.xp i 7/. Every common
domain.
divisor of x C i 7 and x i 7 also divides their difference 2i 7. Since
1 C ip7 1 ip7
2D ;
2 2
we deduce that there exist rational integers a; b; and " in f˙1g such that
p 1 C "ip7 p
xCi 7D .a C ib 7/q : (4.29)
2
Again, the quantity
ˇ 1 C "ip7 a C ib p7 q ˇ ˇ x C ip7 ˇ
ƒ7 WD ˇ p p 1ˇ D ˇ p 1ˇ
ˇ ˇ ˇ ˇ
1 "i 7 a ib 7 x i 7
is very small, bounded from above by x9 . As above, we check that
a C ib p7 log x
h p log.a2 C b 2 / :
a ib 7 q
As already mentioned before the statement of Theorem 2.2, the fact that ƒ7 is smaller
than x9 does not imply that the linear form
ˇ 1 C "ip7 a C ib p7 ˇ
p C q log p ˇ
ˇ ˇ
ˇlog
1 "i 7 a ib 7
is smaller than 18
x
. It only implies the existence of an integer b0 with jb0 j 1 C q such
that
ˇ 1 C "ip7 a C ib p7 ˇ 18
ˇb0 log. 1/ C log p C q log p ˇ :
ˇ ˇ
1 "i 7 a ib 7 x
We then deduce from Theorem 2.2 that
log x
log 18 log x log ƒ7 .log q/:
q
Again, we get an absolute upper bound on q, but it is much bigger than in the case of
the equation x 2 C 2 D y q , since we were forced to use a lower bound for three (instead
of two) logarithms. A careful reader may observe that an upper bound for q can also be
deduced from Theorem 4.1, thus, ultimately follows from Theorem 2.3. However, if we
proceed along these lines, then we end up with a much larger upper bound than the one
obtained by applying Theorem 2.2 in the way explained above.
Let x; y; q be such that x 2 C 25 D y q . Observe that x 2 C 25 D .x C 5i/.x 5i/.
Arguing modulo 2 we deduce that x must be even. If 5 does not divide x, then the greatest
common divisor of x C 5i and x 5i divides 2, thus, since x is even, x C 5i and x 5i
are coprime. Consequently, there exist rational integers a; b such that
x C 5i D .a C ib/q :
We then proceed as in the case of the equation x 2 C 2 D y q and derive an upper bound
for q from an estimate of a linear form in two logarithms. We omit the details.
The difficulty occurs when 5 divides x. Indeed, 5 must then divide y as well, and we
end up with the Diophantine equation
x 2 C 1 D 5q 2 q
y : (4.30)
Noticing that 5 D .2 C i/.2 i/, we factor (4.30) in the ring of Gaussian integers and
eventually derive a linear form in three logarithms. This allows us to compute an upper
bound for q. Again, we leave the details to the reader.
2 q
p resolution of x CC D y
For an arbitrary value of C , a further difficulty occurs in the
when the class number of the quadratic field generated by C is divisible by the prime
number q, because such a factoring as in (4.28) or (4.29) is not possible anymore. In
this case, the standard method consists in reducing x 2 C C D y q to Thue equations of
degree q; see e.g. [138, 301].
4.7. Perfect powers at integer values of polynomials

As already observed in Section 3.7, the theory of linear form in complex logarithms also
allows us to bound effectively the size of the solutions to Diophantine equations, even when
exponents are unknown. The next result was proved by Schinzel and Tijdeman [356] and,
subsequently, generalized by Shorey [369], with a simpler proof; see also [61, 95, 100].
A proof of Theorem 4.9 in the special case where f .X / has at least two simple rational
zeros was given in a survey paper of Tijdeman [417]. While ineffective versions of
Theorems 4.5 and 4.7 were established before the theory of linear forms in logarithms, no
proof of Theorem 4.9 based on the ideas of Thue and Siegel is known.
Theorem 4.9. Let f .X / be an integer polynomial of degree n 2 without multiple roots.
Set z0 D 3 if n D 2 and z0 D 2 otherwise. Let t be a non-zero integer. Then, every
integer solution .x; y; z/ with z z0 and jyj 2 to
f .x/ D ty z (4.31)
satisfies maxfjxj; jyj; zg t;f 1.

4.7. Perfect powers at integer values of polynomials 61
Proof. We follow an argument of Brindza, Evertse, and Győry [95]. Multiplying both
sides of (4.31) by the .n 1/-th power of the leading coefficient of f .X /, we can assume
that f .X / is monic. Our goal is to bound z in terms of f .X / and t since, in view of
Theorem 4.7, this is sufficient to conclude.
Let H be the maximum of the absolute values of the coefficients of f .X /. Let
˛1 ; : : : ; ˛n denote its roots and observe that they are bounded in modulus by H C 1.
Let .x; y; z/ be a solution of (4.31). If jxj 2.H C 1/, then we have
2z jty z j D jx ˛1 j jx ˛n j .3H C 3/n :
We can therefore assume that jxj > 2.H C 1/, which implies that jx ˛j j jxj 2
> 1 and
jx ˛j j jty z j for j D 1; : : : ; n. We can as well assume that jxj > jyjz=.2n/ . Indeed,
otherwise, we would get jx ˛j j 2jxj 2jyjz=.2n/ for j D 1; : : : ; n, thus
jy z j jf .x/j 2n jyjz=2 ;
giving z log jyj 2n log 2 and z 2n.
Assume first that f .X / has an irrational root ˛1 . Let f1 .X / be the minimal defining
polynomial of ˛1 . Set K D Q.˛1 / and let hK denote its class number. By Proposition C.6,
there exist algebraic integers ˇ1 ; 1 ; ı1 in K and a unit 1 in K such that
ı1 .x ˛1 /hK D 1 ˇ1 1z (4.32)
and
˚
max h.ı1 /; h.ˇ1 / t;f 1: (4.33)
Let ˛2 denote a root of f1 .X /, distinct from ˛1 , and denote by the complex embedding
sending ˛1 to ˛2 . For simplicity, for any in Q.˛1 /, we write instead of . /. Then,
by applying to (4.32), we get
ı1 .x ˛2 /hK D 1 ˇ1 . 1 /z : (4.34)
We may assume that
4hK .H C 1/ < 2z=.4n/ ; (4.35)
jxj z=.2n/
since otherwise the theorem is proved. From jx ˛1 j 2
and jxj > jyj , we get
ˇ˛
ˇ 1 ˛2 ˇˇ 4.H C 1/
ˇ :
x ˛1 jyjz=.2n/
ˇ
Without any loss of generality, we may assume that jx ˛2 j jx ˛1 j. Then, by (4.35),

we obtain the upper estimate
ˇ x ˛ hK
2
ˇ ˇx ˛
2
ˇ 4hK .H C 1/ 1
ƒ WD ˇ 1ˇ ˇ 1ˇ hK < : (4.36)
ˇ ˇ ˇ ˇ
x ˛1 x ˛1 jyjz=.2n/ jyjz=.4n/
Let f1 ; : : : ; r g be a fundamental system of units in Q.˛1 /. By (4.32) and (4.34),
there exist 1 in K and integers k1 ; : : : ; kr of absolute values less than z such that
x ˛ hK
2 ˇ ı1 1 k1 r kr 1 z
D 1 : (4.37)
x ˛1 ˇ1 ı1 1 r 1
Furthermore, the height of x ˛1 being t z log jyj, we get
h.1 / t;f log jyj: (4.38)
If ƒ D 0, then .˛1 ˛2 /=.x ˛1 / is an algebraic integer and we derive that jxj f 1.

It then follows from (4.31) that z f 1.
If ƒ ¤ 0, then we deduce from Theorem 2.2, (4.33), (4.37), and (4.38) that

log ƒ t;f maxf1; h.1 /g .log z/ t;f . log jyj log z/: (4.39)
Comparing (4.36) and (4.39), we obtain
z t;f log z:
This proves that z is bounded in terms of t and f .X /. It then only remains to apply
Theorem 4.7 to conclude the proof of the theorem when f .X / has an irrational root.
When f .X / has only rational roots, the proof follows the same lines, but is much
simpler. Since f .X / is monic, its roots are integers. Then, there exist rational numbers
d1 ; d2 and integers y1 ; y2 such that
x ˛1 D d1 y1z ; x ˛2 D d2 y2z ;
where numerators and denominators of d1 and d2 are t;f 1. The quantity j.x ˛1 /=.x ˛2 /j
is close to 1 when jxj is large, and we directly apply Theorem 2.2 to bound z from above.
We leave the details to the reader.
4.8. The Catalan equation and the Pillai conjecture

In a letter written in 1844 to August Leopold Crelle, Eugène Catalan [145] addressed
the following question: do there exist consecutive positive integers other than 8 and 9
which are both perfect powers? Said differently, is 32 23 D 1 the only solution to the
exponential Diophantine equation
xm y n D 1; (4.40)
in the four integer unknowns x; y; m; n all greater than 1? This problem has been solved
completely in 2002 by Mihăilescu [71, 302].
Theorem 4.10. The integers 8 and 9 are the only consecutive positive integers which are
both perfect powers.
Even if Mihăilescu managed to remove any use of estimates for linear forms in loga-
rithms in his latest proof of Theorem 4.10, it remains of interest to study how the theory
of linear forms in logarithms can be applied to show that (4.40) has only finitely many
solutions, a result which was established in 1976 by Tijdeman [416]. The reader interested
in a complete proof of Mihăilescu’s theorem may consult [73, 152, 154, 362].
Theorem 4.11. If x; y; m; n are integers at least equal to 2 such that x m y n D 1, then
maxfx; y; m; ng 1.
4.8. The Catalan equation and the Pillai conjecture 63
Proof. By Theorem 4.7 the equations y 2 C 1 D x m and x 2 1 D y n have only finitely

many solutions, which can be effectively bounded. Thus, we may assume that m and n
are odd. It is sufficient to consider the equation
xm y n D ";
where " D ˙1 and x, y, m, n are positive integers with n > m > 2.

Since, for every integer c 2, every odd integer ` 3, and in f 1; 1g we have
!
c` C X
` k 1 `
D ` C .c C / . 1/ .c C /k 1 ;
cC kC1
1k` 1
`
it follows that the greatest common divisor of ccCC
and c C divides `. Thus, there exist
positive integers m ; n ; u; v, with u 2 and v 2, such that
un vm
x "D ; yC"D ;
m n
and m (resp., n ) is a divisor of m (resp., of n). In particular, we get that
un un
xC1 > ;
m n
thus
log x n log u m: (4.41)
It follows from the assumption n > m that x > y. By (4.41), we get
ˇ y n ˇ ˇ yn ˇ ˇ xm " ˇ m
ƒ1 WD ˇ m .m /m 1ˇ D ˇ mn m 1ˇ D ˇ 1 ˇ ;
ˇ ˇ ˇ ˇ ˇ ˇ
u u .m / .x "/ m x
thus
log ƒ1 log x:
Since
y n x m C 1 .un C 1/m C 1 2m unm ;
we get the upper bound y 2um . By Theorem 2.2, we then obtain
log ƒ1 .m log u/ .log m/ .log n/;

thus,
log x m .log m/ .log n/ .log u/:
By the first inequality of (4.41), this gives
n m.log m/ .log n/: (4.42)
We deduce from (4.42) that

vm vm
yC1 2 v m=2 ;
n m
thus, again by (4.42),

p
log y m log v n log v: (4.43)
Consequently,
ˇ u mn ˇ ˇ umn .n /n ˇ ˇ .x "/m ˇ n
ƒ2 WD ˇ .m / m
.n /n 1ˇ D ˇ m mn 1ˇ D ˇ 1ˇ
ˇ ˇ ˇ ˇ ˇ ˇ
v .m / v .y C "/n y
and
log ƒ2 log y:
Then, we deduce from u 3v (whose proof is left to the reader) and Theorem 2.2 that
log ƒ2 .log v/ .log m/ .log n/ .log mn/;

thus,
log y .log mn/ .log m/ .log n/ .log v/:
Hence, by the first inequality of (4.43), we obtain
m .log m/ .log n/2 : (4.44)
Combining inequalities (4.42) and (4.44) we find that n and m are bounded by an absolute
real number. The fact that x and y are also bounded from above follows from general
results on superelliptic Diophantine equations given in Theorem 4.7.
Shortly after the publication of Tijdeman’s proof, Langevin [251] computed explicit
(albeit huge) upper bounds for the solutions of the Catalan equation (4.40); he showed that
y n ; x m < exp exp exp exp 730:
The proof of Theorem 4.11 involves estimates for linear forms in two logarithms and
in three logarithms. Since, regarding the numerical constant, the current state-of-the-art
for linear forms in three logarithms is not as good as in the case of two logarithms, we
cannot hope to get very small upper bounds for the exponents in (4.40). Nevertheless,
Mignotte [297] established that, if p and q are odd prime numbers such that x p y q D 1
has a solution in positive integers x; y, then
minfp; qg < 7:15 1011 :
There are various criteria to ensure that, for a given pair of odd prime numbers .p; q/, the
equation x p y q D 1 has no solution in positive integers x; y. By means of these criteria,
Mignotte [297] proved that minfp; qg > 107 . This lower bound was improved to 3:2 108
by extensive computations made by Grantham and Wheeler (see [46]). However, there
remains a gap and it is not clear whether an alternative proof of Theorem 4.10 can be
obtained along these lines.
4.9. Exercises 65
The Pillai conjecture is a natural generalization of Catalan’s problem.

Conjecture 4.12 (Pillai’s conjecture). Let a, b, and k be non-zero integers. Then, the
Diophantine equation
ax m by n D k; in integers m, n, x, y with m 3, n 2, x 2, and y 2; (4.45)
has only finitely many solutions.

In 1936, at the end of [331], the Indian mathematician Pillai formulated the special
case a D b D 1 of the above conjecture (see also his paper [332]), which is, however,
commonly referred to as “Pillai’s conjecture”. Also, equation (4.45) is usually called the
Pillai equation. We emphasize that there is no triple of relatively prime integers a, b, k
with jabkj 2 for which the corresponding equation (4.45) is known to have only finitely
many solutions. In particular, we still do not know whether the difference between two
consecutive perfect powers tends to infinity.
As a consequence of Theorems 3.10 and 4.9, we get the following theorem, whose
proof is left as an exercise.
Theorem 4.13. The Pillai equation (4.45) has only finitely many solutions when one
among the four variables x, y, m, n is fixed.
4.9. Exercises
Exercise 4.1. Follow the proof of Theorem 4.9 to establish that, if f .X / is an integer
polynomial of degree at least two (resp., three) and without multiple roots and if t is a
non-zero rational number, then every integer solution .x; y; z/ with z 3 (resp., z 2)
to (4.31) satisfies z f maxfh.t /; 1g. [This result is apparently new; it slightly improves
Theorem 2.1 on page 145 of [386].]
Exercise 4.2. Prove Theorem 4.13.
4.10. Notes
F Baker and Stewart [36] established effective irrationality measures for cubic real
numbers.
F Bombieri [84] (see also [86, 87] and his survey [85]) gave an alternative proof of
Theorem 4.6, independent of the theory of linear forms in logarithms; see [107] for a
combination of both methods.
F For specific algebraic numbers ˛, it is sometimes possible to derive a much better
effective irrationality exponent than the one given in Theorem 4.6. The first result of this
type (although a forerunner is implicit in early works of Thue [412, 413]) was obtained
p by Alan Baker [15], who established that 2:955 is an effective irrationality exponent
in 1964
of 3 2. More recent works include [43, 82, 83, 89, 90, 148, 165].
F Solving unit equations, Thue equations, and superelliptic equations is essentially equiv-
alent; see [261]. The theory of linear forms in logarithms has been used by Baker and
Coates [33] and, subsequently, by Schmidt [358] to give effectively computable upper
bounds for the height of integer points on curves of genus one. Effective results for linear
equations in two unknowns from a multiplicative division group have been given in [60];
see also [182]. Effective upper bounds for the size of solutions to further families of
Diophantine equations can be found in the monographs [182, 183, 376, 386] and in the
papers [68–70].
F Let K be a number field. Shafarevich’s theorem asserts that, for a given finite set
of places S of K, there are only finitely many elliptic curves E defined over K with
good reduction outside of S . An effective version of this theorem, whose proof rests
on the theory of linear forms in logarithms, has been given by Fuchs, von Känel, and
Wüstholz [188]. It has been extended to hyperelliptic curves by von Känel [235]; see also
the monograph [183].
F Two monic integer polynomials f .X /; g.X / are called equivalent if g.X / D f .X C a/
for some integer a. Equivalent polynomials have the same discriminant. Using effective
results on unit equations, Győry [209] proved that for any given non-zero integer D,
there are only finitely many inequivalent monic polynomials with discriminant D and
they can be effectively determined. For quantitative versions, generalisations, and various
applications, see e.g. [183, 210].
F The only reference on the equation x 2 C D y n , where C is a fixed positive integer,
seems to be the unpublished PhD thesis of Carlos Barros [41].
F For the algorithmic resolution of Diophantine equations, the reader is directed to the
monographs [153,154,191,382] and Chapter 5 of [182]. Efficient algorithms to solve Thue
equations of high degree and superelliptic equations are given and discussed in [74–76].
The complete resolution of equations of the form f .x/ D y n with n fixed remains rather
difficult when n is not very small; see, however, [127].
F Starting with a seminal paper of Emery Thomas [411], there is an extensive literature
on parametric families of Thue equations; see for example the survey of Heuberger [227].
Hoshi [233] managed to solve completely a parametric family of Thue equations of
degree 12.
F Pethő [322] used an explicit estimate of Sprindžuk [386] for the integer solutions of
equations of the form f .x/ D ty 2 to establish that, for every quadratic number ˛ and
every integer q 2, we have kq 2 ˛k ˛ .log q/0:0005 q 2 .
F An explicit lower bound for the difference between two perfect powers is given in [429].
For more information on the Pillai equation, the reader is referred to the survey [435].
Theorem 4.11 has been extended to number fields by Brindza, Győry, and Tijdeman [97].
F Let F .X; Y / be an irreducible, homogeneous polynomial of degree three with integer
coefficients. Bennett and Dahmen [55] established that the generalized superelliptic equa-
tion F .x; y/ D z ` , in integers x; y; z; ` with gcd.x; y/ D 1, jzj 2, and ` 4, has finitely
many solutions. They also obtained analogous results for polynomials of degree 4, 6,
and 12. Their proof does not use estimates for linear forms in logarithms. It proceeds via
construction of Frey curves corresponding to putative solutions to F .x; y/ D z ` and rests
on the so-called modular method, based upon modularity of Galois representations.
Chapter 5
Further applications
We have pointed out in Chapter 2 that estimates for linear forms in logarithms are con-
siderably better when the algebraic numbers involved are very close to 1. This chapter is
devoted to some spectacular applications of this theoretical refinement.
5.1. Effective irrationality measures for quotients

of logarithms of rational numbers
Theorem 3.3 asserts that the quotient of the logarithms of two positive, multiplicatively
independent rational numbers aa21 ; bb12 is not a Liouville number and, more precisely, that
its irrationality exponent is bounded from above by some absolute real number times the
product of log maxfa1 ; a2 g and log maxfb1 ; b2 g. The next result shows that we get an
absolute upper bound for this irrationality exponent when aa12 and bb12 are very close to 1.
Its proof and a slightly more general statement are given in [116]; see Exercise 5.1.
Theorem 5.1. Let a1 ; a2 ; b1 ; b2 be integers such that
p p p
36 a2 < a1 < a2 C a1 ; 36 b2 < b1 < b2 C b1 ; and b1 < a1 < b12 :
a1 b1
If a2
and b2
are multiplicatively independent, then we have
log aa21

eff 221105: (5.1)
log bb12
Proof. We show only that the irrationality exponent in (5.1) is bounded from above by an
absolute, effectively computable real number. Let yx be a convergent to .log aa21 /=.log bb12 /
with y 3. Set X WD maxfjxj; jyjg. We need to estimate from below the quantity
ˇ log a1 x ˇˇ 1 ˇˇ a1 b1 ˇˇ
a2
ˇ D ˇy log x log
ˇ
log bb21 y y a2 b2
ˇ ˇ:
We apply Theorem 2.1 with the parameter E given by

p p
E WD minf a1 ; b1 g:
68 5. Further applications
Observe that
p
a
1
a1 a2 a1 a2 a1
log D log 1 C ;
a2 a2 a2 a2
thus
log a1 a2 .log a1 / p a1
a1 p a1 e and log a1 E log :
log a2 a1 a2
Likewise, we check that

log b1 p b1
b1
b1 e and log b1 E log :
log b2
b2
It then follows from Theorem 2.1 applied with A1 D a1 ; A2 D b1 ; and E D E that

ˇ a1 b1 ˇˇ
x log ˇ .log a1 /.log b1 /.log X /.log E/ 2 :
ˇ
log ˇy log (5.2)
a2 b2
Consequently, we have proved that
log aa12

.log a1 /.log b1 / .log a1 /.log b1 /
eff 2
1:
b1
log b2 .log E/ minflog a1 ; log b1 g2
This completes the proof of Theorem 5.1, up to the value of the numerical constant.
It is crucial in the proof of Theorem 5.1 that the dependence on X occurs in (5.2)
through the factor log X, and not .log X /2 , which would have been given by Theorem 2.4.
Actually, in [116], we have applied an estimate of Gouillon [202], which improves the
numerical constants of Theorem 2.1 in the case of two logarithms. This explains the
(relatively) small size of the numerical value in (5.1).
5.2. Effective irrationality measures for n-th roots

of rational numbers
By Theorem 3.6, for positive
p integers a; b; n with n 3 and a > b, the effective
irrationality exponent of n a=b satisfies
p
eff . n a=b / .log a/.log n/; (5.3)
which improves Liouville’s bound when n is large enough. If, moreover, a exceeds 2n ,
then the factor .log n/ in (5.3) can be removed.
It was observed in [117] that, as an immediate consequence of Theorem 2.5, the upper
bound (5.3) can be dramatically improved when the rational number ab is very close to 1.
6b
Theorem 5.2. Let a; b; n be integers with n 3 and 1 < b < a 5
. Define in .0; 1
by a b D ba . Then, we have
p 35:2 n log n o2
eff . n a=b / max 1 C ; 10 :
log a
5.3. The Thue equation ax n by n D c 69
p1 and a n2=9 , then we get

Consequently, if b < a < b 1 C a
p
eff . n a=b / 7040:
Proof. We proceed as in the proof of Theorem 3.6. Let a; b; n; be as above and pq a

convergent to n a=b with q 100. Since ab D 1 C a , it follows from Theorem 2.5 that
p
ˇ q n a ˇ 35:2 n log n o2

1ˇ .log p/ max 1 C ; 10 :
ˇ ˇ
logˇ
p b log a
This gives the first assertion of the theorem. The second one follows by choosing D 12 .
Theorem 5.2 should be compared with a result of Bombieri and Mueller p [89], who
established that, if a; b; n are positive integers with b 2 and n 3 such that n a=b is of
degree n, then, defining 0 by
0
ja bj D b 1
and assuming that n > 2=0 , we get
s
p 2 n5 log n
eff . n
a=b /
0
C63 : (5.4)
log b
p from (5.4) that, for any positive real number " and any integer n 3, we
It follows
have eff . n a=b / < 2 C " if b is sufficiently large in terms of n and if ab is sufficiently close
to 1. Such a remarkable result does not follow from Theorem 5.2. However, we stress
that (5.4) improves Liouville’s bound only when b is very large compared to n; namely,
one requires that b satisfies
2
b > n216n ;
while the much weaker assumption b > n2=9 is sufficient to apply Theorem 5.2.
5.3. The Thue equation ax n by n D c

In this section we consider the binary Thue equation
ax n by n D c; (5.5)
where a; b; c; and n are non-zero fixed integers, with a; b positive, n 3, and where x
and y are unknown integers. We show that it has no non-trivial solutions x; y (that is, no
solutions with jxyj 2) when a and b are close to each other, jcj is small, and n is large
enough. We start with a more general result.
Theorem 5.3. Let a, b, and c be non-zero integers with a > b 1 and jcj b. If
n jcj o
n max 2 log ; 15000 log a ;
a b
then the Thue equation

ax n by n D c (5.6)
has no integer solutions x; y with jxyj 2.
Proof. Let x; y be integers with jxyj 2 and satisfying (5.6). If jxj jyj, then
jcj D jax n by n j .a b/jxjn .a b/2n ;
giving n < 2 log.jcj=.a b//.

Thus, we assume that jyj > jxj. Dividing both sides of (5.6) by by n , we get
ˇ a x n ˇ jcj 1
1ˇ :
ˇ ˇ
b y bjyj n jyjn
ˇ
Theorem 2.3 implies that

n n o2
n log jyj 25:4 .log maxfe; ag/ log jyj max log C 0:21; 20 ;
log maxfe; ag
which gives
n < 15000 log a:
We see from the proof of Theorem 5.3 that if a is very close to b, then the rational
numbers ab and yx are both very close to 1. We may thus expect to get a stronger result in
this particular case. This is indeed what happens, as was first proved by Mignotte [296].
We display a special case of a result from [296].
Theorem 5.4. If there exist positive integers n, b, c, x, y with n 3, maxfx; yg > 1,
c 5, and b c 2 , satisfying
j.b C c/x n by n j c;
then n < 7500.

We point out that the upper bound for n in Theorem 5.4 is independent of b and c.
Proof. We follow the proof of Theorem 5.3, but instead of applying Theorem 2.3, we use
Theorem 2.5. Write
b C c x n
ƒ WD 1
b y
and observe that y > x and
log jƒj n log y: (5.7)
Apply Theorem 2.5 with x2 D b C c and y2 D b. Let 0 be defined by
log bc
0 D
log.b C c/
5.4. A generalization of Diophantine quadruples 71
and note that

log b log 25
0 0:47:
2 log.b C c/ 2 log 30
Then, we get
35:2 ˚ 2
log jƒj .log y/ max 1 C log n; 10 :
0:47
n < 7500:
Equation (5.5) with c D 1 has been considered by Bennett and de Weger [59] and,
subsequently, by Bennett [45], who established the following definitive result.
Theorem 5.5. Let a; b; n be positive integers with n 3. Then, the equation
jax n by n j D 1
has at most one solution in positive integers x and y.

Theorem 5.5 gives us the complete list of solutions to the two-parametric family of
Thue equations .b C 1/x n by n D ˙1.
5.4. A generalization of Diophantine quadruples

In the present section, we study a variant of the problem of Diophantus discussed in
Section 3.8, namely the existence of triples fa; b; cg of positive integers such that the three
numbers ab C 1, ac C 1 and bc C 1 are perfect k-th powers, for a fixed integer k 3.
Examples of such triples for k D 3 and k D 4 are given, respectively, by f2; 171; 25326g
and f1352; 9539880; 9768370g. To our knowledge, no example of such a triple is known
for an integer k 5. The next theorem was established in [120].
Theorem 5.6. Let k 3 and 0 < a < b < c < d be integers such that the four numbers
ac C 1; ad C 1; bc C 1; and bd C 1
are perfect k-th powers. Then we have k 176. Consequently, for any integer k 177,
there exist no set of four positive integers such that the product of any two of them increased
by 1 is a perfect k-th power.
Proof. We content ourselves to prove that k is bounded by 80000. Let a; b; c; d be integers

with 0 < a < b < c < d such that there exist positive integers r; s; t; u; and k 2 with
ac C 1 D r k ; ad C 1 D s k ; bc C 1 D t k ; and bd C 1 D uk :
Set
ur b ac C 1
˛1 D ; ˛2 D :
st a bc C 1
Observe that ˛1 > 1 since the product .b a/.d c/ is positive. Furthermore, ˛2 and
˛1 k ˛2 are very close to 1, namely,
b a 1
j˛2 1j D ˛2 1D < (5.8)
abc C a c
and
ˇ k ˇ
ˇ st b ac C 1 ˇ b.ad C 1/ b a 1
ƒ WD ˇˇ 1ˇˇ D 1D < : (5.9)
ur a bc C 1 a.bd C 1/ a.bd C 1/ d
We apply Theorem 2.5 with yx11 D ˛1 and x2

y2
D ˛2 in order to bound ƒ from below.
By (5.8), the parameter satisfies
log.˛2 1/ log c
WD > :
log b.ac C 1/ log b.ac C 1/
It then follows from Theorem 2.5 that

log b.ac C 1/ ˚ 2
log ƒ 35:2 .log ur/ max 1 C log k; 10 : (5.10)
log c
Using that b.ac C 1/ c 3 , we derive from (5.9) and (5.10) that

˚ 2
k log d 105:6 k.log ur/ max 1 C log k; 10 :
Since
k.log ur/ log.2abcd / 5 log d;

we obtain
˚ 2
k 528 max 1 C log k; 10 ;
giving an absolute upper bound for k, namely k 80000.
5.5. Exercises
Exercise 5.1 (see [116]). Let " be a real number with 0 < " < 1. Let a1 ; a2 ; b1 ; b2 be
integers such that
2 a2 < a1 < a2 C a11 " ; 2 b2 < b1 < b2 C b11 " ; and b1" < a1 < b11=" :
Prove that
log aa12

4
eff b1
" :
log b2
Exercise 5.2 (see [47, 112]). Prove that all the solutions to the Diophantine equation
.x k 1/.y k 1/ D .z k 1/, in positive integers x; y; z; k with z 2, satisfy k 1.
5.6. Notes 73
Exercise 5.3 (see [114]). Let be a positive real number. Let ˛; ˇ be real algebraic
integers such that ˛ C ˇ, ˛ˇ are rational integers and ˛ jˇj1C . Prove that all the
solutions in positive integers n; q; y of
˛n ˇn
D yq ;
˛ ˇ
3
with q prime, n congruent to 1 modulo q, and jyj 2, satisfy q .
5.6. Notes
F The first applications of Shorey’s idea [367, 368] appeared in [366]. Subsequently,
Ramachandra, Shorey, and Tijdeman [341, 342] applied linear forms in logarithms of
algebraic numbers close to 1 to investigate Grimm’s conjecture on factorisation of blocks
of consecutive integers (see Problem 13.21).
F Further applications of estimates for linear forms in logarithms of algebraic numbers
close to 1 can be found in the survey [114].
F In various Diophantine problems, linear forms in logarithms of algebraic numbers
close to 1 are used to give an explicit upper bound for an unknown exponent. When this
exponent is assumed to be fixed, it is often possible to use the hypergeometric method to
show the finiteness of the number of solutions to the corresponding equation. This was
done e.g. in [120], where the authors established, for any integer k with 11 k 176,
the finiteness of the set of quadruples fa; b; c; d g of positive integers such that the product
of any two of them increased by 1 is a k-th power. Other papers combining estimates for
linear forms in logarithms of algebraic numbers close to 1 and the hypergeometric method
include [112, 136, 140, 231, 350].
F Theorem 5.6 suggests to consider sets fa1 ; : : : ; am g of positive integers such that
ai aj C 1 is a perfect power for all i; j with 1 i < j m; see [123, 207, 208, 275, 397].
Estimates for linear forms in four logarithms were used by Luca in [275]. The main result
of [208] was obtained by a modification of Luca’s argument allowing the authors to apply
linear forms in only two logarithms. Finally, Stewart [397] managed to use estimates for
simultaneous linear forms in logarithms (see Chapter 9) to prove that the cardinality of a
subset A of f1; 2; : : : ; ng such that ab C 1 is a perfect power for every distinct a; b in A
cannot exceed some absolute constant times .log n/2=3 .log log n/1=3 .
Chapter 6
Applications of linear forms in p-adic logarithms
In the present chapter, we emphasize several applications of estimates for linear forms
in p-adic logarithms, often in combination with estimates for linear forms in complex
logarithms. Among other results, we extend the effective estimates for the size of the
solutions of classical families of Diophantine equations obtained in Chapter 4: we give
effective upper bounds not only for the integer solutions, but also for the solutions in
rational numbers, whose denominators are composed of prime numbers from a given,
finite set.
6.1. On the p-adic distance between two integral S -units

The following result can be viewed as a p-adic analogue to Theorem 3.4. It shows that
two distinct integral S -units cannot be p-adically too close.
Theorem 6.1. Let S denote a finite, non-empty set of prime numbers and .xj /j 1 the
increasing sequence of all positive integers whose prime factors belong to S . Let p be a
prime number not in S . There exists a real number c, which can be explicitly computed
in terms of the prime numbers in S , such that
cp=.log p/
jxj xh jp .log xj / ; 1 h < j: (6.1)
We point out that the dependence on p in Theorem 6.1 is rather weak. It is very
likely that the exponent of .log xj / in (6.1) should be independent of p. However,
this exponent cannot be smaller than 1. To see this, let J be a large integer and set
` WD d.log J /=.log p/e 1. By the Dirichlet Schubfachprinzip, there exist integers h; k
such that 1 h < k J and xh and xk are in the same congruence class modulo p ` .
Since .xj /j 1 grows at most exponentially fast, there exists a real number , which can
be explicitly computed in terms of the prime numbers in S , such that
` 1 1
jxh xk jp p pJ p .log xJ / :
Proof of Theorem 6.1. Write S D fq1 ; : : : ; qs g, where q1 ; : : : ; qs are prime numbers with
q1 < < qs . Let u; v with 2 u < v be two elements of the sequence .xj /j 1 and
write
s s
qiui ; v D qivi :
Y Y
uD
i D1 i D1
76 6. Applications of linear forms in p-adic logarithms
Without any loss of generality, we can assume that u and v are coprime, thus, for i D
1; : : : ; s, at least one of ui ; vi is zero. Set
s
v
qivi ui
Y
ƒ WD 1D 1:
u
iD1
For i D 1; : : : ; s, put bi D vi ui . Set B D maxf3; jb1 j; : : : ; jbs jg and observe that
B 3 log v: (6.2)
By Theorem 2.9 there exists an effectively computable, absolute, real number C such that
s
s
Y p
vp .ƒ/ D vp .v u/ C log qi .log B/:
.log p/2
i D1

s
Y p
log jv ujp Cs log qi log.3 log v/:
log p
iD1
Taking the exponential of both sides, we get the assertion of the theorem.
6.2. Waring’s problem

Let n 2 be an integer. Hilbert [228] proved that there exists an integer s.n/ such that
every positive integer can be expressed as the sum of at most s.n/ positive integers, all of
which being n-th powers. Let g.n/ denote the smallest integer with this property. Observe
that the integer 2n b.3=2/n c 1 is less than 3n , thus it can only be represented by using
the n-th powers 1n and 2n . Since
2n b.3=2/n c 1 D .b.3=2/n c 1/2n C .2n 1/1n ;
we deduce the lower bound
g.n/ 2n C b.3=2/n c 2: (6.3)
It is known (see e.g. Chapter XXI of [220] for a review of such results) that equality holds
in (6.3) when
2n f.3=2/n g C b.3=2/n c 2n : (6.4)
Furthermore, if (6.4) fails, then
g.n/ D 2n C b.3=2/n c C b.4=3/n c ;
where is equal to 2 or 3, according as the integer
b.4=3/n c b.3=2/n c C b.4=3/n c C b.3=2/n c

6.3. On the b-ary expansion of an algebraic number 77
is equal to 2n or is larger than 2n . A quick check shows that, to prove that (6.4) holds for
the integer n, it is sufficient to establish that
k.3=2/n k .3=4/n 1
: (6.5)
The theory of linear forms in p-adic logarithms allows us to get an effective improvement
of the trivial estimate k.3=2/n k 2 n . This was first proved by Baker and Coates [34].
Theorem 6.2. Let a and b be coprime integers with a > b 2. Then, there exists an
effectively computable positive real number ı, depending only on a and b, such that
a n 1
.1 ı/n ; for n 1:

b

2b
Proof. Let n be a positive integer and An denote the nearest integer to .a=b/n . Set
mn WD an An b n : (6.6)
Let p be the smallest prime divisor of b. It follows from Theorems 2.8 (when an and mn
are multiplicatively dependent) and 2.12 that
n 2
n vp .an mn / a;b .log j2mn j/ log ;
log j2mn j
giving that
n a;b log j2mn j:
This implies the existence of a positive real number ı, depending only on a and b, such
that
2jmn j b ın :
Combined with (6.6) we conclude that
a n jmn j 1
D n .1 :

b b ı/n

2b
According to [111], in the case ab D 32 , Theorem 6.2 holds with ı D 0:0005, thus we
are very far from proving (6.5) by means of the theory of linear forms in p-adic logarithms.
By using a totally different approach, Beukers [65] proved that k.3=2/n k 2 9n=10
for n > 5000. This estimate has been subsequently refined by several authors, including
Zudilin [453], who established that k.3=2/n k 0:5803n for every n greater than an
effectively computable number. Note that Kubina and Wunderlich [246] have checked that
Inequality (6.4) holds for n up to 471; 600; 000.
6.3. On the b-ary expansion of an algebraic number

Throughout this section, we denote by b an integer at least equal to 2. Let be a real
algebraic number of degree d at least 2. It follows from Liouville’s Theorem A.5 that
d C1
kqk q ; for q 1: (6.7)
When d 3, we have established in Theorem 4.6 a slight improvement upon (6.7),

namely that there exists an effectively computable positive number ./, depending only
on , such that
kqk q d C1C ./ ; for q 1:
Taking q D b n , we get
kb n k b .d 1 .//n
; for n 1: (6.8)
It is then natural to ask whether (6.8) also holds when is of degree two. In this case,
the first effective lower bound for kb n k was given by Schinzel [354] in a landmark
paper published in 1967, in which he developed the theory of linear forms in two p-adic
logarithms, making completely explicit the bounds of Gelfond [199, 200]. He established
that, for every integer b 2 and every quadratic real number , we have
kb n k > b n exp c.; b/n1=7 ; for n 1;

where c.; b/ is a positive, effectively computable, real number depending only on

and b. By following his approach and using Theorem 2.12, Bennett and Bugeaud [51]
extended (6.8) to quadratic real numbers.
Theorem 6.3. For every integer b 2 and every quadratic real number , there exists
a positive, effectively computable, real number .; b/, depending only on and b, such
that
kb n k ;b b .1 .;b//n ; for n 1:
Unlike in (6.8), the real number .; b/ in Theorem 6.3 depends on the base b.
It follows from Ridout’s Theorem A.8 that, for every integer b 2 and every positive ",
every irrational algebraic number satisfies
kb n k ineff
;b;" b
"n
:
The implicit numerical constant is, however, not effectively computable.

Proof of Theorem 6.3. We follow Schinzel’s p argument and the proof of Theorem 1.2
p n a, where a 2 is a non-square integer.
of [51]. It is sufficient to treat the case D
Let x; n be positive integers with jx ab j 1. Write
p p n
x 2 ab 2n D .x C ab n / .x ab / DW :
p
Let denote the fundamental unit (see Appendix C) of the quadratic field generated by a.
Write p n
x ab D ım ;
p
where ı in Q. a/ and the integer m are such that
jj1=2 1=2
< jıj jj1=2 1=2 ; ıı D : (6.9)
Here and below, ı denotes the Galois conjugate of ı. Observe that m 0 and
p n 2 p
.x 2 ab 2n / .x ab / D ı 2 2m D 2. ab n x ab 2n /: (6.10)
6.3. On the b-ary expansion of an algebraic number 79
Let us factor b D q1`1 qs`s as a product of powers of distinct prime numbers.

Choose j so that `j D maxf`1 ; : : : ; `s g, and put p D qj and ` D `j . Set M WD maxf2; jmjg.
Since jı j 1=2 and
p p
jı m
j D jx C ab n j 1 C 2 ab n a b n ;
we get
M a n log b: (6.11)
If m D 0, then we immediately derive that n a;b log j2j. If m < 0 and jj D 1,
then ı D ˙1 and we deduce from (6.10), (6.11), and Theorem 2.8 that
n n` vp .2m ˙ 1/ vp .4m 1/ a;p log M a;b log n;
which gives an effective upper bound for n, depending only on a and b.

We assume now that jj 2 and m < 0. We wish to bound
ı
vp . ı 2 2m / D vp 2m C vp .ı 2 /
ı
from above. It follows from (6.9) that 1 jı=ı j . Since ı=ı is an algebraic
integer, the height of ı =ı is a log jj. By Theorem 2.12, we get
ı M 2
vp 2m a;p .log jj/ log 2 C : (6.12)
ı log jj
On the other hand, we derive from (6.10) that

ı
vp 2m n` vp .ı 2 /;
ı
while vp .ı/ a;b log jj, by (6.9). Combined with (6.11) and (6.12), this gives
n.log b/ 2
n n` a;p .log jj/ log 2 C C log jj
log jj
n.log b/ 2
a;p .log jj/ log 2 C ;
log jj
whence
n a;b log jj:
p n
Taking for x the nearest integer to ab , all this proves the existence of an effectively
computable, positive, real number , depending only on a and b, such that
p p n jj b n
kb n ak D jx ab j p n p n :
2 ab C 1 2 ab C 1
This concludes the proof of the theorem.

6.4. Repunits and perfect powers

We continue this chapter with a special case of an equation already discussed in Corol-
lary 3.11. In the proof of the next theorem, the use of estimates for linear forms in p-adic
logarithms yields better numerical upper bounds than the use of estimates for linear forms
in complex logarithms. A repunit is an integer all of whose decimal digits are 1.
Theorem 6.4. The Diophantine equation
10n 1
D yq (6.13)
10 1
has no solution in integers n; y; q with n 2 and q 2. Said differently, no repunit
greater than 1 is a perfect power.
In 1976, Shorey and Tijdeman [375] established that (6.13) has only finitely many
solutions and solved it for q < 23. Its full resolution was completed in 1999 in [132];
see also [133, 135]. We point out that (6.13) has a solution for every q 2 given
by 101 D 9 1q C 1, which makes its resolution a difficult problem (a naïve sieve for
fixed q does not yield any contradiction). There are much similarities between the proofs
of Theorem 6.4 and Theorem 3.12.
Outline of the proof of Theorem 6.4. Rewriting (6.13) in the form
9y q 10 n
1D 10 n
;
we can obtain an upper bound for q by using estimates for linear forms in three complex
logarithms; see Section 3.7 for details. Here, we rather use the fact that the binary
m
recurrence sequence . 10 9 1 /m0 has a dominant root for the 5-adic absolute value in
order to obtain a much sharper bound for q, which follows from an estimate for linear
forms in two 5-adic logarithms. Namely, we consider the 5-adic valuation of
10n D 9y q C 1:
On the one hand, it is trivially equal to n. On the other hand, Theorem 2.12 allows us to
bound v5 .9y q C 1/ from above.
Observe that 3h C 1 is congruent to 2 or 4 modulo 8 for h 1. Thus, if we have
10 D 9y q C 1 with y being a power of 3, then n must be at most equal to 2 and we
n
conclude that the only solution is given by y D 1. Hence, we may assume that 9 and y
are multiplicatively independent and we derive from Theorem 2.12 that
n D v5 .9y q C 1/ .log y/.log q/2 ;
where the constant implied by is reasonably small. Combined with the obvious bound
y q 10n , we derive that
q log y .log y/.log q/2 ;
and we get at once an upper bound for q. This upper bound is much smaller than the one
which can be derived from estimates for three complex logarithms. By using the full force
of the main result of [129], the authors of [132] proved that q 2063; thus, since q may
6.5. Perfect powers with few binary digits 81
clearly be assumed to be prime, they replaced Equation (6.13) in three unknowns by about
three hundred equations in only two unknowns.
Therefore, we may assume that q is a fixed prime number in (6.13). Taking into
account the result of Shorey and Tijdeman [376] mentioned above, we use congruences
for each value of the prime number q in the range 23 q 2063. The idea goes as
follows. Consider a prime number ` of the form ` D hq C 1 for some even integer h.
Then, .y q /h is congruent to 0 or to 1 modulo `, giving us that y q can take at most h C 1
different values modulo `. Since 10n D 9y q C 1, we deduce that n takes at most h C 1
different values modulo ` 1, hence at most h C 1 different values modulo q, since q
divides ` 1. Combining the information given by a few different prime numbers `
congruent to 1 modulo q, we conclude after some computation that n must be congruent
to 1 modulo q.
Hence, there exists a positive integer such that n D q C 1 and we get the equality
10 .10 /q 9y q D 1:
This means that X D 10 ; Y D y is a solution of the Thue equation
10X q 9Y q D 1;
which has already been considered in Section 5.3. By Theorem 5.5, this equation has
no solution with Y 2. We conclude that y D 1, D 0, and n D 1, which has been
excluded. This completes the outline of the proof of the theorem.
6.5. Perfect powers with few binary digits

Theorem 3.16 implies that a large integral power of 3 cannot have too few digits 1 in its
binary representation. In this section, we study a more general question and ask whether
an arbitrarily large perfect power can have only a bounded number of digits 1 in its binary
representation. For a given integer x 2, equations of the form
xa C xb C xc C 1 D yq ;
where a > b > c > 0, q 2, y 2 are unknown integers, have been considered
in an handful of papers, including [48, 49, 52–54, 159, 274, 409]. We reproduce a result
established in [53].
Theorem 6.5. If the integers a; b; y; q satisfy a > b 1, q 2, and
2a C 2b C 1 D y q ; (6.14)
then .a; b; y; q/ D .5; 4; 7; 2/; .9; 4; 23; 2/; .6; 4; 3; 4/; or .a; b; y; q/ D .2t; t C 1;
2t C 1; 2/, for some integer t 2. If there exist positive integers a; b; c; y; and q
such that a > b > c 1 and
2a C 2b C 2c C 1 D y q ; (6.15)
then q 4.
It is clear from the proof of Theorem 6.5 that the C1 in (6.14) and (6.15) is crucial:
we do not know how to prove, for instance, whether or not the equation 2a C 2b C 7 D y q
has no solution when q is sufficiently large.
A crucial ingredient in the proof of the second statement of the theorem is the refined
estimate for p-adic linear forms in logarithms, given in Theorem 2.13, in the special case
where the rational numbers involved are p-adically very close to 1.
Outline of the proof of Theorem 6.5. We content ourselves to explain how to derive an
effectively computable upper bound for q, assuming that q is an odd prime. We do not
discuss the final sieve, performed for fixed values of q. We first show that, when q is
sufficiently large, equations (6.14) and (6.15) have no solutions satisfying a 2b.
Let .a; b; y; q/ (resp., .a; b; c; y; q/) be a solution of (6.14) (resp., of (6.15)) with
a > b 1 (resp., a > b > c 1). Observe that
a q.log y/ a: (6.16)
Assume that a 2b. Then, we have
jy q 2 a
1j < 22 .a b/
22 a=2
: (6.17)
Since y is odd, the integers 2 and y are multiplicatively independent and (6.16) and
Theorem 2.3 then give that
a 2
log jy q 2 a
1j .log y/ log q C
log y
.log q/2
.log y/ .log q/2 a :
q
Combined with (6.17), we immediately obtain the upper bound
q .log q/2 ;
and we conclude that q 1 for every solution .a; b; y; q/ (resp., .a; b; c; y; q/) of (6.14)
(resp., of (6.15)) with a > b 1 (resp., a > b > c 1) and a 2b.
Assume that 2b > a.
First, consider Equation (6.14). It is immediate that 2b divides y q 1. Since y and q
are odd, 1 C y C C y q 1 is also odd, hence 2b divides y 1 and y 2b C 1. Thus,
we get
2a C 2b C 1 .2b C 1/q > 2bq C 2b C 1
and derive the inequality
2b bq < a:
This shows that (6.14) has no solution with 2b > a for any odd prime number q.
Secondly, consider Equation (6.15). The same argument implies that 2c divides y 1.
Note that
b D v2 .2a C 2b / D v2 .y q .2c C 1//: (6.18)
6.6. The S -unit equation 83
If 2c C 1 and y are multiplicatively dependent, then, after a little work, we find that
either 2a C 2b C 9 D 3q ; or 2a C 2b C C D C q ;
where C D 2c C 1 and c ¤ 3. In the first case, arguing modulo 4 contradicts our

assumption that q is odd. In the second case, we write q 1 D 2k Q where Q is odd,
k
so that v2 .C 2 1/ D b. By expanding via the binomial theorem, it is easy to see that
2k
v2 .C 1/ c C k C 2, whereby b c C k C 2. Since 2a C 2b C 2c C 1 D .2c C 1/q
implies that a cq, we thus have
a log q a log a
b cCkC2 C C2 C C 2:
q log 2 q log 2
Since 2b > a, we deduce that q 1.
Thus, we assume that 2c C 1 and y are multiplicatively independent. If c D 1 or 2,
then we derive from Theorem 2.12 upper bounds for v2 .y q 3/ and v2 .y q 5/, which,
by (6.18), give in both cases
b .log y/ .log q/2 :
Since b > a2 , it then follows from (6.16) that q 1. If c 3, then we apply Theorem 2.13
with the parameter E D v2 ..2c C 1/ 1/ D c and, recalling that v2 .y 1/ c, we
get y > 2c and
.log.2c C 1// .log y/
v2 .y q .2c C 1// maxflog q; cg2 .log y/ .log q/2 :
c3
Combined with (6.16), (6.18), and b > a2 , this gives q 1. This shows that (6.15) has
no solutions with 2b a when q is sufficiently large. We have established that every
solution to (6.14) or (6.15) satisfies q 1.
6.6. The S -unit equation

Let K be an algebraic number field and S be a finite set of places on K containing the
set of infinite places. An algebraic number x in K is an S -unit if, by definition, jxjv D 1
for every place v not in S . The S -unit equation is a natural extension of the unit equation
studied in Section 4.2.
Theorem 6.6. Let K be an algebraic number field of degree d and discriminant DK .
Let S be a finite set of places on K containing the infinite places. Let ˛, ˇ be non-zero
elements of K. The equation
˛x C ˇy D 1; in S -units x; y in K, (6.19)
has only finitely many solutions, and all of them satisfy
maxfh.x/; h.y/g d;DK ;S maxfh.˛/; h.ˇ/; 1g: (6.20)
When S is the set of infinite places, then (6.19) is an (ordinary) unit equation. The first
explicit version of Theorem 6.6 was established by Győry [211]. For more recent explicit
versions, generalisations, and applications, see [124, 182–184, 217].
Proof. Let x, y be a solution to (6.19). We assume that h.x/ h.y/. Let s be the
cardinality of S . The case s D 1 being trivial, we assume that s 2. Let f1 ; : : : ; s 1 g
be a fundamental system of S -units in K satisfying the properties specified in Lemma C.4.
Then we can write
y D b11 : : : bs s 11 ; (6.21)
where is a root of˚unity in K and b1 ; : : : ; bs 1 are rational integers.
Put B D max jb1 j; : : : ; jbs 1 j; 3 and let v1 ; : : : ; vs denote non-equivalent places
in S . Then (6.21) implies
s 1
X
log jyjvj D bi log ji jvj ; j D 1; : : : ; s 1:
i D1
Using Cramer’s rule and the definition of the height, we get from (6.21) and Lemma C.4
that
B d;DK ;S h.y/ d;DK ;S B: (6.22)
Let v be a place in S for which jxjv is minimal. It then follows from Definition B.4
that 1
h.x/ D h log jxjv : (6.23)
x
Assume, without any loss of generality, that h.x/ 2dh.˛/. Then, we deduce from (6.22)
and (6.23) that
h.x/ h.y/
log j˛xjv dh.˛/ h.x/ d;DK ;S . B/: (6.24)
2 2
Setting ˛s D ˇ and bs D 1, we get from (6.19) that
j˛xjv D jb11 : : : bs s 11 ˛sbs 1jv : (6.25)
We shall derive a lower bound for j˛xjv . Set H WD exp.maxfh.˛/; h.ˇ/; 1g/.
First assume that the place v is infinite. We can proceed as in the proof of Theorem 4.3,
but we rather apply Theorem 2.2 to (6.25). Combined with Lemma C.4, we derive that
B
log j˛xjv d;DK ;S log H log : (6.26)
log H
Hence it follows from (6.26) and (6.24) that
B B
d;DK ;S log :
log H log H
This gives
B d;DK ;S log H:
Next, assume that the place v is finite. Let p be the prime number below v. We apply
the last assertion of Theorem 2.11 with n D s in the case bs D 1 to get that either
B d;DK ;S 1
6.7. The Thue–Mahler equation and other classical equations 85
or
log jb11 : : : bs s 11 ˛sbs vp .b11 : : : sbs 11 ˛sbs 1/

1jv d;p
B
d;DK ;S log H log :
log H
By (6.24) and (6.25), we have in both cases
B d;DK ;S log H: (6.27)
Since
h.x/ h.y/ C 2 log H C 2 log 2;
we deduce from (6.22) and (6.27) that (6.20) holds. It then follows from Northcott’s
Theorem B.7 that (6.19) has only finitely many solutions.
6.7. The Thue–Mahler equation and other classical equations

In 1933 Mahler [279] established the finiteness of the number of coprime rational solutions
to the Thue equation, whose denominators are composed only of prime numbers from
a given, finite set. The theory of linear forms in logarithms allows us to get effective
upper bounds for the size of these solutions and to prove the following result, which was
established under a weaker form by Coates [150].
Theorem 6.7. Let s be a positive integer and q1 ; : : : ; qs distinct prime numbers. Let
F .X; Y / be an homogeneous polynomial with integer coefficients such that F .X; 1/ has
at least three distinct roots. Let m be a non-zero integer. The Diophantine equation
F .x; y/ D mq1z1 : : : qszs ; in integers x; y; z1 ; : : : ; zs (6.28)

with gcd.x; y/ D 1 and z1 ; : : : ; zs 0,
has only finitely many solutions, and all of these can be effectively determined. Moreover,
there exists an explicitely computable real number C , depending only on F; q1 ; : : : ; qs ,
such that every solution of (6.28) satisfies
maxfjxj; jyj; q1z1 : : : qszs g j2mjC :
Equation (6.28) is called the Thue–Mahler equation. For explicit versions, further ref-
erences, applications, and generalisations to norm form and decomposable form equations,
see e.g. [125, 212, 217] and [182, 183].
There are similar p-adic extensions of the superelliptic equation, the hyperelliptic
equation, and further classical equations; see e.g. [61, 101, 376, 386]. We are able to
establish effective upper bounds for the size of the solutions to these equations in rational
numbers whose denominators are composed only of prime numbers from a given, finite
set. We state below extensions of Theorems 4.7 and 4.9, established (under a slightly
different formulation and over algebraic number fields) in [376].
Theorem 6.8. Let S be a finite, non-empty set of prime numbers. Let f .X / be an integer
polynomial of degree m without multiple roots, and let n be an integer. Assume that
m 2, n 2, and mn 6. Let t be a non-zero integer. If the rational numbers x; y
satisfy
f .x/ D ty n
and if the denominator of x is composed only of prime numbers from S , then
maxfh.x/; h.y/g f;t;n;S 1:
Theorem 6.9. Let S be a finite, non-empty set of prime numbers. Let f .X; Z/ be a
homogeneous, integer polynomial of degree n 2 without multiple roots. Set m0 D 3
if n D 2 and m0 D 2 otherwise. If there exist integers x; y; z; t; m with m m0 , jyj 2,
gcd.x; z/ D 1, such that t z is composed only of prime numbers from S and
f .x; z/ D ty m ;
then maxfjxj; jyj; jtj; jzj; mg f;S 1.

Lower bounds for the greatest prime factor of quantities of the form ax m C by n follow
from explicit versions of Theorems 6.8 and 6.9.
˚ such that m 2, n 2, and mn 6.
Theorem 6.10. Let a; b; m; n be non-zero integers
For any coprime integers x; y, setting X WD max jxj; jyj; 16 , we have
P Œax m C by n a;b;m;n log log X;
and, in the special case m D n,

log log log X
P Œax n C by n a;b;n log log X :
log log log log X
Theorem 6.11. Let a, b, x, y; and n be integers satisfying
ab ¤ 0; jxj > 1; y ¤ 0; gcd.ax; by/ D 1; and n 2:
Then, for any integer m 2, we have
P Œax m C by n a;b;n log m:
Theorems 6.10 and 6.11, along with more general statements, have been established
in [106, 108, 110, 122, 217, 221].
6.8. Perfect powers in binary recurrence sequences

In Section 3.7 we established that, under the dominant root condition, a recurrence se-
quence of integers does not contain any integer of the form y q , where jyj 2 and q is a
sufficiently large prime number. In the case of binary recurrence sequences, the dominant
root condition can be removed. The next result was proved independently by Pethő [323]
and Shorey and Stewart [372].
6.8. Perfect powers in binary recurrence sequences 87
Theorem 6.12. Let u D .un /n0 be a non-degenerate binary recurrence sequence of

integers. Then, the equation
un D y q ; (6.29)
in integers n; y; q with jyj 2 and q a prime number, implies that
maxfn; jyj; qg u 1:
Proof. First, we establish that q is bounded. Write

u0 ˇ u1 u1 u0 ˛
un D a˛ n C bˇ n ; n 0; where a D ; bD ;
ˇ ˛ ˇ ˛
and ˛; ˇ are roots of a monic, integer quadratic polynomial. If ˛ and ˇ are real numbers,
then either j˛j > jˇj, or jˇj > j˛j and we apply Theorem 3.10 to conclude that q u 1.
Thus, we may assume that ˛ and ˇ are complex non-real. In particular, j˛j D jˇj and ˛
is the Galois conjugate of ˇ.
Observe that ˛=ˇ and ˇ=˛ are conjugate algebraic numbers of degree two and of
absolute value one. Since ˛=ˇ is not a root of unity and any unit in Q.˛/ is a root of
unity, neither ˛=ˇ nor ˇ=˛ are algebraic integers. Thus, there exists a prime number p,
depending only on u, such that vp .˛=ˇ/ or vp .ˇ=˛/ is positive. Without any loss of
generality, we assume that vp .˛=ˇ/ is positive. Let n be a large integer, q a prime number
greater than p, and y an integer such that jyj 2 and (6.29) holds. Then, we have
a ˛
vp .y q b 1 ˇ n 1/ D vp C nvp (6.30)
b ˇ
and
q log y u n: (6.31)
Since vp .q/ D 0, we apply Theorem 2.11 with n D 3, B3 D b3 D q, and ı D 12 , to get

that n no
vp .y q b 1 ˇ n 1/ u max .log y/.log q/; :
q
Combined with (6.30) this gives
n no
n u max .log y/.log q/; : (6.32)
q
It then follows from (6.31) and (6.32) that
q u 1:
It remains for us to prove that, for any fixed value of q at least 2, Equation (6.29) has
only finitely many solutions. Write
vn D .˛ ˇ/ .a˛ n bˇ n /; n 0:
Clearly, we have
.˛ ˇ/2 u2n vn2 D 4ab.˛ ˇ/2 .˛ˇ/n :
Observe that .˛ ˇ/2 , ˛ˇ, and 4ab.˛ ˇ/2 are rational integers. If (6.29) holds, then
there exists an integer r with 0 r < 2q such that the hyperelliptic equation
.˛ ˇ/2 X 2q 4ab.˛ ˇ/2 .˛ˇ/r D Y 2
has a solution in rational numbers X; Y , with X being an S -unit, where S is the set of
prime divisors of ˛ˇ. Since the polynomial .˛ ˇ/2 X 2q C has 2q distinct complex
roots for any non-zero integer C , it follows from Theorem 6.8 that all the solutions of each
of these 2q hyperelliptic equations satisfy maxfh.X /; h.Y /g u 1. This completes the
proof of the theorem.
6.9. Perfect powers as sum of two integral S -units

Let S denote a finite, non-empty set of prime numbers. We establish below that the sum
of two integral S -units cannot be an arbitrarily large power of an integer greater than or
equal to 2.
Theorem 6.13. Let S be a finite, non-empty set of prime numbers. If there exist integers
x; y; z; ` with ` prime, x; y coprime and composed only of prime numbers from S , jzj 2,
and such that
x C y D z`;
then
maxfjxj; jyj; jzj; `g S 1:
Proof. This is (a particular case) of Theorem 9.2 of [376]; see also [50]. Let q1 ; : : : ; qs
denote the elements of S . Assume there exist non-negative integers u1 ; : : : ; us ; v1 ; : : : ; vs ,
a prime number `, and an integer z with jzj 2, such that
q1u1 : : : qsus ˙ q1v1 : : : qsvs D z ` (6.33)
and minfuj ; vj g D 0 for j D 1; : : : ; s. Set M WD maxfu1 ; : : : ; us ; v1 ; : : : ; vs g. Assume
that ` exceeds q1 ; : : : ; qs . Since ` is prime, we have vqj .`/ D 0 for j D 1; : : : ; s.
Assume that u1 1. By applying Theorem 2.11 with
˛n D z; bn D `; f˛1 ; : : : ; ˛n 1 g D fqj W vj 1; 1 j sg; p D q1 ; Bn D `;
1
and ı D 2
, we get
n Mo
u1 S max .log jzj/.log `/; :
`
Arguing similarly to bound u2 ; : : : ; us ; v1 ; : : : ; vs , we obtain
n Mo
M D maxfu1 ; : : : ; us ; v1 ; : : : ; vs g S max .log jzj/.log `/; ;
`
and, consequently,
` S 1 or M S .log jzj/.log `/: (6.34)
However, we deduce from (6.33) that
` log jzj S M:
We conclude from (6.34) that ` S log ` holds in both cases, thus ` S 1. The theorem
then follows from Theorem 6.7. We leave the details to the reader.
6.10. On the digital representation of integral S -units 89
6.10. On the digital representation of integral S -units

Let S be a finite set of prime numbers and b; k integers at least equal to 2. The next result,
proved by Bugeaud and Kaneko [128], shows that every sufficiently large integer which is
divisible only by prime numbers in S , and is not a multiple of b, has more than k non-zero
digits in its representation in base b.
Theorem 6.14. Let b 2 be an integer. Let S be a finite, non-empty set of prime numbers.
Then, there exist effectively computable positive integers n0 and C , depending only on b
and S , such that any integral S -unit n n0 which is not divisible by b has more than
log log n
C C log log log n
non-zero digits in its representation in base b.

Let a 2; b 2 be coprime integers. By taking for S the set of prime divisors of a,
Theorem 6.14 allows us to recover a consequence of Theorem 3.16 in the particular case
where m is a power of a (for the case where a and b are multiplicatively independent
and not coprime, the proof of Theorem 6.14 can be easily adapted) and both proofs are
different.
Proof. Let N > b be an integral S -unit and k the number of non-zero digits in its
representation in base b. We assume that b does not divide N , thus k 2 and we write
N DW dk b nk C C d2 b n2 C d1 b n1 ;
where
nk > > n2 > n1 D 0; d1 ; : : : ; dk 2 f1; : : : ; b 1g:
Let s be the cardinality of S and q1 ; : : : ; qs its elements written in increasing order. There
exist non-negative integers r1 ; : : : ; rs such that
N D q1r1 qsrs :
Observe that
b nk N < b nk C1 : (6.35)
First we assume that nk 2nk 1. This covers the case k D 2. Since
s
ˇ Y ˇ k
X1
ƒa WD ˇ qiri dk 1 b nk
1ˇ D dk 1 b nk
dh b nh b 1Cnk 1 nk
b .nk 2/=2
;
ˇ ˇ
i D1 hD1
we get n
k
log ƒa 1 log b: (6.36)
2
By using the inequality
rj log qj .nk C 1/ log b; for j D 1; : : : ; s; (6.37)

we deduce from Theorem 2.2 that
log ƒa b;S . log nk /: (6.38)
Combining (6.35), (6.36), and (6.38), we obtain
N b;S 1:
Now, we assume that nk < 2nk 1 . In particular, we have k 3. If there exists an

integer j with 1 j k 3 and n1Cj nj=.k
k
2/
, then put
3; n1Cj nj=.k 2/
˚
` WD min j W 1 j k k
:
Otherwise, set ` WD k 2. We see that, by construction,
1 `=.k 2/
n`C1 n and n` nk.` 1/=.k 2/
: (6.39)
2 k
Let p be the smallest prime divisor of b. Put
s ` k `
Y X 1 X X 1
ƒu WD qiri dh b nh
1D dh b nh
dh b nh :
iD1 hD1 hD`C1 hD1
We get by (6.39) and (6.35) that
log b 1Cn` 1
vp .ƒu / n`C1 n`=.k 2/
; (6.40)
log p 4 k
if N b;S 1. We deduce from Theorem 2.9, (6.39), and (6.37) that
vp .ƒu / b;S nk.` 1/=.k 2/

log nk : (6.41)
By combining (6.40) and (6.41), we get
n1=.k
k
2/
b;S .k 2/ log n1=.k
k
2/
;
which, by (6.35), establishes the conclusion of Theorem 6.14.
6.11. Exercises
Exercise 6.1. Let S denote a finite set of prime numbers and .xj /j 1 the increasing
sequence of all positive integers whose prime factors belong to S . Let h; j be integers
with 1 h < j and j 3. If xh and xj are coprime, establish that
P Œxh C xj S .log xj /1=2 .log log xj /1=2 :

6.11. Exercises 91
Exercise 6.2 (see Lemma 4.1 of [376]). Let f .X / be a non-constant polynomial with
algebraic coefficients and ˛ a non-zero algebraic number. Apply Theorems 2.8 and 2.1 to
show that the equation
f .m/˛ m D f .n/˛ n ; m > n > 0;
implies that m f;˛ 1.

Exercise 6.3. Let P 2 be an integer and ˛ an irrational, algebraic number. Prove that
there exists an effectively computable real number C , depending only on ˛ and P , such
that, for every rational number pq with q 2 and P Œpq P , we have
ˇ p ˇˇ C
ˇ .log q/ :
ˇ
q
ˇ˛
Compare with Ridout’s Theorem A.8.

Exercise 6.4 (see [102, 103]). Consider the Diophantine equation x 2 2m D y n with
x andpy coprime and n > 2. Prove that m and n must be odd. Factor the left-hand side
in ZŒ 2 and combine the use of estimates for linear forms in two complex and in two
2-adic logarithms to prove that n is bounded. Let p be an odd prime number. Show
how the same method applies to the Diophantine equation x 2 p m D y n to bound the
exponent n.
Exercise 6.5 (see [373]). Establish the p-adic analogue of Theorem 3.10, that is, its
analogue for recurrence sequences having a dominant root for some p-adic absolute value
(see the proof of Theorem 6.4).
Exercise 6.6. Recall that .pk /k1 denotes the increasing sequence of prime numbers.
According to Guy [206, Section A2], Erdős and Stewart conjectured that the only solutions
of the equation
nŠ C 1 D pka pkC1
b
; for some a; b 0 and pk 1 n < pk , (6.42)
are obtained for n 5. This was confirmed by Luca [273]. Use linear forms in two 2-adic
logarithms to show that every solution .a; b; k; n/ of (6.42) satisfies n 1.
Exercise 6.7 (see [118]). Let S be a finite, non-empty set of prime numbers. For a positive
integer n, let ŒnS denote the greatest divisor of n, all of whose prime factors are in S .
Let b 2 be an integer and .un /n1 the increasing sequence of all the positive integers
which are not divisible by b and have at most three non-zero digits in their representation
in base b. Use estimates for complex and p-adic linear forms in logarithms to prove that
there exist effectively computable positive numbers c and n0 , depending only on S and b,
such that Œun S < .un /1 c , for every n > n0 .
Exercise 6.8 (see [333]). Let q1 ; : : : ; qs be distinct prime numbers, and denote by P
their maximum. Combine Theorems 2.3 and 2.12 to show that, for every solution to the
x 2 C .q1z1 : : : qszs /2 D 2y n ;
in non-negative integers x, y, z1 ; : : : ; zs , n, with n 3 odd and gcd.x; y/ D 1, we have
n sP 2 log P . Compare with Exercise 3.6.
Exercise 6.9. Let ˛1 and ˛2 be positive real numbers. Let p be a prime number. Let x1
and x2 be positive integers with jxi jp 1 xi˛i for i D 1; 2. Prove that, for every odd prime
number q which is coprime to p 1 and every non-negative integers a; b; y such that
x1a C x2b C 1 D y q ;
1
we have q .˛1 ˛2 / .
Exercise 6.10. Assume that there exist integers a; b; y; q and a non-zero rational number t
such that a > b > 0, y 2, q 2, and
2a C 2b C 1 D ty q :
Prove that q maxf1; h.t /g.

Exercise 6.11 (see [117]). Prove the p-adic analogues of Theorems 3.6 and 5.2, that is,
establish effective irrationality measures for p-adic n-th roots of rational numbers.
Exercise 6.12 (see [103, 126]). Let p be a prime number and a; b positive integers
not divisible by p. Let x; y; m; n be integers such that n 2, gcd.x; y/ D 1, and
p m D ax n C by n . Prove that n p.log p/.log maxfa; b; 2g/. Discuss the extension
of this result to the equation q1u1 qsus D ax n C by n , where q1 ; : : : ; qs are given prime
numbers and u1 ; : : : ; us are non-negative unknown integers.
Exercise 6.13. Establish that
log log m
P Œx m ˙ 2n log m
log log log m
holds for any odd integer x and any integer m with m 16.
6.12. Notes
F Brumer [99] used his p-adic analogue of Baker’s theorem [16] to establish, following
Ax [14], that the p-adic regulator of any abelian extension of the field of rationals (or of
an imaginary quadratic field) does not vanish.
F Mahler [282] showed that (6.5) holds for every sufficiently large n, but his proof, based
on Ridout’s Theorem A.8, is ineffective in the sense that it does not yield an explicit
value n0 with the property that (6.5) holds for every n greater than n0 . Corvaja and
Zannier [158] extended Mahler’s result to lower bounds for k˛ n k, where ˛ is a real
algebraic number greater than 1 satisfying some necessary conditions. Their proof rests
on the Schmidt subspace theorem and is thus not effective; see also [249]. It would be
very interesting to have effective versions of these results.
F Corvaja and Zannier [159] managed to avoid the use of Theorem 2.13 in their proof
that (6.15) has only finitely many solutions. They combined lower bounds for linear forms
in two p-adic logarithms (say, Theorem 2.12) with the hypergeometric method of Thue
and Siegel, based on Padé approximation to the binomial function. Actually, linear forms
in p-adic logarithms close to 1 combine well with the hypergeometric method; see [52]
and [109, 255, 350].
6.12. Notes 93
F Arithmetical properties of weighted sums of S -units have been studied in [215]. Lower
bounds for the greatest prime factor of solutions to decomposable form equations are
discussed in [178].
F For applications of S -unit equations, see e.g. the survey [184] and the recent mono-
graph [182].
F Levesque and Waldschmidt [262] used the theory of linear forms in logarithms to
construct parametric families of Thue–Mahler equations of arbitrary degree having only
finitely many solutions; see also [117]. By means of a totally different method, Bennett
and Dahmen [55] solved completely some specific families of Thue–Mahler equations of
small degree and where the corresponding set of prime numbers is unbounded.
F In 1980, by using the theory of linear forms in complex and p-adic logarithms, Agrawal,
Coates, Hunt, and van der Poorten [2] solved completely a cubic Thue–Mahler equation.
This enables them to determine all elliptic curves defined over Q of conductor 11. The
first general practical method for solving a Thue–Mahler equation was given by Tzanakis
and de Weger [422] in 1992; see also [382].
F The resolution of S -unit equations, Thue–Mahler equations, and other classical Dio-
phantine equations using modularity of Galois representation is discussed in [50, 55, 237,
240].
F Theorem 4.2 of Koymans [244] extends Theorem 6.13.
F A p-adic generalization of Theorem 4.11 has been worked out by van der Poorten [334];
see also Chapter 12 of [376], Chapter VII of [386], Brindza [93], and Koymans [244].
F Let p; q be distinct prime numbers and a; b; c; d non-zero integers. By a subtle
combination of estimates for linear forms in complex and p-adic logarithms, Vojta [425]
(see also [263]), Skinner [381], Mo and Tijdeman [304], and Mo [303], among other
authors, have established effectively computable upper bounds for the solutions w; x; y; z
in non-negative integers to the exponential equations ap x q y C bp z C cq w C d D 0 and
ap x C bq y C cp z C dq w D 0.
F Let S be a finite set of places on an algebraic number field and assume that S contains
the infinite places. Levin [263] applied the theory of linear forms in logarithms to deduce
effective results for S -integral points on certain higher-dimensional varieties when the
cardinality of S is sufficiently small. He generalized an effective result of Vojta [425]
on the three-variable unit equation by giving an effective solution of the polynomial unit
equation f .u; v/ D w, where u; v, and w are S -units, the cardinality of S is at most 3,
and f is a polynomial satisfying certain conditions, which are generically satisfied.
F Bugeaud and Kaneko [128] established, in a quantitative form, that any sufficiently large
odd integer cannot have simultaneously only very small prime factors and few non-zero
binary digits.
Chapter 7
Primitive divisors and the greatest prime factor of 2n 1
This chapter is devoted to the study of primitive divisors in some special linear recurrence
sequences of algebraic integers, with a special emphasis on Lucas and Lehmer sequences.
Furthermore, we establish that the greatest prime factor of 2n 1 tends to infinity faster
than any linear function of n, thereby solving a conjecture formulated by Erdős [177]
in 1965.
Throughout this chapter, we use several classical arithmetical functions, including the
Möbius function and Euler’s totient function '. Their definitions and some of their
basic properties are recalled in Appendix D. For every positive integer d , we denote
by ˆd .X; Y / the d -th homogeneous cyclotomic polynomial defined by
Y
ˆd .X; Y / WD .X Y /;

where the product is taken over all the primitive d -th roots of unity . Observe that '.d /
is the degree of ˆd .X; Y /.
7.1. Primitive divisors

We begin with the general definition of a primitive divisor.
Definition 7.1. Let K be an algebraic number field. Let .an /n1 be a sequence of non-
zero algebraic integers in K. Let n 2 be an integer. A prime ideal p of K is called a
primitive divisor of an if p divides the principal ideal generated by an , but does not divide
any principal ideal generated by am , where 1 m < n.
The first general result about the existence of primitive divisors dates back to 1892,
when Zsigmondy [452] proved that, if a and b are coprime non-zero integers with jabj 2,
then an b n has a primitive divisor for n > 6. The particular case b D 1 was done a
few years earlier by Bang [39]. Zsigmondy’s result is best possible, since 26 1 D
.22 1/2 .23 1/. It was rediscovered by Birkhoff and Vandiver [78] in 1904.
The next result has been established by Schinzel [355]; see also [393].
Theorem 7.2. Let a and b be non-zero algebraic integers in an algebraic number field K
such that ab is not a root of unity and the principal ideals generated by a and b in K are
coprime. Then, there exists an effectively computable integer n0 , depending only on the
degree of ab , such that an b n has a primitive divisor for every n greater than n0 .
96 7. Primitive divisors and the greatest prime factor of 2n 1
Proof. For simplicity, we content ourselves to prove the existence of an integer n0 de-
pending on the degree d of K and with the property stated in the theorem. A similar
result where n0 depends on a and b was established earlier in [338]. Without any loss of
generality, we assume that jaj jbj. Let m 2 be an integer. If jaj > jbj, then
ˇ b m ˇ ˇ b ˇ2
jam b m j jam j ˇ1 ˇ jam j 1 ˇ ˇ :
ˇ ˇ ˇ ˇ
a a
Since
ˇ b ˇ2 ˇ b ˇ2 ˇ b ˇ a
log 1 ˇ ˇ dh 1 ˇ ˇ 2dh ˇ ˇ d log 2 d 4h ;
ˇ ˇ ˇ ˇ ˇ ˇ
a a a b
by Theorems B.5, B.8, and 10.1, we derive that
a
log jam bmj m log jaj d 4h :
b
If jaj D jbj, then ab is a complex number on the unit circle and we deduce from Theo-
rems 2.6 and 10.1 that
ˇ b m ˇ a
1ˇ d 5 h .log m/2 :
ˇ ˇ
logˇ
a b
Thus, in both cases, we get
a
log jam bmj m log jaj d 5h .log m/2 : (7.1)
b
Let n 2 be an integer. By the Möbius inversion formula, we have
Y
ˆn .X; Y / D .X m Y m /.n=m/ ; (7.2)
mjn
where denotes the Möbius function. We will estimate from below and from above the
absolute value of the norm of ˆn .a; b/.
Let !.n/ denote the number of distinct prime factors of n. For any infinite place v
of K, set dv D 1 if v is real and dv D 2 otherwise. By (7.1) applied with a and b replaced
by their Galois conjugates, there exists an effectively computable absolute real number c
such that
X X n
log j NormK=Q ˆn .a; b/j D log jam b m jdv v
1
m
v2MK mjn
X X n
m log max jajdv v ; jbjdv v
˚

1
m
v2MK mjn
a X
cd 6 h .log m/2
b
mjn;
.n=m/¤0
a a
'.n/ h cd 6 h 2!.n/ .log n/2 ;
b b
7.1. Primitive divisors 97
n
/m D '.n/, by Theorem D.6. As ab is not a root
P
since j NormK=Q bj 1 and mjn . m
of unity, its height is positive, by Theorem B.6. It then follows from Theorem 10.1 and
Theorems D.4 and D.5, on the behaviour of the functions ' and !, that there exists an
effectively computable positive integer n0 .d /, depending only on d , such that
p
n
j NormK=Q ˆn .a; b/j e ; for n > n0 .d /: (7.3)
Let p be a prime ideal of K. By Lemma 7.3 below, if n 4d and p is not a primitive
divisor of an b n , then vp .ˆn .a; b// vp .n/, where p is the prime number lying below p.
Consequently, if the ideal generated by an b n has no primitive divisor and n 4d , then
j NormK=Q ˆn .a; b/j nd : (7.4)
Let n be an integer exceeding n0 .d / and 4d . It then follows from (7.3) that (7.4) cannot
hold, hence, the ideal generated by an b n must have a primitive divisor. This proves the
theorem.
Lemma 7.3. Let a and b be non-zero algebraic integers in a number field K of degree d
such that ab is not a root of unity and the principal ideals generated by a and b in K are
coprime. Let n be a positive integer and p a prime ideal in K such that p divides ˆn .a; b/
but p is not a primitive divisor of an b n . If n 4d , then we have
vp .ˆn .a; b// vp .n/;
where p is the prime number lying below p.
Proof. We follow Postnikova and Schinzel [338]; see also [355]. For i 1, let i denote
the smallest positive such that pi divides a b . Let e be the ramification index
of p (see Section B.1) and set k WD b p e 1 c. For a non-zero algebraic number ˛ in K,
let vp .˛/ denote the exponent of p in the decomposition of the fractional ideal ˛OK in
v .˛/
a product of prime ideals, and recall that vp .˛/ D pe . Let m be a positive integer.
m m
If m is not a multiple of 1 , then vp .a b / D 0. If m is a multiple of 1 , but not
of 2 , then vp .am b m / D 1. And so on. If m is a multiple of k , but not of kC1 ,
then vp .am b m / D k. Furthermore, if m is a multiple of kC1 , then, using the binomial
theorem as in the proof of Lemma F.2, we get
m
vp .am b m / D vp .akC1 b kC1 / C vp ;
kC1
e
since k C 1 > .
Consequently, we obtain
p 1
X n
vp .ˆn .a; b// D vp .am b m /
m
mjn
k X
X n X n
D C .vp .akC1 b kC1 / k/
m m (7.5)
i D1 mjn mjn
i jm kC1 jm
X n m
C vp :
m kC1
mjn
kC1 jm
If k D 0, then we get 1 < n since p divides the ideal generated by an b n but is not
a primitive divisor of an b n . If k 1, then p e C 1 and, as pkC1 divides the ideal
generated by a` b ` , where ` denotes the cardinality of the group of invertible elements
of OK =pkC1 , we have
kC1 NormK=Q pk NormK=Q p 1 :

Since the norm of p over Q is at most equal to p d=e , we obtain
kC1 p kd=e ..e C 1/d=e 1/ p d=.p 1/

.2d 1/ < 4d n:
From the inequalities 1 kC1 < n, we get by Theorem D.6 that

X n
D 0; i D 1; : : : ; k C 1: (7.6)
m
mjn
i jm
By plugging (7.6) into (7.5), we obtain

n=
kC1
X
vp .ˆn .a; b// D .`/ vp :
`
`j.n=kC1 /
This last sum is smaller than its term corresponding to ` D 1, giving

n
vp .ˆn .a; b// vp vp .n/:
kC1
This proves the lemma.
7.2. Primitive divisors of Lucas–Lehmer sequences

Theorem 7.2 addresses arbitrary sequences of the form .an b n /n1 , where a and b are
non-zero algebraic integers. In this section, we focus on two families of integer sequences
of a similar type, the Lucas and Lehmer sequences. Their primitive divisors have been
extensively studied since more than one century.
A Lucas pair is a pair .˛; ˇ/ of algebraic integers such that ˛ C ˇ and ˛ˇ are non-zero
coprime rational integers and ˛=ˇ is not a root of unity. Given a Lucas pair .˛; ˇ/, the
corresponding sequence of Lucas numbers (or Lucas sequence, not to be confused with
the sequence .Ln /n0 defined in Section 3.7) is defined by
˛n ˇn
un D un .˛; ˇ/ D ; n 0:
˛ ˇ
A Lehmer pair is a pair .˛; ˇ/ of algebraic integers such that .˛ C ˇ/2 and ˛ˇ are
non-zero coprime rational integers and ˛=ˇ is not a root of unity. For a Lehmer pair .˛; ˇ/,
the corresponding sequence of Lehmer numbers (or Lehmer sequence) is defined by
˛n ˇn
zn D u
u zn .˛; ˇ/ D ; if n 1 is odd,
˛ ˇ
7.2. Primitive divisors of Lucas–Lehmer sequences 99
˛n ˇn
zn D u
u zn .˛; ˇ/ D ; if n 0 is even.
˛2 ˇ2
Notice that every Lucas pair .˛; ˇ/ is also a Lehmer pair, and
un D u
zn ; if n 1 is odd,
un D .˛ C ˇ/z
un ; if n 0 is even.
Let .˛; ˇ/ be a Lehmer pair and n 2 an integer. A prime number p is a primitive
divisor of u zn .˛; ˇ/ if p divides uzn but does not divide .˛ 2 ˇ 2 /2 u
z1 u
zn 1 . This
provides a proper (“not merely automatical”, as written by Schinzel [353]) generalization
of the notion of primitive divisors introduced in Definition 7.1.
Similarly, let .˛; ˇ/ be a Lucas pair and n 2 an integer. A prime number p is a
primitive divisor of un .˛; ˇ/ if p divides un but does not divide .˛ ˇ/2 u1 un 1 . To
illustrate this notion, we observe that, among the first Fibonacci numbers
1; 1; 2; 3; 5; 8; 13; 21; 34; 55; 89; 144; 233; 377; 610; 987; 1597; : : :
the underlined ones have no primitive divisors. Actually, these are all the Fibonacci
numbers without a primitive divisor, as was proved by Carmichael [142] in 1913.
The following problem goes back to the beginning of the twentieth century, though it
does not seem to have been ever formulated explicitly.
Problem 7.4. List all Lucas and Lehmer numbers without primitive divisors. Classify
all the triples .˛; ˇ; n/ such that .˛; ˇ/ is a Lucas (resp.,Lehmer) pair and un .˛; ˇ/
zn .˛; ˇ/) has no primitive divisors.
(resp., u
Throughout the end of this section, it is convenient to use the following terminology.
A Lucas (resp., Lehmer) pair .˛; ˇ/ such that un .˛; ˇ/ (resp., uzn .˛; ˇ/) has no primitive
divisors is called an n-defective Lucas (resp., Lehmer) pair. Note that for any integer
n 3, a Lucas pair is an n-defective Lucas pair if and only if it is an n-defective Lehmer
pair.
The very first results towards the resolution of Problem 7.4 have been listed below
Definition 7.1. In 1913 Carmichael [142] proved that if .˛; ˇ/ is a Lucas pair composed of
real numbers, then un .˛; ˇ/ has a primitive divisor for n > 12. Since the 12-th Fibonacci
number, namely 144, has no primitive divisors, this is best possible. Carmichael’s result
was extended by Ward [436] (see [393, Lemma 8]) to Lehmer pairs .˛; ˇ/ such that ˛ 2
and ˇ 2 are real numbers; see also Durst [174, 175].
The situation is much more complicated for the remaining Lucas and Lehmer pairs.
“Nothing appears to be known about the intrinsic divisors of Lucas and Lehmer numbers
when ˛ and ˇ are complex,”— wrote Ward [436, p. 230] in 1955. In 1962 Schinzel [352]
completed the proof of the non-existence of n-defective Lucas and Lehmer pairs for n
exceeding an effectively computable absolute integer n0 . While Carmichael and Ward
used skillful but, in principle, elementary arguments, Schinzel’s proof relies upon the
theory of linear forms in (two) complex logarithms; see the proof of Theorem 7.2.
The next statement reproduces a result of Stewart [393], making explicit Schinzel’s
statement.
Theorem 7.5. For n > e452 267 , there are no n-defective Lucas pairs. For n > e452 467 ,
there are no n-defective Lehmer pairs.
Since there are infinitely many Lucas and Lehmer pairs, this result is not sufficient
to reduce the complete determination of all n-defective Lucas–Lehmer pairs to a finite
amount of computation. To do this, one needs an additional argument, which has been
given by Stewart in [393].
Theorem 7.6. For any integer n in f7; 9; 11g or satisfying n 13, there are only finitely
many n-defective Lucas and Lehmer pairs.
The condition on n is best possible, since, as proved in [393], for any integer n
in f1; 2; 3; 4; 5; 6; 8; 10; 12g, there exist infinitely many n-defective Lehmer pairs.
We gather in a lemma several useful auxiliary results established in [392] and whose
(elementary, but not easy) proofs are left to the reader. Recall that P Œ denotes the
greatest prime factor.
Lemma 7.7. Let .˛; ˇ/ be a Lehmer pair. If n 5 and n ¤ 6; 12, then P Œn= gcd.3; n/
divides ˆn .˛; ˇ/ to at most the first power. All other prime factors of ˆn .˛; ˇ/ are con-
gruent to 1 or 1 modulo n. For n in f7; 9; 11g or n 13 the algebraic integers uzn and un
have a primitive divisor whenever ˆn .˛; ˇ/ is different from ˙1 and ˙P Œn= gcd.3; n/.
Proof of Theorem 7.6. The key point is to reduce the problem to a family of Thue equa-
tions. Let .˛; ˇ/ be a Lehmer pair. Let denote by .z un /n0 the corresponding Lehmer
sequence and, if ˛ C ˇ is an integer, by .un /n0 the corresponding Lucas sequence.
Let n 7 be an integer. Since
Y
˛n ˇn D ˆd .˛; ˇ/;
d jn
any primitive divisor of uzn or un also divides ˆn .˛; ˇ/. Assume that n does not belong
to f8; 10; 12g and, in view of Theorem 7.5, that n e452 467 .
Set WD e2i=n . Let k be the largest integer smaller than n2 and such that k and n are
coprime. Then,
Y
ˆn .˛; ˇ/ D .˛ j ˇ/.˛ j ˇ/
1j k;
gcd.j;n/D1
Y
D .˛ 2 C ˇ 2 . j C j
/˛ˇ/:
1j k;
gcd.j;n/D1
Observe that ˛ 2 C ˇ 2 D .˛ C ˇ/2 2˛ˇ is an integer. Put ˛ 2 C ˇ 2 D v and ˛ˇ D w.

Define the homogeneous binary form fn .X; Y / by
Y
ˆn .˛; ˇ/ D fn .v; w/ D .v . j C j /w/:
1j k;
gcd.j;n/D1
The polynomial fn .X; 1/ has degree '.n/=2 and distinct roots. Our assumption on n
implies that '.n/ 6, thus the degree of the homogeneous binary form fn .X; Y / is
greater than or equal to 3. By Theorem 4.5, the Thue equation
fn .v; w/ D a;
7.3. The Diophantine equation x 2 C C D y n , continued 101
where a is in f˙1; ˙P Œn= gcd.3; n/g, has only finitely many solutions. Each solution
.v; w/ to this equation gives rise to two solutions .˛; ˇ/ and . ˛; ˇ/ to the equation
ˆn .˛; ˇ/ D a: (7.7)
We then deduce from Lemma 7.7 that we can effectively find all the solutions to the
equations (7.7) and establish the finite list of all Lucas–Lehmer pairs .˛; ˇ/ whose n-th
term has no primitive divisor, for some integer n 7 with n … f8; 10; 12g.
The lower bound in Theorem 7.5 was reduced to 30030 by Voutier [427, 428]. In view
of Theorem 7.6 and its proof, solving Problem 7.4 reduces to the resolution of finitely
many Thue equations of degree less than 30030. This programm was successfully carried
out by Bilu, Hanrot, and Voutier [77].
Theorem 7.8. For every integer n greater than 30, the n-th term of any Lucas or Lehmer
sequence has a primitive divisor.
The proof of Theorem 7.8 required the complete resolution of several Thue equations
of very large degree [74].
Furthermore, the authors of [77] also listed for every n 30 all Lucas and Lehmer se-
quences whose n-th term has no primitive divisors. Four lists were composed: n-defective
Lucas sequences for n 30 and '.n/ > 2, as well as n-defective Lehmer sequences
for n 30 and '.n/ > 4, were listed by Voutier [426], while n-defective Lucas sequences
for n D 1; 2; 3; 4; 6 and n-defective Lehmer sequences for n D 1; : : : ; 6; 8; 10; 12 were
listed in [77]. There are several omissions in these lists and the reader is referred to [1]
for the complete list of n-defective Lucas sequences for n D 1; 2; 3; 4; 6 and n-defective
Lehmer sequences for n D 1; : : : ; 6; 8; 10; 12.
Theorem 7.8 has many important applications to Diophantine equations.
7.3. The Diophantine equation x 2 C C D y n , continued

The equation x 2 C C D y n was already discussed in Section 4.6. We content ourselves
to explain how its resolution is deeply connected to Theorem 7.8.
We focus on the case C D 2 and let x; y; q be positive integers with q an odd prime
such that
x2 C 2 D yq :
Arguing as in the proof of Theorem 4.8, there exist rational integers a; b such that
p p
x C i 2 D .a C ib 2/q :
By equating the imaginary parts of both members of the previous equality, we deduce that
b D ˙1, thus
ˇ .a C ib p2/q .a ib p2/q ˇ ˇ 2ip2 ˇ
p p ˇ D ˇ p ˇ D 1:
ˇ ˇ ˇ ˇ
ˇ
.a C ib 2/ .a ib 2/ 2i 2
This implies that the q-th term of the Lucas sequence defined by
p p
.a C i 2/n .a i 2/n
un D p p ; n 0;
.a C i 2/ .a i 2/
has no primitive divisor and it follows from Theorem 7.8 that q 29.
Clearly, some assumptions on C are required to apply Theorem 7.8 to the equation
x 2 C C D y q ; see [138, 155] for precise statements.
7.4. On the number of solutions to the Diophantine equation

x 2 C D D pn
We have already mentioned the famous Ramanujan–Nagell equation x 2 C 7 D 2n in
Section 4.6. In this section, we briefly discuss some closely related equations. We begin
by a result of Beukers [64], who improved an earlier work by Apéry [12].
Theorem 7.9. Let D be an odd, positive integer. Then, the equation
x 2 C D D 2n
has at most one solution in positive integers x and n, unless D D 7, 23 or 2m 1 for some
m 4. The solutions in these exceptional cases are given by
D D 7; .x; n/ D .1; 3/; .3; 4/; .5; 5/; .11; 7/; .181; 15/I
D D 23; .x; n/ D .3; 5/; .45; 11/I
D D 2m 1 .m 4/; .x; n/ D .1; m/; .2m 1
1; 2m 2/:
While the proof of Theorem 7.9 does not rest on the theory of linear forms in logarithms,
Theorem 7.8 is a crucial ingredient in the proof of Theorem 7.10 below, established by
Bugeaud and Shorey [139]. It improves an earlier work by Apéry [13] and complements
Theorem 7.9.
Theorem 7.10. Let D be a positive integer and p an odd prime, not dividing D. Then,
the Diophantine equation
x2 C D D pn (7.8)
has at most one solution in positive integers x and n, unless .p; D/ D .3; 2/ or .p; D/ D
.4a2 C 1; 3a2 C 1/ for some positive integer a. In these exceptional cases, there are
precisely two such solutions.
If we remove the coprimeness assumption gcd.p; D/ D 1 in Theorem 7.10, then the
exceptional pairs are given by .p; D/ D .3; 2 32j / or .p; D/ D .4a2 C 1; .3a2 C 1/
.4a2 C 1/2j / for some positive integer a and some non-negative integer j ; see Theo-
rem 12.1 of [42].
Theorems 7.9 and 7.10 show that the Ramanujan–Nagell equation x 2 C 7 D 2n is
exceptional, in the sense that it has much more solutions than the other equations of similar
type.
As already noted by Beukers [63] (see also [139]), distinct solutions in positive inte-
gers x and n to (7.8) correspond to integers ` 2 for which
` . /`
u` WD D ˙1; (7.9)

p
where is an algebraic integer in Q. D/ and denotes its Galois conjugate. Here, u`
is the `-th term of a Lucas or Lehmer sequence. The key tool for the proof of Theorem 7.10
7.5. On the greatest prime factor of 2n 1 103
(and for the results from [139]) is Theorem 7.8, which applies to (7.9). It is interesting to
note that this result, which depends ultimately on estimates for linear forms in logarithms,
is used to get a quantitative result, namely an upper bound for a number of solutions.
Actually, a more general result has been obtained in [139] (see also [259]) on equations
of the form
D1 x 2 C D2 D bk n ; in integer unknowns x 1, n 1; (7.10)
when D1 and D2 are positive integers, the positive integer k is coprime with the prod-
uct D1 D2 , and b is in f1; 2; 4g, such that b D 4 when k is even. Then, Equation (7.10)
has at most one solution, except in some cases, which are explicitly listed (and the exact
number of solutions in these cases is known).
7.5. On the greatest prime factor of 2n 1

Let n 7 be an integer and p a primitive divisor of 2n 1. Then, the order of 2 modulo p
is equal to n. Since this order divides p 1, we deduce that p is congruent to 1 modulo n
and, consequently, that P Œ2n 1 is at least equal to nC1. In 1965 Erdős [177] conjectured
that
P Œ2n 1
lim D C1:
n!C1 n
In this section, we combine an estimate of Yamada [442] on linear forms in two p-adic
logarithms with an approach of Stewart [400] to prove Erdős’ conjecture.
Specifically, we establish the following result.
Theorem 7.11. Let a; b be integers with a > b 1. For any integer n 3, the greatest
prime factor of an b n satisfies
s
n n log n
P Œa b b n :
log log n
This solves the above mentioned conjecture of Erdős. Actually, a stronger and more
general result was proved in 2013 by Stewart [400]; see Theorem 7.13 below.
Birkhoff and Vandiver [78] attributed to Euler the first assertion of Lemma 7.12, but
gave no precise reference. It extends the observation on P Œ2n 1 made at the beginning
of this section.
Lemma 7.12. Let a; b be coprime integers with a > b 1. For n 2, every primitive
prime divisor of an b n is congruent to 1 modulo n. For any odd prime number p not
dividing ab and any positive integer n we have
vp .an b n / vp .ap 1
bp 1
/ C vp .n/: (7.11)
Proof. We follow the argument of Stewart given in Section 6 of [400]. For simplicity,
write zn D an b n for every positive integer n. Let p be an odd prime number not
dividing ab. Let ` D `.p/ be the smallest positive integer h for which p divides zh . It
is equal to the order of ab modulo p, hence ` divides p 1 and ` divides every positive
integer n such that p divides zn . This implies that, for n 2, every primitive divisor of zn
is congruent to 1 modulo n.
Let n be a positive integer such that p divides zn . Assume that p divides ˆn .a; b/.
Write n D `tp k , where k is a non-negative integer and t is a positive integer not divis-
ible by p. Observe that p divides zn=t since ` divides n=t . If t > 1, then ˆn .a; b/
divides zn =zn=t and we get a contradiction by using Exercise 7.1. We conclude that t D 1
and n D `p k . Since p is odd and, for any positive integer m, we have
!
zmp .zm C b m /p b mp m.p 1/ p m.p 2/ p 1
D D pb C b zm C C zm ;
zm zm 2
we deduce that vp .zmp / D vp .zm / C 1 if p divides zm . We have proved that

vp ˆ`pk .a; b/ D 1; for k 1,
if m is not of the form `p k for k 0.

and vp ˆm .a; b/ D 0;
Q
Let n be a positive integer. Since zn D d jn ˆd .a; b/, this implies that
n
vp .zn / D vp .z` / C vp ; if ` divides n;
`
while we have vp .zn / D 0 if ` does not divide n. As ` divides p 1, we derive that z`
divides zp 1 and vp .z` / vp .zp 1 /. Recalling that vp .`/ D 0, we get the desired
result.
Proof of Theorem 7.11. We proceed as in the proof of Theorem 7.2, but the situation is
much simpler since a and b are positive integers. For any positive integer m, we have
b m a b
0 log.am bm/ m log a D log 1 log log a:
a a
Let n > 12 be an integer. It follows from (7.2) that
ˇ X n ˇ
ˇlog ˆn .a; b/ m log aˇ 2!.n/ log a;
ˇ ˇ
m
mjn
where !.n/ denotes the number of disctinct prime factors of n. By Theorems D.4, D.5,
and D.6, we get
'.n/
log ˆn .a; b/ '.n/ log a 2!.n/ log a log a; (7.12)
2
when n is sufficiently large. Let p be a prime number dividing ˆn .a; b/. Since a and b
are coprime, p does not divide ab and
vp .ˆn .a; b// vp .an b n /: (7.13)

7.5. On the greatest prime factor of 2n 1 105
It then follows from Lemma 7.12 and Theorem 12.3 (note that we can as well apply
Theorem 2.9) that
vp .an b n / vp .ap 1
b p 1 / C vp .n/
p (7.14)
132 105 .log 5a/ .log 5b/ C vp .n/:
.log p/2
Let Pn denote the greatest prime factor of ˆn .a; b/. By Lemma 7.7 there is at most
one prime factor of ˆn .a; b/ which is not congruent to ˙1 modulo n. This prime factor
divides n and it divides ˆn .a; b/ to at most the first power. Bounding this prime factor
by n, we get X
log ˆn .a; b/ log n C .log p/ vp .ˆn .a; b//: (7.15)
pPn ; p6 j n
Combining (7.12), (7.14), and (7.15), we obtain

X p
'.n/ log n C .log 5b/:
log p
pPn ;
p˙1.n/
Using Theorem D.5 and Theorem D.7, which bounds from above the number of primes
which are less than Pn and congruent to ˙1 modulo n, we obtain
Pn Pn Pn2
'.n/ b b :
'.n/ log.Pn =n/ log Pn '.n/.log Pn /.log.Pn =n//
It then follows from Theorem D.5 that there exists a positive real number C such that
Pn b n log n.log log n/ C :

p
However, by using (D.3) instead of Theorem D.5, we derive the larger lower bound
s
log n
Pn b n :
log log n
which proves the theorem, since P Œan b n Pn , by (7.13).

Stewart [400] established the following stronger and more general result.
Theorem 7.13. If .un /n0 is a Lucas or Lehmer sequence, then there exists an effectively
computable integer n0 such that, for any integer n greater than n0 , we have
log n
P Œun > n exp :
104 log log n
The case of Lucas and Lehmer sequences with irrational roots is much more delicate
than the special case considered in Theorem 7.11, since we have to deal with linear forms
in logarithms of algebraic numbers lying in a quadratic number field. The dependence
on p in Theorems 2.9 and 2.12 is then of order p 2 , thus, we cannot adapt the proof
of Theorem 7.11. New ideas are needed: Stewart [400] inflated artificially the number
of terms which occur in the p-adic linear form in logarithms and applied a refined estimate
of Yu [450].
Similar ideas allowed Stewart [401] to establish the following theorem, which improves
an earlier result of Yu and Hung [451] (see also [126]).
Theorem 7.14. Let .un /n0 be a non-degenerate binary recurrence sequence of integers.
Then, there exists an effectively computable integer n0 such that, for any integer n greater
than n0 , we have
p log n
P Œun > n exp :
104 log log n
We display a consequence of Theorem 7.14 to arithmetic properties of the convergents
of quadratic numbers.
Corollary 7.15. Let be a quadratic real number and let .pn =qn /n0 denote the
sequence of its convergents. Then, for every sufficiently large integer n, we have
p log log qn
P Œqn log qn exp :
104 log log log qn
Proof. Write D Œa0 I a1 ; : : : ; ar 1 ; b0 ; b1 ; : : : ; bs 1 ; b0 ; : : : ; bs 1 ; : : :. In the course of

the proof of Theorem 1 in [258], it is established that there exists an integer t such that
qnC2s t qnCs C . 1/s qn D 0;
for every n r. Set M WD maxfa1 ; : : : ; ar 1 ; b1 ; : : : ; bs 1 g. Then, for n 1, an easy

induction shows that qn .M C 2/n and, thus, n .log qn /=.log.M C 2//. Combined
with Theorem 7.14, this proves the corollary.
7.6. Exercises
Exercise 7.1 (see [392]). Let .˛; ˇ/ be a Lehmer pair and d; m; n positive integers. Prove
that if d divides n, then gcd.un =ud ; ud / divides n=d . Prove that gcd.un ; um / D ugcd.n;m/ .
Exercise 7.2. Let a; b be coprime integers with a > b 1. For an integer t 1, let Qt
denote the product of the t first prime numbers. Prove that we have
Qt
log P ŒaQt b Qt log a:
log log Qt
7.7. Notes
F With a non-torsion rational point P D .x.P /; y.P // on an elliptic curve written
y 2 D x 3 C ax C b in minimal form we can associate an elliptic divisibility sequence
.Bn .P //n1 defined, by using the addition law on the curve, by x.nP / D An .P /=Bn .P /2
(in lowest terms). Silverman [380] established the analogue of Zsigmondy’s theorem,
namely that, for any given elliptic divisibility sequence .Bn /n1 , there exists an integer n0
such that Bn has a primitive divisor for every n n0 .
7.7. Notes 107
F The only perfect powers in the Fibonacci sequence are 0; 1; 8, and 144; see Theo-
rem 3.14, the paper [137] and the references therein. Perfect powers in other Lucas and
Lehmer sequences have been studied in [131, 379].
F Stiller [406] established that the Diophantine equation x 2 C 119 D 15 2n has exactly
six solutions in positive integers; see Ulas [423] for further examples of similar equations
with many solutions.
F Let a; b be coprime positive integers. Murty and Wong [309] established that, for every
positive real number ", there exist real numbers c1 and c2 , depending only on a; b; ",
such that the greatest prime power divisor of an b n exceeds c1 n2 " and, under the
abc-conjecture (see Chapter 8), such that P Œan b n exceeds c2 n2 " .
F Let .un /n0 be a Lucas or Lehmer sequence. Győry [213] pestablished that, if n > 30
or if 6 < n 30 and jun j > exp exp.7040/, then P Œun > 14 log log jun j.
Chapter 8
The abc-conjecture
In June 1985 Oesterlé [318] gave a lecture at the Max-Planck-Institute in Bonn. Motivated
by conjectures about elliptic curves, he asked whether there is an absolute real number
such that, for every positive, coprime integers a; b, we have
Y
aCb p ;
pjab.aCb/
where the product is taken over the distinct prime divisors of ab.a C b/. A few weeks
thereafter, at a conference in London in honour of Roth’s sixtieth birthday, Masser [285,
288] proposed the following refinement of the conjecture formulated by Oesterlé.
Conjecture 8.1 (abc-conjecture). For every positive real number ", there exists a positive
real number ."/, which depends only on ", such that, for all positive integers a; b; and c
with
aCb Dc and gcd.a; b; c/ D 1; (8.1)

we have
Y
c < ."/G 1C" ; where G D p: (8.2)
pjabc
In (8.2) and below, the product is taken over the distinct prime divisors of abc. The
abc-conjecture is a very deep conjecture, with profound consequences; see Chapter 12
of [88] and the references given therein. We show in this chapter how the theory of linear
forms in complex and p-adic logarithms applies to get a weaker version of (8.2).
8.1. Effective results towards the abc-conjecture

In 1986, applying an estimate for linear forms inp-adic logarithms of van der Poorten [335],
Stewart and Tijdeman [402] established that, for all positive integers a; b; and c with (8.1),
we have Y 15
log c p : (8.3)
pjabc
Subsequently, in 1991, Stewart and Yu [404] strengthened (8.3) by means of a subtle

combination of estimates for complex and p-adic linear forms in logarithms. We state
below the further improvement they obtained in 2001 in [405].
110 8. The abc-conjecture
Theorem 8.2. For all positive integers a; b; and c with a C b D c and gcd.a; b; c/ D 1,
we have Y
log c G 1=3 .log G/3 ; where G D p:
pjabc
In the same paper [405], the authors derived another estimate in the direction of the
abc-conjecture. Recall that P Œ denotes the greatest prime factor.
Theorem 8.3. There exists an effectively computable real number such that, for all
positive integers a; b; and c with a C b D c and gcd.a; b; c/ D 1, we have
log c < p 0 G .log log log.8G//= log log G ;
where
Y
GD p and p 0 D minfP Œa; P Œb; P Œcg:
pjabc
We point out a consequence of Theorem 8.3, which should be compared with the lower
bound for P Œf .x; y/ established in [217]; see also Sections 3.5 and 6.7.
Corollary 8.4. For all positive integers a; b with a < b and b 40, we have
log log log b
P Œab.a C b/ log log b :
log log log log b
Proof. Set P WD P Œab.a C b/. Observe that
Y X
log p log p .log P / Cardfp P W p primeg P;
pjab.aCb/ pP
by Theorem D.2. It follows from Theorem 8.3 that there exists an effectively computable
real number such that
P log log P
log log b < log P C :
log P
This implies the corollary.
8.2. Proofs of Theorems 8.2 and 8.3

Without any loss of generality, we suppose that a < b. To shorten the notation, write
pa ; pb ; pc instead of P Œa; P Œb; P Œc, respectively. Throughout this proof, c1 ; c2 ; : : :
denote effectively computable positive numbers. Observe that G 6 and put
n G o
G D max ; 16 and r D !.abc/;
pa pb pc
the number of distinct prime factors of abc.
Put m D r 2 if a D 1 (in which case pa D 1) and m D r 3 otherwise. Observe
that, by Theorem D.2,
r 3
X
log G log pj r log r;
j D1
8.2. Proofs of Theorems 8.2 and 8.3 111
hence,
log G
mr : (8.4)
log log G
If m 1, then the arithmetic-geometric mean inequality gives
m
log G m

Y 1 X
log p log p :
m m
pjabc; pjabc;
p…fpa ;pb ;pc g p…fpa ;pb ;pc g
We deduce that
Y log G log log log G
log p < exp c1 : (8.5)
log log G
pjabc;
p…fpa ;pb ;pc g
With the usual convention that the empty product is equal to 1, we see that (8.5) also holds
if m D 0.
Let p be a prime divisor of abc. Our aim is to estimate vp .abc/. First, we assume
that
2
p < e.log r/ : (8.6)
Suppose that p divides c, the other two cases being analogous.
Then, p does not divide b and
c a a 2
vp .c/ D vp D vp C 1 vp 1 :
b b b
Write
a 2 !.ab/
u
Y
1D qj j 1;
b
j D1
for some non-zero integers uj and distinct prime numbers q1 ; : : : ; q!.ab/ . Note that we
have juj j 3 log c for j D 1; : : : ; !.ab/.
Y
vp .c/ pc2!.ab/ log ` log log c: (8.7)
`jab;` prime
By (8.4), (8.5), and (8.6), this gives

log G log log log G
vp .c/ < exp c3 .log.2pa // .log pb / log log c: (8.8)
log log G
The fact that, in Theorem 2.9, the dependence on the number n of algebraic numbers is
exponential and not of the form ncn is crucial here. Indeed, by using an earlier estimate we
would have c2!.ab/ replaced by !.ab/2!.ab/ in (8.7) and would then get an upper bound
of the form
vp .c/ < .G /c4 .log.2pa // .log pb / log log c;
which is much weaker than (8.8).
Similarly, we get
log G log log log G
vp .b/ < exp c5 .log.2pa // .log pc / log log c (8.9)
log log G
and
log G log log log G
vp .a/ < exp c6 .log pb / .log pc / log log c: (8.10)
log log G
We now define the quantities
0
Y 0
Y 0
Y
Ra D `v` .a/ ; Rb D `v` .b/ ; Rc D `v` .c/ ;
`ja;`6Dpa `jb;`6Dpb `jc;`6Dpc
Q0 2
where means that ` runs through the prime numbers less than e.log r/ . Observe that
log Ra < !.a/ .log r/2 max v` .a/

`ja;`<e.log r/2
log G log log log G
r.log r/2 exp c7 .log pb / .log pc / log log c
log log G
log G log log log G
exp c8 .log pb / .log pc / log log c;
log log G
by (8.10) and (8.4), and similarly for Rb and Rc .
Put q1 D Ra =Rb and set
a
D q1 q2b2 qnbn ; (8.11)
b
where b2 ; : : : ; bn are non-zero rational integers and q2 ; : : : ; qn are distinct prime numbers
2
exceeding e.log r/ or in the set fpa ; pb g. Observe that, by Theorem B.5, the height of q1
satisfies
log G log log log G
h.q1 / exp c9 .log maxfpa ; pb g/ .log pc / log log c: (8.12)
log log G
Using (8.11) to bound n, we get
.n 3/.log r/2 log G ;
thus, if r .log G /1=2 , we obtain

log G
n2n < exp c10 : (8.13)
log log G
Since n r, Inequality (8.13) also holds when r < .log G /1=2 .

We are now in position to estimate vp .abc/ for an arbitrary prime factor p of abc.
By (8.11), we have
vp .c/ D vp .q1 q2b2 qnbn 1/:
Set
log G log log log G p
Wp D exp c11
log log G .log p/4
Y
log maxfp; `g .log log c/2 : (8.14)
`2fpa ;pb ;pc g
Using Theorem 2.10, it follows from a careful computation based on (8.4), (8.5), (8.12),
and (8.13) that
vp .c/ < Wp log maxfpa ; pb g: (8.15)
The crucial point with the use of Theorem 2.10 here is that we do need a good dependence
on the prime number p, while the weaker dependence on n does not cause any trouble in
view of (8.13).
Similarly, we obtain
vp .b/ < Wp log maxfpa ; pc g

and vp .a/ < Wp log maxfpb ; pc g:
These upper bounds for vp .a/; vp .b/, and vp .c/ are weaker than the upper bounds obtained
in (8.8), (8.9), and (8.10), which are valid under the assumption (8.6). It would be much
desirable to remove the factor p in (8.14).
Having estimated vp .abc/ for every prime divisor p of abc, we get
X
log c D vp .c/ log p r max fvp .c/ log pg; (8.16)
pjc
pjc
and similar upper bounds hold for log a and log b. Set
L D log maxfpa ; pb g log maxfpb ; pc g log maxfpa ; pc g:
By (8.4), (8.14), (8.15), and (8.16), we then get
log c log G log log log G pc

2
< exp c12 L: (8.17)
.log log c/ log log G .log pc /2
c
Since b > 2
and c 3, we have
log c
log b > log c log 2 > ; (8.18)
4
thus
log c < 4 log b 4r max fvp .b/ log pg;
pjb
and, arguing as above, we deduce that
log c log G log log log G pb

2
< exp c13 L: (8.19)
.log log c/ log log G .log pb /2
p
We distinguish two cases. First, assume that a b. Then, by (8.18),
log b log c
log a >
2 8
and, similarly as above, we deduce from
log c < 8 log a 8r max fvp .a/ log pg

pja
that log G log log log G

log c pa
< exp c14 L: (8.20)
.log log c/2 log log G .log.2pa //2
p
Second, assume that a < b, in which case
a C b r
1 1 2
log < log 1 C p < p < : (8.21)
b b b c
Put q10 D Rc =Rb and write
c 0 0
D q10 .q20 /b2 .qn0 /bn ;
b
where b20 ; : : : ; bn0 are non-zero rational integers and q20 ; : : : ; qn0 are distinct prime numbers
2
exceeding e.log r/ or in the set fpb ; pc g. Then,
a C b c
0 < log D log D log q10 C b20 log q20 C C bn0 log qn0 :
b b
By applying Theorem 2.2, we get
a C b log G log log log G
log log > exp c15 log maxfpb ; pc g
b log log G
.log.2pa // .log pb / .log pc / .log log c/2 : (8.22)
We deduce from (8.21) and (8.22) that
log c log G log log log G

< exp c16 log maxfpb ; pc g
.log log c/2 log log G
.log.2pa // .log pb / .log pc /:
Since pa
log maxfpb ; pc g .log.2pa // .log pb / .log pc / L;
.log.2pa //2
p
we see that (8.20) still holds when a < b.
Let p 0 ; p 00 , and p 000 be such that p 0 < p 00 < p 000 and fpa ; pb ; pc g D fp 0 ; p 00 ; p 000 g. It
follows from (8.17), (8.19), and (8.20) that
log c log G log log log G p0
2
< exp c17
.log p 00 / .log p 000 /2 : (8.23)
.log log c/ log log G .log.2p 0 //2
8.3. Exercise 115
Proceeding as in the proof of (8.5) we get

Y 2 log G log log log maxfG; 16g
.log p 00 / .log p 000 /2 log p < exp c18 ;
log log G
pjabc
thus, by (8.23),
log c log G log log log maxfG; 16g p0

< exp c19 : (8.24)
.log log c/2 log log G .log.2p 0 //2
This proves Theorem 8.3.
To establish Theorem 8.2, we multiply (8.17), (8.19), and (8.20) together to get
log c 3 log G log log log G
< exp c20
.log log c/2 log log G
pa pb pc
.log p 00 /3 .log p 000 /6 : (8.25)
.log.2pa / log pb log pc /2
We may assume that
p 0 > G 1=4 ;
since otherwise Theorem 8.2 follows directly from (8.24). Then, we get from (8.25) that
log c 3
G pa pb pc .log G/3 G.log G/3 ;
.log log c/2
and so
log c G 1=3 .log G/3 ;
as asserted.
8.3. Exercise
Exercise 8.1. Let a; b, and c be positive integers as in (8.1) and define G as in (8.2).
Let P be the greatest prime factor of abc. Let B denote the largest integer t such that p t
divides abc for some prime number p. Apply Theorem 2.9 with the prime number P
to bound B from above in terms of G. Bound c from above in terms of B and G, and
conclude that there exists an effectively computable real number such that
log c < G G .log log log.8G//= log log G :
8.4. Notes
F Robert, Stewart, and Tenenbaum [348] have given very precise refinements of the abc-
conjecture, which rest on the sole heuristic assumption that, if a and b are coprime, then
the greatest squarefree factors of a; b, and c D a C b are statistically independent.
F Keep the notation of Theorems 8.2 and 8.3. Using a different (and more direct) app-
roach, also based on lower bounds for linear forms in complex and p-adic logarithms,
Győry and Yu [217] have established that
P
log c < 223 G 653.log log log maxfG;16g/=.log log maxfG;16g/ ; where P D P Œabc.
log P
Chi [146], following the proof of [405], has proved that
z
log c < p 0 G 13:6.log log log G/=.log log G/ ; z D maxfG; 9699690g.
where G
F By using the modularity of elliptic curves, Murty and Pasten [307] established that all
positive integers a; b; c with a C b D c satisfy
log c 4:8G log G C 13G C 25;
where G is as in (8.2); see also von Känel [236] for a slightly weaker result.
F Masser [286] extended the abc-conjecture to algebraic number fields; see also Brow-
kin [98], Győry [214], and the references quoted therein.
F Let a; b; c be coprime positive integers satisfying
P a C b D c (called an abc-triple in the
sequel). Put h D log c and r D r.a; b; c/ D pjabc log p. The abc-conjecture asserts
that for any positive " there exists a real number K."/ such that h r C "h C K."/.
The choice .a; b; c/ D .1; 2n ; 2n C 1/ shows that there exists an infinite sequence
of abc-triples such that h r log 2. Stewart and Tijdeman [402] obtained a much
p positive real number ı, there exist infinitely many abc-triples
sharper result: for any
with h r C .4 ı/ h= log h. Van Frankenhuijsen [187] showed that, in the latter
result, 4 ı can be replaced by 6:068.
F Baker [29–31] suggested modifications and strengthenings of the abc-conjecture which
he related to earlier conjectures on lower bounds for linear forms in logarithms of algebraic
numbers formulated in the Introduction to Chapters X and XI of [250]. An interesting
connection is pointed out with conjectures on simultaneous complex and p-adic estimates
for the same linear combination of logarithms; see also Philippon’s papers [327,328]. Von
Känel and Matschke [237] Q verified an explicit abc-conjecture of Baker ([30, Conjecture 4])
for all abc-triples with pjabc p 107 .
F Additional information on and around the abc-conjecture can be found in the sur-
vey [283] and on the website maintained by Abderrahmane Nitaj www.math.unicaen.
fr/~nitaj/abc.html.
Chapter 9
Simultaneous linear forms in logarithms and applications
A longstanding open problem in the theory of linear forms in logarithms asks whether in
the lower bounds established so far the product of the heights of the algebraic numbers
can be replaced by their maximum. We have seen that this is indeed the case when the
algebraic numbers are very close to 1; see the discussion below Theorem 2.1. This is
also not far from being the case when we have several independent small linear relations
involving the same logarithms of algebraic numbers. Theorem 9.1 shows that, under this
assumption, the product log A1 log An occurring in (1.7) can be replaced by its c-th
power, for some real number c with n1 < c < 1.
9.1. A theorem of Loxton

The next result was established in 1986 by Loxton [270]. Recall that the naïve height of an
algebraic number is the maximum of the absolute values of the coefficients of its minimal
defining polynomial over the rational integers.
Theorem 9.1. Let n 1 be an integer and ˛1 ; : : : ; ˛n non-zero multiplicatively indepen-
dent algebraic numbers. Let t 2 be an integer and, for i D 1; : : : ; t and j D 0; : : : ; n,
let ˇi;j be an algebraic number. Consider the linear forms
ƒi D ˇi;0 C ˇi;1 log ˛1 C C ˇi;n log ˛n ; i D 1; : : : ; t;
where log is the principal determination of the logarithm. Assume that the t .n C 1/
matrix whose entries are the ˇi;j has rank t. Let B 4 and, for j D 1; : : : ; n, let Aj 4
be real numbers such that the naïve height of ˛j (resp., of ˇ1;j ; : : : ; ˇt;j ) is at most Aj
(resp., at most B). Let D be an upper bound for the degree of the field generated by all
the ˛j and ˇi;j . Then, setting
D log A1 : : : log An ;
we have
.16nD/200n . log /1=t log B:
˚
max log jƒ1 j; : : : ; log jƒt j >
This is Theorem 4 in [270]. A result in the same spirit has been obtained by
Ramachandra [340] at an early stage of the development of the theory of linear forms
in logarithms.
118 9. Simultaneous linear forms in logarithms and applications
The first estimates for simultaneous linear forms in p-adic logarithms were given by
Dong [169]. Subsequently, Gaudron [195] established an explicit lower bound for a system
of homogeneous and inhomogeneous linear forms in logarithms of algebraic numbers with
algebraic coefficients, both in a complex and in a p-adic framework. His result extends
Loxton’s estimate.
9.2. Perfect powers in short intervals

Loxton [270] applied Theorem 9.1 to give an upper bound for the number of perfect powers
in short intervals.
p
Theorem 9.2. For any sufficiently large integer N , the interval ŒN; N C N contains
at most
p
exp 30 log log N log log log N
perfect powers.
p
The interval ŒN; N C N is the appropriate interval to consider in Theorem 9.2, since
for any integer r 2, it contains at p most one r-th power. Furthermore, for any positive
real number ", the interval ŒN; N C N N " contains at least N " . 21 C o.1// squares. It
p
is believed that, if N is sufficiently large, then the interval ŒN; N C N contains at most
three distinct perfect powers. Actually, the computations in [407] give just one example,
namely 112 ; 53 , and 27 , of three prime powers in such an interval.
Loxton’s proof of Theorem 9.1 was not complete in the case where the integers he was
working with were multiplicatively dependent. Bernstein [62] overcame this difficulty and
an alternative proof of Theorem 9.2 was subsequently given by Stewart [398], who also
established the following extension of Theorem 9.2, that we state without proof.
Theorem 9.3. Let k be an integer with k 2. For any sufficiently large integer N , the
interval ŒN; N C N .k 1/=k contains at most
p
exp 30 log log N log log log N
integers which are perfect r-th powers, for some integer r k.

Again, the interval ŒN; N C N .k 1/=k is the right interval to consider in Theorem 9.3.
A more general result of [398] deals with perfect powers of rational numbers in short
intervals.
For the proof of Theorem 9.2 we need an auxiliary result, which is Theorem 1 in [398].
Lemma 9.4. Let " be a real number with 0 < " < 1. Let t and k be integers with t 2
and
k t 1 C d" 1 et 1 :

(9.1)
If ˛1 ; : : : ; ˛k are non-zero algebraic numbers with the property that any t of them are
multiplicatively dependent, then there exist distinct indices i0 ; : : : ; it in f1; : : : ; kg for
which ˛
i
h 0 ".h.˛i2 / C C h.˛it //:
˛i1
9.2. Perfect powers in short intervals 119
Proof. By assumption, for every t-tuple .i1 ; : : : ; it / with 1 i1 < i2 < < it k,
there are integers ì1 ; : : : ; ìt not all zero such that
ì `
˛i1 1 ˛itit D 1:
We associate with .i1 ; : : : ; it / the .t 1/-tuple obtained by removing the j -th coordinate ij ,
where ij is an index such that jìj j is maximal, that is,
jìj j jìn j; for 1 n t .
Since there are t k 1 .t 1/-tuples and kt different tuples .i1 ; : : : ; it / as above, at least

one of the .t 1/-tuples is associated with m distinct t -tuples, where

& ! !'
k k k t C1
mD = D :
t t 1 t
Consider one .t 1/-tuple with this property. By reordering the ˛i ’s if necessary,

we may assume that this .t 1/-tuple is .1; : : : ; t 1/ and the m associated t-tuples
are .1; : : : ; t 1; j C t 1/, for j D 1; : : : ; m. Let j D 1; : : : ; m. There are integers
`1;j ; : : : ; `t 1;j and `j with `j > 0, j`1;j j; : : : ; j`t 1;j j `j , and
` ` `
˛11;j ˛t t 11;j D ˛j jCt 1:
Put bi;j D ì;j =`j for i D 1; : : : ; t 1. Put Bj D .b1;j ; : : : ; bt 1;j / and observe that Bj
is in Œ0; 1t 1 . Since, by the assumption (9.1), we have
1 t 1
m > d" e ;
it follows from Dirichlet’s Schubfachprinzip that there exist two .t 1/-tuples Bu and Bs
with 1 u < s m whose entries satisfy
jbi;u bi;s j "; for i D 1; : : : ; t 1: (9.2)
Since ˛ ù `s
` `1;u ù `1;s ` ` ù `t uCt 1
˛1s ˛t s 1t 1;u 1;s
D ;
˛sCt 1
we get
˛ t 1
uCt 1
X
ù `s h j`s ì;u ù ì;s j h.˛i /;
˛sCt 1
iD1
by Theorem B.5. We conclude by (9.2) that
˛ t 1
uCt 1
X
h " h.˛i /:
˛sCt 1
i D1
This proves the lemma.

Proof of Theorem 9.2. Let N 100 be an integer and put

& s '
1 log log N
tD ; L D t .1 C .9.t 1//t 1
/;
15 log log log N
and p
M D exp 29 log log N log log log N :
p minfb1 ; : : : ; bL g
Suppose that there are positive integers x1 ; : : : ; xL ; b1 ; : : : ; bL with
M such that x1b1 ; : : : ; xL
bL
are distinct and lie in the interval ŒN; N C N . Assume that
any t integers among x1b1 ; : : : ; xL
bL
are multiplicatively dependent. Then, by Lemma 9.4
1
applied with " D .9.t 1// , there exist distinct indices i0 ; : : : ; it for which
bi bi bi0
maxfxi0 0 ; xi1 1 g xi0 bi b " p
xi2 2 xitit .2N /".t 1/ 9
bi0 bi1
D exp h bi1
D 2N : (9.3)
gcd.xi0 ; xi1 / xi1
bi bi p
Since the integers xi0 0 and xi1 1 are lying in the interval ŒN; N C N , their greatest
p
prime divisor
p cannot exceed N . Consequently, the left hand side of (9.3) is at least
equal to N and we have derived a contradiction since N 4.
Thus, by reordering the powers, we may assume, without any loss of generality, that
x1 ; : : : ; xtbt are multiplicatively independent and that bi < bt for i D 1; : : : ; t 1. Put
b1
ƒi WD bi log xi bt log xt ; i D 1; : : : ; t 1;
and observe that

ˇ x bi ˇˇ 1 1
jƒi j D ˇlog ib ˇ log 1 C p <p ; i D 1; : : : ; t 1:
ˇ
(9.4)
xt t
N N
log jƒi j > 1010 .log xi /.log xt /.log bt /; i D 1; : : : ; t 1;
and, using that

xj .2N /1=bj N 2=M ; j D 1; : : : ; t; (9.5)
we get
log jƒi j > 1010 .log N 2=M /2 .log bt /; i D 1; : : : ; t

1: (9.6)
Since x1 ; : : : ; xt are multiplicatively independent and the .t 1/ t matrix
b1 0 0 bt
::
0 :
:: ::
: :
0 0 bt 1 bt
9.3. Simultaneous Pellian equations with at most one solution 121
has rank t 1, we can apply Theorem 9.1 to show that the lower bound (9.6) can be con-
siderably improved (when N is large) for (at least) one of the quantities jƒ1 j; : : : ; jƒt 1 j.
Indeed, since bt M 4, it follows from Theorem 9.1 and (9.5) that
log maxfjƒ1 j; : : : ; jƒt 1 jg

200t
t =.t 1/ 1=.t 1/ t
> .16t / log N 2=M t log log N 2=M log log N 2=M bt :
2 log N
Combined with (9.4) and the upper bound bt log 2
< 4 log N , we then get
2 log N t 2 log N 1=.t 1/ 2 log N t
log N < 2.16t /200t t log log 4.log N /
M M M
2 log N t 1=.t 1/
200t
2.t C 1/.16t / t log.2 log N / log.4 log N /;
M
which gives
M < .16t /200t .log N /1=t log.4 log N /:
By the definitions of tpand M , this cannot happen if N is large enough. Consequently,
the interval ŒN; N C N contains at most L integers of the form x r with r pM .
Furthermore, as already noticed, for any integer r with r 2, the interval ŒN; N C N
pmost one r-th power. We conclude that the total number of perfect powers in
contains at
ŒN; N C N is at most equal to L C M . This completes the proof.
9.3. Simultaneous Pellian equations with at most one solution

Our second application of Loxton’s result, which was obtained by Bennett and Pintér [58],
gives a quantitative upper bound for the number of solutions to certain systems of Pellian
equations. Classical results on quadratic fields are given in Appendix C. Recall that the
fundamental unit of a real quadratic field K is the unit in K such that > 1 and any unit
in K can be written ˙m , where m is a rational integer.
Theorem 9.5. Let a and b p be non-square p positive integers and let "a and "b denote
the fundamental units in Q. a/ and Q. b/, respectively. There exists an effectively
computable real number > 1 such that, if
log "b .log a/.log "a /.log log.2a"a //3 ; (9.7)
then the system of simultaneous Pellian equations
jx 2 ay 2 j D jy 2 bz 2 j D 1 (9.8)
has at most one solution in positive integers x; y; and z.

Proceeding as in the proof of Theorem 3.13, it is easy to show that, for any given
non-square positive integers a; b, the equations (9.8) have only finitely many solutions in
integers x; y; z.
In [58] the factor log log.2a"a / in (9.7) is replaced by log log.3"a /. The authors
overlooked in their proof that a can be much larger than "a , a case which may occur
when a has a large square factor. The result of [58] addresses, more generally, intersection
of recurrence sequences having a dominant root (the authors assumed that this root is
simple, but their argument also works if this is not the case).
Proof. Let denote by "a and "b the Galois conjugates of "a and "b , respectively. Note that
p
log "a log "b 1C 5
h."a / D ; h."b / D ; and "a ; "b :
2 2 2
Assume that (9.7) holds for some > 1 and that Equation (9.8) has two solutions in
positive integers, say .x1 ; y1 ; z1 / and .x2 ; y2 ; z2 /, with x1 < x2 . Then, there are positive
integers m1 ; m2 ; n1 , and n2 such that
"nai ." /ni "mi C ."b /mi
yi D pa D b ; i D 1; 2:
2 a 2
Let i D 1; 2. Observe that
"nai "mi p
p ¤ b ; ni 2mi ; and j"m i
b a
" ni a 1j 1:1"a 2ni ; (9.9)
2 a 2
by (9.7). It follows from (9.7) and (9.9) that the linear form
p
ƒi WD mi log "b ni log "a C log a
satisfies
log "b
log jƒi j < log 2:2 2ni log "a < : (9.10)
2
By (9.7),p the fundamental units "a and "b are distinct. Thus, the three algebraic numbers
"a ; "b ; a are multiplicatively independent. By applying Theorem 2.2, we get from (9.9)
that
log jƒi j .log a/.log "a /.log "b /.log 2ni /:
We conclude that
ni .log a/.log "b /.log 2ni /;
thus, by (9.7),
ni .log a/.log "b /.log log 2"b /: (9.11)
The matrix !
m1 n1 1
m2 n2 1
has rank two. Observe that the naïve height of "a is equal to "a C "a , thus is less than 2"a .
Likewise, the naïve height of "b is less than 2"b . By Theorem 9.1 applied with the quantity
D .log 2a/.log 3"a /.log 3"b /;
we deduce from (9.11) that
maxflog jƒ1 j; log jƒ2 jg . log /1=2 log maxfn1 ; n2 g
1=2 .log /3=2 :
9.4. Exercise 123
Combining this with (9.10) and (9.7), we get
log "b 1=2 .log /3=2 .log a/1=2 .log "a /1=2 .log "b /1=2 .log log "b /3=2 ;
whence
log "b .log a/.log "a /.log log "b /3 :
This shows that the system (9.8) cannot have two solutions if in (9.7) is taken sufficiently
large. This completes the proof of the theorem.
9.4. Exercise
Exercise 9.1 (Continuation of Exercise 3.7; see [104]). Use (3.36) and Theorem 9.1 to
prove that, if a; b; c are positive integers with a b > c, then
P Œa.ab C 1/.bc C 1/.ca C 1/ log log a:

acC1 .acC1/.bcC1/
[Hint. Consider the quantities ƒ1 WD ac
1 and ƒ2 WD c 2 .abC1/
1].
9.5. Notes
F Theorem 9.1 has been applied to Thue equations by Brindza [94], to unit equations
by Brindza and Győry [96], and to sets of integers whose shifted products are powers by
Stewart [397]; see the Notes at the end of Chapter 5.
F It does not seem that any application of simultaneous linear forms in p-adic logarithms
has been found yet.
F Corvaja and Zannier [157] and, independently and simultaneously, Hernández and
Luca [226] applied the Schmidt subspace theorem to show that the greatest prime factor
of .ab C 1/.bc C 1/.ca C 1/ tends to infinity as the maximum of the pairwise distinct
positive integers a; b; c tends to infinity. Actually, a stronger result is proved in [157],
namely that the greatest prime factor of .ab C 1/.ac C 1/ tends to infinity as the maximum
of the pairwise distinct positive integers a, b, and c goes to infinity; see also [130].
Chapter 10
Multiplicative dependence relations
between algebraic numbers
By definition, if ˛1 ; : : : ; ˛m are multiplicatively dependent algebraic numbers, then there

exist integers n1 ; : : : ; nm not all zero such that ˛1n1 ˛m
nm
D 1. The main aim of this
chapter is to show that n1 ; : : : ; nm can be chosen not too large in absolute value.
We largely follow the presentation of Waldschmidt [432] and begin with a brief study
of lower bounds for the height of an algebraic number expressed in terms of its degree.
10.1. Lower bound for the height of an algebraic number

By definition of the height, if ˛ is zero or a root of unity, then its height is equal to zero.
Kronecker’s Theorem B.6 asserts that these are the only algebraic numbers whose height
is zero. The next theorem, used in the proof of Theorem 10.5 and in various other parts
of this monograph, gives an explicit version of Theorem B.6.
Theorem 10.1. Let d be a positive integer and ˛ a non-zero algebraic number of degree
at most d . If ˛ is not a root of unity, then
1
h.˛/ > :
11d 3
A much larger lower bound for h.˛/ than the one given in Theorem 10.1 is known.
Namely, Louboutin [269] established that, for every positive real number ", there exists an
integer d0 ."/ such that every algebraic number of degree d greater than d0 ."/, and which
is not a root of unity, satisfies
9 " log log d 3

h.˛/ > : (10.1)
4d log d
This slight improvement of a seminal theorem of Dobrowolski [167] is currently the best
known result towards the celebrated Lehmer problem.
Problem 10.2 (Lehmer’s problem). To prove or to disprove that there exists a positive
absolute real number c such that every non-zero algebraic number ˛ which is not a root
of unity satisfies
c
h.˛/ > :
deg.˛/
126 10. Multiplicative dependence relations between algebraic numbers
A comprehensive list of partial results towards Lehmer’s problem is given in Section 3.6
of [432]; see also [143,306,383] and Chapter 16 of [287] for a complete proof of a slightly
weaker lower bound than (10.1).
We need two auxiliary lemmas for the proof of Theorem 10.1. The first one is the
following consequence of Fermat’s little theorem.
Lemma 10.3. Let p be a prime number and f .X1 ; : : : ; Xk / a polynomial in k variables
with integer coefficients. Then, every coefficient of the polynomial
f .X1p ; : : : ; Xkp / f .X1 ; : : : ; Xk /p
is divisible by p.
Proof. For non-negative integers i1 ; : : : ; ik and a non-zero integer a, the desired re-
sult holds for the monomial aX1i1 Xkik , since p divides the integer a ap . Let
f1 .X1 ; : : : ; Xk / and f2 .X1 ; : : : ; Xk / be integer polynomials such that every coefficient of
the polynomials
f1 .X1p ; : : : ; Xkp / f1 .X1 ; : : : ; Xk /p and f2 .X1p ; : : : ; Xkp / f2 .X1 ; : : : ; Xk /p
is divisible by p. Observe that the coefficients of the polynomial

pX1 p
!
p p
.f1 C f2 / p
f1 f2 D f hf p h
h 1 2
hD1
are rational integers which are all divisible by p. Consequently, every coefficient of the
polynomial p
.f1 C f2 /.X1p ; : : : ; Xkp / .f1 C f2 /.X1 ; : : : ; Xk /
is a multiple of p. The lemma follows by linearity.
We further need an easy lemma on roots of unity.
Lemma 10.4. Let ˛ be a non-zero algebraic number. Assume that there exist two distinct
positive integers h and ` such that ˛ h and ˛ ` are Galois conjugate. Then ˛ is a root of
unity.
Proof. Let K be the number field generated over Q by ˛ and its Galois conjugates. From
the assumption that ˛ h and ˛ ` are conjugate, we deduce that there exists an element '
in the Galois group of K over Q such that '.˛ h / D ˛ ` . By induction, we establish that
n n
' n .˛ h / D ˛ ` for n 1. Denoting by m the order of ' in the Galois group of K over Q,
m m
we deduce that ˛ h D ˛ ` . Since h and ` are distinct, we conclude that ˛ is a root of
unity.
Proof of Theorem 10.1. Let ˛ be a non-zero algebraic number of degree d . If ˛ is not
an algebraic integer, then its height is at least equal to logd 2 . Consequently, there is no
restriction in assuming that ˛ is an algebraic integer. Let ˛ D ˛1 ; ˛2 ; : : : ; ˛d denote its
Galois conjugates. Let ` be a positive integer and p a prime number. The sum
S` D ˛1` C C ˛d`
10.2. Existence of small multiplicative dependence relations 127
(which is the trace of ˛ ` from Q.˛/ to Q) is a rational integer. By Fermat’s little theorem,
S` and S`p are congruent modulo p. Furthermore, by applying Lemma 10.3 with k D d
for the polynomial f` D X1` C C Xd` , we deduce that there exists an integer polynomial
g` .X1 ; : : : ; Xd / such that
S`p S`p D p g` .˛1 ; : : : ; ˛d /:
Since g` .˛1 ; : : : ; ˛d / is an algebraic integer, it must be a rational integer and we deduce

that S`p and S`p are congruent modulo p. This shows that the three integers S`p , S` ,
and S`p are congruent modulo p. Recalling that, for j D 1; : : : ; d ,
d
1 X 1
h.˛/ D log maxf1; j˛h jg log maxf1; j˛j jg;
d d
hD1
we get j˛j j edh.˛/ , hence

jS` j d e`dh.˛/ :
We now assume edh.˛/ 1 C 1=.4ed 2 /. By Theorem D.1, we may assume that the prime
number p satisfies 2ed < p < 4ed . For ` D 1; : : : ; d , the estimates
1 d 1 4ed 2
jS` j d 1 C de and jS`p j d 1 C de
4ed 2 4ed 2
imply that
jS` S`p j 2d e < p:
Since S` and S`p are congruent modulo p, it follows that S` D S`p for ` D 1; : : : ; d . This
implies that ˛ and ˛ p have the same minimal defining polynomial, thus they are Galois
conjugates. We then deduce from Lemma 10.4 that ˛ is a root of unity.
Consequently, if ˛ is a non-zero algebraic number which is not a root of unity, then
edh.˛/ > 1 C 1=.4ed 2 /. Noticing that the inequality
1 11d 2
1C >e
4ed 2
holds for d 2, we deduce that
1 1 1
h.˛/ log 1 C > :
d 4ed 2 11d 3
This completes the proof of Theorem 10.1.
10.2. Existence of small multiplicative dependence relations

The main result of this section is the following theorem.
128 10. Multiplicative dependence relations between algebraic numbers
Theorem 10.5. Let m 2 be an integer and ˛1 ; : : : ; ˛m multiplicatively dependent non-

zero algebraic numbers. Let log ˛1 ; : : : ; log ˛m be any determination of their logarithms.
Let D be the degree of the number field generated by ˛1 ; : : : ; ˛m over Q. For j D 1; : : : ; m,
let Aj be a real number satisfying
n j log ˛j j o
log Aj max h.˛j /; ;1 :
D
Then, there exist integers n1 ; : : : ; nm , not all zero, such that
n1 log ˛1 C C nm log ˛m D 0; ˛1n1 ˛m

nm
D 1;
and
m 1 .log A1 / .log Am /
jnk j 11.m 1/D 3 ; for k D 1; : : : ; m. (10.2)
log Ak
Let ˛1 ; : : : ; ˛m be multiplicatively dependent algebraic numbers. The set G of integer
m-tuples .n1 ; : : : ; nm / such that ˛1n1 ˛m
nm
D 1 is a subgroup of Zm of rank at least
one. Theorem 10.5 provides us with an upper bound for the first minimum of the discrete
subgroup G in Rm .
m 1
Matveev [289,292] proved that the factor 11.m 1/D 3 in (10.2) can be replaced
by C m D log D, for some absolute real number C ; see also Loher and Masser [267] and
Loxton and van der Poorten [271].
Proof. We assume, as we may without loss of generality, that m 2, and that any m 1
elements among ˛1 ; : : : ; ˛m are multiplicatively independent. Thus there exists a unique
(up to a factor ˙1) set of relatively prime non-zero integers n1 ; : : : ; nm such that
n1 log ˛1 C C nm log ˛m D 0:
Fix an integer k with 1 k m. Define c1 ; : : : ; cm by
cj D 11.m 1/D 3 log Aj / 1

; 1 j m; j ¤ k;
and m 1
Y
ck D 11.m 1/D 3 log Aj ;
1j m;
j ¤k
in such a way that c1 cm D 1. Using Minkowski’s theorem (see e.g. [357, Chapter II]
or [113, Appendix B]), we deduce that there exist integers 1 ; : : : ; m , not all zero, such
that ˇ k nj ˇˇ
ˇ cj ; 1 j m; j ¤ k; and jk j ck :
ˇ
ˇj
nk
In order to prove the relation 1 log ˛1 C C m log ˛m D 0, we first show that the
number ˛ D ˛11 ˛m
m
is a root of unity. Observe that the number
m m
nk nk k nj
Y Y
˛ nk D ˛j j D ˛j j
j D1 j D1
10.2. Existence of small multiplicative dependence relations 129
satisfies X
jnk j h.˛/ jj nk k nj j h.˛j /;
1j m;
j ¤k
hence
X m 1 1
h.˛/ cj h.˛j / D :
11.m 1/D 3 11D 3
1j m;
j ¤k
It then follows from Theorem 10.1 that ˛ is a root of unity. Let M be the smallest
positive integer such that ˛ M D 1. Then, ˛ is a primitive M -th root of unity, hence of
degree ŒQ.˛/ W Q D '.M / over Q, where ' denotes Euler’s totient function. Write
M D q1u1 qrur , where q1 ; : : : ; qr are distinct prime numbers and u1 ; : : : ; ur are positive
integers. Then,
'.M / D q1u1 1 .q1 1/ qrur 1 .qr 1/
and
'.M / q1 1 qr 1 1
p D q1.u1 2/=2
.q1 1/ qr.ur 2/=2
.qr 1/ p ::: p p :
M q1 qr 2
Since the degree of ˛ is at most equal to D, we get '.M / D, giving that M 2D 2

(note that sharper estimates hold; see Exercise 7.2.a of [432]) and
m
X
M j log ˛j is in 2iZ: (10.3)
j D1
Observe that
m m
ˇ X ˇ ˇ X k nj ˇ
j log ˛j ˇ D ˇM j log ˛j ˇ
ˇ ˇ ˇ ˇ
nk
ˇM
j D1 j D1
X M 2
M cj j log ˛j j < 2:
11D 2 11
1j m;
j ¤k
Combined with (10.3), this shows that

m
X
j log ˛j D 0:
iD1
Therefore, there exists a non-zero rational integer ` such that .1 ; : : : ;m / D .`n1 ; : : : ;`nm /.
We deduce the inequality jnk j jk j, from which the desired upper bound jnk j ck
readily follows.
Chapter 11
in two complex logarithms: proofs
The purpose of this chapter is to present complete proofs of versions of Theorems 2.3,
2.4, and 2.6, with the same dependence in all the parameters, but with larger numerical
constants. Not to focus on the size of these constants allows some simplification in the
proofs.
11.1. Three estimates for linear forms in two complex logarithms

Let ˛1 ; ˛2 be non-zero complex algebraic numbers and log ˛1 ; log ˛2 any determinations
of their logarithms. Set D D ŒQ.˛1 ; ˛2 / W Q. Let A1 ; A2 be real numbers with
n j log ˛j j 1 o
log Aj max h.˛j /; ; ; j D 1; 2:
D D
Let b1 ; b2 be non-zero integers such that b1 log ˛1 C b2 log ˛2 is non-zero and set
jb1 j jb2 j
B0 D C :
D log A2 D log A1
Theorem 11.1. Under the above notation and assumption, we have the lower bound
log jb1 log ˛1 C b2 log ˛2 j 21600 D 4 .log A1 /.log A2 /.maxf10; log B 0 g/2 :
A complete proof of a version of Theorem 11.1 when ˛1 and ˛2 are real numbers was
given by Laurent [252]; see also [253, 254] for the complex case and some improvements.
As will be apparent in its proof, Theorem 11.1 can be considerably sharpened when
˛1 and ˛2 are both real numbers very close to 1.
Theorem 11.2. Keep the above notation and assumption. Suppose furthermore that ˛1
and ˛2 are real numbers greater than 1 and set
n D log A D log A o
1 2
E D 1 C min ; :
log ˛1 log ˛2
Assume that E minfA3D=2

1 ; A3D=2
2 g. Then, we have the lower bound
.log A1 /.log A2 / n 3 log B 0 o2
log jb1 log ˛1 C b2 log ˛2 j 78500 D 4 max ; :
log E D log E
132 11. Lower bounds for linear forms in two complex logarithms: proofs
By taking E D 2 in Theorem 11.2, we get a slightly weaker version of Theorem 11.1.

Unlike in Theorem 2.3, we do not assume in Theorem 11.1 that ˛1 and ˛2 are
multiplicatively independent. In particular, Theorem 11.1 applied with ˛1 D 1 and

log A1 D D implies the following result, which is similar to Theorem 2.6.
Theorem 11.3. Let ˛ be an algebraic number on the unit circle and which is not a root of
unity. Let D be its degree and A a real number such that
n j log ˛j 1 o
log A max h.˛/; ; :
D D
Let b1 ; b2 be non-zero integers and set
jb1 j jb2 j
B0 D C :
D log A
Then,
log jb1 i b2 log ˛j 68000 D 3 .log A/.maxf10; log B 0 g/2 :
Apart from the value of the numerical constant, Theorem 11.3 is as strong as Theo-
rem 2.6.
11.2. An auxiliary inequality involving several parameters

For convenience, we slightly change the notation of Theorem 11.1. Without loss of
generality, we assume that b1 and b2 are positive and that the moduli of ˛1 and ˛2 are at
least equal to 1. We then write in all what follows
ƒ D b2 log ˛2 b1 log ˛1 :
This is not restrictive. Indeed, if j˛1 j and j˛2 j are both less than 1, then replacing ˛1
and ˛2 by their inverses does not affect jƒj. Also, if j˛1 j < 1 and j˛2 j > 1, then the real
part of ƒ is the sum of two positive terms and can easily be bounded using Liouville’s
inequality Theorem B.10. Furthermore, by permuting the indices 1 and 2 if necessary, we
can suppose that
b1 j log ˛1 j b2 j log ˛2 j: (11.1)
We keep this assumption until the end of this chapter.
Theorem 11.4. Let K 3, L, R1 , R2 , S1 , and S2 be integers at least equal to 2. Set
N D KL, R D R1 C R2 1, and S D S1 C S2 1. Assume that the conditions
Cardf˛1r ˛2s W 0 r < R1 ; 0 s < S1 g L; (11.2)

Cardfrb2 C sb1 W 0 r < R2 ; 0 s < S2 g > .K 1/L (11.3)
are satisfied. Set

KY1 2=.K 2 K/

b D .R 1/b2 C .S 1/b1 kŠ : (11.4)
kD1
11.3. Proof of Theorem 11.4 133
Let > 1 be a real number. If the inequality
.N K/ log > 2.D C 1/ log N C .D C 1/K log b

C 2LR j log ˛1 j C D h.˛1 / C 2LS j log ˛2 j C D h.˛2 / (11.5)
holds, then we have

n ejƒjLR=b1 LR ejƒjLS=b2 LS o
N
jƒj max ; > :
b1 b2
Theorem 11.4 is slightly weaker than Théorème 1 of [254]. The next section is devoted
to its proof. Then, we deduce Theorems 11.1 and 11.2 from Theorem 11.4 in Sections 11.4
and 11.5.
Let us briefly discuss (11.5) and convince the reader that it can be satisfied, even for
large values of . Assume for convenience that ˛1 or ˛2 is not a root of unity. Then, (11.2)
holds for R1 D S1 D L. Assuming that K L > 10, we see that (11.5) holds if
K.L 1/ log > 2.D C 1/K log b

C 2. C 1/DL .L C R2 / log A1 C .L C S2 / log A2 : (11.6)
This forces L to be D log b. By assumption (11.3), the product R2 S2 exceeds K.L 1/.
Thus, the left hand side of (11.6) involves the product R2 S2 , while its right hand side is
linear in R2 and S2 . This shows that (11.6) can be satisfied, even for large values of .
For instance, we can take
R2 DL.log A2 /; S2 DL.log A1 /; and K 2 D 2 L.log A1 /.log A2 /;
keeping in mind that the conclusion of Theorem 11.4 is better when KL is small.
To replace (11.5) by (11.6), we have bounded j log ˛1 j (resp., j log ˛2 j) from above
by D log A1 (resp., D log A2 ). If ˛1 and ˛2 are real numbers greater than 1, then,
by taking equal to the quantity E defined in Theorem 11.2, we get the stronger upper
bounds log ˛1 2D log A1 and log ˛2 2D log A2 , showing that the factor 2. C 1/
occurring in (11.6) can then be replaced by 6. For large values of , the inequality

K.L 1/ log > 2.D C 1/K log b C 6DL .L C R2 / log A1 C .L C S2 / log A2 :
is clearly much easier to be satisfied than (11.6). This roughly explains why the lower
bound in Theorem 11.1 can be considerably improved when the parameter E can be taken
to be large.
11.3. Proof of Theorem 11.4

We keep the notation of Section 11.2 and Theorem 11.4. Assume that the inequalities (11.2)
and (11.3) are satisfied. Observe that
RS R1 S1 C R2 S2 > KL:
It follows from Lemma E.1, applied with

E D f.r; s/ W 0 r < R1 ; 0 s < S1 g
and E 0 D f.r 0 ; s 0 / W 0 r 0 < R2 ; 0 s 0 < S2 g;
that the KL RS matrix M whose coefficients are the complex numbers
!
rb2 C sb1
˛1 `r ˛2 `s
k
is of maximal rank, equal to N D KL. To see this, observe that if the KL lines of M
were linearly dependent, then there would exist a non-zero integer polynomial P .X; Y /,
whose degree in X is at most K 1 and whose degree in Y is at most L 1, vanishing
at all the points .rb2 C sb1 ; ˛1r ˛2s / with 0 r < R; 0 s < S . This would contradict
Lemma E.1.
Thus, we can extract from M a non-zero minor of size N N , which can be written,
with a suitable ordering of the columns,
! !
rj b2 C sj b1 ` i rj ì sj
D det ˛1 ˛2 :
ki 1i;j N
We index the lines by pairs .k; `/ and the columns by pairs .r; s/. For convenience, we fix
the numbering of the lines by using the formulas
ki D i 1 Kb.i 1/=Kc; ì D b.i 1/=Kc; for i D 1; : : : ; KL.
The columns are indexed by the N distinct pairs of integers .r1 ; s1 /; : : : ; .rN ; sN /, where
0 rj < R and 0 sj < S for j D 1; : : : ; N . In the sequel, we will use repeatedly that,
for any permutation of the set f1; : : : ; N g, we have
N
X N
X
ì r .i / LRN and ì s .i / LSN: (11.7)
i D1 i D1
We stress that it is very important to use, in the definition of the determinant , the
binomial coefficients rb2 Csb , and not the integers .rb2 C sb1 /k . Indeed, otherwise, we
1

k
would have to take b equal to Rb2 C S b1 and we would ultimately get a weaker version
of Theorem 11.1 with, in particular, B 0 replaced by maxf3; jb1 j; jb2 jg. The idea to use
binomial coefficients rather than integral powers goes back to Feldman [185].
The strategy of the proof goes as follows. Recall that, by construction, is non-zero.
We bound jj from below by means of Liouville’s inequality (Theorem B.10) and from
above using complex analysis (namely, the maximum principle). Then, we show that the
lower bound exceeds the upper bound if jƒj is small enough. This contradiction implies
that jƒj cannot be too small.
As explained in [252], the determinant is the interpolation determinant of the N
functions in the two variables x; y,
b2ki ki ì x ì y
'i .x; y/ D x ˛1 e ; 1 i N;
ki Š
evaluated at the N points

b1 sj
rj C sj ;ƒ ; 1 j N:
b2 b2
To bound such a determinant, the general method consists in expanding it as a Taylor series
(around the origin or any other point) of the 2N variables xj ; yj ; 1 j N , determined
by the coordinates of the given N points. In our special case, the second coordinate y is
small, so it appears to be sufficient to expand 'i .x; y/ at order 1 in y. The source of the
numerical improvement in [253] is precisely the use of the whole Taylor series expansion.
Proceeding as in [252] and [254], we introduce the quantity
jƒjLR=b1
LR ejƒjLS=b2 LS

e
ƒ0 WD ƒ max ; :
b1 b2
Observe that
ƒ b1
log ˛2 D ˇ log ˛1 C ; with ˇ D :
b2 b2
Since
!
rj b2 C sj b1 b2ki
D .rj C sj ˇ/ki C (terms of degree < ki ); 1 i; j n;
ki ki Š
the multilinearity of the determinant implies that
ki
b2 ki ` i rj ì sj
D det .rj C sj ˇ/ ˛1 ˛2 :
ki Š 1i;j N
Let i; j be integers with 1 i; j N and write

˛1 ì rj ˛2 ì sj D ˛1 ì .rj Csj ˇ / eì sj ƒ=b2 D ˛1 ì .rj Csj ˇ / .1 C ƒ0 i;j /; (11.8)
where
eì sj ƒ=b2 1
i;j WD
ƒ0
satisfies
b2 b2
ji;j j jeì sj ƒ=b2 1j .eì sj jƒj=b2 1/;
LS jƒjeLSjƒj=b2 LS jƒjeLSjƒj=b2
since
jeu 1j ejuj 1; for every complex number u:
Then, using that
eì sj jƒj=b2 1 eLS jƒj=b2 1 and eu 1 ueu ; for u 0;
we deduce that
ji;j j 1:
The fact that all the i;j are bounded by 1 in modulus explains why it is better to work
with ƒ0 in place of ƒ.
We see from (11.8) that the determinant tends to 0 when ƒ0 tends to 0; see (11.12)
below.
– Arithmetic lower bound for jj. We apply the Liouville inequality Theorem B.10 to
bound jj from below. Our lower bound, which involves the quantity b defined in (11.4),
is slightly less precise than Lemme 6 of [254].
Lemma 11.5. We have
log b
log jj DN log N D.K 1/N DLN Rh.˛1 / C S h.˛2 / : (11.9)
2
Proof. Consider the integer polynomial
N
!
X Y r .i / b2 C s .i / b1 PN
ì r .i /
PN
ì s .i /
P .X; Y / WD sgn. / X iD1 Y iD1 ;
2SN i D1
ki
where sgn. / denotes the signature of the permutation . Using the obvious bound
rj b2 C sj b1 .R 1/b2 C .S 1/b1 ; j D 1; : : : ; N;
we observe that the length L.P / of the polynomial P .X; Y / (see Definition B.9) satisfies
N
k1 CCkN Y 1
L.P / N Š .R 1/b2 C .S 1/b1 ki Š :
iD1
.K 1/N
Noticing that k1 C C kN D 2
, we get the upper bound
L.P / N N b .K 1/N=2
; (11.10)
with b defined in (11.4). Since, by (11.7), the degrees in X and in Y of P .X; Y / do not
exceed LRN and LSN , respectively, it follows from (11.10) and Theorem B.10 that
log b
log jP .˛1 ; ˛2 /j DN log N D.K 1/N DLN Rh.˛1 / C S h.˛2 / :
2
Since D P .˛1 ; ˛2 /, this proves the lemma.
– Analytic upper bound of jj.

Lemma 11.6. Let > 1 be a real number. If jƒ0 j N
, then
N.N K/ log KN log b

log jj C N log N C
2 2 (11.11)
C LN Rj log ˛1 j C S j log ˛2 j :
Proof. Plugging (11.8) in the expression of the determinant , we get

X
D .ƒ0 /N jI j
I ; (11.12)
I f1;:::;N g
where the summation is taken over all the 2N subsets I of f1; : : : ; N g and
ˇ 'i .z1 / 'i .zN /

i 2I
I WD det ;

i …I
i;1 'i .z1 / i;N 'i .zN /
with
z ki ì z
'i .z/ D b2ki ˛ ; zj D rj C ˇsj ; 1 i; j N:
ki Š 1
For a subset I of f1; : : : ; N g define the entire function Î of the variable x by
ˇ 'i .xz1 / 'i .xzN /

i 2I
Î .x/ WD det (11.13)

i …I
i;1 'i .xz1 / i;N 'i .xzN /
and observe that I D Î .1/. For i D 1; : : : ; N , let

X
'i .z/ D pi;n z n
n0
be the Taylor series expansion of the function 'i . Plugging these expressions into (11.13),
ˇ
we obtain
z1ni ni
zN
i 2I
X Y
ni
Î .x/ D pi;ni x det ;
ni Wi2I i 2I

i …I
i;1 'i .xz1 / i;N 'i .xzN /
where the summation is taken over all the jI j-tuples .n1 ; : : : ; njI j / of non-negative integers.
If an jI j-tuple is such that there are distinct indices h; k in I for which nh D nk , then
the rows h and k in the determinant are equal, thus the corresponding term in the sum is
zero. Consequently, the summation is taken over all the jI j-tuples of distinct non-negative
integers. For such an jI j-tuple I , we have
X jI j2 jI j
ni 0 C C .jI j 1/ D :
2
i 2I
Consequently, the function Î has at the origin a zero of multiplicity at least .jI j2 jI j/=2.
Lemma F.1 then implies that
.jI j2 jI j/=2
jI j D jÎ .1/j max jÎ .x/j:
jxjD
Using this for every subset I of f1; : : : ; N g and inserting the upper bound jƒ0 j N
in (11.12), we obtain
N.N i / .i 2 i /=2
jj 2N

max max max jÎ .x/j
0i N I jxjD
.N 2 N /=2
2N max max jÎ .x/j;
I jxjD
where the last but one maximum is taken over all the subsets I of f1; : : : ; N g.
To bound jÎ .x/j from above, we expand the determinant Î .x/ and use that
ji;j j 1. Then, for every complex number x of modulus , we have
N
1/b2 C .S 1/b1 //ki

Y .jxj..R
jÎ .x/j N Š
ki Š
iD1
XN
max exp ì .r .i / C ˇs .i / /jxj j log ˛1 j

i D1
N Š .b/.K 1/N=2

exp LRN j log ˛1 j C LSN j log ˛2 j ;
by (11.1), (11.7), and (11.4). Since N 6, we get 2N N Š N N and we obtain
N 2 log N log .K 1/N log .K 1/N log b

log jj C C N log N C C
2 2 2 2
C LN Rj log ˛1 j C S j log ˛2 j :
This completes the proof of Lemma 11.6.
2
– Completion of the proof of Theorem 11.4. Multiplying (11.9) and (11.11) by N
, we see
that jƒ0 j > N holds as soon as
.N K/ log > 2.D C 1/ log N C .D C 1/K log b

C 2LR. j log ˛1 j C D h.˛1 // C 2LS. j log ˛2 j C D h.˛2 //:
This establishes Theorem 11.4.
11.4. Deduction of Theorem 11.1 from Theorem 11.4
First, we give an upper bound for the quantity b defined by (11.4).

Lemma 11.7. For any K 2, the quantity b satisfies
KY1 2=.K 2 K/
Rb2 C S b1
b D .R 1/b2 C .S 1/b1 kŠ 5 : (11.14)
K 1
kD1
11.4. Deduction of Theorem 11.1 from Theorem 11.4 139
Proof. For K D 2; : : : ; 7, this is an immediate computation. Note that, for any integer
k 2, we have
k
X Z k
log kŠ D log j log t dt D Œt log t t k1 k log k k
j D1 1
and
k k
1 2 k2 k2
X Z
j log j t log t dt Œ2t log t t 2 k1 log k :
1 4 2 4
j D1
Thus, we have the lower bound

KY1 2 KY1 2 KY1 2
K2 K K2 K K2 K
k k
kŠ e k
kD1 kD1 kD1
1/2 log.K .K 1/2

1 .K 1/
e exp
K2 K 2.K 2 K/
1 .K 1/=.2K/ 1=K K 1
e e .K 1/ .K 1/ ;
5
for K 8. This completes the proof of the lemma.
We proceed as in [252] to choose our parameters. The parameter is usually called

the radius. It could be a priori taken very large, but in view of the terms j log ˛1 j
and j log ˛2 j in the right hand side of (11.5), this inequality cannot be satisfied if is
too large, unless j log ˛1 j and j log ˛2 j happen to be simultaneously very small. This is the
case when ˛1 and ˛2 are both very close to 1 and this will be considered separately in the
next section.
By the definitions of A1 and A2 , we get j log ˛1 j D log A1 , j log ˛2 j D log A2 ,
and it follows from Theorem 11.4 that jƒ0 j > N holds as soon as
K.L 1/ log > 2.D C 1/ log N C .D C 1/K log b

(11.15)
C 2L. C 1/D.R log A1 C S log A2 /:
Set B D maxf10; log B 0 g, with B 0 defined in Section 11.1, and
K D b4900BD 3 .log A1 /.log A2 /c; L D b4BDc; R1 D d4BD 2 .log A2 /e;

S1 D d4BD 2 .log A1 /e; R2 D d140BD 2 .log A2 /e; S2 D d140BD 2 .log A1 /e:
Note that, by (11.4),

.R1 C R2 /b2 C .S1 C S2 /b1
log b log 5 C log
K 1
(11.16)
b2 b1
log C B:
D log A1 D log A2
Observe that R1 L and S1 L, since we have assumed D log A1 1 and

D log A2 1. Thus, if ˛1 and ˛2 are not both roots of unity (in which case the proof
of the theorem follows straightforwardly from Theorem B.10), then condition (11.2) is
satisfied. Observe also that R2 S2 > .K 1/L. Consequently, if we furthermore assume
that all the elements of the set
frb2 C sb1 W 0 r < R2 ; 0 s < S2 g
are distinct, then condition (11.3) is satisfied.

We check that (11.15) is satisfied with the choice D 3. Thus, Theorem 11.4 implies
the lower bound
log jƒ0 j N log 21550 D 4 .log A1 /.log A2 /B 2 :
By the definition of ƒ0 and the crude upper bound
1 C log L C log R C log S D 4 .log A1 /.log A2 /B 2 ;
we conclude that
log jƒj 21600 D 4 .log A1 /.log A2 /B 2 ; (11.17)
provided that (11.3) is satisfied.
It remains for us to consider the case where (11.3) is not satisfied. Assume that there
exist integers r and s such that jrj R 1, jsj S 1 and rb2 C sb1 D 0. We then have
b1
jƒj D js log ˛2 C r log ˛1 j: (11.18)
r
It follows from Theorem B.10 that
j˛1r ˛2s

1j D log 2 C Rh.˛1 / C S h.˛2 / ;
which, combined with (11.18), implies a lower bound sharper than (11.17). This concludes
the proof of the theorem.

We assume that ˛1 and ˛2 are real numbers greater than 1. We have established at the end
of Section 11.3 that jƒ0 j > N holds as soon as
K.L 1/ log 2.D C 1/ log N C .D C 1/K log b

C 2LR. log ˛1 C D h.˛1 // C 2LS. log ˛2 C D h.˛2 //: (11.19)
To derive (11.15) in the previous section, we have bounded D h.˛j / and log ˛j by
D log Aj , for j D 1; 2. This is far from being sharp when log ˛j is very small compared
to D h.˛j /, in which case the radius can be chosen quite large.
We introduce a new parameter E defined by

D log A1 D log A2
E D 1 C min ; ;
log ˛1 log ˛2
and, using the inequality log ˛j D log Aj , for j D 1; 2, we check that
E log ˛1 2D log A1 ; E log ˛2 2D log A2 :
We assume that
E minfA3D=2
1 ; A3D=2
2 g: (11.20)
With D E, the condition (11.19) is satisfied if
2D
K.L 1/ > 2 log N C K log b C 3L.R log A1 C S log A2 / ;
log
since D C 1 2D.
Then, we choose
n log o
B D max 3 ; log B 0 (11.21)
D
and, since log b B, Inequality (11.19) is satisfied if
2D
K.L 1/ > 2 log N C KB C 3L.R log A1 C S log A2 / : (11.22)
log
Set
D
D D
log
and observe that, by (11.20) and (11.21), we have
2
D minflog A1 ; log A2 g ; D B 3:
3
Set
K D b4900B.D /3 .log A1 /.log A2 /c; L D b16BD c;

R1 D d32B.D /2 .log A2 /e; S1 D d32B.D /2 .log A1 /e;
R2 D d280B.D /2 .log A2 /e; S2 D d280B.D /2 .log A1 /e:
We check that (11.22) is satisfied. Consequently, Theorem 11.4 implies the lower bound
log jƒ0 j N log 78450.D /4 .log A1 /.log A2 /B 2 .log /:
By the definition of ƒ0 and the crude upper bound
1 C log L C log R C log S .D /4 .log A1 /.log A2 /B 2 .log /;
we conclude that
n log o2
log jƒj 78500D 4 .log A1 /.log A2 / max 3 ; log B 0 .log / 3
;
D
provided that (11.3) is satisfied. We argue as at the end of Section 11.4 if (11.3) is not
satisfied. This ends the proof of Theorem 11.2.
Chapter 12
in two p-adic logarithms: proofs
The purpose of this chapter is to present complete proofs of versions of Theorems 2.12
and 2.13, with the same dependence in all the parameters, but with larger numerical
constants. Not to focus on the size of these constants allows some simplification in the
proofs.
12.1. Three estimates for linear forms in two p-adic logarithms

Let p be a prime number and ˛1 and ˛2 complex algebraic numbers. Let p be a prime
ideal of the ring of integers of Q.˛1 ; ˛2 / lying above p. We denote by vp the p-adic
valuation on Q.˛1 ; ˛2 / induced by vp and which extends the p-adic valuation vp on Q
normalized such that vp .p/ D 1.
The next theorem gives an upper bound for the p-adic valuation of the difference
between integral powers of ˛1 and ˛2 . It is a corollary to the main result of [129]. As
noted below Theorem 2.12, the assumption vp .˛1 / D vp .˛2 / D 0 is not restrictive.
Theorem 12.1. Let p be a prime number. Let ˛1 and ˛2 be multiplicatively independent
complex algebraic numbers with vp .˛1 / D vp .˛2 / D 0. Set D D ŒQ.˛1 ; ˛2 / W Q.
Let A1 and A2 be real numbers with
n 1o n 1o
log A1 max h.˛1 /; ; log A2 max h.˛2 /; :
D D
Let b1 ; b2 be non-zero integers and set
jb1 j jb2 j
B0 D C :
D log A2 D log A1
pD 1 4
vp .˛1b1 ˛2b2 / 132000 D log A1 log A2
.log p/4
n 10 log p o2
max log B 0 C log log p; :
D
144 12. Lower bounds for linear forms in two p-adic logarithms: proofs
As in Theorem 11.1, the factor log B 0 occurs with an exponent two: this is a conse-
quence of the method.
Unlike in Theorem 2.3, we do not assume in Theorem 12.1 that log A1 and log A2
exceed logDp , but only that they exceed D
1
. This refinement, which was first noticed by
Yamada [442], allows us to deduce Theorem 12.3 below directly from Theorem 12.1.
Theorem 12.2 below refines Theorem 12.1 in the special case where ˛1 and ˛2 are
multiplicatively independent non-zero rational numbers which are p-adically close to 1.
It involves an extra parameter E, which plays a similar rôle as the parameter also called E
in Theorem 11.2.
Theorem 12.2. Let yx11 and yx22 be non-zero multiplicatively independent rational numbers.
Let b1 and b2 be non-zero integers. Let p be a prime number. Assume that there exist a
positive integer g and a real number E 3 such that
x g
1
vp 1 E
y1
and at least one of the two following conditions
x g
2
vp 1 E
y2
or
x g
2
vp 1 1 and vp .b2 / vp .b1 /
y2
is satisfied. Let A1 ; A2 be real numbers with
log Aj maxflog jxj j; log jyj j; E log pg; j D 1; 2;
and put
jb1 j jb2 j
B0 D C :
log A2 log A1
Then we have the upper estimate
x b1 x b2 g
1 2
vp 88000 log A1 log A2
y1 y2 E 3 .log p/4
2
maxflog log p C log B 0 ; 10E log pg :
We display an important consequence of Theorem 12.1, which is a key ingredient in

the proof of a conjecture of Erdős given in Chapter 7.
Theorem 12.3. Let p be an odd prime number and a; b integers with a > b 1. If p
does not divide ab, then
p
vp .ap 1
bp 1
/ 132 105 .log 5a/ .log 5b/:
.log p/2
A similar result was first established by Yamada [442] by means of a slight modification
of the choice of the auxiliary parameters in [129]. Theorem 12.3 is proved in Section 12.6.
12.2. Two auxiliary inequalities involving several parameters 145
12.2. Two auxiliary inequalities involving several parameters

Keep the notation and assumption of Theorem 12.1. Assume also that b1 and b2 are
positive. This is not at all restrictive since ˛1 and ˛2 are p-adic units and the height
of a non-zero algebraic number is equal to the height of its inverse. Our results depend
among other things on several parameters related to ˛1 ; ˛2 , and p. We denote by e the
ramification index of p and by t the non-negative integer satisfying the inequalities
3e
pt 1
< pt : (12.1)
2.p 1/
Put D D ŒQ.˛1 ; ˛2 / W Qj. Let ‰ denote the isomorphism between the completion
of Q.˛1 ; ˛2 / with respect to p and a sub-field of Cp , which we write Qp .˛1 ; ˛2 /. Then,
we have vp .˛/ D vp .‰.˛// for every non-zero ˛ in Q.˛1 ; ˛2 /; see Section B.1.
Let g denote the smallest positive integer such that
vp .˛1g 1/ > 0 and vp .˛2g 1/ > 0:
The assumption vp .˛1 / D vp .˛2 / D 0 ensures that g exists. We recall that g divides
p D 1 (actually, g divides p f 1, where f is the residue degree of p; see Lemma F.4).
S p . Let m1 ; m2 be integers such that, defining 1
We fix a g-th primitive root of unity in Q
and 2 by
˛1 D m1 1 and ˛2 D m2 2 ;
we have vp .1 1/ D vp .˛1g 1/ > 0 and vp .2 1/ D vp .˛2g 1/ > 0. By Lemma F.3,
the choice of t ensures that
t 1 t 1
vp .1p 1/ > ; vp .2p 1/ > :
p 1 p 1
The main result of [129] is the following.
Theorem 12.4. Let K 3, L, R1 , R2 , S1 , S2 be integers at least equal to 2. Set
R D R1 C R2 1; S D S1 C S2 1, and N D KL. Set
KY1 2=.K 2 K/

b D .R 1/b2 C .S 1/b1 kŠ :
kD1
Let t be defined by (12.1). Assume that there are two residue classes c1 and c2 modulo g
such that
˚ t t
Card ˛1p r ˛2p s W 0 r < R1 ; 0 s < S1 ; m1 r C m2 s c1 modulo g L;

˚
Card rb2 C sb1 W 0 r < R2 ; 0 s < S2 ; m1 r C m2 s c2 modulo g > .K 1/L:
(12.2)
Assume furthermore that

.N K/.log p/ > 3D 2 log N C .K 1/ log b C LRh.˛1 / C LS h.˛2 / : (12.3)
Denoting by p u the greatest power of p which divides simultaneously b1 and b2 , we have
vp .˛1b1 ˛2b2 / 2KL C u:

The next theorem is a variant of Theorem 12.4 and a version of it was established
in [109]. It is of interest in the particular case where ˛1g and ˛2g are p-adically very close
to one.
Theorem 12.5. Let K, L, R1 , R2 , S1 , S2 , R, S , N , and b be as in the statement of
Theorem 12.4. Assume that there exist two residue classes c1 and c2 modulo g such
that (12.2) holds, where t is defined by (12.1). Denote by p u the greatest power of p which
divides simultaneously b1 and b2 . Assume furthermore that there exists a real number E
such that
1
vp .˛1g 1/ E > 1 C
p 1
and that either
1
vp .˛2g 1/ E > 1 C ;
p 1
or
b2
vp .˛2g 1/ > 0 and p does not divide :
pu
Under the condition
.N K/.E C t 1/.log p/ > D 2 log N C .K 1/ log b

C 2LRp t h.˛1 / C 2LSp t h.˛2 / ;

we have
vp .˛1b1 ˛2b2 / < .E C t /KL C u:
If ˛1 and ˛2 are rational numbers, then the condition
.N K/.E 1/.log p/ > 2 log N C .K 1/ log b C 4LR h.˛1 / C 4LS h.˛2 /
implies that
vp .˛1b1 ˛2b2 / < .E C 1/KL C u:
Our assumption on E in Theorem 12.5 is slightly stronger than in [109]. This is due to
the fact that we do not use the Schur polynomials as in [129]. The quantity E is analogous
to the parameter log in Chapter 11.
In comparison with the complex case, an extra difficulty in the p-adic setting is caused
by the fact that the radius of convergence of the p-adic exponential function is p 1=.p 1/ ,
thus it is not infinite as for the complex exponential function. Consequently, to be able to
apply p-adic analysis, we have first to find some trick in order to place us within the disc
of convergence of the p-adic exponential function. A naïve answer consists in working
with vp .˛1g ˛2g / instead of vp .˛1 ˛2 /, in which case no congruence conditions would
be needed in (12.2), but ˛1 and ˛2 would have to be replaced by their g-th power in (12.3).
Consequently, we would have a factor gL.Rh.˛1 / C S h.˛2 // and, ultimately, an upper
bound for vp .˛1 ˛2 / involving g 2 and not g (since g divides p D 1, we would get
eventually a bound involving .p D 1/2 ). A more clever way to proceed consists in
working with vp .1 2 /, since 1 and 2 are principal units. The price to pay is then
the congruence conditions occurring in (12.2), which impose the inequalities R1 S1 gL
and R2 S2 > g.K 1/L and, ultimately, cost (only) a factor g. This explains the presence
of the factor p D 1 in Theorem 12.1.
To remove this factor for abitrary ˛1 and ˛2 , or to replace it by a quantity significantly
smaller, remains a major open problem. However, the preceding discussion shows that
the factor p D 1 can indeed be removed when ˛1 and ˛2 are both principal units (that is,
if g D 1). If, moreover, these algebraic numbers are p-adically very close to 1, then the
functions z 7! ˛1z and z 7! ˛2z are analytic on a large disc (see Lemma F.7) and we can
take advantage of this to improve the conclusion of Theorem 12.1 in this particular case:
this is precisely the purpose of Theorem 12.2.
12.3. Proofs of Theorems 12.4 and 12.5

We keep the notation of Section 12.2 and assume that all the hypotheses of Theorem 12.4
until (12.2) are satisfied.
Our proof is very close to that of Theorem 11.3 and, at one step, it differs from the
proof in [129], where the authors used Schur polynomials instead of the Schwarz lemma
to obtain the analytic estimate. We have decided not to include this improvement, which
is however crucial for obtaining very good numerical constants. For this reason, we had
to define the quantity t in a slightly different way as in [129].
Observe that RS > KL. By Lemma E.1 applied with
˚
E D .r; s/ W 0 r < R1 ; 0 s < S1 ; m1 r C m2 s c1 modulo g
E 0 D .r 0 ; s 0 / W 0 r 0 < R2 ; 0 s 0 < S2 ; m1 r 0 C m2 s 0 c2 modulo g ;
˚
and
the KL RS matrix M whose coefficients are the complex numbers

!
rb2 C sb1 t t
˛1 p `r ˛2 p `s
k
is of maximal rank, equal to N D KL. To see this, observe that if the KL lines of M
were linearly dependent, then there would exist a non-zero integer polynomial P .X; Y /,
whose degree in X is at most K 1 and whose degree in Y is at most L 1, vanishing at
t t
all the points .rb2 C sb1 ; ˛1p r ˛2p s / with 0 r < R, 0 s < S . This would contradict
Lemma E.1.
We can thus extract from M a non-zero minor of size N N , which can be written,
with a suitably ordering of the columns,
!
rj b2 C sj b1

p t ì rj p t ` i sj
D det ˛1 ˛2 :
ki 1i;j N
We index the lines by pairs .k; `/ and the columns by pairs .r; s/. For convenience, we fix
the numbering using the formulas
ki D i 1 Kb.i 1/=Kc; ì D b.i 1/=Kc; for i D 1; : : : ; KL:

The columns are indexed by the N distinct pairs of integers .r1 ; s1 /; : : : ; .rN ; sN /, where
0 rj < R and 0 sj < S for j D 1; : : : ; N . In the sequel, we will use repeatedly that,
for any permutation of the set f1; : : : ; N g we have
N
X N
X
ì r .i / LRN and ì s .i / LSN: (12.4)
i D1 iD1
The strategy of the proof goes as follows. By construction, we know that is non-zero.
We bound vp ./ from above by means of Liouville’s inequality Theorem B.10 and from
below by using p-adic analysis, namely the p-adic maximum principle. Then, we show
that the lower bound exceeds the upper bound if vp .˛1b1 ˛2b2 / is large enough. This
contradiction implies that vp .˛1b1 ˛2b2 / cannot be too large.
We establish Theorems 12.4 and 12.5 simultaneously. In view of their assumptions,
we can suppose without any loss of generality that p does not divide b2 =p u .
– Arithmetic upper bound for vp ./. We have the following analogue of Lemma 11.5.
Lemma 12.6. We have the upper bound
DN
vp ./ 2 log N C .K 1/ log b C 2p t L Rh.˛1 / C S h.˛2 / :
2e.log p/
Proof. This is a variant of Lemma 11.5. Consider the polynomial
N
!
X Y r .i / b2 C s .i / b1 PN PN
P .X; Y / D sgn. / X iD1 ì r .i / Y iD1 ì s .i / ;
2S i D1
ki
N
where sgn. / denotes the signature of the permutation . Since, by (12.4), the degrees in X
and in Y of P .X; Y / do not exceed LRN and LSN , respectively, it follows from (11.10)
and Theorem B.10 that
t t D log b
log jP .˛1p ; ˛2p /jv N log N C .K 1/N C p t NL Rh.˛1 / C S h.˛2 / ;
e 2
where j jv is the absolute value associated with p and e is the ramification index of p. This
t t
completes the proof of the lemma, since D P .˛1p ; ˛2p / and log jjv D vp ./ log p,
by (B.3).
– Analytic lower bound for vp ./. There is an extra difficulty when b1 and b2 are both
divisible by p. We introduce a quantity E to deal simultaneously with the proofs of
t
Theorems 12.4 and 12.5. For the proof of Theorem 12.4 we set E D p3e C p 1 1 , while
we set E D E C t with E > 1 C p 1 1 in the proof of Theorem 12.5. In both cases, we
have E > p 1 1 .
Lemma 12.7. Assume that
vp .˛1b1 ˛2b2 / E N C u:
Then,
1 N.N K/
vp ./ E :
p 1 2
Proof. Assume first that u 1. Some extra arguments are needed in this case. Since
m1 rj C m2 sj c1 C c2 mod g; j D 1; : : : ; N;
and ˛1 D m1 1 ; ˛2 D m2 2 , we get
t
D .c1 Cc2 /p .`1 CC`N /
0 ;
where ! !
0 rj b2 C sj b1 p t ì rj p t ` i sj
D det 1 2 :
ki 1i;j N
In particular, we have vp ./ D vp .0 /. Furthermore, since the p-adic valuation of
ƒ WD ˛1b1 ˛2b2 D m1 b1 1b1 m2 b2 2b2
is positive, the integers m1 b1 and m2 b2 are in the same class modulo g. Setting
z D b1
ƒ 2b2 ;
1
z Define the quantities

we have vp .ƒ/ D vp .ƒ/.
b0
b1 b2 1
b10 D u
; b20 D u ; D 1b 0 :
p p 2 2
Observe that
u
z
ƒ 1
vp p 1 D vp b D vp .ƒ/ E N C u >

C u;
2 2 p 1
by assumption, thus
X 1
vp . / D vp .ƒ/ E N C u > C u;
p 1

Sp .
where the sum is taken over all the p u -th roots of unity in Q
We then show that is near one, and exactly one, of these roots. Order the p u -th roots
of unity 1 ; : : : ; pu in such a way that
vp . 1 / vp . pu /:
It follows that

j
vp 1 D vp .j 1 / vp . j /; j D 2; : : : ; p u :
1
Thus,
X 1
vp . 1 / C vp . 1/ E N C u > C u: (12.5)
u
p 1
p D1
¤1
1
It follows from Lemma F.5 that the sum in (12.5) is equal to u. This gives vp . 1 / > p 1
.
If vp . 1 / D vp . 2 /, then, by the first assertion of Lemma F.5, we have
1
2
1
D vp 1 D vp .2 1 / > ;
pw 1 .p 1/ 1 p 1
2
where p w denotes the exact order of the root of unity 1
. Since 1 ¤ 2 , we have
w 1. We get a contradiction and conclude that vp . 1 / > vp . j / holds for
j D 2; : : : ; p u .
Lemma F.6 then shows that 1 belongs to the field Qp . /. Assume that 1 ¤ 1 and
denote by p w its order. Then, we have
2 t
pw 1
.p 1/ e < p .p 1/;
3
since, by Lemma F.5, the ramification index of the extension Qp .1 /=Qp is equal
t
to p w 1 .p 1/. We derive that p w p t , thus 1p D 1, which is obviously true
also when 1 D 1. Set
b0 b0
ƒ0 D 1 1 1 2 2 :
1
Since 1 is a p u -th root of unity and vp . 1 / > p 1
, Lemma F.2 implies that
p u
z D vp pu

vp .ƒ/ D vp .ƒ/ 1 D vp 1
1
D vp . 1 / C u D vp .ƒ0 / C u:
We then get the lower bound
vp .ƒ0 / vp .ƒ/ u E N:
z and 1 D 1.
This also holds when u D 0 by setting ƒ0 D ƒ
b1 b10
Recall that p does not divide b20 . Set ˇ D b2
D b20
and write
b0 b0 b10
b0 1 1 ƒ0 1 1 .1 1 ƒ0 /
2 2 D D :
1 1
t
Let i D 1; : : : ; N and j D 1; : : : ; N . Since vp .b20 / D 0 and 1p D 1, we get
p t ì sj t b10 t
ì sj =b20
2 D .1p /ˇ ì sj .1 1 ƒ0 /p
t
D .1p /ˇ ì sj .1 C i;j ƒ0 /;
S p . All these expressions are well-defined since the p-adic valuations

where i;j is in Q
t b0
of ˇ and p t =b2 are non-negative and 1p 1 and 1 1 ƒ0 are in the disc of convergence
of the p-adic exponential function; see Section F.3. Furthermore, since vp .ì sj =b20 / 0,
we deduce that ji;j jp 1.
We can now substitute the above expressions in the determinant 0 . By multilinearity

of the determinant, we get
.rj b2 C sj b1 /ki pt ì .rj Csj ˇ /

0 0
D det 1 .1 C i;j ƒ /
ki Š 1i;j N
ki
.rj C sj ˇ/ p ì .rj Csj ˇ /
t
D b2k1 CCkN det 1 .1 C i;j ƒ0 / (12.6)
ki Š 1i;j N
X
D b2N.K 1/=2 .ƒ0 /N jI j 0I ;
I f1;:::;N g
where, for every subset I of f1; : : : ; N g, we have written

0 1
'i .z1 / ;:::; 'i .zN / i 2I
0I D det B
B C
C
@ A
i;1 'i .z1 / ; : : : ; i;N 'i .zN / i …I
with
z k i p t ì z
'i .z/ D ; zj D rj C ˇsj ; 1 i; j N:
ki Š 1
t t
It follows from Lemmas F.3 (for E D p3e ) and F.2 (for E D E Ct) that vp .1p 1/
E . Since E > p 1 1 , we infer from Lemma F.7 that, for i D 1; : : : ; N , the function 'i
is analytic for z with vp .z/ E C p 1 1 .
Unlike in [129], we do not use here the Schur determinants, which would yield a better
numerical estimate. We keep the same steps as in the proof of Theorem 11.3 and apply
the p-adic Schwarz lemma.
Let I be a subset of f1; : : : ; N g. Consider the function Î of the variable x defined
for x in Cp with vp .x/ E C p 1 1 by
ˇ 'i .xz1 / ;:::; 'i .xzN /

i 2I
Î .x/ ´ det

i …I
i;1 'i .xz1 / ;:::; i;N 'i .xzN /
in such a way that 0I D Î .1/. Expanding in Taylor series the functions 'i for the
indices i in I , we argue as in Section 11.3 to see that the function Î has at the point
x D 0 a zero of multiplicity at least .jI j2 jI j/=2, where jI j is the cardinality of I .
Set

D pE 1=.p 1/
: (12.7)
Since the function Î is analytic on fz W jzjp g, we deduce from Lemma F.8 that
.jI j2 jI j/=2
jI jp D jÎ .1/jp max jÎ .x/jp :
jxjp D
Using this for every subset I of f1; : : : ; N g and inserting the upper bound
E N
jƒ0 jp p N
in (12.6), we obtain by means of the ultrametric inequality that
N.N i / .i 2 i /=2
j0 jp max

max max jÎ .x/jp
1iN I jxjp D
(12.8)
.N 2 N /=2
max max jÎ .x/jp ;
I jxjp D
where the last but one maximum is taken over all the subsets I of f1; : : : ; N g. It remains
for us to bound jÎ .x/jp from above. To this end, we expand the determinant Î .x/
and we use that ji;j jp 1, for 1 i; j N . Then, for every x with jxjp , the
ultrametric inequality implies that
.k1 CCkN / N.K 1/=2
jÎ .x/jp D ;
hence, by (12.8),
.N 2 N CNKCN /=2
jjp D j0 jp D N.N K/=2
:
We conclude from (12.7) that

1 N.N K/
vp ./ E :
p 1 2
This completes the proof of the lemma.

t
– Choice of the parameters. We first deal with Theorem 12.4, in which case E 1
p 1
D p3e .
It then follows from Lemmas 12.6 and 12.7 that the assumption
vp .˛1b1 ˛2b2 / E N C u
yields a contradiction as soon as
pt D
1/ log b C 2p t L Rh.˛1 / C S h.˛2 / :

.N K/ > 2 log N C .K
3e e.log p/
Consequently, under the assumptions of Theorem 12.4, the inequality
3D
.N K/ 2 log N C .K 1/ log b C 2L R log A1 C S log A2
log p
implies
vp .˛1b1 ˛2b2 / < 2N C u;
since E 2. This establishes Theorem 12.4.
Likewise, using that E D E C t in the setting of Theorem 12.5, it follows from

Lemmas 12.6 and 12.7 that the assumption
vp .˛1b1 ˛2b2 / E N C u
yields a contradiction as soon as
D
1/ log b C 2p t L Rh.˛1 / C S h.˛2 / :

.E C t 1/.N K/ > 2 log N C .K
e.log p/
In the special case where ˛1 and ˛2 are rational numbers, we have e D D D 1. Further-
more, t D 0 if p 3 and t D 1 if p D 2. We crudely bound E C t 1 from below by
E 1 and p t from above by 2. This completes the proof of Theorem 12.5.

We proceed essentially as in the proof of Theorem 11.1 to choose the parameters. Define
D
D D and g D maxfg; .log p/2 g:
log p
The introduction of the quantity g allows us to replace the assumptions log A1 logDp and
log A2 logDp occurring in [129] by the weaker (when p 3) assumptions log A1 D 1
1
and log A2 D .
Set n 10 o
B D max log log p C log B 0 ;
D
and
K D b212 Bg .D /3 .log A1 /.log A2 /c; L D b16BD c;

R1 D d4Bg .D /2 .log A2 /e; S1 D d4Bg .D /2 .log A1 /e;
R2 D d28 Bg .D /2 .log A2 /e; S2 D d28 Bg .D /2 .log A1 /e:
Since
BD 10; g D D; and g .D /2 D 2 ;
we deduce from the assumptions D log A1 1 and D log A2 1 that K; L; R1 ; S1 ; R2 ; S2
are greater than 3.
Observe that R1 S1 gL. Thus, since ˛1 and ˛2 are multiplicatively independent,
there exists a residue class c1 modulo g such that the first inequality of (12.2) is satisfied.
Observe also that R2 S2 > g.K 1/L. Consequently, if we furthermore assume that
all the elements of the set frb2 C sb1 W 0 r < R2 ; 0 s < S2 g are distinct, then there
exists a residue class c2 modulo g such that the second inequality of (12.2) is satisfied.
Arguing as in the proof of (11.16), we see that
log b B:
Since (12.3) is satisfied, Theorem 12.4 implies the upper bound
vp .˛1b1 ˛2b2 / 2N C u 217 g .D /4 .log A1 /.log A2 /B 2 C u:
By using the upper bounds g p D 1 and
log minfb1 ; b2 g
u ;
log p
we conclude that
vp .˛1b1 ˛2b2 / 132000.p D 1/.D /4 .log A1 /.log A2 /B 2 ; (12.9)
provided that (12.3) is satisfied.

It remains for us to consider the case where the second inequality of (12.2) is not
satisfied. Assume that there is a class c modulo g such that
˚
Card b1 r C b2 s W 0 r < R2 ; 0 s < S2 ; m1 r C m2 s c modulo g
˚
< Card .r; s/ W 0 r < R2 ; 0 s < S2 ; m1 r C m2 s c modulo g :
In that situation, we establish a smaller upper bound for vp .˛1b1 ˛2b2 / than the right hand
side of (12.9).
By Dirichlet’s Schubfachprinzip, there exists a pair .r; s/ ¤ .0; 0/ of integers such that
jrj < R2 ; jsj < S2 ; b2 r C b1 s D 0; m1 r C m2 s 0 modulo g:
Write
gcd.b1 ; b2 / D n D n0 p u ;
where p does not divide n0 , and note that
r s
b1 D nr 0 ; b2 D ns 0 ; where r 0 D , s0 D :
gcd.r; s/ gcd.r; s/
Observe that
0 0 0 0
˛1b1 ˛2b2 D ˛1nr
Y
˛2 ns D .˛1r ˛2 s /:
W n D1
It follows from Lemma F.5 that

X
vp . 1/ D u:
W n D1
¤1
Sp
Arguing as in the proof of Lemma 12.7, there exists a unique n-th root of unity in Q
such that
0 0
vp .˛1r ˛2 s / vp .˛1b1 ˛2b2 / u;
0
s0 0
s0
vp .˛1r ˛2 / > vp .˛1r ˛2 /; for every ¤ such that n D 1;
as soon as vp .˛1b1 ˛2b2 / > p 1 1 C u. Lemma F.6 then shows that belongs to the
0 0
subfield Qp .˛1r ˛2s / of Qp .˛1 ; ˛2 /. Thus, we can write D m , for an integer m and a
p t -th root of unity (recall that is a root of unity of order exactly g). To see this, we
use that the class in U=U 1 (here, U denotes the group of units of Qp .˛1 ; ˛2 / and U 1 its
group of principal units) of the component of of order prime to p is generated by the
classes of ˛1 and ˛2 , and that the order of the p-primary component of divides p t , by
the argument based on ramification already used in the proof of Lemma 12.7. Since is
0 0
in U 1 and the valuation of ˛1r ˛2 s is positive, the reduction modulo U 1 implies the
congruence
m1 r 0 m2 s 0 C m modulo g:
Consequently,
m gcd.r; s/ m1 r C m2 s 0 modulo g
0
and, setting g D g= gcd.m; g/, the above congruence shows that gcd.r; s/ is an integer
multiple of g 0 . Thus, we get the upper bounds
R2 1 S2 1
jr 0 j ; js 0 j :
g0 g0
Applying Liouville’s inequality Theorem B.10 to the polynomial X Y , it follows
from (B.3) that
0 0 ŒQ.˛1 ; ˛2 ; / W Q R2 1 S2 1
vp .˛1r ˛2 s / log 2 C h.˛ 1 / C h.˛ 2 / :
e.log p/ g0 g0
Furthermore, since m is a root of unity of order exactly g 0 , we get

ŒQ.˛1 ; ˛2 ; / W Q ŒQ.˛1 ; ˛2 / W Q ŒQ. m / W Q ŒQ./ W Q

e.log p/ e.log p/
0 t 1
Dg p .p 1/ 3D 0
g:
e.log p/ 2
This gives
0 0 3D
vp .˛1r ˛2 s / .g log 2 C R2 log A1 C S2 log A2 /
2
210 gB.D /3 .log A1 /.log A2 /:
Since g p D 1, we obtain a smaller upper bound than in (12.9). We have established

Theorem 12.1.

We have established that, under the assumptions of Theorem 12.5, if the inequality
D
1/ log b C p t L Rh.˛1 / C S h.˛2 /

.E C t 1/.N K/ 2 log N C .K
2.log p/
is satisfied, then
vp .˛1b1 ˛2b2 / < .E C t /N C u:
We give an application to the rational case and establish Theorem 12.2. Below, ˛1
and ˛2 are the rational numbers yx11 and yx22 , respectively. Recall that t D 0 if p is odd and
that t D 1 if p D 2. We have shown that, if E > 2 and the parameters satisfy
1
.E 1/.N K/ 2 log N C .K 1/ log b C 4L Rh.˛1 / C S h.˛2 / ;
log p
then
vp .˛1b1 ˛2b2 / < .E C 1/N C u:
2E
Using the inequality E 1 3
(which follows from the assumption E 3) and
log A1 maxflog jx1 j; log jy1 j; E log pg; log A2 maxflog jx2 j; log jy2 j; E log pg;
we choose the various parameters to satisfy the weaker assumption

2
.N K/ 2 log N C .K 1/ log b C 4L R log A1 C S log A2 :
E.log p/
Define
1 n 10 o
D D and B D max log log p C log B 0 ; :
E log p D
Check that log b B and set
K D b212 Bg.D /3 .log A1 /.log A2 /c; L D b16BD c;

R1 D d4Bg.D /2 .log A2 /e; S1 D d4Bg.D /2 .log A1 /e;
R2 D d28 Bg.D /2 .log A2 /e; S2 D d28 Bg.D /2 .log A1 /e:
We proceed as in the proof of Theorem 12.1. We omit the details.

If a and b are multiplicatively independent, then Theorem 12.3 follows straightforwardly
from Theorem 12.1.
If b D 1, then choose d in f2; 3; 5g such that a and d are multiplicatively independent
and p does not divide d . Since the integers ad and d are multiplicatively independent,
Theorem 12.3 follows straightforwardly from Theorem 12.1.
If b > 1 and a and b are multiplicatively dependent, then there exist positive integers
d; r; s such that a D d r and b D d s . We then have
vp .ap 1
bp 1
/ D vp .d .r s/.p 1/
1/;
and we are now in the case b D 1. Since d r s

a, the result follows.
Chapter 13
Open problems
We collect in this chapter a list of open questions and conjectures in the theory of linear
forms in logarithms or somehow related to their applications. Some of them have been
extracted from Waldschmidt’s survey [434] of open Diophantine problems, to which the
reader is also directed. We take the opportunity to highlight several questions which are
slightly aside the main topic of this book.
13.1. Classical conjectures in transcendence theory

We start with a conjectural analogue of Theorem 1.2 for the logarithm function, gen-
eralising Theorem 1.4. Recall that the complex numbers z1 ; : : : ; zn are algebraically
independent if no non-zero polynomials in n variables with algebraic coefficients vanish
at the point .z1 ; : : : ; zn /.
Conjecture 13.1. Let n 2 be an integer. Let ˛1 ; : : : ; ˛n be non-zero algebraic numbers
and log ˛1 ; : : : ; log ˛n any determinations of their logarithms. If log ˛1 ; : : : ; log ˛n are
linearly independent over the rationals, then log ˛1 ; : : : ; log ˛n are algebraically indepen-
dent over the rationals.
It is not yet known whether log 2 and log 3 are algebraically independent.
We display two celebrated extensions of Conjecture 13.1. A first one, the Schanuel
conjecture, contains any reasonable transcendence conjecture dealing with values of the
exponential function.
Conjecture 13.2 (Schanuel). Let z1 ; : : : ; zn be complex numbers, linearly independent
over the rationals. Then at least n of the 2n numbers z1 ; : : : ; zn ; ez1 ; : : : ; ezn are alge-
braically independent over the rationals.
Conjecture 13.1 is a special case of Schanuel’s conjecture (assume that the n numbers
ez1 ; : : : ; ezn are algebraic). Another special case is the Lindemann–Weierstrass Theo-
rem 1.2, which corresponds to the case where z1 ; : : : ; zn are algebraic.
Another kind of open problems deals with measures of linear independence. We point
out a conjecture of Lang, quoted as Conjecture 1.11 in [432].
Conjecture 13.3. For any positive ", there exists a positive real number c."/ such that,
for any non-zero rational integers a1 ; : : : ; an ; b1 ; : : : ; bn with a1b1 anbn ¤ 1, we have
c."/n
ja1b1 anbn 1j ;
B n 1C" AnC"
where A D maxfja1 j; : : : ; jan j; 2g and B D maxfjb1 j; : : : ; jbn j; 2g.
158 13. Open problems
Conjecture 13.3 is likely to be out of reach. Recall that, under its assumption, the
current best known estimate (Theorem 2.2) takes the form
log ja1b1 anbn 1j C n .log j2a1 j/ : : : .log j2an j/ .log B/; (13.1)
for an effectively computable positive real number C (note that c."/ is not required to be
effectively computable in Conjecture 13.3). It would be of greatest interest to replace the
product .log j2a1 j/ : : : .log j2an j/ in (13.1) by some smaller quantity (say, by log A), even
if we get a weaker dependence on B (as long as it remains o.B/).
We conclude this short section with a particular case of the four exponential conjecture.
Problem 13.4. Does there exist an irrational number t such that 2t and 3t are both
rational integers?
We know that there are no irrational number t for which 2t , 3t , and 5t are all rational
integers. This statement is a particular case of the six exponentials theorem, which occurred
for the first time in a paper by Alaoglu and Erdős [4] and can also be deduced from a more
general result of Schneider [361]; see Section 1.4 of [432] for additional references.
13.2. Diophantine equations
By applying estimates for linear forms in logarithms, Stark [390] proved that, for every
positive ", we have
jx 3 y 2 j " .log x/1 " :
The current best lower bound for the difference between squares and cubes, namely
log x
jx 3 y2j ;
.log log x/4
was obtained by Juricevic [234]. This lower bound is considerably far away from what is
predicted by a conjecture of Hall [219] (see Lang [250, p. 213] for a generalization).
Problem 13.5 (Hall’s conjecture). There exists a positive real number C such that for all
positive integers x and y with x 3 ¤ y 2 we have
p
jx 3 y2j > C x:
p
Danilov [160] proved that 0 < jx 3 y 2 j < 0:97 x has infinitely many solutions
in positive integers x; y; see also [173] and the references given therein. According to
Elkies [176], Hall’s conjecture is likely not to hold, and should be replaced by the weaker
formulation
“For any positive ", there exists a positive number C."/ such that for all positive integers x
p
and y with x 3 ¤ y 2 we have jx 3 y 2 j > C."/ x 1 " .”
For the difference between higher powers, see [66, 161, 435].
13.2. Diophantine equations 159
We reproduce Problem 1.1 of [434].

Problem 13.6. Let f .X; Y / be an integer polynomial such that the equation f .x; y/ D 0
has only finitely many solutions in integers x; y. Give an effectively computable upper
bound, depending only on the degree of f .X; Y / and of the maximal absolute value of its
coefficients, for the absolute values of the integers x and y satisfying f .x; y/ D 0.
We have seen in Chapter 4 partial solutions of Problem 13.6, for instance when f .X; Y /
is of the form g.X; Y / C a, where a is a non-zero integer and g.X; Y / is homogeneous
and such that g.X; 1/ has at least three distinct roots.
The following open problem has been posed by Schinzel and Tijdeman [375].
Problem 13.7. Let P .X / be a polynomial with rational coefficients and at least three
simple roots. Then, the equation P .x/ D y 2 z 3 has only finitely many solutions in non-zero
integers x; y; z.
We formulate under a slightly different form a conjecture of Shorey [371].
Problem 13.8 (Shorey’s conjecture). Let L 2 be an integer and m1 ; : : : ; mL 1 integers
with m1 > m2 > > mL 1 > 0. Let f .X / D b1 X m1 C C bL 1 X mL 1 C bL be a
polynomial of degree m1 with rational coefficients. Assume that f .X / has distinct roots
and that bL is non-zero. Denote by H an upper bound for the absolute values of the
numerators and denominators of b1 ; : : : ; bL . Let x, y, and n be integers with n 2,
jyj 2 and
f .x/ D y n :
Then, there exists an integer n0 , depending only on H and L, such that either
n n0
or
yn f .x/ D y n b1 x m1 bL 1x
mL 1
bL
has a proper subsum which vanishes.

Shorey observed that the assumptions on f .X / in the above conjecture are necessary,
and that the integer n0 has to depend on H and on L. It is not difficult to see that
Shorey’s conjecture is a generalization of the Pillai conjecture, mentioned in Section 4.8
and reproduced below.
Problem 13.9 (Pillai’s conjecture). Let a, b, and k be non-zero integers. Then, the
ax m by n D k; in integers m, n, x, y, with m 3, n 2, x 2, and y 2;

Nair [314] established that the Pillai conjecture is true if there exist positive real
numbers C and ı such that jx 3 y 2 j > C x ı holds for any positive integers x and y with
x3 ¤ y2.
Equal binomial coefficients have been considered by de Weger [437]; see also [80]
and the references given therein.
Problem 13.10. Find all the integers n; k; m; ` with

! !
n m n m
2k ; 2` ; k < l; and D :
2 2 k `
Theorem 3.14 asserts that 0, 1, 8, and 144 are the only perfect powers in the Fibonacci
sequence. The following question is much more complicated.
Problem 13.11. Determine all the perfect powers in the Tribonacci sequence .Tn /n2Z
defined by T0 D T1 D 0, T2 D 1, and TnC3 D TnC2 C TnC1 C Tn for n in Z.
We know that 27 is the only perfect power equal to a square plus 2 (see Section 4.6),
but we do not know whether there exist perfect powers equal to a square minus 2.
Problem 13.12. Determine all the integer solutions .x; y; n/ of the equation x 2 2 D y n .
Problem 13.12 is difficult since x 2 2 D y n has the solution 12 2 D . 1/n for every
odd integer n. Siksek [378] showed how estimates for linear forms in two logarithms can
be combined with other arguments to show that if x 2 2 D y n holds with n prime and
y 2, then n is less than 1237.
There are only very few results on the distribution of perfect powers. As already
pointed out in Section 4.8, the following question remains unsolved.
Problem 13.13. Does the difference between two consecutive perfect powers tend to
infinity?
The next problem remains open, even though conjecturally 25 is the only perfect
power n such that n C 2 is also a perfect power.
Problem 13.14. Prove that there exists a positive constant k for which there are only
finitely many integers n such that n, n C 2; : : : ; n C 2k are perfect powers.
It is very likely that
35 1 74 1 183 1
D 112 ; D 202 ; and D 73
3 1 7 1 18 1
are the only solutions to the Nagell–Ljunggren equation; see the survey [133].
Problem 13.15. The Nagell–Ljunggren equation
xn 1
D yq ; in integers n 3, q 2, x 2, y 2,
x 1
The next open question asks for an extension of some results established in Section 6.5.
Problem 13.16. If a; b; c; d; y; and q are integers with a > b > c > d > 0, y 2, and
2a C 2b C 2c C 2d C 1 D y q ;
then prove that q is bounded. Solve the Diophantine equation
2a C 2b C 7 D y q ;
in integers a; b; y; q with a > b > 0, y 2, and q 2.

13.3. Miscellaneous 161
We conclude this section with two open questions on upper bounds for the number of
solutions to certain Diophantine equations.
Problem 13.17. Let D, k, and p be positive integers with p prime and gcd.D; kp/ D 1.
Does there exist an upper bound for the number of solutions of the Diophantine equation
x2 C D D k pn; in integers x 1, n 1,
which does not depend on D, k, and p?

The next problem reproduces a conjecture of Schmidt [358].
Problem 13.18. Let F .X; Y / be an integer polynomial of naïve height H and total
degree d , such that the equation F .x; y/ D 0 defines an irreducible curve of positive
genus. Let N.F / denote the number of pairs .x; y/ of integers with F .x; y/ D 0. Prove
that, for every positive real number ", there exists a real number c.d; "/, depending only
on d and ", such that N.F / < c.d; "/ H " .
Problem 13.18 remains open even for very special families of polynomials, including
the polynomials Y 2 X 3 A with varying integer A; see [67, 170, 225, 339].
13.3. Miscellaneous
The next problem addresses the number of perfect powers in short intervals.
Problem 13.19. For any positive real number x and any positive integer n, let „.n; x/
denote the number of perfect powers in Œn; n C x and set
„.x/ D lim sup „.n; x/:

n!C1
Give an upper bound for „.x/.

Using sieve methods, it is possible to prove that „.x/ x=.log x/. Presumably, this
upper estimate is very far from the true order of magnitude of „. It is even likely that
„.x/ D 1 for any x 1. Such a result would follow from Pillai’s conjecture. Slightly
related problems have been considered by de Weger and van de Woestijne [438, 439].
Mahler [281] proved that ken k > n 40n and k log nk n 40 log log n hold for every
sufficiently large integer n; see [224] for the best known lower bounds to date, [248, 431],
and Section 4 of [433].
Problem 13.20. Prove that there exists a real number C such that
ken k > e Cn
and k log nk > n C
hold for every integer n 2.

To solve Problem 13.20 we need to improve the current lower bounds for the inhomo-
geneous linear form jb0 log b1 j, where b0 and b1 are integers greater than 2.
We reproduce Grimm’s conjecture [204].
Problem 13.21 (Grimm’s conjecture). Let n; k be positive integers such that n C 1; : : : ;
nCk are composite. Then, there exist k distinct prime numbers p1 ; : : : ; pk such that nCj
is divisible by pj for j D 1; : : : ; k.
The theory of linear forms in logarithms has been applied to Grimm’s conjecture by
Ramachandra, Shorey, and Tijdeman [341, 342].
In the statement of Theorem 6.3, the effectively computable real number .; b/
depends on and b. It is likely that a stronger statement holds.
Problem 13.22. For every integer b 2 and every quadratic real number , there exists
a positive, effectively computable real number , depending only on (and not on b), such
that
kb n k ;b b .1 /n ; for n 1:
Approximation to algebraic numbers by rational numbers whose denominator is a
perfect power was considered by Pethő [322].
Problem 13.23. Let ˛ be an irrational algebraic number. Let k 2 be an integer. Prove
that there exist effectively computable positive real numbers q0 ; " such that
kq k ˛k > q kC"
; for q > q0 .
As mentioned in Sections 6.2 and 6.12, Inequality (6.5) and, consequently, the equality
g.n/ D 2n C b.3=2/n c 2 hold for every sufficiently large integer n.
Problem 13.24. Give an upper bound for the number of positive integers n such that
3 n 3 n 1
< :

2 4

Upper bounds for the number of solutions to (A.2), given for instance in [181], do not
seem to be sufficient to solve Problem 13.24.
Theorem 6.2 gives an effective lower bound for the fractional part of powers of rational
numbers. It would be of interest to extend it to powers of algebraic numbers.
Problem 13.25. Let ˛ be a real algebraic irrational number larger than 1 and n a positive
integer. Give an effective lower bound for k˛ n k.
The next problem is Question 6.2 of [249], which proposes an extension of Theo-
rem 6.2.
Problem 13.26. Let k 2 and q; a1 ; : : : ; ak be integers with gcd.q; a1 ; : : : ; ak / D 1 and
2 q < ai for i D 1; : : : ; k. Is there an effectively computable positive real number
such that an C C an
1 k 1
n
> .1 /n ; for n 1?
q

2q
We point out an open problem from [115]; see also [370].
Problem 13.27. Let be a real algebraic number of degree at least three. Let .pn =qn /n1
denote the sequence of its convergents. Find an effective lower bound for P Œpn and
for P Œqn .
An important open problem addresses effective estimates for the growth of integer
linear recurrence sequences. Obviously, if .un /n0 is a non-degenerate recurrence se-
quence of integers with a dominant root ˛, then jun j grows as n` j˛ n j for some integer `.
When there is no dominant root and ˛ is a root of maximal modulus, Evertse [180] and,
independently, van der Poorten and Schlickewei [337] established that, for any positive ",
13.3. Miscellaneous 163
there exists an integer n0 such that jun j > j˛j.1 "/n for every n > n0 . The proof is based
on the Schmidt subspace theorem and, consequently, does not yield an effective estimate
for n0 .
Problem 13.28. Let .un /n0 be a non-degenerate recurrence sequence of integers. Find
an effective lower bound for jun j.
When there are no more than three roots of maximal modulus, Problem 13.28 has
been solved by Mignotte [293] (when these roots are simple) and by Mignotte, Shorey,
and Tijdeman [299] (in the general case).
The preceding problem is closely connected to the well-known Skolem problem.
Problem 13.29 (Skolem’s problem). Is the question “does a given non-degenerate recur-
rence sequence of integers have a zero?” decidable?
The answer to Problem 13.29 is positive for recurrence sequences of order at most 4,
see [299, 424], but not yet known for recurrence sequences of order 5. For example, it
is pointed out in [164] that we still cannot find an effectively computable integer n0 such
that un is non-zero for every integer n greater than n0 , where
un D .8 C i/n C .8 i/n .7 C 4i/n .7 4i/n 1:
Other decision problems for linear recurrence sequences have been investigated, in-
cluding the Positivity problem (“Given a non-degenerate recurrence sequence of inte-
gers, are all of its terms positive?”) and the Ultimate positivity problem (“Given a
non-degenerate recurrence sequence of integers, are all but finitely many of its terms
positive?”); see [319, 320] and the references therein.
We reproduce Lehmer’s problem, discussed in Section 10.1.
Problem 13.30 (Lehmer’s problem). There exists a positive real number c such that every
non-zero algebraic number ˛ of degree d which is not a root of unity satisfies
c
h.˛/ :
d
Amoroso and David [8] have extended Lehmer’s problem to simultaneous approxima-
tion and established a slightly weaker version of it.
Problem 13.31. Let n be a positive integer. There exists a positive real number c.n/ such
that if ˛1 ; : : : ; ˛n are multiplicatively independent algebraic numbers and D denotes the
degree of the extension Q.˛1 ; : : : ; ˛n / over Q, then
n
Y c.n/
h.˛i / :
D
iD1
See [10] for the currently best known result towards Problem 13.31 and [9, 344] for
related results and problems.
Appendices
Appendix A
Approximation by rational numbers
This appendix is devoted to classical results on the approximation to real numbers by

rational numbers. We mostly omit the proofs and refer the reader to a text of van der
Poorten [336] and to the books of Bugeaud [113], Cassels [144], Hardy and Wright [220],
Khintchine [239], Perron [321], and Schmidt [357], among many others.
Let a0 ; a1 ; : : : be integers with a1 ; a2 ; : : : positive. A finite continued fraction denotes
any expression of the form
1
Œa0 I a1 ; a2 ; : : : ; an D a0 C
1
a1 C
1
a2 C
1
C :
an
We call any expression of the above form or of the form
1
Œa0 I a1 ; a2 ; : : : D a0 C D lim Œa0 I a1 ; a2 ; : : : ; an
1 n!C1
a1 C
1
a2 C
:::
a continued fraction. The above limit always exists.
Any rational number r has exactly two different continued fraction expansions. These
are Œr and Œr 1I1 if r is an integer and, otherwise, one of them reads Œa0 I a1 ; : : : ; an 1 ; an
with an 2, and the other one is Œa0 I a1 ; : : : ; an 1 ; an 1; 1. Any irrational number has
a unique expansion in continued fraction.
Theorem A.1. Let D Œa0 I a1 ; a2 ; : : : be an irrational number. Let n be a positive
integer. Set pqnn WD Œa0 I a1 ; a2 ; : : : ; an with pn and qn non-negative and coprime. Putting
p 1 D 1; q 1 D 0; p0 D a0 ; and q0 D 1;
we have
pn D an pn 1 C pn 2; qn D an qn 1 C qn 2;
and
pn 1 qn pn qn 1 D . 1/n :
168 A. Approximation by rational numbers
Furthermore, we have
1 1 ˇ pn ˇˇ 1 1 1
< < ˇ< < 2:
ˇ
(A.1)
.anC1 C 2/qn2 qn .qn C qnC1 / qn qn qnC1 anC1 qn2 qn
ˇ
With the notation of Theorem A.1, the rational number pqnn is called the n-th convergent
to . The next theorem, established by Legendre [257], gives a sufficient condition for a
rational number to be a convergent of a given real number.
a
Theorem A.2. Let be a real number. Any non-zero rational number b
with
ˇ a ˇˇ 1
ˇ < 2:
ˇ
b 2b
ˇ
is a convergent of .
The irrationality exponent of an irrational number measures the quality of approxi-
mation to by rational numbers.
Definition A.3. Let be an irrational real number. The real number is an irrationality
measure for if there exists a positive real number C./ such that every rational number
p
q
with q 1 satisfies
ˇ p ˇˇ C./
ˇ> :
ˇ
q q
ˇ
If, moreover, the constant C./ is effectively computable, then is an effective irrationality
measure for . We denote by ./ (resp., eff ./) the infimum of the irrationality
measures (resp., effective irrationality measures) for and call it the irrationality exponent
(resp., effective irrationality exponent) of .
Clearly, eff ./ is always larger than or equal to ./. It follows from (A.1) and
Theorem A.2 that the irrationality exponent of can be read off its continued fraction
expansion. Many authors attribute to Dirichlet [166] the first statement of Theorem A.4,
which was, however, known long before 1842, since it directly follows from the theory of
continued fractions. Throughout this chapter, “almost every” always refers to the Lebesgue
measure.
Theorem A.4. Every irrational real number satisfies ./ 2, with equality for almost
every .
Proof. The first assertion follows from (A.1). Let " be a positive real number. The set
q
\ [ [ p 1 p 1
2C"
; C 2C"
pD0
q q q q
Q1 qQ
has zero Lebesgue measure, since the sum qQ q 1 " tends to 0 as Q tends to infinity.
P
This shows that the set of real numbers in .0; 1/ whose irrationality exponent exceeds 2 C "
has zero Lebesgue measure. This implies the second assertion of the theorem.
169
The next result was proved in 1844 by Liouville [264].

Theorem A.5 (Liouville’s theorem). Let be an irrational algebraic number of degree d .
The inequality
ˇ p ˇˇ 1
ˇ d
ˇ
q
ˇ
q
p
holds for all rational numbers q
with q 1. Consequently, we have
eff ./ d:
Proof. Let pq be a rational number with j pq j < 1. Let P .X / be the minimal defining
polynomial of over the rational integers. Since it is irreducible, we have P . pq / 6D 0 and
jq d P . pq /j 1. By Rolle’s theorem, there exists a real number t0 , lying between and pq ,
such that ˇ p ˇ ˇ p ˇ ˇ p ˇˇ
ˇ D ˇP ./ P ˇ D ˇ ˇ jP 0 .t0 /j:
ˇ ˇ ˇ ˇ ˇ
q q q
ˇP
Consequently, we have jt0 j 1 and

ˇ p ˇˇ 1
ˇ d :
ˇ
q 0
ˇ
q maxfjP .t /j W 1 t C 1g
This concludes the proof.
Definition A.6. A real number is called a Liouville number if ./ is infinite.

It follows from Theorem A.5 that every Liouville number is transcendental. By
combining Theorems A.4 and A.5, we immediately get that the effective irrationality
exponent of every quadratic real number is equal to 2.
Roth’s celebrated theorem [349], established in 1955, asserts that, from the point of
view of rational approximation, an irrational real algebraic number behaves like almost
every real number.
Theorem A.7 (Roth’s theorem). Every irrational real algebraic number satisfies ./ D 2.
We point out that we do not know any example of a real algebraic number of degree at
least three and whose effective irrationality exponent is equal to 2. However, for any given
positive ", there are explicit examples of real algebraic numbers with eff ./ 2 C ";
see e.g. [89, 90].
The next theorem, established in 1957 by Ridout [346], is aside the main topic of this
book; however, it is referred to at several places. Recall that for a prime number ` and a
non-zero integer a, we write jaj` for the inverse of the greatest power of ` which divides a.
Theorem A.8 (Ridout’s theorem). Let be a real algebraic number, S a finite set of
prime numbers, and " a positive real number. Then, there are only finitely many rational
numbers pq with q 1 and
Y ˇ p ˇˇ 1
jpqj` ˇ ˇ < 2C" :
ˇ
(A.2)
q q
`2S
170 A. Approximation by rational numbers
TheoremQA.7 corresponds to Theorem A.8 applied with an empty set S . Observe that
the product `2S jpqj` in (A.2) is always at most equal to 1.
The Roth theorem has been extended by Schmidt to systems of n linear forms in n
variables with algebraic coefficients (Theorem A.7 deals with the case n D 2): this is
the Schmidt subspace theorem. A complete proof of it can be found in [357]; see also
Chapter 7 of [88].
Appendix B
Heights
This appendix starts with basic results on absolute values and the elementary theory of
heights. We omit most of the proofs and direct the reader to Chapter 1 of [88], Chapter 1
of [183], or Chapter 3 of [432] for a more detailed exposition; see also [315] for the
theory of valuations. We largely follow the presentation of [432] to define the Weil height
of a complex algebraic number. The second section is devoted to a general form of the
Liouville inequality and this appendix ends with an elementary estimate for a linear form
in one p-adic logarithm.
B.1. Definitions
Let ˛ be a complex algebraic number of degree d 1 with minimal defining polynomial
f .X / WD ad X d C C a1 X C a0 D ad .X ˛1 / .X ˛d / (B.1)
over Z, where ˛ D ˛1 and ˛1 ; : : : ; ˛d are the Galois conjugates of ˛. Set k D Q.˛/

and let Mk denote the set of places of k. To each complex embedding W k ! C we
associate a complex absolute value v defined by jxjv D j .x/j for x in k. Conversely,
let v be a complex absolute value on k, which coincides with the ordinary absolute value
on the set of positive rational numbers. The completion kv of k is an extension of the
completion R of Q, hence it is equal to R or C. We denote by dv the degree Œkv W R and
observe that (
1; if v is real;
dv D
2; if v is complex:
To a real absolute value v corresponds one and only one real embedding of k, while two
(complex conjugate) embeddings of k into C correspond to a complex non-real absolute
value v. We deduce that the number of elements in the set Mk1 of infinite places of k
(i.e. the number of non-equivalent Archimedean absolute values of k) is r D r1 C r2 ,
where r1 is the number of real roots of f .X / and r2 is the number of pairs of conjugate
complex roots of f .X /. Observe that d D r1 C 2r2 and
X
dD dv :
v2Mk1
172 B. Heights
The d -tuple j˛1 j; : : : ; j˛d j consists of the elements j˛jv , v 2 Mk1 , where each j˛jv

is repeated dv times. It follows from (B.1) that

d ˇa ˇ
Y Y ˇ 0ˇ
j˛jdv v D j˛i j D ˇ ˇ (B.2)
ad
v2Mk1 iD1
and
Y d
Y
maxf1; j˛jv gdv D maxf1; j˛i jg:
v2Mk1 i D1
Let p be a prime number. Since the p-adic field Qp is complete, the absolute value jjp
on Qp normalized such that jpjp D p 1 has a unique extension to any finite extension K
of Qp . This extension is given as follows. For in K, let NormK=Qp . / denote the norm
of the Qp -endomorphism of K which maps x onto x. If d is the degree of K over Qp ,
then the extension j jp of the p-adic absolute value of Qp to K is defined by
j jp D jNormK=Qp . /j1=d
p :
Furthermore, if is non-zero, its p-adic valuation is given by
log j jp 1 log jNormK=Qp . /jp

vp . / D D ;
log p d log p
which is a non-negative rational number whose denominator divides d . The ramification

index eK of K over Qp is the index of Z in the subgroup vp .K n f0g/ of d1 Z. The residue
degree of K over Qp is the quotient fK WD d=eK .
Denote by Q S p the algebraic closure of Qp , equipped with the absolute value j jp .
S p is not complete (this makes a difference with the Archimedean situation).
The field Q
We denote by Cp the completion of Q S p for j jp . This is a complete field in which Q
S p is
dense, and moreover Cp is algebraically closed.
Let ˛ be a complex algebraic number of degree d 1 with minimal defining polyno-
mial
f .X / WD ad X d C C a1 X C a0 D ad .X ˛1.p/ / .X ˛d.p/ /
over Z, where ˛1.p/ ; : : : ; ˛d.p/ denote the roots of f .X / in Cp . Set k D Q.˛/. There are d
distinct embeddings of k into Cp (each of them maps a root of f .X / onto another root
of f .X /). With each such embedding W k ! Cp we associate an ultrametric absolute
value v dividing p defined by jxjv D j .x/jp .
Let v be an absolute value on k which extends the p-adic absolute value of Q. We
view the completion kv of k as an extension of Qp and denote by ˛v the image of ˛
in kv . Then Qp .˛v / is a finite extension of Qp . We can say more about the degree of
this extension. Write the factorisation of f .X / into irreducible polynomials in Qp ŒX
as f .X / D fx1 .X / fxr .X /. Notice that the number r of irreducible factors depends
on the prime number p and that, since f .X / is irreducible in QŒX , the polynomials
fx1 .X /; : : : ; fxr .X / in Qp ŒX are pairwise distinct. Since ˛v is a root of f .X / in Cp ,
there is a unique i, with 1 i r, such that ˛v is a root of fxi .X /. Therefore fxi .X /
B.1. Definitions 173
has a root in the field Qp .˛v /, which is an extension of Qp of degree dv D deg.fxi /, and
kv D Qp .˛v /. The integer dv is called the local degree at v. From this it follows that kv
is isomorphic to a subfield of Cp , and we get an embedding v of k into Cp such that
vv D v. Hence the mapping 7! v is surjective.
The number of distinct embeddings of k into Cp associated to a given absolute value v
dividing p is the local degree dv D Œkv W Qp of v, and the number of places v in Mk
with v dividing p is the number of irreducible factors of f .X / over Qp . Let ev and fv
denote respectively the ramification index and the residue degree of kv over Qp . Then,
we have
dv D ev fv
and
X X
dD dv D ev fv :
v2Mk ;vjp v2Mk ;vjp
Let Ok be the ring of integers of k. The integer ideal pOk factors in k as a product
of prime ideals and can be written as pe11 per r , where ej is the ramification index of the
prime ideal pj for j D 1; : : : ; r and the integer r is the number of irreducible polynomials
in the factorisation of f .X / in Qp ŒX . Let j D 1; : : : ; r and ˛ be a non-zero algebraic
number in k. Let vpj .˛/ denote the exponent of pj in the factorisation of the fractional
ideal ˛Ok in a product of prime ideals. Let vj be the absolute value of k associated with
the prime ideal pj . The completion kpj D kvj of k with respect to the absolute value vj
is a finite extension of the p-adic field Qp . Its ramification index is the ramification index
of the ideal pj and its degree is the degree of the polynomial fxvj .X /. Furthermore, fvj is
the degree of the field Ok =pj over Z=pZ and it is called the residue degree of kpj . The
norm over Q of the ideal pj , denoted by Normk=Q pj , is equal to p fvj . The field kpj is
isomorphic to the extension of the field Qp by a root of the irreducible polynomial fxvj .X /
and we have
dv
j˛jvj j D .Normk=Q pj / vpj .˛/ : (B.3)
If ‰ denotes this isomorphism, then the valuation v.j /

p defined by
vpj .˛/
v.j /
p .˛/ D ;
ej
which extends to the algebraic number field k the valuation vp defined on Q, satisfies
v.j /
p .˛/ D vp .‰.˛//. With a slight abuse of notation, we denote the latter quantity
by vp .˛/.
Warning B.1. Through this book, we use vp , without any superscript, to denote a p-adic
valuation on a complex algebraic number field. We also use the same notation vp to denote
the unique extension to Cp of the p-adic valuation. This should not cause any confusion.
Any ideal a of k factors uniquely as a product of powers of prime ideals of Ok and
its norm Normk=Q a is then defined by multiplicativity. Furthermore, if ˛ is a non-zero
algebraic number in k, then the norm (over Q) of the ideal ˛Ok is the absolute value of
the norm (over Q) of ˛, defined as the product of its Galois conjugates.
174 B. Heights
The d -tuple j˛1.p/ jp ; : : : ; j˛d.p/ jp consists of the elements j˛jv , where v in Mk

divides p and each j˛jv is repeated dv times. In particular, we get
d ˇa ˇ
ˇ 0ˇ
j˛i.p/ jp D ˇ ˇ
Y Y
j˛jdv v D (B.4)
ad p
v2Mk ;vjp iD1
and
d
maxf1; j˛i.p/ jp g:
Y Y
maxf1; j˛jv gdv D
v2Mk ;vjp i D1
The proof that the latter quantity is equal to jad1jp is left as an exercise for the reader; see
Lemma 3.1 of [432].
We are now in position to prove the very useful product formula.
Lemma B.2. Let k be an algebraic number field and ˛ a non-zero element of k. Then,
Y
j˛jdv v D 1:
v2Mk
Proof. This formula is true for k D Q. The lemma then follows from (B.2) and (B.4).
Theorem B.3. Let k be an algebraic number field and ˛ in k. Let K be a finite extension
of k. Let Mk (resp., MK ) be the set of places on k (resp., on K) and dv (resp., Dv ) the
local degree at the place v of Mk (resp., of MK ). Then,
1 X 1 X
Dv log maxf1; j˛jv g D dv log maxf1; j˛jv g:
ŒK W Q Œk W Q
v2MK v2Mk
Proof. It is sufficient to show that for each place v in Mk we have

X
Dw D ŒK W kdv ;
wjv
where the sum is taken over the places w of K which divide v. Let be in K such
that K D Q. /. Note that K D k. /. The irreducible polynomial g.X / of over k,
of degree ŒK W k, factors into irreducible polynomials in kv ŒX : let us write
which is Q
g.X / D wjv gw .X /, where gw is of degree ŒKw W kv . Therefore, for each v in Mk , we
have X
ŒKw W kv D ŒK W k:
wjv
Since Dw D ŒKw W kv dv , our claim follows.
We immediately get from Theorem B.3 that the following definition of the (logarithmic)
Weil height of an algebraic number does not depend on the number field in which it is
contained.
B.1. Definitions 175
Definition B.4. Let ˛ be a complex algebraic number in a number field k. The (loga-
rithmic) Weil height of ˛ is
1 X
h.˛/ D dv log maxf1; j˛jv g; (B.5)
Œk W Q
v2Mk
where dv denotes the local degree at the place v of Mk .

It follows from the definition of the height that, if a and b are coprime non-zero
integers, then a
h D log maxfjaj; jbjg
b
and, if a and d are positive integers, then
p log a
h. d a/ D :
d
Theorem B.5. For all algebraic numbers ˛1 and ˛2 , we have
h.˛1 ˛2 / h.˛1 / C h.˛2 /

and h.˛1 C ˛2 / log 2 C h.˛1 / C h.˛2 /:
Furthermore, for any non-zero algebraic number ˛ and any integer n, we have
h.˛ n / D jnjh.˛/: (B.6)
Proof. The first inequalities follow from Definition B.4 and the upper bounds
maxf1; xyg maxf1; xg maxf1; yg; for all x 0, y 0;

and maxf1; x C yg 2 maxf1; xg maxf1; yg; for all x 0, y 0:
In view of (B.5), it is sufficient to establish (B.6) for n D 1. Let d denote the degree of ˛,
set k D Q.˛/, and, for v in Mk , let dv denote the local degree at the place v. It follows
from Lemma B.2 that
1 X 1 X 1
dv log j˛jv D dv log :
d d j˛jv
v2Mk ; v2Mk ;
j˛jv 1 j˛jv 1
The left hand side of the above equality is equal to h.˛/, while the right hand side is equal
to h.˛ 1 /. Thus, we get h.˛/ D h.˛ 1 /.
The next statement characterises the algebraic numbers whose height is equal to zero.
It is often referred to as Kronecker’s theorem [245].
Theorem B.6. The only non-zero algebraic numbers of height equal to zero are the roots
of unity.
Proof. Let ˛ be a non-zero algebraic number of height equal to zero. Then, we have
j˛jv 1 for every place v of Q.˛/. Let d denote the degree of ˛ and ad X d C Ca1 X Ca0
its minimal defining polynomial over Z. Let p be a prime number. For j D 0; : : : ; d 1,
176 B. Heights
a
expressing adj as a symmetric function of the Galois conjugates of ˛ and using that the
p-adic absolute value is ultrametric, we deduce that jaj jp jad jp . Since a0 ; : : : ; ad are
relatively prime, we get that jad jp D 1 and conclude that ˛ is an algebraic integer.
Let ` be a positive integer. All the Galois conjugates of ˛ ` have modulus at most
equal to one. Consequently, ˛ ` is a root of a monic integer polynomial of degree d and
whose coefficients are in absolute value bounded from above by 2d . Since the set of such
0
polynomials is finite, there exist distinct positive integers ` and `0 such that ˛ ` D ˛ ` . We
conclude that ˛ is a root of unity.
We conclude this section with a result often referred to as Northcott’s theorem [317].
Theorem B.7. There are only finitely many algebraic numbers of bounded degree and
bounded height.
Proof. Let ˛ be an algebraic number of degree at most d and height at most H . Then,
the leading coefficient a of the minimal defining polynomial P˛ .X / of ˛ over the integers
and all the Galois conjugates of ˛ are bounded in absolute value in terms of d and H .
By expressing the coefficients of P˛ .X / in terms of a and the symmetric functions of the
Galois conjugates of ˛, we see that they are all bounded in absolute value in terms of d
and H . Consequently, the set of algebraic numbers of degree at most d and height at
most H is finite.
B.2. The Liouville inequality

By “Liouville’s inequality”, we mean a non-trivial lower bound for the value of a poly-
nomial with integer (or algebraic) coefficients evaluated at a tuple of algebraic numbers
(when this value is non-zero), and expressed in terms of the degrees and the heights of
the polynomial and of the algebraic numbers involved. A typical example is given by
Theorem A.5.
One of the most useful inequalities of Liouville’s type is the following one.
Theorem B.8. For every non-zero algebraic number ˛ of degree d and every place v
of MQ.˛/ of local degree dv , we have
dv log j˛jv dh.˛/: (B.7)
In particular, we get vp .˛/ log p dh.˛/ for every prime number p.

1
Proof. By the definition of the height of ˛ , we have
1 1 1
dv log j˛ jv ŒQ.˛ / W Qh.˛ /:
1 1
Since h.˛/ D h.˛ / and ŒQ.˛ / W Q D d , this gives (B.7).
Definition B.9. The length L.f / of the polynomial f .X1 ; : : : ; Xt / in t variables and
with complex coefficients is the sum of the moduli of its coefficients.
We can now state a version of Liouville’s inequality (for a sharper version, with the
L1 -norm in place of the length, see Exercise 3.2 of [432]).
B.3. Linear forms in one p-adic logarithm 177
Theorem B.10. Let f .X1 ; : : : ; Xt / be a non-zero polynomial in t variables with integer

coefficients. Let 1 ; : : : ; t be algebraic numbers in an algebraic number field K of
degree d . Then
t
X

h f . 1 ; : : : ; t / log L.f / C .degXi f /h. i / (B.8)
i D1
and, for every place v of K,

t
d X
log jf . 1 ; : : : ; t /jv log L.f / C .degXi f /h. i / ; (B.9)
dv
i D1
where dv denotes the local degree at the place v.
Proof. Let v be a place of K. If v is non-Archimedean, then
`
X
log maxf1; jf . 1 ; : : : ; t /jv g .degXi f / log maxf1; j i jv g:
i D1
If v is Archimedean, then
`
X
log maxf1; jf . 1 ; : : : ; t /jv g log L.f / C .degXi f / log maxf1; j i jv g:
i D1
Since the sum of the local degrees over all the Archimedean places is equal d , we easily
deduce (B.8) and, by applying (B.7), we get (B.9).
B.3. Linear forms in one p-adic logarithm

The next statement is an explicit version of Lemma A.8 from [376]; see also Lemma 1.4
of [444].
Theorem B.11. Let p be a prime number, m 2 an integer, and ˛ an algebraic number
of degree d such that ˛ m is not equal to 1. Then, we have
log m pd 1 log 2
vp .˛ m 1/ C 2d h.˛/ C 2d : (B.10)
log p log p log p
Proof. If ˛ is a root of unity, then so is ˛ m and the height of ˛ m 1 is, by Theorem B.5,
bounded from above by log 2. The upper bound (B.10) follows from Theorem B.8. We
can assume that ˛ is not a root of unity and that vp .˛ m 1/ is positive. This implies that
vp .˛/ D 0. Let s be the smallest positive integer such that vp .˛ s 1/ is positive. By
Lemma F.4, it satisfies s p d 1. Since
vp .˛ m 1/ vp ..˛ s /m 1/;
178 B. Heights
we may assume that vp .˛ 1/ D 0. Setting

a
m D p a m0 ; gcd.p; m0 / D 1; ˇ D ˛p ;
we get
ˇ m0 1 ˇ 1
m
˛ 1D .˛ 1/:
ˇ 1 ˛ 1
Since p does not divide m0 and vp .ˇ 1/ > 0, we obtain that
ˇ m0 1 0
vp D vp .1 C ˇ C C ˇ m 1
/ D 0:
ˇ 1
Furthermore, since vp .˛ 1/ > 0, we get

ˇ 1 log m
vp a C vp .˛ C 1/ C vp .˛ C 1/:
˛ 1 log p
By Theorems B.8 and B.5, we have
vp .˛ ˙ 1/ log p dh.˛ ˙ 1/ d.h.˛/ C log 2/:

Appendix C
Auxiliary results on algebraic number fields
In a first section, we display without proofs classical results on Pellian equations. The
second section is devoted to results from algebraic number theory, which are very useful
for the effective resolution of classical families of Diophantine equations.
C.1. Real quadratic fields and Pellian equations

Classical references include Nagell’s book [312] and LeVeque’s monograph [260]; see
also [11]. We omit the proofs. Equation (C.1) below is usually called the Pell equation.
Theorem C.1. Let D be a positive integer which is not a perfect square. Then, the
x 2 Dy 2 D 1 (C.1)
has infinitely many solutions in integers x; y. Furthermore, all the solutions to (C.1) in
positive integers x; y are given by the formula
p p
xn C yn D D .x1 C y1 D/n ; n 1; (C.2)
p
where x1 C y1 D is the fundamental solution of (C.1), that is, the solution with the
smallest positive x.
It follows from (C.1) and (C.2) that the integer sequences .xn /n1 and .yn /n1 satisfy
the same linear recurrence relation, namely
ZnC1 D 2x1 Zn Zn 1; n 2:
If in (C.1) the right hand side 1 is replaced by a non-zero integer N , then the cor-
responding equation x 2 Dy 2 D N has either no solutions (consider, for example, the
equation x 2 3y 2 D 1 and argue modulo 4), or infinitely many solutions. In the latter
case, the set of solutions can be precisely described. Equations (C.3) below are usually
called Pellian equations.
Theorem C.2. If D and N are positive integers and D is not a perfect square, then the
Diophantine equations
x2 Dy 2 D N and x 2 Dy 2 D N (C.3)
180 C. Auxiliary results on algebraic number fields
p
have a finite number of classes of solutions. If x C y D denotes the fundamental
solution of a given class (that is, the solution with minimal positive x ), then all the
solutions in this class are obtained by the formula
p p p
xn C yn D D .x C y D/.x C y D/;
where .x; y/ runs through the set of solutions of (C.1).
Finally, we define the fundamental unit of a real quadratic field.
Definition C.3. Let D be a squarefree integer with D 2. Let .t; u/ be the solution
with smallest positive t and u 1 to the equation
x2 Dy 2 D ˙4:
p
Then, setting x1 WD 2t ; y1 WD u2 , the unit WD x1 C y1 D is the fundamental unit of the
p p
quadratic field Q. D/, that is, is greater than 1 and all the units of Q. D/ are of the
form ˙m , for some rational integer m.
It is not difficult to check
p that the smallest fundamental unit of a real quadratic field is
the Golden Ratio .1 C 5/=2.
C.2. Algebraic number fields

We gather in this section several classical results and four auxiliary statements on alge-
braic number fields, which are used when applying linear forms in logarithms to derive
effective upper bounds for the size of the solutions to classical families of Diophantine
equations, including S -unit equations, Thue equations, superelliptic equations, etc. We
omit most of the proofs. The reader is directed to Chapter A of [376], the monograph of
Narkiewicz [315], and Chapter 1 of [182].
Let K be an algebraic number field of degree d . Let denote by OK its ring of integers
and by DK its discriminant. Let S be a finite set of places of K containing all the infinite
places and s denote its cardinality. Let v1 ; : : : ; vs 1 be a subset of S and f1 ; : : : ; s 1 g
a fundamental system of S -units in K (that is, every S -unit in K can be written uniquely
as the product of a root of unity in K times b11 sbs 11 , for some integers b1 ; : : : ; bs 1 ).
d
Denote by RS the absolute value of the determinant of the matrix .log ji jvjj /i;j D1;:::;s 1 ,
where dj denotes the local degree at the place vj . It is easy to check that RS is positive
and independent of the choice of v1 ; : : : ; vs 1 and of the fundamental system of S -units
f1 ; : : : ; s 1 g. It is called the S -regulator of K. If S is the set of infinite places of K,
then RS is the regulator of K.
Lemma C.4. Keep the above notation. If S is the set of infinite places of K, then put
P D e; otherwise, let P denote the largest prime number lying below a finite place of S .
There exists a fundamental system f1 ; : : : ; s 1 g of S -units in K such that
sY1
h.i / d;s RS d;DK ;s .log P /s ;
iD1
h.i / d;s RS d;DK ;s .log P /s ; i D 1; : : : ; s 1;
d
and the absolute values of the entries of the inverse matrix of .log ji jvjj /i;j D1;:::;s 1
are d;s 1.
C.2. Algebraic number fields 181
Proof. This follows from Lemma 3 in [124]; see also Proposition 4.3.9 of [182].
Proposition C.5. Keep the above notation with S being the set of infinite places of K. Let r
denote the unit rank of K. Let n be a positive integer. Then, for every non-zero algebraic
integer ˇ in K, there exist integers b1 ; : : : ; br such that, setting D ˇ1 nb1 r nbr
and B D maxfjb1 j; : : : ; jbr jg, we have
ˇ log j NormK=Q .ˇ/j ˇˇ
ˇ d;DK ;n 1
ˇ
d
ˇh./
and
B d;DK maxfh.ˇ/; 1g:
Proof. This is Lemma 2 in [124]; see also Proposition 4.3.12 of [182].
Proposition C.5 asserts that, for any given algebraic integer ˇ in K, there exists a unit
in K such that the height of ˇ is controlled in terms of the degree and the discriminant
of K and in terms of the norm of ˇ. Said differently, in any integer ideal, there exists an
algebraic number of small height.
The next proposition is a key ingredient in the proofs of Theorems 4.7 and 4.9.
Proposition C.6. Let f .X / be a monic integer polynomial without multiple roots and
of degree at least equal to 2. Let K denote the splitting field of f .X / and hK its class
number. Let n be an integer with n 2. Let t be a non-zero integer. Let x and y be
integers satisfying
f .x/ D ty n :
Let ˛ be a root of f .X /. Then, there are algebraic integers ˇ; ; ı in K such that
ı.x ˛/ D ˇ n and h.ˇ/; h.ı/ f;t;n 1:
Furthermore, there are algebraic integers ˇ1 ; 1 ; ı1 in K and a unit 1 in K such that
ı1 .x ˛/hK D 1 ˇ1 1n and h.ˇ1 /; h.ı1 / f;t 1: (C.4)
We stress that the upper bounds for the heights of ˇ1 and ı1 in (C.4) do not depend
on n.
Proof. The first assertion is Lemma 6.1 in [376] and the second one is established on
page 22 of [95]. Their proofs can be summarized as follows. Assume that x 6D ˛ and
let ˛1 WD ˛; ˛2 ; : : : ; ˛m be the roots of f .X /. The greatest common divisor of the
0
principal ideals .x ˛/O K and .x ˛2 / : : : .x ˛m /OK divides the ideal f .˛/OK .
0
Since NormK=Q f .˛/ divides the discriminant of f .X /, there exist integral ideals a, b,
and c of OK such that
a .x ˛/OK D b cn (C.5)
and
NormK=Q .a/ f;t 1; NormK=Q .b/ f;t 1:
Then, we apply a theorem of Minkowski (see Theorem A.1 and Lemma A.11 in [376]) to
deduce the existence of integral ideals b0 , c0 in OK whose norms over Q are f;t 1 and
182 C. Auxiliary results on algebraic number fields
which are such that the ideals bb0 and cc0 are principal. By multiplying both sides of (C.5)
by b0 .c0 /n , we deduce that the ideal ab0 .c0 /n is principal. Furthermore, its norm over Q
is f;t;n 1. The first assertion of the lemma then follows from Proposition C.5. For the
second assertion we raise (C.5) to the power hK and recall that the hK -th power of any
integral ideal in OK is principal. Then, we again apply Proposition C.5 to conclude.
Lemma C.7. Let K be a number field and L a finite extension of K. Let ˛ be such that
L D K.˛/. Then, the absolute value of the discriminant of L is bounded from above in
terms of the degree and discriminant of K and of the degree and height of ˛.
Proof. This follows from Corollary A.7 of [376].
Appendix D
Classical results on prime numbers
We denote by p1 D 2; p2 D 3; p3 D 5; : : : the increasing sequence of prime numbers.

In this short appendix, we recall several useful results on prime numbers and arithmetical
functions, most of which are proved in [220].
We begin with Bertrand’s postulate [220, Theorem 418].
Theorem D.1. For every integer k with k 2, there exists at least one prime number p
satisfying k < p < 2k.
We continue with the Prime number theorem [220, Theorem 8] and two of its immediate
consequences.
Theorem D.2. We have
Pt Pt
pt j D1 log pj j D1 log log pj
lim D 1; lim D 1; and lim D 1:
t!C1 t log t t !C1 t log t t !C1 t log log t
In the course of the present book, three classical arithmetical functions appear naturally.
We recall their definitions and some of their properties.
Definition D.3. Let n 2 be an integer. We denote by !.n/ the total number of distinct
prime factors of n and by '.n/ the number of positive integers less than n and coprime
to n. The function ' is called the Euler totient function. The Möbius function is
defined by .n/ D 0 if the integer n is divisible by the square of a prime number, and
.q1 qk / D . 1/k if the prime numbers q1 ; : : : ; qk are distinct. Furthermore, we set
!.1/ D 0; '.1/ D 1, and .1/ D 1.
We begin with an upper bound for the function !.
Theorem D.4. For every integer n 3 we have
log n
!.n/ :
log log n
Proof. The number of distinct prime factors of n is at most equal to the integer k defined
by the inequalities
Yk kC1
Y
pj n < pj :
j D1 iD1
It follows from Theorem D.2 that
k log k log n;
and the theorem follows.
184 D. Classical results on prime numbers
Theorem D.5. There exists a real number C such that, for every integer n 3, we have
n
'.n/ :
.log log n/C
Proof. Observe that
Y 1
'.n/ D n 1 ; (D.1)
p
pjn
where the product is taken over all the distinct prime divisors of n. We estimate
!.n/ !.n/
'.n/ X 1 X 1
log log 1 log log !.n/ : (D.2)
n pj j log.2j /
j D1 j D1
We deduce from (D.2) and Theorem D.4 that there exist real numbers c1 and c2 such that
'.n/ c1 c2
.log !.n// .log log n/ :
n
Much more is known on the Euler totient function, namely (see Theorem 328 of [220])
that it satisfies
'.n/ log log n
lim inf D e ; (D.3)
n!C1 n
where 1 1 1
WD lim 1 C C C C log n
n!C1 2 3 n
is the Euler constant. It is possible to deduce from Theorem D.2 that the lim inf in (D.3)
is positive; see [220].
The next result establishes a relation between the Euler totient function and the Möbius
function.
Theorem D.6. For any integer n 2, we have
X n X
'.n/ D m and .m/ D 0:
m
mjn mjn
Proof. It follows from (D.1), the definition of the Möbius function, and a change of
variables that
X .m/ X n
'.n/ D n D m :
m m
mjn mjn
Furthermore, if q1 ; : : : ; qk denote the prime divisors of n, then

k X X
0 D 1 C . 1/ D .m/ D .m/:
mjq1 qk mjn

185
The last result is much less classical.

Theorem D.7. Let x and n be integers with 3 n < x. The number of prime numbers
6x
which are not greater than x and congruent to ˙1 modulo n does not exceed '.n/ log.x=n/ :
Proof. This follows from a version of the Brun–Titchmarsch theorem; see Theorem 3.8
of [218].
Appendix E
A zero lemma
Most of the proofs in transcendence theory go as follows, an emblematic example being

given by the proof of Theorem A.5. We define a suitable quantity as a combination of
algebraic numbers. We bound its absolute value from above by means of analytic tools
and from below by using the Liouville inequality. We reach a contradiction when the
lower bound exceeds the upper bound. But we need also to prove that our quantity is not
zero, which is often a very delicate part of the proof. To do this step, we require so-called
zero lemmas.
In this appendix, we establish a zero lemma for polynomials in two variables.
Lemma E.1. Let ˛1 ; ˛2 ; b1 ; b2 be complex numbers with ˛1 ˛2 ¤ 0. Let K; L be positive
integers and E; E 0 finite subsets of Z2 . Let P .X; Y / be a non-zero complex polynomial of
degree at most K 1 in X and degree at most L 1 in Y . Assume that
Card ˛1r ˛2s W .r; s/ 2 E L

˚
(E.1)
Card b2 r 0 C b1 s 0 W .r 0 ; s 0 / 2 E 0 > .K
˚
and 1/ L: (E.2)
Then, at least one of the numbers

0 0
P b2 .r C r 0 / C b1 .s C s 0 /; ˛1rCr ˛2sCs ; .r; s/ 2 E; .r 0 ; s 0 / 2 E 0 ; (E.3)
is non-zero.
Lemma E.1 is close to be best possible. Before proving it, we state an auxiliary result.
Lemma E.2. Let n be a positive integer and K1 ; : : : ; Kn ; L integers with
0 K1 < K2 < < Kn < L:
Let E be a set of at least L complex numbers. There exist a1 ; : : : ; an in E such that the
determinant of the matrix .ajKi /1i;j n is non-zero.
Proof. We proceed by induction on n. The case n D 1 is immediate. Let n 2 be an

integer and assume that the lemma holds for n 1. Let K1 ; : : : ; Kn ; L be integers with
0 K1 < K2 < < Kn < L. By the induction hypothesis, there exist a1 ; : : : ; an 1
188 E. A zero lemma
in E such that the determinant A of the matrix .ajKi /1i;j n 1 is non-zero. Then, the
polynomial
ˇa K1
anK11 z K1

1
K2
a1 anK21 z K2
P .z/ WD det D Az Kn C

a1Kn anKn1 z Kn
is of degree Kn and it can be written as P .z/ D .z a1 / .z an 1 /Q.z/, for a complex

polynomial Q.z/. Since the set E n fa1 ; : : : ; an 1 g contains at least
L .n 1/ > Kn .n 1/ D deg Q
elements, there exists an in E n fa1 ; : : : ; an 1g such that Q.an / is non-zero. Thus, P .an /
is non-zero as well.
Proof of Lemma E.1. Without any loss of generality, we assume that Y does not divide
P .X; Y /. Assume that all the numbers in (E.3) are zero. Define the integers K1 D 0 <
K2 < < Kn < L by writing P .X; Y / as a polynomial in Y with coefficients in CŒX :
n
X
P .X; Y / D Qi .X /Y Ki ; Qi .X / ¤ 0; i D 1; : : : ; n:
iD1
By (E.1) and Lemma E.2, there exists a subset L of E of cardinality n such that
B WD det..˛1r ˛2s /Ki /1in;.r;s/2L ¤ 0:
For .r; s/ in L, consider the polynomial
Pr;s .X; Y / WD P .X C b2 r C b1 s; ˛1r ˛2s Y /

n
X (E.4)
D Qi .X C b2 r C b1 s/.˛1r ˛2s /Ki Y Ki ;
i D1
and set
.X / D det Qi .X C b2 r C b1 s/.˛1r ˛2s /Ki

1in; .r;s/2L
:
For i D 1; : : : ; n, write Qi .X / D qi X mi C with qi ¤ 0 and observe that
.X / D q1 qn BX m1 CCmn C
with q1 : : : qn B ¤ 0. Viewing (E.4) as a linear system of n equations in the n variables

Y K1 ; : : : ; Y Kn , there exist, by Cramer’s rule, polynomials Sr;s .X / in CŒX , indexed
by .r; s/ in L, such that
X
Y K1 .X / D .X / D Pr;s .X; Y / Sr;s .X /:
.r;s/2L
189
By assumption, we have
0 0
Pr;s .b2 r 0 C b1 s 0 ; ˛1r ˛2s / D 0; .r; s/ 2 L; .r 0 ; s 0 / 2 E 0 :
It then follows that .b2 r 0 C b1 s 0 / D 0 for .r 0 ; s 0 / in E 0 and, by (E.2), the non-zero

polynomial .X / has more than .K 1/L roots. Hence, we get
.K 1/L < deg D m1 C C mn n.K 1/ .K 1/L:
This contradiction proves the lemma.

Appendix F
Tools from complex and p-adic analysis
In this appendix, we gather auxiliary results from complex and p-adic analysis used to
establish lower bounds for linear forms in two complex and p-adic logarithms.
F.1. The Schwarz lemma in complex analysis

We need only a basic version of Schwarz’ lemma, namely for analytic functions of a single
variable with a single (multiple) zero. For a positive real number R and a function f
analytic in the closed disc centered at 0 and of radius R, we denote by jf jR the maximum
of jf .z/j, where z runs over this disc. Obviously, if r satisfies 0 < r < R, then jf jr
jf jR . The purpose of the Schwarz lemma is to improve this inequality when f has many
zeros, counted with multiplicity, in the closed disc centered at 0 and of radius r.
Lemma F.1. Let T be a non-negative integer, r and R real numbers satisfying 0 < r R,
and ‰ a function of one complex variable which is analytic in the closed disc centered at
the origin and of radius R. Assume that ‰ has a zero of multiplicity at least T at 0. Then
R T
j‰jr j‰jR :
r
Proof. The function z 7! ˆ.z/ WD z T ‰.z/ is analytic in the disc fz W jzj Rg. It
follows from the maximum modulus principle that
T T
jˆjr D r j‰jr and jˆjR D R j‰jR : (F.1)
Since r R, we have jˆjr jˆjR . Combined with (F.1), this proves the lemma.
F.2. Auxiliary results on p-adic fields

Throughout this section, p denotes a prime number and K a finite extension of Qp . We
state below several elementary or classical results.
Let e be the ramification index of K and f its residue degree (see Appendix B for the
definitions).
Lemma F.2. Let be in K such that vp . 1/ > 0. Then
1
vp . p 1/ D pvp . 1/; if vp . 1/ < ;
p 1
192 F. Tools from complex and p-adic analysis
1
vp . p 1/ D vp . 1/ C 1; if vp . 1/ > ;
p 1
p 1 1
vp . p 1/ D C 1; if vp . 1/ D :
p 1 p 1 p 1
Proof. It follows from the binomial identity

p
X1
!
p p p p 1
x 1 D .x 1/ C .x 1/j
j j 1
j D1
that
vp . p
˚
1/ min pvp . 1/; 1 C vp . 1/ ;
with equality if the two terms in the minimum are distinct. Observe that these two terms
are equal when vp . 1/ D p 1 1 . The assertions of the lemma then follow.
Lemma F.3. Let be in K such that vp . 1/ > 0. Let t 0 be the integer satisfying
the inequalities
3e
pt 1
< pt :
2.p 1/
Then
t pt 1
vp p 1 > C :
3e p 1
t
Proof. The conclusion of the lemma holds if vp . 1/ > pp 1 , since p3e 2.pp 1/ 1.
Thus, we assume that vp . 1/ pp 1 . Let u be the greatest integer less than or equal
to t such that p u 1 vp . 1/ p 1 1 . Then, by induction and using Lemma F.2, we get
u
vp p 1 p u vp .

1/
pt
1 p u vp .

and vp 1/ C .t u/: (F.2)
1
If u D t , then the inequality vp . 1/ e
and (F.2) imply that
t pt pt 2p t pt 1
vp p 1 D C > C :
e 3e 3e 3e p 1
1
If u < t, then p u vp . 1/ > p 1
and we deduce from (F.2) that
t 1 1 pt
vp p

1 > C1 C ;
p 1 p 1 3e
pt p
since 3e
2.p 1/
1. This completes the proof of the lemma.
F.3. The Schwarz lemma in p-adic analysis 193
We further need classical properties of p-adic fields. A good reference is the mono-
graph of Robert [347]. Let U (resp., U 1 ) denote the (multiplicative) group of units
(resp., of principal units, that is, of units such that vp . 1/ > 0) of K.
Lemma F.4. The quotient group U=U 1 is cyclic of cardinality p f 1 and each class
contains a unique .p f 1/-th root of unity.
Proof. This is Proposition 8 of Chapter 2 of [365].
Lemma F.5. Let u be a positive integer. For any root of unity of order p u , the ramification
index of the extension Qp ./=Qp is equal to p u 1 .p 1/. Furthermore, for any positive
integer m not divisible by p, we have
X
vp . 1/ D u:
u
mp D1
6D1
Proof. The first assertion and the case m D 1 of the second one follow from the theorem
on page 107 of [347]. For m 2, see Proposition 17 of Chapter 4 of [365].
Lemma F.6 (Krasner’s lemma). Let and 0 be in an algebraic closure of Qp . Let
1 WD ; 2 ; : : : ; d denote the Galois conjugates of over Qp . If
vp . 0 / > vp . 0 i /; i D 2; : : : ; d;
then is an element of the field Qp . 0 /.

Proof. See e.g. [347, p. 130].
F.3. The Schwarz lemma in p-adic analysis

Let p be a prime number. We start this section by recalling informally some classical
results in p-adic analysis. Good references include [203, 242, 347, 351] and Chapter 17
of [223].
The p-adic logarithm function logp is defined in the open disc B.1; 1/ of Cp centered
at 1 and of radius 1 by the series
C1
X .x 1/n
logp .x/ D logp 1 C .x 1/ D . 1/nC1
nD1
n
and is analytic in B.1; 1/. Furthermore, every x in B.1; 1/ satisfies vp .logp .x// D
vp .x 1/.
The p-adic exponential function expp is defined in the open disc of Cp centered at the
origin and of radius p 1=.p 1/ by the series
C1
X xn
expp .x/ D
nD0
nŠ
and is analytic in this disc.

194 F. Tools from complex and p-adic analysis
1 1
Let a; x be in Cp with vp .x 1/ > p 1
and vp .a/ C vp .x 1/ > p 1
. Then, the
expression x a is defined by
x a WD expp a logp .x/ :

Lemma F.7. Let x be in Cp such that there exists 0 with vp .x 1/ > C p 1 1 . Then,
the function z 7! x z is analytic on the set fz 2 Cp W vp .z/ g.
Proof. See e.g. Chapter 17 of [223].
n
P
Let R be a positive real number and f W z !
7 n0 an z a formal power series with
coefficients in Qp . Set
jf jR WD sup jan jp Rn : (F.3)
n0
We use the same notation as in Section F.1. This should not cause any confusion. If jf jR
is finite, then the series f converges in the disc fz 2 Cp W jzjp < Rg. It converges in
the closed disc fz 2 Cp W jzjp Rg when jan jp Rn tends to 0 as n tends to infinity.
Moreover, for every z in the disc of convergence of f , we have
jf .z/jp jf jR :
The next statement is a p-adic version of Lemma F.1; see the Appendix of [364].
Lemma F.8. Let p be a prime number. Let T be a non-negative integer, r and R real
numbers satisfying 0 < r R, and ‰ a power series with coefficients in Qp which
converges in the closed disc centered at the origin and of radius R. Assume that ‰ has a
zero of multiplicity at least T at 0. Then
R T
j‰jr j‰jR :
r
T
Proof. The function z 7! ˆ.z/ D z ‰.z/ is analytic in the disc fz 2 Cp W jzj Rg.
We deduce from (F.3) that
T T
jˆjr D r j‰jr and jˆjR D R j‰jR :
Combined with the inequality jˆjr jˆjR , this completes the proof.
Bibliography
[1] M. Abouzaid, Les nombres de Lucas et Lehmer sans diviseur primitif , J. Théor.
Nombres Bordeaux 18 (2006), 299–313. (Cited in Chapter 7.)
[2] A. K. Agrawal, J. H. Coates, D. C. Hunt, and A. J. van der Poorten, Elliptic curves
of conductor 11, Math. Comp. 35 (1980), 991–1002. (Cited in Chapter 6.)
[3] S. D. Adhikari, N. Saradha, T. N. Shorey, and R. Tijdeman, Transcendental infinite
sums, Indag Math. 12 (2001), 1–14. (Cited in Chapter 3.)
[4] L. Alaoglu and P. Erdős, On highly composite and similar numbers, Trans. Amer.
Math. Soc. 56 (1944), 448–469. (Cited in Chapter 13.)
[5] Yu. M. Aleksentsev, The Hilbert polynomial and linear forms in logarithms of
algebraic numbers, Izv. Ross. Akad. Nauk Ser. Mat. 72 (2008), 6–52 (in Russian);
English translation in Izv. Math. 72 (2008), 1063–1110. (Cited in Chapter 2.)
[6] J. Amila, Au balcon d’Hiroshima, Gallimard, 1985.
[7] F. Amoroso, Algebraic numbers close to 1 and variants of Mahler’s measure, J.
Number Theory 60 (1996), 80–96. (Cited in Chapter 2.)
[8] F. Amoroso and S. David, Le problème de Lehmer en dimension supérieure, J.
Reine Angew. Math. 513 (1999), 145–179. (Cited in Chapter 13.)
[9] F. Amoroso and D. Masser, Lower bounds for the height in Galois extensions, Bull.
Lond. Math. Soc. 48 (2016), 1008–1012. (Cited in Chapter 13.)
[10] F. Amoroso and E. Viada, Small points on rational subvarieties of tori, Comment.
Math. Helv. 87 (2012), 355–383. (Cited in Chapter 13.)
[11] T. Andreescu and D. Andrica, Quadratic Diophantine equations. Developments in
Mathematics, 40. Springer, New York, 2015. (Cited in Appendix C.)
[12] R. Apéry, Sur une équation diophantienne, C. R. Acad. Sci. Paris Sér. A 251 (1960),
1263–1264. (Cited in Chapter 7.)
[13] , Sur une équation diophantienne, C. R. Acad. Sci. Paris Sér. A 251 (1960),
[14] J. Ax, On the units of an algebraic number field, Illinois J. Math. 9 (1965), 584–589.
(Cited in Chapter 6.)
p
[15] A. Baker, Rational approximations to 3 2 and other algebraic numbers, Quart. J.
Math. Oxford Ser. 15 (1964), 375–383. (Cited in Chapter 4.)
196 Bibliography
[16] , Linear forms in the logarithms of algebraic numbers. I, Mathematika 12

(1966), 204–216. (Cited in Chapters 1, 3 & 6.)
[17] , Linear forms in the logarithms of algebraic numbers. II, Mathematika 14
(1967), 102–107. (Cited in Chapter 1.)
[18] , Linear forms in the logarithms of algebraic numbers. III, Mathematika 14
(1967), 220–228. (Cited in Chapter 1.)
[19] , Linear forms in the logarithms of algebraic numbers. IV, Mathematika 15
(1968), 204–216. (Cited in Chapter 1.)
[20] , Contributions to the theory of Diophantine equations. I, On the repre-
sentation of integers by binary forms, Phil. Trans. R. Soc. London A 263 (1968),
[21] , Contributions to the theory of Diophantine equations. II, The Diophantine
equation y 2 D x 3 C k, Phil. Trans. R. Soc. London A 263 (1968), 193–208. (Cited
in Chapter 4.)
[22] , Bounds for solutions of hyperelliptic equations, Proc. Cambridge Philos.
Soc. 65 (1969), 439–444. (Cited in Chapter 4.)
[23] , On the periods of the Weierstrass }-function. In: Symposia Mathematica,
Vol. IV (INDAM, Rome, 1968/69), pp. 155–174, Academic Press, London, 1970.
[24] , Imaginary quadratic fields with class number 2, Ann. of Math. 94 (1971),
[25] , A sharpening of the bounds for linear forms in logarithms I, Acta Arith.
21 (1972), 117–129. (Cited in Chapters 1 & 2.)
[26] , A sharpening of the bounds for linear forms in logarithms II, Acta Arith.
24 (1973), 33–36. (Cited in Chapter 1.)
[27] , A sharpening of the bounds for linear forms in logarithms III, Acta Arith.
27 (1975), 247–252. (Cited in Chapter 1.)
[28] , Transcendental number theory. Cambridge University Press, London-New
York, 1975. (Cited in Chapters 1 & 3.)
[29] , Logarithmic forms and the abc-conjecture. In: Number theory (Eger,
1996), 37–44, de Gruyter, Berlin, 1998. (Cited in Chapter 8.)
[30] , Experiments on the abc-conjecture, Publ. Math. Debrecen 65 (2004),
[31] , On an arithmetical function associated with the abc-conjecture. In: Dio-
phantine geometry, 25–33, CRM Series, 4, Ed. Norm., Pisa, 2007. (Cited in Chap-
ter 8.)
[32] A. Baker, B. J. Birch, and E. A. Wirsing, On a problem of Chowla, J. Number
Theory 5 (1973), 224–236. (Cited in Chapter 3.)
[33] A. Baker and J. Coates, Integer points on curves of genus 1, Proc. Cambridge Philos.
Bibliography 197
[34] , Fractional parts of powers of rationals, Math. Proc. Cambridge Philos.

Soc. 77 (1975), 269–279. (Cited in Chapters 1 & 6.)
[35] A. Baker and H. Davenport, The equations 3x 2 2 D y 2 and 8x 2 7 D z 2 , Quart.
J. Math. Oxford (2) 20 (1969), 129–137. (Cited in Chapter 3.)
[36] A. Baker and C. Stewart, On effective approximations to cubic irrationals, New
Advances in Transcendence Theory, A. Baker ed., Cambridge University Press,
1988, 1–24. (Cited in Chapter 4.)
[37] A. Baker and G. Wüstholz, Logarithmic forms and group varieties, J. Reine Angew.
Math. 442 (1993), 19–62. (Cited in Chapter 2.)
[38] , Logarithmic forms and Diophantine geometry. New Mathematical Mono-
graphs, 9. Cambridge University Press, Cambridge, 2007. (Cited in Chapters 1, 2
& 3.)
[39] A.S. Bang, Taltheoretiske Undersøgelser, Tidsskrift for Mat. (5), 4 (1886), 70–80,
[40] G. Barat, R. F. Tichy, and R. Tijdeman, Digital blocks in linear numeration systems.
In: Number theory in progress, Vol. 2 (Zakopane-Kościelisko, 1997), 607–631, de
Gruyter, Berlin, 1999. (Cited in Chapter 3.)
[41] C. F. Barros, On the Lebesgue-Nagell equation and related subjects. PhD Thesis,
University of Warwick, 2010. https://homepages.warwick.ac.uk/~maseap/
theses/carlosphd.pdf (Cited in Chapter 4.)
[42] M. Bauer and M. A. Bennett, Applications of the hypergeometric method to the
generalized Ramanujan–Nagell equation, Ramanujan Math. J. 6 (2002), 209–270.
[43] M. A. Bennett, Explicit lower bounds for rational approximation to algebraic
numbers, Proc. London Math. Soc. 75 (1997), 63–78. (Cited in Chapter 4.)
[44] , On the number of solutions of simultaneous Pell equations, J. Reine Angew.
[45] , Rational approximation to algebraic number of small height : The Dio-
phantine equation jax n by n j D 1, J. Reine Angew. Math. 535 (2001), 1–49.
[46] , Review of ‘Catalan’s equation with a quadratic exponent’ by T. Metsän-
kylä; MR1816462 (2002a:11021). (Cited in Chapter 4.)
[47] , The Diophantine equation .x k 1/.y k 1/ D .z k 1/t , Indag. Math.
(N.S.) 18 (2007), 507–525. (Cited in Chapter 5.)
[48] , Perfect powers with few ternary digits, Integers 12A, 8pp., 2012. (Cited in
Chapter 6.)
[49] , The polynomial-exponential equation 1 C 2a C 6b D y q , Period. Math.
Hungar. 75 (2017), 387–397. (Cited in Chapter 6.)
[50] M. A. Bennett and N. Billerey, Sums of two S -units via Frey-Hellegouarch curves,
Math. Comp. 86 (2017), 1375–1401. (Cited in Chapter 6.)
198 Bibliography
[51] M. A. Bennett and Y. Bugeaud, Effective results for restricted rational approxima-
tion to quadratic irrationals, Acta Arith. 155 (2012), 259–269. (Cited in Chapter 6.)
[52] , Perfect powers with three digits, Mathematika 60 (2014), 66–84. (Cited in
Chapters 2 & 6.)
[53] M. A. Bennett, Y. Bugeaud, and M. Mignotte, Perfect powers with few binary digits
and related Diophantine problems, II, Math. Proc. Cambridge Philos. Soc. 153
(2012), 525–540. (Cited in Chapter 6.)
[54] , Perfect powers with few binary digits and related Diophantine problems,
Ann. Sc. Norm. Super. Pisa Cl. Sci. (5) 12 (2013), 941–953. (Cited in Chapters 2
& 6.)
[55] M. A. Bennett and S. R. Dahmen, Klein forms and the generalized superelliptic
equation, Ann. of Math. 177 (2013), 171–239. (Cited in Chapters 4 & 6.)
[56] M. A. Bennett, M. Filaseta, and O. Trifonov, Yet another generalization of the Rama-
nujan-Nagell equation, Acta Arith. 134 (2008), 211–217. (Cited in Chapter 3.)
[57] , On the factorization of consecutive integers, J. Reine Angew. Math. 629
(2009), 171–200. (Cited in Chapter 3.)
[58] M. A. Bennett and A. Pintér, Intersections of recurrence sequences, Proc. Amer.
[59] M. A. Bennett and B. M. M. de Weger, On the Diophantine equation jax n by n j D
1, Math. Comp. 67 (1998), 413–438. (Cited in Chapter 5.)
[60] A. Bérczes, J.-H. Evertse, and K. Győry, Effective results for linear equations in two
unknowns from a multiplicative division group, Acta Arith. 136 (2009), 331–349.
[61] , Effective results for hyper- and superelliptic equations over number fields,
Publ. Math. Debrecen 82 (2013), 727–756. (Cited in Chapters 4 & 6.)
[62] D. J. Bernstein, Detecting perfect powers in essentially linear time, Math. Comp.
67 (1998), 1253–1283. (Cited in Chapter 9.)
[63] F. Beukers, The generalised Ramanujan–Nagell Equation, Dissertation, 1979, R. U.
Leiden. (Cited in Chapter 7.)
[64] , On the generalized Ramanujan–Nagell equation, I, Acta Arith. 38 (1980/
1981), 389–410. (Cited in Chapter 7.)
[65] , Fractional parts of powers of rationals, Math. Proc. Cambridge Philos.
[66] F. Beukers and C. L. Stewart, Neighboring powers, J. Number Theory 130 (2010),
660–679. Addendum: J. Number Theory 130 (2010), 1571. (Cited in Chapter 13.)
[67] M. Bhargava, A. Shankar, T. Taniguchi, F. Thorne, J. Tsimerman, and Y Zhao,
Bounds on 2-torsion in class groups of number fields and integral points on elliptic
curves. Preprint. (Cited in Chapter 13.)
[68] Yu. Bilu, Effective analysis of integral points on algebraic curves, Israel J. Math.
90 (1995), 235–252. (Cited in Chapter 4.)
Bibliography 199
[69] , Quantitative Siegel’s theorem for Galois coverings, Compositio Math. 106
(1997), 125–158. (Cited in Chapter 4.)
[70] , Baker’s method and modular curves. In: A panorama of number theory
or the view from Baker’s garden (Zürich, 1999), 73–88, Cambridge Univ. Press,
Cambridge, 2002. (Cited in Chapter 4.)
[71] , Catalan’s conjecture (after Mihăilescu). Astérisque No. 294 (2004), vii,
[72] Yu. Bilu et Y. Bugeaud, Démonstration du théorème de Baker-Feldman via les
formes linéaires en deux logarithmes, J. Théor. Nombres Bordeaux 12 (2000),
13–23. (Cited in Preface and Chapter 4.)
[73] Yu. Bilu, Y. Bugeaud, and M. Mignotte. The Problem of Catalan. Springer, 2014.
[74] Yu. Bilu and G. Hanrot, Solving Thue Equations of High Degree, J. Number Theory
[75] , Solving superelliptic Diophantine equations by Baker’s method, Compo-
sitio Math. 112 (1998), 273–312. (Cited in Chapter 4.)
[76] , Thue equations with composite fields, Acta Arith. 88 (1999), 311–326.
[77] Y. Bilu, G. Hanrot, and P. M. Voutier, Existence of primitive divisors of Lucas and
Lehmer numbers. With an appendix by M. Mignotte. J. Reine Angew. Math. 539
(2001), 75–122. (Cited in Preface and Chapter 7.)
[78] G.D. Birkhoff and H.S. Vandiver, On the integral divisors of an b n , Ann. of Math.
5 (1904), 173–180. (Cited in Chapter 7.)
[79] R. Blecksmith, M. Filaseta, and C. Nicol, A result on the digits of an , Acta Arith.
64 (1993), 331–339. (Cited in Chapter 3.)
[80] A. Blokhuis, A. Brouwer, and B. de Weger, Binomial collisions and near collisions.
Integers 17 (2017), Paper No. A64, 8 pp. (Cited in Chapter 13.)
[81] R. Bolaño, 2666, Editorial Anagrama, 2004.
[82] E. Bombieri, On the Thue-Siegel-Dyson Theorem, Acta Math. 148 (1982) 255–296.
[83] , Lectures on the Thue principle. In: Analytic Number Theory and Diophan-
tine Problems (Stillwater, OK, 1984), Progr. Math. 70, Birkhäuser Boston, Boston,
MA, 1987, 15–52. (Cited in Chapter 4.)
[84] , Effective Diophantine approximation on Gm , Ann. Scuola Norm. Sup.
Pisa Cl. Sci. 20 (1993), 61–89. (Cited in Chapter 4.)
[85] , Forty years of effective results in Diophantine theory. In: A panorama of
number theory or the view from Baker’s garden (Zürich, 1999), 194–213, Cam-
bridge Univ. Press, Cambridge, 2002. (Cited in Chapter 4.)
[86] E. Bombieri and P. B. Cohen, Effective Diophantine Approximation on Gm , II ,
Ann. Scuola Norm. Sup. Pisa Cl. Sci. 24 (1997), 205–225. (Cited in Chapter 4.)
200 Bibliography
[87] , An elementary approach to effective Diophantine approximation on Gm .

In: Number Theory and Algebraic Geometry, London Math. Soc. Lecture Note Ser.
303, Cambridge Univ. Press, Cambridge, 2003, 41–62. (Cited in Chapter 4.)
[88] E. Bombieri and W. Gubler, Heights in Diophantine geometry. New Mathematical
Monographs, 4. Cambridge University Press, Cambridge, 2006. (Cited in Chapter 8
and Appendices A & B.)
p
[89] E. Bombieri and J. Mueller, On effective measures of irrationality for n a=b and
related numbers, J. Reine Angew. Math. 342 (1983), 173–196. (Cited in Chapters 4
& 5 and Appendix A.)
[90] E. Bombieri, A. J. van der Poorten, and J. D. Vaaler, Effective Measures of Irra-
tionality for Cubic Extensions of Number Fields, Ann. Scuola Norm. Sup. Pisa Cl.
Sci. 23 (1996), 211–248. (Cited in Chapter 4 and Appendix A.)
[91] J. Bourgain, E. Lindenstrauss, Ph. Michel, and A. Venkatesh, Some effective results
for a b, Ergodic Theory Dynam. Systems 29 (2009), 1705–1722. (Cited in
Chapter 3.)
[92] B. Brindza, On S -integral solutions of the equation y m D f .x/, Acta Math.
[93] , On S -integral solutions of the Catalan equation, Acta Arith. 48 (1987),
[94] , Thue equations and multiplicative independence. In: Number Theory and
Cryptography (Sydney 1989), London Math. Soc. Lecture Note Ser., 154, Cam-
bridge Univ. Press, Cambridge, 1990, 213–220. (Cited in Chapter 9.)
[95] B. Brindza, J. H. Evertse, and K. Győry, Bounds for the solutions of some dio-
phantine equations in terms of discriminants, J. Austral. Math. Soc. (Series A) 51
(1991), 8–26. (Cited in Chapter 4 and Appendix C.)
[96] B. Brindza and K. Győry, On unit equations with rational coefficients, Acta Arith.
53 (1990), 367–388. (Cited in Chapter 9.)
[97] B. Brindza, K. Győry, and R. Tijdeman, On the Catalan equation over algebraic
number fields, J. Reine Angew. Math. 367 (1986), 90–102. (Cited in Chapter 4.)
[98] J. Browkin, The abc-conjecture for algebraic numbers, Acta Math. Sin. (Engl. Ser.)
22 (2006), 211–222. (Cited in Chapter 8.)
[99] A. Brumer, On the units of algebraic number fields, Mathematika 14 (1967), 121–
124. (Cited in Chapters 1 & 6.)
[100] Y. Bugeaud, Sur la distance entre deux puissances pures, C. R. Acad. Sci. Paris 322
(1996), 1119–1121. (Cited in Chapter 4.)
[101] , Bounds for the solutions of superelliptic equations, Compositio Math. 107
(1997), 187–219. (Cited in Chapters 4 & 6.)
[102] , On the Diophantine equation x 2 2m D ˙y n , Proc. Amer. Math. Soc.
125 (1997), 3203–3208. (Cited in Chapter 6.)
[103] , On the diophantine equation x 2 p m D ˙y n , Acta Arith. 80 (1997),
Bibliography 201
[104] , On the greatest prime factor of .ab C 1/.bc C 1/.ca C 1/, Acta Arith. 86
(1998), 45–49. (Cited in Chapter 9.)
[105] , Algebraic numbers close to 1 in non-Archimedean metrics, Ramanujan J.
2 (1998), 449–457. (Cited in Chapter 2.)
[106] , Sur le plus grand facteur premier de ax n C by m , C. R. Acad. Sci. Paris
326 (1998), 661–665. (Cited in Chapter 6.)
[107] , Bornes effectives pour les solutions des équations en S -unités et des équa-
tions de Thue-Mahler, J. Number Theory 71 (1998), 227–244. (Cited in Chapter 4.)
[108] , On the greatest prime factor of ˛x n C ˇy m , Proceedings of the Number
Theory Conference, Eger 1996, Eds. Győry, Pethő, Sós. De Gruyter (1998), 115–
122. (Cited in Chapter 6.)
n
[109] , Linear forms in p-adic logarithms and the Diophantine equation xx 11 D
q
y , Math. Proc. Cambridge Philos. Soc. 127 (1999), 373–381. (Cited in Chapters 2,
6 & 12.)
[110] , On the greatest prime factor of ax m C by n , II , Bull. London Math. Soc.
32 (2000), 673–678. (Cited in Chapter 6.)
[111] , Linear forms in two m-adic logarithms and applications to Diophantine
problems, Compositio Math. 132 (2002), 137–158. (Cited in Chapters 2 & 6.)
[112] , On the Diophantine equation .x k 1/.y k 1/ D .z k 1/, Indag. Math.
(N.S.) 15 (2004), 21–28. (Cited in Chapter 5.)
[113] , Approximation by algebraic numbers. Cambridge Tracts in Mathematics
160, Cambridge, 2004. (Cited in Chapters 4 & 10 and Appendix A.)
[114] , Linear forms in the logarithms of algebraic numbers close to 1 and appli-
cations to Diophantine equations, Proceedings of the Number Theory conference
DION 2005, Mumbai, pp. 59–76, Narosa Publ. House, 2008. (Cited in Chapter 5.)
[115] , On the convergents to algebraic numbers. In: Analytic number theory,
133–143, Cambridge Univ. Press, Cambridge, 2009. (Cited in Chapters 3 & 13.)
[116] , Effective irrationality measures for quotients of logarithms of rational
numbers, Hardy-Ramanujan J. 38 (2015), 45–48. (Cited in Chapter 5.)
[117] , Effective irrationality measures for real and p-adic roots of rational num-
bers close to 1, with an application to parametric families of Thue–Mahler equa-
tions, Math. Proc. Cambridge Phil. Soc. 164 (2018), 99–108. (Cited in Chapters 2,
5 & 6.)
[118] , On the digital representation of integers with bounded prime factors, Osaka
Math. J. To appear. (Cited in Chapter 6.)
[119] Y. Bugeaud, M. Cipu, and M. Mignotte, On the representation of Fibonacci and
Lucas numbers in an integer base, Ann. Math. Qué. 37 (2013), 31–43. (Cited in
Chapter 3.)
[120] Y. Bugeaud and A. Dujella, On a problem of Diophantus for higher powers, Math.
Proc. Cambridge Philos. Soc. 135 (2003), 1–10. (Cited in Chapter 5.)
202 Bibliography
[121] Y. Bugeaud and J.-H. Evertse, On S -parts of linear recurrence sequences, Mathe-
matika 63 (2017), 840–851. (Cited in Chapter 3.)
[122] Y. Bugeaud, J.-H. Evertse, and K. Győry, S -parts of values of univariate polyno-
mials, binary forms and decomposable forms at integral points. Preprint. (Cited in
Chapters 3 & 6.)
[123] Y. Bugeaud and K. Gyarmati, On generalizations of a problem of Diophantus,
Illinois J. Math. 48 (2004), 1105–1115. (Cited in Chapter 5.)
[124] Y. Bugeaud and K. Győry, Bounds for the solutions of unit equations, Acta Arith.
74 (1996), 67–80. (Cited in Chapters 4 & 6 and Appendix C.)
[125] , Bounds for the solutions of Thue–Mahler equations and norm form equa-
tions, Acta Arith. 74 (1996), 273–292. (Cited in Chapters 4 & 6.)
[126] , On binomial Thue–Mahler equations, Period. Math. Hungar. 49 (2004),
25–34. (Cited in Chapters 6 & 7.)
n
[127] Y. Bugeaud, G. Hanrot et M. Mignotte, Sur l’équation diophantienne xx 1
1
D yq ,
III , Proc. London Math. Soc. 84 (2002), 59–78. (Cited in Chapter 4.)
[128] Y. Bugeaud and H. Kaneko, On the digital representation of smooth numbers, Math.
Proc. Cambridge Philos. Soc. To appear. (Cited in Chapter 6.)
[129] Y. Bugeaud et M. Laurent, Minoration effective de la distance p-adique entre
puissances de nombres algébriques, J. Number Theory 61 (1996), 311–342. (Cited
in Chapters 2, 6 & 12.)
[130] Y. Bugeaud and F. Luca, A quantitative lower bound for the greatest prime factor of
.ab C 1/.bc C 1/.ca C 1/, Acta Arith. 114 (2004), 275–294. (Cited in Chapter 9.)
[131] Y. Bugeaud, F. Luca, M. Mignotte, and S. Siksek, On perfect powers in Lucas
sequences, Int. J. Number Theory 1 (2005), 309–332. (Cited in Chapter 7.)
[132] Y. Bugeaud and M. Mignotte, On integers with identical digits, Mathematika 46
(1999), 411–417. (Cited in Chapter 6.)
n
[133] , L’équation de Nagell-Ljunggren xx 1
1
D y q , Enseign. Math. (2) 48 (2002),
[134] Y. Bugeaud, M. Mignotte et F. Normandin, Nombres algébriques de petite mesure
et formes linéaires en un logarithme, C. R. Acad. Sci. Paris Sér. I Math. 321 (1995),
xn 1
[135] Y. Bugeaud, M. Mignotte, and Y. Roy, On the Diophantine equation x 1
D yq ,
Pacific J. Math. 193 (2000), 257–268. (Cited in Chapter 6.)
[136] Y. Bugeaud, M. Mignotte, Y. Roy, and T. N. Shorey, The diophantine equation
.x n 1/=.x 1/ D y q has no solution with x square, Math. Proc. Cambridge
Philos. Soc. 127 (1999), 353–372. (Cited in Chapter 5.)
[137] Y. Bugeaud, M. Mignotte, and S. Siksek, Classical and modular approaches to
exponential Diophantine equations I. Fibonacci and Lucas perfect powers, Ann. of
Math. 163 (2006), 969–1018. (Cited in Chapters 3 & 7.)
Bibliography 203
[138] , Classical and modular approaches to exponential Diophantine equations

II. The Lebesgue–Nagell Equation, Compos. Math. 142 (2006), 31–62. (Cited in
Chapters 4 & 7.)
[139] Y. Bugeaud and T. N. Shorey, On the number of solutions of the generalized
Ramanujan–Nagell equation, J. Reine Angew. Math. 539 (2001), 55–74. (Cited
in Chapter 7.)
[140] , On the Diophantine equation .x m 1/=.x 1/ D .y n 1/=.y 1/,
Pacific J. Math. 207 (2002), 61–75. (Cited in Chapter 5.)
[141] E. Burger and R. Tubbs, Making transcendence transparent. An intuitive approach
to classical transcendental number theory. Springer-Verlag, New York, 2004. (Cited
in Chapter 1.)
[142] P. D. Carmichael, On the numerical factors of the arithmetic forms ˛ n ˙ ˇ n , Ann.
of Math. 15 (1913), 30–70. (Cited in Chapter 7.)
[143] M. Carrizosa, Survey on Lehmer problems, São Paulo J. Math. Sci. 3 (2009), 317–
[144] J. W. S. Cassels, An introduction to Diophantine Approximation. Cambridge Tracts
in Math. and Math. Phys., vol. 99, Cambridge University Press, 1957. (Cited in
Chapter 3 and Appendix A.)
[145] E. Catalan, Note extraite d’une lettre adressée à l’éditeur, J. Reine Angew. Math.
27 (1844), 192. (Cited in Chapter 4.)
[146] C. K. Chi, New explicit result related to the abc-conjecture. MPhil. thesis, Hong
Kong Univ. of Science and Technology, 2005. (Cited in Chapter 8.)
[147] K. C. Chim and V. Ziegler, On Diophantine equations involving sums of Fibonacci
numbers and powers of 2. Preprint. (Cited in Chapter 3.)
[148] G. V. Chudnovsky, On the method of Thue-Siegel, Ann. of Math. 117 (1983),
[149] J. Coates, An effective p-adic analogue of a theorem of Thue, Acta Arith. 15 (1969),
[150] , An effective p-adic analogue of a theorem of Thue, II, Acta Arith. 16
(1970), 399–412. (Cited in Chapter 6.)
[151] J. Coates and S. Lang, Diophantine approximation on Abelian varieties with com-
plex multiplication, Invent. Math. 34 (1976), 129–133. (Cited in Chapter 1.)
[152] H. Cohen, Démonstration de la conjecture de Catalan. In: Théorie algorithmique
des nombres et équations diophantiennes, 1–83, Ed. Éc. Polytech., Palaiseau, 2005.
[153] , Number Theory. Vol. I. Tools and Diophantine equations. Graduate Texts
in Math. 239, Springer, 2007. (Cited in Chapter 4.)
[154] , Number Theory. Vol. II. Analytic and Modern Tools. Graduate Texts in
Math. 240, Springer, 2007. (Cited in Chapter 4.)
204 Bibliography
[155] J. H. E. Cohn, The Diophantine equation x 2 C C D y n , Acta Arith. 65 (1993),

[156] P. Corvaja and U. Zannier, Finiteness of integral values for the ratio of two linear
recurrences, Invent. Math. 149 (2002), 431–451. (Cited in Chapter 3.)
[157] , On the greatest prime factor of .ab C 1/.ac C 1/, Proc. Amer. Math. Soc.
131 (2003), 1705–1709. (Cited in Chapter 9.)
[158] , On the rational approximations to the powers of an algebraic number:
solution of two problems of Mahler and Mendès France, Acta Math. 193 (2004),
[159] , Finiteness of odd perfect powers with four nonzero binary digits, Ann.
Inst. Fourier (Grenoble) 63 (2013), 715–731. (Cited in Chapter 6.)
[160] L. V. Danilov, The Diophantine equation x 3 y 2 D k and Hall’s conjecture, Math.
Notes Acad. Sci. USSR 32 (1982), 617–618. (Cited in Chapter 13.)
[161] , The Diophantine equations x m Ay n D k, Mat. Zametki 46 (1989),
38–45, 126 (in Russian); English translation in Math. Notes 46 (1989), 914–919.
[162] S. David, Minorations de formes linéaires de logarithmes elliptiques. Mém. Soc.
Math. France (N.S.) No. 62 (1995). (Cited in Chapter 1.)
[163] S. David and N. Hirata-Kohno, Linear forms in elliptic logarithms, J. Reine Angew.
[164] H. Derksen and D. Masser, Linear equations over multiplicative groups, recur-
rences, and mixing II, Indag. Math. (N.S.) 26 (2015), 113–136. (Cited in Chap-
ter 13.)
[165] G. A. Dill, Effective approximation and Diophantine applications, Acta Arith. 177
(2017), 169–199. (Cited in Chapter 4.)
[166] L. G. P. Dirichlet, Verallgemeinerung eines Satzes aus der Lehre von den Kettenbrü-
chen nebst einige Anwendungen auf die Theorie der Zahlen, S.-B. Preuss. Akad.
Wiss. (1842), 93–95. (Cited in Appendix A.)
[167] E. Dobrowolski, On a question of Lehmer and the number of irreducible factors of
a polynomial, Acta Arith. 34 (1979), 391–401. (Cited in Chapter 10.)
[168] P. Dong, Minoration de combinaisons linéaires de deux logarithmes p-adiques,
Ann. Fac. Sci. Toulouse Math. (5) 12 (1991), 195–250. (Cited in Chapters 1 & 2.)
[169] , Minorations de combinaisons linéaires de logarithmes p-adiques de nom-
bres algébriques, Dissertationes Math. (Rozprawy Mat.) 343 (1995), 97 pp. (Cited
in Chapters 1, 2 & 9.)
[170] K. A. Draziotis, On the number of integer points on the elliptic curve y 2 D x 3 CAx,
Int. J. Number Theory 7 (2011), 611–621. (Cited in Chapter 13.)
[171] A. Dubickas, On algebraic numbers close to 1, Bull. Austral. Math. Soc. 58 (1998),
Bibliography 205
[172] A. Dujella, There are only finitely many Diophantine quintuples, J. Reine Angew.
[173] , On Hall’s conjecture, Acta Arith. 147 (2011), 397–402. (Cited in Chap-
ter 13.)
[174] L. K. Durst, Exceptional real Lehmer sequences, Pacific J. Math. 9 (1959), 437–441.
[175] , Exceptional real Lucas sequences, Pacific J. Math. 11 (1961), 489–494.
[176] N. D. Elkies, Rational points near curves and small nonzero jx 3 y 2 j via lattice
reduction, Algorithmic number theory (Leiden, 2000), 33–63, Lecture Notes in
Comput. Sci. 1838, Springer, Berlin, 2000. (Cited in Chapter 13.)
[177] P. Erdős, Some recent advances and current problems in number theory. In: Lectures
on Modern Mathematics, Vol. III, pp. 196–244, Wiley, New York, 1965. (Cited in
Chapter 7.)
[178] G. Everest and K. Győry, On some arithmetical properties of solutions of decom-
posable form equations, Math. Proc. Cambridge Philos. Soc. 139 (2005), 27–40.
[179] G. Everest, A. van der Poorten, I. Shparlinski, and T. Ward, Recurrence sequences.
Mathematical Surveys and Monographs, 104. American Mathematical Society,
Providence, RI, 2003. (Cited in Chapter 3.)
[180] J.-H. Evertse, On sums of S -units and linear recurrences, Compositio Math. 53
[181] J.-H. Evertse and R. G. Ferretti, A further improvement of the quantitative subspace
theorem, Ann. of Math. 177 (2013), 513–590. (Cited in Chapter 13.)
[182] J. H. Evertse and K. Győry, Unit equations in Diophantine number theory. Cam-
bridge Studies in Advanced Mathematics 146, Cambridge University Press, 2015.
(Cited in Preface, Chapters 4 & 6, and Appendix C.)
[183] , Discriminant equations in Diophantine number theory. New Mathematical
Monographs 32, Cambridge University Press, 2016. (Cited in Chapters 2, 4 & 6
and Appendix B.)
[184] J. H. Evertse, K. Győry, C. L. Stewart, and R. Tijdeman, S -unit equations and their
applications. In: New Advances in Transcendence Theory (Ed. A. Baker), 110–174.
Cambridge University Press, 1988. (Cited in Chapter 6.)
[185] N. I. Feldman, Improved estimate for a linear form of the logarithms of algebraic
numbers, Mat. Sb. 77 (1968), 256–270 (in Russian). English translation in Math.
USSR. Sb. 6 (1968) 393–406. (Cited in Chapters 1, 4 & 11.)
[186] , Une amélioration effective de l’exposant dans le théorème de Liouville,
Izv. Akad. Nauk 35 (1971), 973–990 (in Russian). English translation in Math.
USSR Izv. 5 (1971), 985–1002. (Cited in Chapter 1.)
[187] M. van Frankenhuysen, A lower bound in the abc conjecture, J. Number Theory 82
(2000), 91–95. (Cited in Chapter 8.)
206 Bibliography
[188] C. Fuchs, R. von Känel, and G. Wüstholz, An effective Shafarevich theorem for
elliptic curves, Acta Arith. 148 (2011), 189–203. (Cited in Chapter 4.)
[189] C. Fuchs and D. H. Pham, Commutative algebraic groups and p-adic linear forms,
Acta Arith. 169 (2015), 115–147. (Cited in Chapter 1.)
[190] C. Fuchs and R. F. Tichy, Perfect powers in linear recurring sequences, Acta Arith.
107 (2003), 9–25. (Cited in Chapter 3.)
[191] I. Gaál, Diophantine equations and power integral bases. New computational meth-
ods. Birkhäuser Boston, Inc., Boston, MA, 2002. (Cited in Chapter 4.)
[192] É. Gaudron, Mesures d’indépendance linéaire de logarithmes dans un groupe al-
gébrique commutatif , Invent. Math. 162 (2005), 137–188. (Cited in Chapter 1.)
[193] , Formes linéaires de logarithmes effectives sur les variétés abéliennes,
Ann. Sci. École Norm. Sup. 39 (2006), 699–773. (Cited in Chapter 1.)
[194] , Étude du cas rationnel de la théorie des formes linéaires de logarithmes,
J. Number Theory 127 (2007), 220–261. (Cited in Chapter 1.)
[195] , Minorations simultanées de formes linéaires de logarithmes de nombres
algébriques, Bull. Soc. Math. France 142 (2014), 1–62. (Cited in Chapter 9.)
[196] J. Gebel, A. Pethő, and H. G. Zimmer, Computing integral points on elliptic curves,
[197] A. O. Gelfond, On Hilbert’s seventh problem, Dokl. Akad. Nauk SSSR 2 (1934),
1–3 (in Russian) and 4–6 (in French) Sur le septième problème de Hilbert, Izv.
Akad. Nauk SSSR 7 (1934), 623–630. (Cited in Chapter 1.)
[198] , Sur les approximations des nombres transcendants par des nombres al-
gébriques, C. R. Acad. Sc. URSS 2 (1935), 177–182. (Cited in Chapter 1.)
[199] , Sur la divisibilité de la différence des puissances de deux nombres entiers
par une puissance d’un idéal premier, Mat. Sb. 7 (1940), 7–25. (Cited in Chapters 1
& 6.)
[200] , Transcendental and algebraic numbers. Dover publ., New York, 1960.
(Cited in Chapters 1 & 6.)
[201] A. Gorodnik and R. Spatzier, Mixing properties of commuting nilmanifold auto-
morphisms, Acta Math. 215 (2015), 127–159. (Cited in Chapter 3.)
[202] N. Gouillon, Explicit lower bounds for linear forms in two logarithms, J. Théor.
Nombres Bordeaux 18 (2006), 125–146. (Cited in Chapters 2 & 5.)
[203] F. Q. Gouvêa, p-adic numbers. Universitext. Springer-Verlag, Berlin, 1993. (Cited
in Appendix F.)
[204] C. A. Grimm, A conjecture on consecutive composite numbers, Amer. Math.
Monthly 76 (1969), 1126–1128. (Cited in Chapter 13.)
[205] S. S. Gross and A. F. Vincent, On the factorization of f .n/ for f .x/ in ZŒx, Int. J.
[206] R. K. Guy, Unsolved problems in number theory. Springer-Verlag, New York, 1994
Bibliography 207
[207] K. Gyarmati, A. Sárközy, and C.L. Stewart, On shifted products which are powers,
Mathematika 49 (2002), 227–230. (Cited in Chapter 5.)
[208] K. Gyarmati and C. L. Stewart, On powers in shifted products, Glas. Mat. 42 (2007),
[209] K. Győry, Sur les polynômes à coefficients entiers et de discriminant donné, Acta
Arith. 23 (1973), 419–426. (Cited in Chapter 4.)
[210] , Sur les polynômes à coefficients entiers et de discriminant donné. II, Publ.
Math. Debrecen 21 (1974), 125–144. (Cited in Chapter 4.)
[211] , On the number of solutions of linear equations in units of an algebraic
number field, Comment. Math. Helv. 54 (1979), 583–600. (Cited in Chapter 6.)
[212] , Explicit upper bounds for the solutions of some Diophantine equations,
Ann. Acad. Sci. Fenn. Ser. A I Math. 5 (1980), 3–12. (Cited in Chapter 6.)
[213] , On some arithmetical properties of Lucas and Lehmer numbers. II, Acta
Acad. Paedagog. Agriensis Sect. Mat. 30 (2003), 67–73. (Cited in Chapter 7.)
[214] , On the abc conjecture in algebraic number fields, Acta Arith. 133 (2008),
[215] K. Győry, M. Mignotte, and T. N. Shorey, On some arithmetical properties of
weighted sums of S -units, Math. Pannon. 1 (1990), 25–43. (Cited in Chapter 6.)
[216] K. Győry and Z. Z. Papp, Effective estimates for the integer solutions of norm form
and discriminant form equations, Publ. Math. Debrecen 25 (1978), 311–325. (Cited
in Chapter 4.)
[217] K. Győry and K. Yu, Bounds for the solutions of S -unit equations and decomposable
form equations, Acta Arith. 123 (2006), 9–41. (Cited in Chapters 3, 4, 6 & 8.)
[218] H. Halberstam and H.-E. Richert, Sieve methods. London Mathematical Society
Monographs, No. 4. Academic Press, London-New York, 1974. (Cited in Ap-
pendix D.)
[219] M. Hall Jr., The Diophantine equation x 3 y 2 D k. In: A. Atkin, B. Birch
(Eds.), Computers in Number Theory, Academic Press, 1971, pp. 173–198. (Cited
in Chapter 13.)
[220] G. H. Hardy and E. M. Wright, An introduction to the theory of numbers, 5th.
edition, Clarendon Press, 1979. (Cited in Chapter 6 and Appendices A & D.)
[221] J. Haristoy, Équations diophantiennes exponentielles. Thèse, University of Stras-
bourg I (Louis Pasteur), Strasbourg, 2003. Prépublication de l’Institut de Recherche
Mathématique Avancée, 2003/29, Strasbourg, 2003. 76 pp. (Cited in Chapters 3
& 6.)
[222] S. Harrap and A. Haynes, The mixed Littlewood conjecture for pseudo-absolute
values, Math. Ann. 357 (2013), 941–960. (Cited in Chapter 3.)
[223] H. Hasse, Number theory. Grundlehren der Mathematischen Wissenschaften, 229.
Springer-Verlag, Berlin-New York, 1980. (Cited in Appendix F.)
208 Bibliography
[224] M. Hata, A lower estimate for ken k, J. Number Theory 130 (2010), 1685–1704.
[225] H. A. Helfgott and A. Venkatesh, Integral points on elliptic curves and 3-torsion in
class groups, J. Amer. Math. Soc. 19 (2006), 527–550. (Cited in Chapter 13.)
[226] S. Hernández and F. Luca, On the largest prime factor of .ab C 1/.ac C 1/.bc C 1/,
Bol. Soc. Mat. Mexicana 9 (2003), 235–244. (Cited in Chapter 9.)
[227] C. Heuberger, Parametrized Thue equations – A survey. In: Proceedings of the
RIMS symposium ‘Analytic Number Theory and Surrounding Areas’, Kyoto, Oct.
18–22, 2004, RIMS Kôkyûroku, vol. 1511, 2006, pp. 82–91. (Cited in Chapter 4.)
[228] D. Hilbert, Beweis für die Darstellbarkeit der ganzen Zahlen durch eine feste Anzahl
nter Potenzen (Waringsches Problem), Math. Ann. 67 (1909), 281–300. (Cited in
Chapter 6.)
[229] N. Hirata-Kohno, Formes linéaires de logarithmes de points algébriques sur les
groupes algébriques, Invent. Math. 104 (1991), 401–433. (Cited in Chapter 1.)
[230] , Approximations simultanées sur les groupes algébriques commutatifs,
Compositio Math. 86 (1993), 69–96. (Cited in Chapter 1.)
[231] N. Hirata-Kohno and T. N. Shorey, On the equation .x m 1/=.x 1/ D y q with
x power. In: Analytic Number Theory (Kyoto, 1996), 119–125, London Math.
Soc. Lecture Note Ser. 247, Cambridge Univ. Press, Cambridge, 1997. (Cited in
Chapter 5.)
[232] N. Hirata-Kohno and R. Takada, Linear forms in two elliptic logarithms in the
p-adic case, Kyushu J. Math. 64 (2010), 239–260. (Cited in Chapter 1.)
[233] A. Hoshi, Complete solutions to a family of Thue equations of degree 12, J. Théor.
Nombres Bordeaux 29 (2017), 549–568. (Cited in Chapter 4.)
[234] R. Juricevic, Explicit estimates of solutions of some Diophantine equations, Funct.
Approx. Comment. Math. 38 (2008), 171–194. (Cited in Chapter 13.)
[235] R. von Känel, An effective proof of the hyperelliptic Shafarevich conjecture, J.
Théor. Nombres Bordeaux 26 (2014), 507–530. (Cited in Chapter 4.)
[236] , Integral points on moduli schemes of elliptic curves, Trans. London Math.
[237] R. von Känel and B. Matschke, Solving S -unit, Mordell, Thue, Thue–Mahler
and generalized Ramanujan–Nagell equations via Shimura–Taniyama conjecture.
Preprint. (Cited in Chapters 6 & 8.)
[238] R. M. Kaufman, An estimate of a linear form of logarithms of algebraic numbers
in a p-adic metric, Vestnik Moskov. Univ. Ser. I Mat. Meh. 26 (1971), 3–10 (in
Russian). (Cited in Chapter 1.)
[239] A. Ya. Khintchine, Continued Fractions. The University of Chicago Press, Chicago
Ill., London, 1964. (Cited in Appendix A.)
[240] D. Kim, A modular approach to cubic Thue–Mahler equations, Math. Comp. 86
(2017), 1435–1471. (Cited in Chapter 6.)
Bibliography 209
[241] J. Kim and C. L. Stewart, Well spaced integers generated by an infinite set of primes,
Proc. Amer. Math. Soc. 143 (2015), 915–923. (Cited in Chapter 3.)
[242] N. Koblitz, p-adic numbers, p-adic analysis, and zeta-functions. Second edition.
Graduate Texts in Mathematics, 58. Springer-Verlag, New York, 1984. (Cited in
Appendix F.)
[243] J. F. Koksma, Diophantische Approximationen, Ergebnisse d. Math. u. ihrer Gren-
zgebiete, vol. 4, Springer, 1936. (Cited in Chapter 1.)
[244] P. H. Koymans, The Catalan equation, Indag. Math. (N. S.) 28 (2017), 321–352.
[245] L. Kronecker, Zwei Sätze über Gleichungen mit ganzzahligen Coefficienten, J. Reine
Angew. Math. 53 (1857), 173–175. (Cited in Appendix B.)
[246] J. F. Kubina and M. C. Wunderlich, Extending Waring’s conjecture to 471; 600; 000,
Math. Comp. 55 (1990), 815–820. (Cited in Chapter 6.)
[247] L. Kühne, An effective result of André–Oort type, Ann. of Math. 176 (2012), 651–
[248] , Logarithms of algebraic numbers, J. Théor. Nombres Bordeaux 27 (2015),
[249] A. Kulkarni, N. M. Mavraki, and K. D. Nguyen, Algebraic approximations to linear
combinations of powers: an extension of results by Mahler and Corvaja–Zannier.
Trans. Amer. Math. Soc. To appear. (Cited in Chapters 6 & 13.)
[250] S. Lang, Elliptic curves: Diophantine analysis, Grundlehren der Mathematischen
Wissenschaften 231, Springer–Verlag, Berlin–New York, 1978. (Cited in Chapters 8
& 13.)
[251] M. Langevin, Quelques applications de nouveaux résultats de Van der Poorten. In:
Séminaire Delange-Pisot-Poitou, 17e année (1975/76), Théorie des nombres: Fasc.
2, Exp. No. G12, 11 pp. Secrétariat Math., Paris (1977). (Cited in Chapter 4.)
[252] M. Laurent, Linear forms in two logarithms and interpolation determinants, Acta
Arith. 66 (1994), 181–199. (Cited in Chapters 2 & 11.)
[253] , Linear forms in two logarithms and interpolation determinants. II, Acta
Arith. 133 (2008), 325–348. (Cited in Chapters 2 & 11.)
[254] M. Laurent, M. Mignotte et Y. Nesterenko, Formes linéaires en deux logarithmes
et déterminants d’interpolation, J. Number Theory 55 (1995), 285–321. (Cited in
Chapters 2, 3, 4 & 11.)
[255] M. Le, A note on the diophantine equation .x m 1/=.x 1/ D y n , Acta Arith. 64
(1993), 19–28. (Cited in Chapter 6.)
[256] V. A. Lebesgue, Sur l’impossibilité en nombres entiers de l’équation x m D y 2 C 1,
Nouv. Ann. Math. 9 (1850), 178–181. (Cited in Chapter 4.)
[257] A.-M. Legendre, Théorie des nombres, troisième édition, Tome 1, Paris, 1830.
(Cited in Appendix A.)
210 Bibliography
[258] H. W. Lenstra and J. O. Shallit, Continued fractions and linear recurrences, Math.
Comp. 61 (1993), 351–354. (Cited in Chapter 7.)
[259] M.-G. Leu and G.-W. Li, The Diophantine equation 2x 2 C 1 D 3n , Proc. Amer.
[260] W. J. LeVeque, Fundamentals of number theory. Addison-Wesley Publishing Co.,
Reading, Mass.-London-Amsterdam, 1977. (Cited in Appendix C.)
[261] C. Levesque and M. Waldschmidt, Some remarks on diophantine equations and
diophantine approximation, Vietnam J. Math. 39 (2011), 343–368. (Cited in Chap-
ter 4.)
[262] , Familles d’équations de Thue–Mahler n’ayant que des solutions triviales,
[263] A. Levin, Linear forms in logarithms and integral points on higher-dimensional
varieties, Algebra Number Theory 8 (2014), 647–687. (Cited in Chapter 6.)
[264] J. Liouville, Remarques relatives à des classes très étendues de quantités dont la
valeur n’est ni algébrique, ni même réductible à des irrationnelles algébriques,
C. R. Acad. Sci. Paris 18 (1844), 883–885 et 910–911. (Cited in Chapter 1 and
Appendix A.)
[265] W. Ljunggren, Über die unbestimmte Gleichung Ax 2 By 4 D C , Arch. f. Mat. og
Naturvid. 41 (1938), No. 10, 18 pp. (Cited in Chapter 3.)
[266] , On the diophantine equation x 2 C 4 D Ay 2 , Det. Kgl. Norske Vid.-Selsk.
Forh. 24 (1951), No. 18, 82–84. (Cited in Chapter 3.)
[267] T. Loher and D. Masser, Uniformly counting points of bounded height, Acta Arith.
111 (2004), 277–297. (Cited in Chapter 10.)
[268] H. London and R. Finkelstein, On Fibonacci and Lucas numbers which are perfect
powers, Fibonacci Quart. 5 (1969), 476–481. (Cited in Chapter 3.)
[269] R. Louboutin, Sur la mesure de Mahler d’un nombre algébrique, C. R. Acad. Sci.
Paris Sér. I Math. 296 (1983), 707–708. (Cited in Chapter 10.)
[270] J. H. Loxton, Some problems involving powers of integers, Acta Arith. 46 (1986),
[271] J. H. Loxton and A. J. van der Poorten, Multiplicative dependence in number fields,
Acta Arith. 42 (1983), 291–302. (Cited in Chapters 2 & 10.)
[272] F. Luca, Distinct digits in base b expansions of linear recurrence sequences, Quaest.
[273] , On a conjecture of Erdős and Stewart, Math. Comp. 70 (2001), 893–896.
[274] , The Diophantine equation x 2 D p a ˙ p b C 1, Acta Arith. 112 (2004),
[275] , On shifted products which are powers, Glas. Mat. 40 (2005), 13–20. (Cited
in Chapter 5.)
Bibliography 211
[276] , Repdigits as sums of three Fibonacci numbers, Math. Commun. 17 (2012),

[277] F. Luca and M. Mignotte, Arithmetic properties of the integer part of the powers of
an algebraic number, Glas. Mat. Ser. III 44 (2009), 285–307. (Cited in Chapter 3.)
[278] K. Mahler, Ein Beweis der Transzendenz der P -adischen Exponentialfunktion, J.
Reine Angew. Math. 169 (1932), 61–66. (Cited in Chapter 1.)
[279] , Zur Approximation algebraischer Zahlen. I, Math. Ann. 107 (1933), 691–
[280] , Über transzendente P -adische Zahlen, Compositio Math. 2 (1935), 259–
[281] , On the approximation of logarithms of algebraic numbers, Philos. Trans.
R. Soc. Lond. Ser. A 245 (1953), 371–398. (Cited in Chapter 13.)
[282] , On the fractional parts of the powers of a rational number, II, Mathematika
4 (1957), 122–124. (Cited in Chapter 6.)
[283] G. Martin and W. Miao, abc triples, Funct. Approx. Comment. Math. 55 (2016),
[284] D. W. Masser, Elliptic functions and transcendence. Lecture Notes in Mathematics,
Vol. 437. Springer-Verlag, Berlin-New York, 1975. (Cited in Chapter 1.)
[285] , Open problems. Proc. Symp. Analytic Number Th. W. W. L. Chen (ed.).
London: Imperial College 1985. (Cited in Chapter 8.)
[286] , On abc and discriminants, Proc. Amer. Math. Soc. 130 (2002), 3141–
[287] , Auxiliary polynomials in number theory. Cambridge Tracts in Mathemat-
ics, 207. Cambridge University Press, Cambridge, 2016. (Cited in Chapter 10.)
[288] , Abcological anecdotes, Mathematika 63 (2017), 713–714. (Cited in Chap-
ter 8.)
[289] E. M. Matveev, Linear and multiplicative relations, Mat. Sb. 184 (1993), 23–40 (in
Russian); English translation in Russian Acad. Sci. Sb. Math. 78 (1994), 411–425.
[290] , Explicit lower estimates for rational homogeneous forms in logarithms of
algebraic numbers, Izv. Akad. Nauk SSSR Ser. Mat. 62 (1998), 81–136 (in Russian);
English translation in Izv. Math. 62 (1998), 723–772. (Cited in Chapter 2.)
[291] , An explicit lower bound for a homogeneous rational linear form in log-
arithms of algebraic numbers. II, Izv. Ross. Acad. Nauk Ser. Mat. 64 (2000),
125–180 (in Russian); English translation in Izv. Math. 64 (2000), 1217–1269.
[292] , On the index of multiplicative groups of algebraic numbers, Mat. Sb. 196
(2005), 59–70 (in Russian); English translation in Sb. Math. 196 (2005), 1307–1318.
212 Bibliography
[293] M. Mignotte, A note on linear recursive sequences, J. Austral. Math. Soc. 20 (1975),
[294] , Intersection des images de certaines suites récurrentes linéaires, Theoret.
Comput. Sci. 7 (1978), 117–122. (Cited in Chapter 3.)
[295] , Sur les entiers qui s’écrivent simplement en différentes bases, European J.
Combin. 9 (1988), 307–316. (Cited in Chapter 3.)
[296] , A note on the equation ax n by n D c, Acta Arith. 75 (1996), 287–295.
[297] , Catalan’s equation just before 2000. In: Number theory (Turku, 1999),
247–254, de Gruyter, Berlin, 2001. (Cited in Chapter 4.)
[298] , A kit on linear forms in three logarithms, http://www-irma.
u-strasbg.fr/~bugeaud/travaux/kit.pdf (Cited in Chapters 2 & 3.)
[299] M. Mignotte, T. N. Shorey, and R. Tijdeman, The distance between terms of an
algebraic recurrence sequence, J. Reine Angew. Math. 349 (1984), 63–76. (Cited
in Chapter 13.)
[300] M. Mignotte and M. Waldschmidt, On algebraic numbers of small height: linear
forms in one logarithm, J. Number Theory 47 (1994), 43–62. (Cited in Chapter 2.)
[301] M. Mignotte and B. M. M. de Weger, On the Diophantine equations x 2 C 74 D y 5
and x 2 C 86 D y 5 , Glasgow Math. J. 38 (1996), 77–85. (Cited in Chapter 4.)
[302] P. Mihăilescu, Primary cyclotomic units and a proof of Catalan’s conjecture,
J. Reine Angew. Math. 572 (2004), 167–195. (Cited in Chapter 4.)
[303] De Ze Mo, Exponential Diophantine equations with four terms. II, Acta Math.
Sinica 37 (1994), 482–490. (Cited in Chapter 6.)
[304] De Ze Mo and R. Tijdeman, Exponential Diophantine equations with four terms,
Indag. Math. (N.S.) 3 (1992), 47–57. (Cited in Chapter 6.)
[305] H. L. Montgomery and P. J. Weinberger, Notes on small class numbers, Acta Arith.
24 (1973/74), 529–542. (Cited in Chapter 3.)
[306] M. J. Mossinghoff, G. Rhin, and Q. Wu, Minimal Mahler measures, Experiment.
[307] M. R. Murty and H. Pasten, Modular forms and effective Diophantine approxima-
tion, J. Number Theory 133 (2013), 3739–3754. (Cited in Chapter 8.)
[308] M. R. Murty and P. Rath, Transcendental Numbers. Springer-Verlag, New York,
[309] M. R. Murty and S. Wong, The abc-conjecture and prime divisors of the Lucas and
Lehmer sequences. In: Number theory for the millennium, III (Urbana, IL, 2000),
43–54, A K Peters, Natick, MA, 2002. (Cited in Chapter 7.)
[310] T. Nagell, Lsning til oppgave nr 2, 1943, s. 29, Nordisk Mat. Tidskr. 30 (1948),
[311] , The Diophantine equation x 2 C 7 D 2n , Ark. Math. 4 (1960), 185–187.
Bibliography 213
[312] , Introduction to number theory. Second edition. Chelsea Publishing Co.,

New York, 1964. (Cited in Appendix C.)
[313] , Collected papers of Trygve Nagell. Vol. 1–4, Edited by Paulo Riben-
boim, Queen’s Papers in Pure and Applied Mathematics 121, Queen’s University,
Kingston, ON, 2002. (Cited in Chapter 4.)
[314] M. Nair, A note on the equation x 3 y 2 D k, Quart. J. Math. Oxford (2) 29 (1978),
[315] W. Narkiewicz, Elementary and analytic theory of algebraic numbers. Third edition.
Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2004. (Cited in
Appendices B & C.)
[316] Yu. Nesterenko, Linear forms in logarithms of rational numbers. In: Diophantine
approximation (Cetraro, 2000), 53–106, Lecture Notes in Math., 1819, Springer,
Berlin, 2003. (Cited in Chapter 2.)
[317] D. G. Northcott, An inequality in the theory of arithmetic on algebraic varieties,
Proc. Cambridge Philos. Soc. 45 (1949), 502–509. (Cited in Appendix B.)
[318] J. Oesterlé, Nouvelles approches du “théorème” de Fermat. Séminaire Bourbaki,
30 (1987-1988), Exposé No. 694, 22 p. (Cited in Chapter 8.)
[319] J. Ouaknine and J. Worrell, Ultimate Positivity is decidable for simple linear re-
currence sequences. In: Automata, languages, and programming. Part II, 330–341,
Lecture Notes in Comput. Sci., 8573, Springer, Heidelberg, 2014. (Cited in Chap-
ter 13.)
[320] , Positivity problems for low-order linear recurrence sequences. In: Pro-
ceedings of the Twenty-Fifth Annual ACM-SIAM Symposium on Discrete Algo-
rithms, 366–379, ACM, New York, 2014. (Cited in Chapter 13.)
[321] O. Perron, Die Lehre von den Ketterbrüchen. Teubner, Leipzig, 1929. (Cited in
Appendix A.)
[322] A. Pethő, Über die nichtlineare Diophantische Approximation von quadratischen,
algebraischen Zahlen, J. Number Theory 12 (1980), 334–338. (Cited in Chapters 4
& 13.)
[323] , Perfect powers in second order linear recurrences, J. Number Theory 15
(1982), 5–13. (Cited in Chapter 6.)
[324] , Perfect powers in second order recurrences. In: Topics in Classical Number
Theory, Proceedings of the Conference in Budapest 1981, Colloq. Math. Soc. János
Bolyai 34, pp. 1217–1227, North–Holland, Amsterdam, 1984. (Cited in Chapter 3.)
[325] , Diophantine properties of linear recursive sequences. II, Acta Math. Acad.
Paedagog. Nyházi. (N.S.) 17 (2001), 81–96. (Cited in Chapter 3.)
[326] , Fifteen problems in number theory, Acta Univ. Sapientiae, Mathematica,
2 (2010), 72–83. (Cited in Chapter 3.)
[327] P. Philippon, Quelques remarques sur des questions d’approximation diophanti-
enne, Bull. Austral. Math. Soc. 59 (1999), 323–334. (Cited in Chapter 8.)
214 Bibliography
[328] , Addendum à quelques remarques sur des questions d’approximation dio-

phantienne, Bull. Austral. Math. Soc. 61 (2000), 167–169. (Cited in Chapter 8.)
[329] P. Philippon and M. Waldschmidt, Lower bounds for linear forms in logarithms.
In: New advances in transcendence theory (Durham, 1986), 280–312, Cambridge
Univ. Press, Cambridge, 1988. (Cited in Chapter 1.)
[330] , Formes linéaires de logarithmes sur les groupes algébriques commutatifs,
Illinois J. Math. 32 (1988), 281–314. (Cited in Chapter 1.)
[331] S. S. Pillai, On ax b y D c, J. Indian Math. Soc. (N.S.) 2 (1936) 119–122, 215.
[332] , On the equation 2x 3y D 2X C 3Y , Bull. Calcutta Math. Soc. 37 (1945),
[333] I. Pink, On the Diophantine equation x 2 C .p1z1 pszs /2 D 2y n , Publ. Math.
Debrecen 65 (2004), 205–213. (Cited in Chapter 6.)
[334] A. J. van der Poorten, Effectively computable bounds for the solutions of certain
Diophantine equations, Acta Arith. 33 (1977), 195–207. (Cited in Chapter 6.)
[335] , Linear forms in logarithms in the p-adic case. In: Transcendence theory:
advances and applications (Proc. Conf., Univ. Cambridge, Cambridge, 1976), pp.
29–57. Academic Press, London, 1977. (Cited in Chapters 1 & 8.)
[336] , An introduction to continued fractions, Diophantine analysis (Kensington,
1985), 99–138, London Math. Soc. Lecture Note Ser. 109, Cambridge Univ. Press,
Cambridge, 1986. (Cited in Appendix A.)
[337] A. J. van der Poorten and H. P. Schlickewei, The growth condition for recurrence
sequences, Macquarie University Math. Rep. 82-0041 (1982). (Cited in Chapters 3
& 13.)
[338] L. P. Postnikova and A. Schinzel, Primitive divisors of the expression an b n in
algebraic number fields, Mat. Sbornik 75 (1968), 171–177 (in Russian); English
translation in Math. USSR Sbornik 4 (1968), 153–159. (Cited in Chapter 7.)
[339] D. Poulakis, A note on Schmidt’s conjecture, Bull. Austral. Math. Soc. 96 (2017),
[340] K. Ramachandra, A note on Baker’s method, J. Austral. Math. Soc. 10 (1969),
[341] K. Ramachandra, T. N. Shorey, and R. Tijdeman, On Grimm’s problem relating to
factorisation of a block of consecutive integers, J. Reine Angew. Math. 273 (1975),
[342] , On Grimm’s problem relating to factorisation of a block of consecutive
integers. II, J. Reine Angew. Math. 288 (1976), 192–201. (Cited in Chapters 5
& 13.)
[343] S. Ramanujan, Question 464, J. Indian Math. Soc. 5 (1913), 120. (Cited in Chap-
ter 4.)
[344] G. Rémond, Généralisations du problème de Lehmer et applications à la conjecture
de Zilber-Pink. Panoramas et synthèses 52 (2017), 243–284. (Cited in Chapter 13.)
Bibliography 215
[345] G. Rémond and F. Urfels, Approximation diophantienne de logarithmes elliptiques

p-adiques, J. Number Theory 57 (1996), 133–169. (Cited in Chapter 1.)
[346] D. Ridout, Rational approximations to algebraic numbers, Mathematika 4 (1957),
125–131. (Cited in Appendix A.)
[347] A. M. Robert, A Course in p-adic Analysis. Graduate Texts in Mathematics 198,
Springer–Verlag, New York, 2000. (Cited in Appendix F.)
[348] O. Robert, C. L. Stewart, and G. Tenenbaum, A refinement of the abc conjecture,
Bull. Lond. Math. Soc. 46 (2014), 1156–1166. (Cited in Chapter 8.)
[349] K. F. Roth, Rational approximations to algebraic numbers, Mathematika 2 (1955),
1–20. (Cited in Appendix A.)
[350] N. Saradha and T. N. Shorey, The equation .x n 1/=.x 1/ D y q with x square,
Math. Proc. Cambridge Philos. Soc. 125 (1999), 1–19. (Cited in Chapters 5 & 6.)
[351] W. H. Schikhof, Ultrametric calculus. Cambridge Studies in Advanced Mathemat-
ics, 4. Cambridge University Press, Cambridge, 1984. (Cited in Appendix F.)
[352] A. Schinzel, The intrinsic divisors of Lehmer numbers in the case of negative
discriminant, Ark. Mat. 4 (1962), 413–416. (Cited in Chapter 7.)
[353] , On primitive prime factors of Lehmer numbers. I, Acta Arith. 8
(1962/1963), 213–223. (Cited in Chapter 7.)
[354] , On two theorems of Gelfond and some of their applications, Acta Arith.
[355] , Primitive divisors of the expression An B n in algebraic number fields,
J. Reine Angew. Math., 268/269 (1974), 27–33. (Cited in Chapter 7.)
[356] A. Schinzel and R. Tijdeman, On the equation y m D P .x/, Acta Arith. 31 (1976),
[357] W. M. Schmidt, Diophantine Approximations and Diophantine equations, Lecture
Notes in Math. 785, Springer, Berlin, 1991. (Cited in Chapters 4 & 10 and Ap-
pendix A.)
[358] , Integer points on curves of genus 1, Compositio Math. 81 (1992), 33–59.
[359] Th. Schneider, Transzendenzuntersuchungen periodischer Funktionen, J. Reine
Angew. Math. 172 (1934), 65–74. (Cited in Chapter 1.)
[360] , Arithmetische Untersuchungen elliptischer Integrale, Math. Ann. 113
(1937), 1–13. (Cited in Chapter 1.)
[361] , Ein Satz über ganzwertige Funktionen als Prinzip für Transzendenzbe-
weise, Math. Ann. 121 (1949), 131–140. (Cited in Chapter 13.)
[362] R. Schoof, Catalan’s Conjecture. Universitext. Springer-Verlag London, Ltd., Lon-
don (2008). (Cited in Chapter 4.)
[363] H. G. Senge and E. G. Straus, PV-numbers and sets of multiplicity, Period. Math.
216 Bibliography
[364] J.-P. Serre, Dépendance d’exponentielles p-adiques. In: Séminaire Delange-Pisot-

Poitou. Théorie des nombres, Tome 7 (1965-1966) no. 2 , Exposé no. 15, p. 1–14.
(Cited in Appendix F.)
[365] , Local fields. Graduate Texts in Mathematics, 67. Springer-Verlag, New
York-Berlin, 1979. (Cited in Appendix F.)
[366] T. N. Shorey, On gaps between numbers with a large prime factor. II, Acta Arith.
25 (1973/74), 365–373. (Cited in Chapter 5.)
[367] , Linear forms in the logarithms of algebraic numbers with small coefficients
I, J. Indian Math. Soc. (N. S.) 38 (1974), 271–284. (Cited in Chapters 2 & 5.)
[368] , Linear forms in the logarithms of algebraic numbers with small coefficients
II, J. Indian Math. Soc. (N. S.) 38 (1974), 285–292. (Cited in Chapters 2 & 5.)
[369] , On the greatest prime factor of ax m C by n , Acta Arith. 36 (1980), 21–25.
[370] , Divisors of convergents of a continued fraction, J. Number Theory 17
[371] , Some conjectures in the theory of exponential Diophantine equations,
Publ. Math. Debrecen 56 (2000), 631–641. (Cited in Chapter 13.)
[372] T. N. Shorey and C. L. Stewart, On the Diophantine equation ax 2t C bx t y C cy 2 D
d and pure powers in recurrence sequences, Math. Scand. 52 (1983), 24–36. (Cited
in Chapters 3 & 6.)
[373] , Pure powers in recurrence sequences and some related Diophantine equa-
tions, J. Number Theory 27 (1987), 324–352. (Cited in Chapter 6.)
[374] T. N. Shorey and R. Tijdeman, On the greatest prime factors of polynomials at
integer points, Compositio Math. 33 (1976), 187–195. (Cited in Chapter 3.)
[375] , New applications of Diophantine approximations to Diophantine equa-
tions, Math. Scand. 39 (1976), 5–18. (Cited in Chapters 6 & 13.)
[376] , Exponential Diophantine equations, Cambridge Tracts in Mathematics
87, Cambridge University Press, Cambridge, 1986. (Cited in Preface, Chapters 3,
4 & 6, and Appendices B & C.)
[377] C. L. Siegel (under the pseudonym X), The integer solutions of the equation y 2 D
ax n C bx n 1 C C k, (Extract from a letter to Prof. L. J. Mordell), J. London
[378] S. Siksek, The modular approach to Diophantine equations. In: Explicit methods
in number theory, 151–179, Panor. Synthèses, 36, Soc. Math. France, Paris, 2012.
[379] J. Silliman and I. Vogt, Powers in Lucas sequences via Galois representations, Proc.
Amer. Math. Soc. 143 (2015), 1027–1041. (Cited in Chapter 7.)
[380] J. H. Silverman, Wieferich’s criterion and the abc-conjecture, J. Number Theory
30 (1988), 226–237. (Cited in Chapter 7.)
Bibliography 217
[381] C. M. Skinner, On the Diophantine equation ap x C bq v D c C dp z q w , J. Number

Theory 35 (1990), 194–207. (Cited in Chapter 6.)
[382] N. P. Smart, The algorithmic resolution of Diophantine equations. London Mathe-
matical Society Student Texts, 41. Cambridge University Press, Cambridge, 1998.
[383] C. Smyth, The Mahler measure of algebraic numbers: a survey. In: Number theory
and polynomials, 322–349, London Math. Soc. Lecture Note Ser., 352, Cambridge
[384] V. G. Sprindžuk, Concerning Baker’s theorem on linear forms in logarithms, Dokl.
Akad. Nauk BSSR 11 (1967), 767–769 (in Russian). (Cited in Chapter 1.)
[385] , Estimates of linear forms with p-adic logarithms of algebraic numbers,
Vesci Akad. Navuk BSSR Ser. Fiz.-Mat. Navuk 1968 (1968), 5–14 (in Russian).
[386] , Classical Diophantine Equations, Lecture Notes in Math. 1559, Springer-
Verlag, Berlin, 1993. (Cited in Preface and Chapters 4 & 6.)
[387] H. M. Stark, A complete determination of the complex quadratic fields of class-
number one, Michigan Math. J. 14 (1967), 1–27. (Cited in Chapter 3.)
[388] , A transcendence theorem for class-number problems, Ann. of Math. 94
(1971), 153–173. (Cited in Chapter 3.)
[389] , A transcendence theorem for class-number problems. II, Ann. of Math.
96 (1972), 174–209. (Cited in Chapter 3.)
[390] , Effective estimates of solutions of some Diophantine equations, Acta Arith.
24 (1973), 251–259. (Cited in Chapter 13.)
[391] , On complex quadratic fields wth class-number two, Math. Comp. 29
(1975), 289–302. (Cited in Chapter 3.)
[392] C. L. Stewart, On divisors of Fermat, Fibonacci, Lucas and Lehmer numbers, Proc.
London Math. Soc. 35 (1977), 425–447. (Cited in Chapter 7.)
[393] , Primitive divisors of Lucas and Lehmer numbers. In: Transcendence The-
ory: Advances and Applications (Cambridge, 1976), pp. 79–92. Academic Press,
London, 1977. (Cited in Chapter 7.)
[394] , On the representation of an integer in two different bases, J. Reine Angew.
[395] , On divisors of terms of linear recurrence sequences, J. Reine Angew.
[396] , A note on the product of consecutive integers. In: Topics in Classical
Number Theory, Vols. I, II, Colloq. Math. Soc. János Bolyai, Vol. 34 (North-
Holland, Amsterdam, 1984), pp. 1523–1537. (Cited in Chapter 3.)
[397] , On sets of integers whose shifted products are powers, J. Combin. Theory
Ser. A 115 (2008), 662–673. (Cited in Chapters 5 & 9.)
218 Bibliography
[398] , On heights of multiplicatively dependent algebraic numbers, Acta Arith.

133 (2008), 97–108. (Cited in Chapter 9.)
[399] , On the greatest square-free factor of terms of a linear recurrence sequence.
In: Diophantine equations, 257–264, Tata Inst. Fund. Res. Stud. Math., 20, Tata Inst.
Fund. Res., Mumbai, 2008. (Cited in Chapter 3.)
[400] , On divisors of Lucas and Lehmer numbers, Acta Math. 211 (2013), 291–
314. (Cited in Preface and Chapter 7.)
[401] , On prime factors of terms of linear recurrence sequences. In: Number
theory and related fields, 341–359, Springer Proc. Math. Stat., 43, Springer, New
York, 2013. (Cited in Chapter 7.)
[402] C. L. Stewart and R. Tijdeman, On the Oesterlé-Masser conjecture, Monatsh. Math.
102 (1986), 251–257. (Cited in Chapter 8.)
[403] , On the greatest prime factor of .ab C 1/.bc C 1/.ca C 1/, Acta Arith. 79
(1997), 93–101. (Cited in Chapter 3.)
[404] C. L. Stewart and K. Yu, On the abc conjecture, Math. Ann. 291 (1991), 225–230.
[405] , On the abc conjecture. II, Duke Math. J. 108 (2001), 169–181. (Cited in
Preface and Chapter 8.)
[406] J. Stiller, The Diophantine equation x 2 C 119 D 15 2n has exactly six solutions,
Rocky Mountain J. Math. 26 (1996), 295–298. (Cited in Chapter 7.)
[407] R. J. Stroeker and R. Tijdeman, Diophantine equations. In: Computational methods
in number theory, Part II, 321–369, Math. Centre Tracts, 155, Math. Centrum,
Amsterdam, 1982. (Cited in Chapter 9.)
[408] R. J. Stroeker and N. Tzanakis, Solving elliptic Diophantine equations by estimat-
ing linear forms in elliptic logarithms, Acta Arith. 67 (1994), 177–196. (Cited in
Chapter 1.)
[409] L. Szalay, The equations 2n ˙ 2m ˙ 2l D z 2 , Indag. Math. (N.S.) 13 (2002),
[410] Sz. Tengely, On the Diophantine equation x 2 C a2 D 2y p , Indag. Math. (N.S.) 15
(2004), 291–304. (Cited in Chapter 3.)
[411] E. Thomas, Complete solutions to a family of cubic Diophantine equations, J.
[412] A. Thue, Ueber Annäherungswerte algebraischer Zahlen, J. Reine Angew. Math.
135 (1909), 284–305. (Cited in Chapter 4.)
[413] , Berechnung aller Lösungen gewisser Gleichungen von der Form ax r
r
by D f , Kra. Vidensk. Selsk. Skrifter. I. Math. Kl. 2 (1918), 9 pages. (Cited in
Chapter 4.)
[414] R. Tijdeman, On integers with many small prime factors, Compositio Math. 26
(1973), 319–330. (Cited in Chapter 3.)
Bibliography 219
[415] , On the maximal distance between integers composed of small primes,

Compositio Math. 28 (1974), 159–162. (Cited in Chapter 3.)
[416] , On the equation of Catalan, Acta Arith. 29 (1976), 197–209. (Cited in
Chapter 4.)
[417] , Applications of the Gel’fond-Baker method to rational number theory. In:
Topics in number theory (Proc. Colloq., Debrecen, 1974), pp. 399–416. Colloq.
Math. Soc. Janos Bolyai, Vol. 13, North-Holland, Amsterdam, 1976. (Cited in
Chapter 4.)
[418] , Exponential Diophantine equations 1986–1996, Proceedings of the Num-
ber Theory Conference, Eger 1996, 523–539, Eds. Győry, Pethő, Sós. De Gruyter,
[419] R. Tubbs, Hilbert’s seventh problem. Solutions and extensions. Institute of Math-
ematical Sciences Lecture Notes, 2. Hindustan Book Agency, New Delhi, 2016.
[420] N. Tzanakis, Solving elliptic Diophantine equations by estimating linear forms in
elliptic logarithms. The case of quartic equations, Acta Arith. 75 (1996), 165–190.
[421] , Elliptic Diophantine equations. A concrete approach via the elliptic loga-
rithm. De Gruyter Series in Discrete Mathematics and Applications, 2. Walter de
Gruyter GmbH & Co. KG, Berlin, 2013. (Cited in Chapter 1.)
[422] N. Tzanakis and B. M. M. de Weger, How to explicitly solve a Thue–Mahler
equation, Compositio Math. 84 (1992), 223–288. (Cited in Chapter 6.)
[423] M. Ulas, Some experiments with Ramanujan–Nagell type Diophantine equations,
Glas. Mat. Ser. III 49 (2014), 287–302. (Cited in Chapter 7.)
[424] N. K. Vereshchagin, The problem of the appearance of a zero in a linear recursive
sequence, Mat. Zametki 38 (1985), 177–189, 347 (in Russian). (Cited in Chap-
ter 13.)
[425] P. A. Vojta, Integral points on varieties. Ph.D. thesis, Harvard University, 1983.
[426] P. M. Voutier, Primitive divisors of Lucas and Lehmer sequences, Math. Comp. 64
(1995), 869–888. (Cited in Chapter 7.)
[427] , Primitive divisors of Lucas and Lehmer sequences, II, J. Théor. Nombres
Bordeaux 8 (1996), 251–274. (Cited in Chapter 7.)
[428] , Primitive divisors of Lucas and Lehmer sequences, III, Math. Proc. Cam-
bridge Philos. Soc. 123 (1998), 407–419. (Cited in Chapter 7.)
[429] M. Waldschmidt, Sur l’équation de Pillai et la différence entre deux produits de
puissances de nombres entiers, C. R. Math. Rep. Acad. Sci. Canada 12 (1990),
[430] , Minorations de combinaisons linéaires de logarithmes de nombres algébri-
ques, Canadian J. Math. 45 (1993), 176–224. (Cited in Chapter 2.)
220 Bibliography
[431] , On a problem of Mahler concerning the approximation of exponentials

and logarithms, Publ. Math. Debrecen 56 (2000), 713–738. (Cited in Chapter 13.)
[432] , Diophantine Approximation on Linear Algebraic Groups. Transcendence
Properties of the Exponential Function in Several Variables, Grundlehren Math.
Wiss. 326, Springer, Berlin, 2000. (Cited in Preface, Chapters 1, 2, 4, 10 & 13, and
Appendix B.)
[433] , Linear independence measures for logarithms of algebraic numbers. In:
Diophantine approximation (Cetraro, 2000), 250–344, Lecture Notes in Math.,
1819, Springer, Berlin, 2003. (Cited in Chapters 2 & 13.)
[434] , Open Diophantine problems, Mosc. Math. J. 4 (2004), 245–305. (Cited in
Chapter 13.)
[435] , Perfect Powers: Pillai’s works and their developments. In: Collected works
of S. Sivasankaranarayana Plliai, Eds. R. Balasubramanian and R. Thangadurai,
Collected Works Series, no. 1, Ramanujan Mathematical Society, Mysore, 2010.
[436] M. Ward, The intrinsic divisors of Lehmer numbers, Ann. Math. (2) 62 (1955),
[437] B. M. M. de Weger, Equal binomial coefficients: some elementary considerations,
J. Number Theory 63 (1997), 373–386. (Cited in Chapter 13.)
[438] B. M. M. de Weger and C. E. van de Woestijne, On the diameter of sets of almost
powers, Acta Arith. 90 (1999), 371–385. (Cited in Chapter 13.)
[439] , On the power-free parts of consecutive integers, Acta Arith. 90 (1999),
[440] G. Wüstholz, A new approach to Baker’s theorem on linear forms in logarithms. III.
In: New advances in transcendence theory (Durham, 1986), 399–410, Cambridge
[441] , One century of logarithmic forms. In: A panorama of number theory
[442] T. Yamada, A note on the paper by Bugeaud and Laurent "Minoration effective de
la distance p-adique entre puissances de nombres algébriques”, J. Number Theory
130 (2010), 1889–1897. (Cited in Chapters 2, 7 & 12.)
[443] K. Yu, Linear forms in p-adic logarithms, Acta Arith. 53 (1989), 107–186. (Cited
in Chapters 1 & 2.)
[444] , Linear forms in p-adic logarithms. II, Compositio Math. 74 (1990), 15–
113. (Cited in Appendix B.)
[445] , Linear forms in p-adic logarithms. III, Compositio Math. 91 (1994), 241–
[446] , p–adic logarithmic forms and group varieties, I, J. Reine Angew. Math.
Bibliography 221
[447] , p–adic logarithmic forms and group varieties, II, Acta Arith. 89 (1999),
[448] , Report on p-adic logarithmic forms. In: A panorama of number theory
[449] , p–adic logarithmic forms and group varieties, III, Forum Math. 19 (2007),
[450] , p–adic logarithmic forms and a problem of Erdős, Acta Math. 211 (2013),
[451] K. Yu and L. Hung, On binary recurrence sequences, Indag. Math. (N.S.) 6 (1995),
[452] K. Zsigmondy, Zur Theorie der Potenzreste, Monatsh. Math. 3 (1892), 265–284.
[453] W. Zudilin, A new lower bound for k.3=2/k k, J. Théor. Nombres Bordeaux 19
(2007), 311–323. (Cited in Chapter 6.)
Index
abc-conjecture, 109 integral S -unit, 25, 75, 88–90

irrationality exponent, 43, 168
Baker’s theorem, 2 irrationality measure, 168
Baker–Davenport lemma, 37
Bertrand’s postulate, 183 Krasner’s lemma, 193
Kronecker’s theorem, 125, 175
Catalan equation, 62
characteristic polynomial, 30 Lehmer’s problem, 125, 163
companion polynomial, 30 length, 176
continued fraction, 167, 167 Lindemann–Weierstrass’ theorem, 1,
convergent, 168 157
linear recurrence sequence, 30, 80, 86,
degenerate recurrence sequence, 31 91, 95, 162
Diophantine tuple, 36, 71, 73 Liouville number, 25, 43, 67, 169
dominant root, 31, 86, 91, 122, 162 Liouville’s inequality, 22, 47, 55, 132,
134, 136, 140, 148, 155, 176,
effective irrationality exponent, 25, 27, 187
55, 65, 67–69, 168 Liouville’s theorem, 1, 27, 68, 69, 77,
effective irrationality measure, 65, 92, 169, 176, 187
168 local degree, 173
Euler totient function, 183 Lucas–Lehmer sequence, 33, 98
Fibonacci sequence, 33 Möbius function, 183

fundamental system of units, 180 Minkowski’s theorem, 48, 128
fundamental unit, 36, 78, 121, 180
Nagell–Ljunggren equation, 160
Gelfond–Schneider’s theorem, 2 norm, 172, 173
Grimm’s conjecture, 73, 161 Northcott’s theorem, 52, 54, 56, 85, 176
Hall’s conjecture, 158 p-adic exponential function, 193

Hermite–Lindemann theorem, 1 p-adic logarithm function, 193
Hilbert’s seventh problem, 2 p-adic Schwarz lemma, 194
homogeneous cyclotomic polynomial, Pell equation, 179
95 Pellian equation, 35, 53, 121, 179
hyperelliptic equation, 36, 55, 85, 88 Pillai equation, 65, 66
hypergeometric method, 73, 92 Pillai’s conjecture, 65, 159
224 Index
Prime number theorem, 183 Schwarz lemma, 151, 191

primitive divisor, 95 Shafarevich’s theorem, 66
principal unit, 147, 193 simultaneous linear forms in
product formula, 174 logarithms, 73, 117
Skolem’s problem, 163
Ramanujan–Nagell equation, 58, 102 superelliptic equation, 55, 65, 66, 85
ramification index, 150, 172, 173, 191
regulator, 180 Thue equation, 35, 52, 56, 65, 66, 69,
repunit, 80 71, 81, 100, 101, 123
residue degree, 172, 173, 191 Thue–Mahler equation, 85, 93
Ridout’s theorem, 78, 91, 92, 169 Tribonacci sequence, 45, 160
Roth’s theorem, 27, 39, 55, 169
unit equation, 50, 57, 65, 123
S -unit equation, 83, 93 Waring’s problem, 76
Schanuel’s conjecture, 157 Weil height, 175
Schmidt subspace theorem, 92, 123, 170
Schubfachprinzip, 3, 75, 119, 154 zero lemma, 187
IRMA Lectures in Mathematics
and Theoretical Physics 28

Yann Bugeaud
Yann Bugeaud
The aim of this book is to serve as an introductory text to the theory of linear
forms in the logarithms of algebraic numbers, with a special emphasis on a
Yann Bugeaud
large variety of its applications. We wish to help students and researchers to
learn what is hidden inside the blackbox ‘Baker’s theory of linear forms in
logarithms’ (in complex or in p-adic logarithms) and how this theory applies to
many Diophantine problems, including the effective resolution of Diophantine
Linear Forms in
equations, the abc-conjecture, and upper bounds for the irrationality measure
of some real numbers. Logarithms
Written for a broad audience, this accessible and self-contained book can be
used for graduate courses (some 30 exercises are supplied). Specialists will
appreciate the inclusion of over 30 open problems and the rich bibliography
and Applications
of over 450 references.
ISBN 978-3-03719-183-5
www.ems-ph.org
Bugeaud IRMA 28 | FONT: Rotis Sans | Farben: Pantone 287, Pantone 116 | 170 x 240 mm | RB: 15??? mm

Linear Forms in Logarithms and Applications

Uploaded by

Linear Forms in Logarithms and Applications

Uploaded by

IRMA Lectures in Mathematics

and Theoretical Physics 28

Linear Forms in Logarithms and Applications

Institut de Recherche Mathématique Avancée

For a complete listing see our homepage at www.ems-ph.org.

© 2018 European Mathematical Society

European Mathematical Society Publishing House

Typeset using the author’s TEX files: Nicole Bloye, Cardiff, UK

— Les maths, m’sieur, ça fonctionne toujours à côté de ses grolles.

In 1748 Leonhard Euler published Introductio in analysin infinitorum where, among

the expression ˛ ˇ for an algebraic

solved in 1934, independently and simultaneously, by Aleksandr Gelfond and Theodor

1. Brief introduction to linear forms in logarithms . . . . . . . . . . . . . . . . . 1

4.8. The Catalan equation and the Pillai conjecture . . . . . . . . . . . . . . . 62

10. Multiplicative dependence relations between algebraic numbers . . . . . . . . . 125

log : unless otherwise specified, log z denotes the principal determination

1.1. Linear forms in complex logarithms

The expression ˛ ˇ for an algebraic base

at least an irrational number.

Theorem 1.3 was generalized to linear combinations of n logarithms of algebraic

ˇ0 C ˇ1 log ˛1 C C ˇn log ˛n ; (1.1)

where ˛1 ; : : : ; ˛n , ˇ1 ; : : : ; ˇn are non-zero algebraic numbers and ˇ0 is algebraic, vanishes

B WD maxf3; jb1 j; : : : ; jbn jg and Aj WD maxf3; jxj j; jyj jg: (1.2)

We consider the rational number

0 < maxfjb1 j; : : : ; jbn jg  B

ja1b1 anbn 1j ineff

ja1b1 anbn 1j  exp. C log B/ D B C

In many applications, especially to classical families of Diophantine equations, we are

0 < j˛1b1 ˛nbn ˛nC1 1j < e ıB

j˛1b1 ˛nbn 1j > B C … log …

and C is an effectively computable real number depending only on n and D.

1.2. Linear forms in p-adic logarithms

1.3. Linear forms in elliptic logarithms

Let n be a positive integer and E1 ; : : : ; En elliptic curves over an algebraic number

2.1. Lower bounds for linear forms in complex logarithms

log jˇ0 C ˇ1 log ˛1 C C ˇn log ˛n j

log jb1 log ˛1 C C bn log ˛n j

In particular, choosing E D e and E D 3D, we get

log jb1 log ˛1 C C bn log ˛n j

D 3 .log B0 /.log Aj /.log E /  .log D/.log E/2 ; 1  j  n;

where B0 D B in the general case and B0 D B 0 in the homogeneous rational case.

B n;D ı 1 log A1 : : : log An 1 log ı 1 log.D n 1 A1 : : : An 1 / log An :

we see that, since B 0  E, Theorem 2.1 implies the lower bound

j˛1b1 : : : ˛nbn 1j; (2.2)

which occurs frequently in Diophantine questions. We choose below to mainly consider

that b1 log ˛1 C C bn log ˛n is close to an integer multiple of 2i. Precisely, we derive

b0 log. 1/ C b1 log ˛1 C C bn log ˛n

 WD jb0 log. 1/ C b1 log ˛1 C C bn log ˛n j

satisfies j˛1b1 : : : ˛nbn 1j  

Then, we have the lower bound

log jb1 log ˛1 C b2 log ˛2 j  25:2 D 04 .log A1 /.log A2 /.log B 0 /2 : (2.5)

numbers such that

Assume furthermore that 2  E  minfA3D=2

log jb1 log ˛1 C b2 log ˛2 j  35:1D 4 .log A1 /.log A2 /.log B 0 /2 .log E/ 3

and the theorem holds.

Furthermore, we may assume that

since x1  6. Consequently, we have checked that

2.2. Multiplicative dependence relations between algebraic numbers

jb1 log ˛1 C C bn log ˛n j;

when it is non-zero. However, in some situations, we derive linear forms in logarithms

b10 log ˛1 C C bn0 log ˛n D 0:

2.3. Lower bounds for linear forms in p-adic logarithms

This defines a valuation vp on K which extends the p-adic valuation vp on Q normalized

Let ˛1 ; : : : ; ˛n be elements of K and b1 ; : : : ; bn non-zero rational integers. We look

vp .˛1b1 : : : ˛nbn 1/;

Yu [449] obtained an improvement of Theorem 2.10, with a better dependence on n,

In particular, we get either

B  2Bn .log A1 / .log An / c0 .n; D/ (2.9)

The formulation of Theorem 2.11 looks complicated, but it essentially corresponds to

Then, we have the upper bound

and at least one of the two following conditions

log Aj  maxflog jxj j; log jyj j; E log pg; j D 1; 2;

Then, we have the upper bound

if p is odd or if p D 2 and v2 . yx22 1/  2.

and equal. Consequently, p k divides yx 1 and we apply Theorem 2.13 with g D 1,

3.1. On the distance between powers of 2 and powers of 3

0 < maxfjb1 j; : : : ; jbn jg B

ja1b1 anbn 1j ineff

ja1b1 anbn 1j exp. C log B/ D B C

D 3 .log B0 /.log Aj /.log E / .log D/.log E/2 ; 1 j n;

B n;D ı 1 log A1 : : : log An 1 log ı 1 log.D n 1 A1 : : : An 1 / log An :

we see that, since B 0 E, Theorem 2.1 implies the lower bound

that b1 log ˛1 C C bn log ˛n is close to an integer multiple of 2i. Precisely, we derive

WD jb0 log. 1/ C b1 log ˛1 C C bn log ˛n j

satisfies j˛1b1 : : : ˛nbn 1j

log jb1 log ˛1 C b2 log ˛2 j 25:2 D 04 .log A1 /.log A2 /.log B 0 /2 : (2.5)

Assume furthermore that 2 E minfA3D=2

log jb1 log ˛1 C b2 log ˛2 j 35:1D 4 .log A1 /.log A2 /.log B 0 /2 .log E/ 3

since x1 6. Consequently, we have checked that

B 2Bn .log A1 / .log An / c0 .n; D/ (2.9)

log Aj maxflog jxj j; log jyj j; E log pg; j D 1; 2;

if p is odd or if p D 2 and v2 . yx22 1/ 2.

Proof. Let n 2 be an integer and define m and m0 by the conditions

2B=3 q1B=3 v 2u: (3.4)

Furthermore, if a; b are integers with 1 b < a, then we have

and, if a 2n , then we get

p n is sufficiently large in terms of , Theorem 3.6 improves the upper bound

Since pj j log j , for j D 2; : : : ; k, by Theorem D.2, we deduce from (3.8) and

log jp1"1 u1 : : : pk"k uk 1j c1k .log p1 / .log pk / .log log n/

maxfja0 j; ja00 j; 2g jnjc4 :

If jnj > 21=c4 , then we get a maxfja0 j; ja00 j; 2g jnjc4 and

un D f1 .n/˛1n C C ft .n/˛tn ; for n 0:

un D f1 .n/˛1n C C ft .n/˛tn ; for n 0;

where r1 ; : : : ; rk are non-negative integers and rk 1. Then, we have

and, since < 1, we get

log ƒ v c1k .log p1 / .log pk / .log n/2 ;

n v .c1 log pk /k .log n/2 ;

un D f1 .n/˛1n C C ft .n/˛tn ; for n 0;

in integers n; y; q with q a prime number, jyj 2, and un ¤ f1 ˛1n , implies that

5Fm2 L2m D 4; for m 1 odd:

log maxfm; ng a;b;u;v log ƒ a;b;u;v . minfm; ng/:

jm0 n C ˇ0 j < A0 C0 m ; (3.25)

jm 0 n C ˇ0j < m 1 "

log m 4c1 .log a/.log log m/: (3.29)

This ensures that k 1.

BC.m; a/ C BC.m; b/ r C t 2 k: (3.30)

.k C 1/ log > log log m log.4 log a/:

Combined with (3.30) and our choice of , this gives