Approximation by Algebraic Numbers PDF

This page intentionally left blank
CAMBRIDGE TRACTS IN MATHEMATICS

General Editors
W. FULTON, A. KATOK, F. KIRWAN, P. SARNAK, B. SIMON

B. BOLLOBAS.
160
Approximation by Algebraic Numbers
Approximation by Algebraic Numbers

YANN BUGEAUD
Universite Louis Pasteur, Strasbourg
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, So Paulo

Cambridge University Press
The Edinburgh Building, Cambridge CB2 2RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521823296
Cambridge University Press 2004
This publication is in copyright. Subject to statutory exception and to the provision of
relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
First published in print format 2004
ISBN-13
ISBN-10
978-0-511-26591-4 eBook (NetLibrary)

0-511-26591-3 eBook (NetLibrary)
ISBN-13
ISBN-10
978-0-521-82329-6 hardback
0-521-82329-3 hardback
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Contents
Preface
Frequently used notation
page ix
xiv
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
Approximation by rational numbers

Dirichlet and Liouville
Continued fractions
The theorem of Khintchine
The DufnSchaeffer Conjecture
Complementary results on continued fractions
Exercises
Notes
1
1
5
12
18
19
21
23
2
2.1
2.2
2.3
2.4
2.5
2.6
Approximation to algebraic numbers

Rational approximation
Effective rational approximation
Approximation by algebraic numbers
Effective approximation by algebraic numbers
Remarks on irrationality and transcendence statements
Notes
27
27
29
31
33
39
40
3
3.1
3.2
3.3
3.4
3.5
3.6
The classications of Mahler and Koksma

Mahlers classication
Some properties of Mahlers classication
Koksmas classication
Comparison between both classications
Some examples
Exponents of Diophantine approximation
41
42
45
47
52
61
62
vi
Contents
3.7
3.8
Exercises
Notes
68
70
4
4.1
4.2
4.3
4.4
4.5
Mahlers Conjecture on S-numbers

Statements of the theorems
An auxiliary result
Proof of Theorem 4.3
Exercise
Notes
74
75
78
80
87
88
5
5.1
5.2
5.3
5.4
5.5
5.6
5.7
5.8
5.9
5.10
Hausdorff dimension of exceptional sets

Hausdorff measure and Hausdorff dimension
Upper bound for the Hausdorff dimension
The mass distribution principle
Regular systems
The theorem of JarnkBesicovitch
Hausdorff dimension of sets of S -numbers
Hausdorff dimension of sets of S-numbers
Restricted Diophantine approximation
Exercises
Notes
90
90
93
95
98
103
105
110
113
114
117
6
6.1
6.2
6.3
6.4
6.5
6.6
6.7
6.8
Deeper results on the measure of exceptional sets

Optimal regular systems
A Khintchine-type result
Hausdorff measure of exceptional sets
Sets with large intersection properties
Application to the approximation by algebraic numbers
Exercises
Notes
122
123
125
129
130
130
131
136
137
7
7.1
7.2
7.3
7.4
7.5
7.6
7.7
On T-numbers and U -numbers

T -numbers do exist
The inductive construction
Completion of the proof of Theorem 7.1
On the gap between wn and wn
Hausdorff dimension and Hausdorff measure
On U -numbers
A method of Guting
139
140
141
149
151
152
153
159
Contents
vii
7.8 Brief summary of the results towards the Main Problem

7.9 Exercises
7.10 Notes
161
162
163
8
8.1
8.2
8.3
166
166
171
8.4
8.5
Other classications of real and complex numbers

Sprindzuks classication
Another classication proposed by Mahler
Transcendence measures and measures of algebraic
approximation
Exercises
Notes
9
9.1
9.2
9.3
9.4
9.5
Approximation in other elds

Approximation in the eld of complex numbers
Approximation in the eld of Gaussian integers
Approximation in the p-adic elds
Approximation in elds of formal power series
Notes
191
191
193
194
199
201
10
10.1
10.2
10.3
Conjectures and open questions

The Littlewood Conjecture
Open questions
Notes
204
204
206
217
A.1
A.2
A.3
A.4
A.5
Appendix A Lemmas on polynomials

Denitions and useful lemmas
Liouvilles inequality
Zeros of polynomials
Exercises
Notes
219
219
222
227
233
233
Appendix B
235
References
Index
Geometry of numbers
180
184
188
240
273
Preface
Und alles ist mir dann immer wieder zerfallen, auf dem Konzentrationshohepunkt
ist mir dann immer wieder alles zerfallen.
Thomas Bernhard
(...) il faut continuer, je vais continuer.
Samuel Beckett
The central question in Diophantine approximation is: how well can a given
real number be approximated by rational numbers, that is, how small can the
difference | p/q| be made for varying rational numbers p/q? The accuracy
of the approximation of by p/q is being compared with the complexity of
the rational number p/q, which is measured by the size of its denominator q. It
follows from the theory of continued fractions (or from Dirichlets Theorem)
that for any irrational number there exist innitely many rational numbers
p/q with | p/q| < q 2 . This can be viewed as the rst general result in
this area.
There are two natural generalizations of the central question. On the one
hand, one can treat rational numbers as algebraic numbers of degree one and
study, for a given positive integer n, how well can be approximated by algebraic numbers of degree at most n. On the other hand, p/q can be viewed
as q p, that is as P( ), where P(X ) denotes the integer polynomial q X p.
Thus, for a given positive integer n, one may ask how small |P( )| can be made
for varying integer polynomials P(X ) of degree at most n. To do this properly,
one needs to dene a notion of size, or complexity, for algebraic numbers
and for integer polynomials P(X ), and we have to compare the accuracy of
ix
Preface
the approximation of by (resp. the smallness of |P( )|) with the size of
(resp. of P(X )). In both cases, we use for the size the naive height H: the
height H(P) of P(X ) is the maximum of the absolute values of its coefcients
and the height H() of is that of its minimal polynomial over Z.
In 1932, Mahler proposed to classify the real numbers (actually, the complex numbers) into several classes according to the various types of answers to
the latter question, while in 1939 Koksma introduced an analogous classication based on the former question. In both cases, the algebraic numbers form
one of the classes. Let be a real number and let n be a positive integer. According to Mahler, we denote by wn ( ) the supremum of the real numbers w
for which there exist innitely many integer polynomials P(X ) of degree at
most n satisfying
0 < |P( )| H(P)w ,
and we divide the set of real numbers into four classes according to the behaviour of the sequence (wn ( ))n1 . Following Koksma, we denote by wn ( )
the supremum of the real numbers w for which there exist innitely many real
algebraic numbers of degree at most n satisfying
0 < | | H()w1 .
It turns out that both classications coincide, inasmuch as each of the four
classes dened by Mahler corresponds to one of the four classes dened by
Koksma. However, they are not strictly equivalent, since there exist real numbers for which wn ( ) and wn ( ) differ for any integer n at least equal
to 2. In addition, it is a very difcult (and, often, still open) question to
determine to which class a given classical number like , e, (3), log 2, etc.
belongs.
The present book is mainly concerned with the following problem: given
two non-decreasing sequences of real numbers (wn )n1 and (wn )n1 satisfying some necessary restrictions (e.g. wn wn wn + n 1), does there
exist a real number with wn ( ) = wn and wn ( ) = wn for all positive integers n? This question is very far from being solved, although we know (see
Chapter 4) that almost all (in the sense of the Lebesgue measure on the line)
real numbers share the same approximation properties, namely they satisfy
wn ( ) = wn ( ) = n for all positive integers n.
There are essentially two different points of view for investigating such a
problem. We may try to construct explicitly (or semi-explicitly) real numbers
with the required properties (Chapter 7) or, if this happens to be too difcult, we may try to prove the existence of real numbers with a given property
by showing that the set of these numbers has positive Hausdorff dimension
Preface
xi
(Chapters 5 and 6). Most often, however, we are unable to exhibit a single explicit example of such a number.
The content of the present book is as follows. Chapter 1 is devoted to the
approximation by rational numbers. We introduce the notion of continued fractions and establish their main properties needed to prove the celebrated metric
theorem of Khintchine saying that, for a continuous function : R1 R>0
such that x x(x) is non-increasing, the equation |q p| < (q) has
innitely many integer solutions ( p, q) with q positive for either almost no
+
(q) conor almost all real numbers , according to whether the sum q=1
verges or diverges.
In Chapter 2, we briey survey the approximation to algebraic numbers by
algebraic numbers, and recall many classical results (including Roths Theorem and Schmidts Theorem). We make a clear distinction between effective
and ineffective statements.
Mahlers and Koksmas classications of real numbers are dened in
Chapter 3, where we show, following ideas of Wirsing, how closely they are related. Some links between simultaneous rational approximation and these classications are also mentioned, and we introduce four other functions closely
related to wn and wn .
In Chapter 4, we establish Mahlers Conjecture to the effect that almost all
real numbers satisfy wn ( ) = wn ( ) = n for all positive integers n. This
result, rst proved by Sprindzuk in 1965, has been rened and extended since
that time and we state the most recent developments, essentially due to a new
approach found by Kleinbock and Margulis.
Exceptional sets are investigated from a metric point of view in Chapters
5 and 6. To this end, we introduce a classical powerful tool for discriminating
between sets of Lebesgue measure zero, namely the notion of Hausdorff dimension. We recall the basic denitions and some well-known results useful
in our context. This allows us to prove the theorem of Jarnk and Besicovitch saying that for any real number 1 the Hausdorff dimension of the
set of real numbers with w1 ( ) 2 1 is equal to 1/ . We also establish its generalization to any degree n (with w1 ( ) 2 1 replaced by
wn ( ) (n + 1) 1) obtained in 1970 by A. Baker and Schmidt. Chapter 6
is devoted to rened statements and contains general metric theorems on sets
of real numbers which are very close to innitely many elements of a xed set
of points which are, in some sense, evenly distributed.
In Chapter 7, we prove, following ideas of Schmidt, that the class formed
by the real numbers with lim supn+ wn ( )/n innite and wn ( ) nite for
all positive integers n is not empty. At the same time, we show that there exist
real numbers for which the quantities wn ( ) and wn ( ) differ by preassigned
xii
Preface
values, a result due to R. C. Baker. The real numbers with the required property are obtained as limits of sequences of algebraic numbers. This illustrates
the importance of results on approximation of algebraic numbers by algebraic
numbers. The remaining part of Chapter 7 is concerned with some other (simpler) explicit constructions.
Mahlers and Koksmas classications emphasize the approximation by algebraic numbers of bounded degree. We may as well exchange the roles played
by degree and height or let both vary simultaneously. We tackle this question in Chapter 8 by considering the classication introduced by Sprindzuk in
1962 and the so-called order functions dened by Mahler in 1971. Further,
some recent results of Laurent, Roy, and Waldschmidt expressed in terms of a
more involved notion of height (namely, the absolute logarithmic height) are
given.
In Chapter 9, we briey discuss approximation in the complex eld, in the
Gaussian eld, in p-adic elds, and in elds of formal Laurent series.
Chapter 10, which begins by a brief survey on the celebrated Littlewood
Conjecture, offers a list of open questions. We hope that these will motivate
further research.
Finally, there are two appendices. Appendix A is devoted to lemmas on
zeros of polynomials: all proofs are given in detail and the statements are the
best known at the present time. Appendix B lists classical auxiliary results
from the geometry of numbers.
The Chapters are largely independent of each other.
We deliberately do not give proofs to all the theorems quoted in the main
part of the text. We have clearly indicated when this is the case (see below).
Furthermore, we try, in the end-of-chapter notes, to be as exhaustive as possible
and to quote less-known papers, which, although interesting, did not yield up
to now to further research. Of course, exhaustivity is an impossible task, and
it is clear that the choice of the references concerning works at the border of
the main topic of this book reects the personal taste and the limits of the
knowledge of the author.
The purpose of the exercises is primarily to give complementary results,
thus they are often an adaptation of an original research work to which the
reader is directed.
There exist already many textbooks dealing, in part, with the subject of the
present one, e.g., by Schneider [517], Sprindzuk [539, 540], A. Baker [44],
Schmidt [510, 512], Bernik and Melnichuk [90], Harman [273], and Bernik
and Dodson [86]. However, the intersection rarely exceeds two or three chapters. Special mention should be made to the wonderful book of Koksma [332],
which contains an impressive list of references which appeared before 1936.
Preface
xiii
Maurice Mignotte and Michel Waldschmidt encouraged me constantly.

Many colleagues sent me comments, remarks, and suggestions. I am very
grateful to all of them. Special thanks are due to Guy Barat and Damien Roy,
who carefully read several parts of this book.
The present book will be regularly updated on my institutional Web page:
http://www-irma.u-strasbg.fr/bugeaud/Book
The following statements are not proved in the present book:

Theorems 1.13 to 1.15, 1.17, 1.20, 2.1 to 2.8, 3.7, 3.8, 3.10, 3.11, 4.4 to 4.7,
Proposition 5.1, Theorems 5.7, 5.9, 5.10, Proposition 6.1, Theorems 6.3 to 6.5,
8.1, 8.5, 8.8 (partially proved), 9.1 to 9.8, Lemma 10.1, Theorems 10.1, B.3,
and B.4.
The following statements are left as exercises:
Theorems 1.16, 1.18, 1.19, 5.4, 5.6, 6.2, 6.9, 6.10, 7.2, 7.3, 7.6, 8.4, 8.6,
8.12, and Proposition 8.1.
deg degree.
positive strictly positive.
N innite set of integers.
[] integer part.
{} fractional part.
|| || distance to the nearest integer.
An empty sum is equal to 0 and an empty product is equal to 1.
the Euler totient function.
logi the i-fold iterated logarithm.
, means that there is an implied constant.

Card the cardinality (of a nite set).
c, ci , constants.
c(var1 , . . . , varm ) constant depending (at most) on the
var1 , . . . , varm .
An the set of real algebraic numbers of degree at most n.
H naive height, Ch. 2.
size, Ch. 8.
h absolute height, Ch. 8.
M Mahlers measure, App. A.
the Lebesgue measure on the real line.
(I ) = |I | the Lebesgue measure of an interval I .
vol the n-dimensional Lebesgue measure.
a measure (not the Lebesgue one).
the number we approximate.
d the degree of (when is algebraic).
the algebraic approximant of .
n (an upper bound for) the degree of the approximant .
H (an upper bound for) the height of the approximant .
xiv
variables
xv
x the n-tuple (x1 , . . . , xn ).

B set of badly approximable real numbers, Ch. 1.
B M set of real numbers whose partial quotients are at most equal to M, Ch. 5.
, x x approximation functions, Ch. 1, 5, 6.
Kn (), Kn ( ), Kn ( ), Kn+ ( ), KS () Ch. 1, 5, 6.
wn (, H ), wn (, H ), wnc (, H ), wn ( ), wn ( ), wnc ( ), w( ), w ( ) Ch. 3
wn ( ), w n ( ), w n ( ), w n ( ) Ch. 3
A-, S-, T -, U -, A -, S -, T -, U -numbers Ch. 3.
Um -numbers Ch. 7.
t ( ), t ( ) the type and the -type of , Ch. 3.
+ Ch. 4.
wn+ ( ) Ch. 5.
Wn ( ), Wn ( ), Wn+ ( ), Wn ( ) Ch. 5.
dim Hausdorff dimension, Ch. 5.
H f Hausdorff H f -measure, Ch. 5.
Hs Hausdorff s-measure, Ch. 5.
order between dimension functions, Ch. 5.
, increasing functions, Ch. 5.
S innite set of real numbers, Ch. 5, 6.
lower order at innity, Ch. 6.
f
M net measure, Ch. 6.
w(,
H ), w(
), (,
H ), (
), w (, H ), w ( ) Ch. 8.
A-, S-, T -, U -numbers Ch. 8.

O(u | ), O (u | ) order functions, Ch. 8.
order between order functions, Ch. 8.
( ) transcendence type of , Ch. 8.
Res the resultant of two polynomials, App. A.
Disc the discriminant of an integer polynomial, App. A.
C a bounded convex body, App. B.
1 , . . . , n the successive minima of a bounded convex body, App. B.
1
Throughout the present Chapter, we are essentially concerned with the following problem: for which functions : R1 R0 is it true that, for a
given real number , or for all real numbers in a given class, the equation
| p/q| < (q) has innitely many solutions in rational numbers p/q? We
begin by stating the results on rational approximation obtained by Dirichlet
and Liouville in the middle of the nineteenth century. In Section 1.2, we dene the continued fraction algorithm and recall the main properties of continued fractions expansions. These are used in Section 1.3 to give a full proof
of a metric theorem of Khintchine. The next two Sections are devoted to the
DufnSchaeffer Conjecture and to some complementary results on continued
fractions.
1.1 Dirichlet and Liouville

Every real number can be expressed in innitely many ways as the limit of
a sequence of rational numbers. Furthermore, for any positive integer b, there
exists an integer a with | a/b| 1/(2b), and one may hope that there
are innitely many integers b for which | a/b| is in fact much smaller than
1/(2b). For instance, this is true when is irrational, as follows from the theory
of continued fractions. In order to measure the accuracy of the approximation
of by a rational number a/b (written in its lowest terms), we have to compare the difference | a/b| with the size, or complexity, of a/b. A possible
denition for this notion is, for example, the number of digits of a plus the
number of digits of b. However, as usual, we dene the size, or the height, of
a/b as the maximum of the absolute values of its denominator and numerator:
this denition is more relevant and can be easily extended (see Denition 2.1).
The rst statement of Theorem 1.1 is often referred to as Dirichlets
Theorem, although it is not explicitly stated under this form in [196], a paper
1
which appeared in 1842. However, it follows easily from the proof of the main
result of [196], which actually provides an extension of the second assertion of
Theorem 1.1 to linear forms and to systems of linear forms.
T HEOREM 1.1. Let and Q be real numbers with Q 1. There exists a
rational number p/q, with 1 q Q, such that

p < 1 .

q qQ
Furthermore, if is irrational, then there exist innitely many rational numbers p/q such that

p < 1 ,
(1.1)

q q2
and if = a/b is rational, then for any rational p/q = a/b with q > 0 we
have

p 1 .

q |b|q
P ROOF. Let t denote the integer part of Q. If is the rational a/b, with a and
b integers and 1 b t, it is sufcient to set p = a and q = b. Otherwise,
the t + 2 points 0, { }, . . . , {t }, and 1 are pairwise distinct and they divide
the interval [0, 1] into t + 1 subintervals. Clearly, at least one of these has its
length at most equal to 1/(t + 1). This means that there exist integers k, and
m k , m with 0 k < t and
|( m ) (k m k )|
1
1
< .
t +1
Q
We conclude by setting p := m m k and q := k, and by noticing that q

satises 1 q t Q. Instead of reasoning with the lengths of the intervals,
we could as well use an argument dating back to Dirichlet [196], now called
Dirichlets Schubfachprinzip (or pigeon-hole principle, or principe des tiroirs,
or principio dei cassetti, or principio de las cajillas, or principiul cutiei, or
skatulya-elv, or lokeroperiaate, or zasada pudekowa). It asserts that at least
two among the t + 2 points 0, { }, . . . , {t }, and 1 lie in one of the t + 1
intervals [ j/(t + 1), ( j + 1)/(t + 1)], where j = 0, . . . , t; hence, the existence
of integers k, , m k , and m as above.
Suppose now that is irrational and let Q 0 be a positive integer. By the
rst assertion of the theorem, there exists an integer q with 1 q Q 0 such
that

p < 1 1

q qQ
q2
0
1.1 Dirichlet and Liouville
holds for some integer p. We may assume that q is the smallest integer between
1 and Q 0 with this property. By the rst assertion of the theorem applied with
Q = 1/| p/q|, there exists a rational number p /q with 1 q 1/|
p/q| such that

1
p

p < 1 p < 1
and < 2 .

q q
q q Q0
p
q
Our choice of q ensures that q is strictly larger than q and we proceed inductively to get an innite sequence of distinct rational numbers satisfying (1.1),
thus the second assertion is proved. The third one is immediate.
Theorem 1.1 provides a useful criterion of irrationality: a real number having
innitely many good rational approximants must be irrational.
Recall that a complex number is an algebraic number if it is root of a
non-zero integer polynomial P(X ). Otherwise, is a transcendental number.
In 1844, two years after Dirichlets paper, Liouville [368, 369] was the rst to
prove that transcendental numbers exist, and, moreover, he constructed explicit
examples of such numbers. Thirty years later, Cantor [152] gave an alternative
proof of the existence of real transcendental numbers: he showed that the set
of real algebraic numbers is countable and that, given a countable set of real
numbers, any real interval of positive length contains points not belonging to
that set. Cantors proof, however, does not yield any explicit example of a real
trancendental number.
A detailed proof of Theorem 1.2 is given in [370] and includes the case
n = 1 (that is, the last assertion of Theorem 1.1). The main idea, however,
already appeared in Liouvilles note [369].
T HEOREM 1.2. Let be a real root of an irreducible integer polynomial P(X )
of degree n 2. There exists a positive constant c1 ( ) such that

p c1 ( )
(1.2)

q
qn
for all rational numbers p/q. A suitable choice for c1 ( ) is
1
.
c1 ( ) :=
1 + max|t |1 |P (t)|
P ROOF. With c1 ( ) dened as above, inequality (1.2) is true when | p/q|
1. Let p/q be a rational number satisfying | p/q| < 1. Since P(X ) is irreducible and has integer coefcients, we have P( p/q) = 0 and |q n P( p/q)|
1. By Rolles Theorem, there exists a real number t lying between and p/q
such that
|P( p/q)| = |P( ) P( p/q)| = | p/q| |P (t)|.
Hence, we have |t | 1 and

1
c1 ( )
p
,

q q n |P (t)|
qn
as claimed.
C OROLLARY 1.1. The number :=
n1 10
n!
is transcendental.
P ROOF. Since its decimal expansion is not ultimately periodic, is irrational.

For any integer n 2, set qn = 10(n1)! and pn = qn (101! +. . .+10(n1)! ).
Then we have

1
2
2
pn =
n! = n

q
q
10m!
10
n
mn
and is not algebraic of degree greater than or equal to 2, by Theorem 1.2.

Consequently, is a transcendental number.
Corollary 1.1 illustrates how Theorem 1.2 can be applied to prove the transcendence of a large class of real numbers, which are now called Liouville
numbers.
D EFINITION 1.1. Let be a real number. If for any positive real number w
there exists a rational number p/q such that

1
p

0 < < w ,
q
q
then is called a Liouville number.
An easy modication of the proof of Corollary 1.1 shows that any real num
ber n1 an 10n! with an in {1, 2} is a Liouville number. Hence, there exist
uncountably many Liouville numbers. Furthermore, Theorem 1.2 provides a
useful transcendence criterion, see Exercise 1.1.
Combining the theorems of Liouville and Dirichlet, we see that the problem
of rational approximation of real quadratic numbers is, in some sense, solved.
C OROLLARY 1.2. Let be a real quadratic algebraic number. Then there
exists a positive real number c2 ( ) such that

p c2 ( ) for all rationals p/q
(1.3)

q
q2
whereas

p < 1

q q2
for innitely many rationals p/q.
1.2 Continued fractions
One may ask whether there exist real numbers, other than quadratic irrationalities, for which the property (1.3) is satised. The answer is afrmative, and a
way to prove it is to use the theory of continued fractions.
Let x0 , x1 , . . . be real numbers with x1 , x2 , . . . positive. A nite continued
fraction denotes any expression of the form
1
.
[x0 ; x1 , x2 , . . . , xn ] = x0 +
1
x1 +
1
x2 +
1
... +
xn
More generally, we call any expression of the above form or of the form
1
= lim [x0 ; x1 , x2 . . . , xn ]
[x0 ; x1 , x2 , . . . ] = x0 +
n+
1
x1 +
1
x2 +
...
a continued fraction, provided that the limit exists.
The aim of this Section is to show how a real number can be expressed as
= [x0 ; x1 , x2 , . . . ], where x0 is an integer and xn a positive integer for any
n 1. We rst deal with the case of a rational number , then we describe an
algorithm which associates to any irrational an innite sequence of integers
(an )n0 , with an 1 for n 1, and we show that the sequence of rational
numbers [a0 ; a1 , a2 , . . . , an ] converges to .
For more results on continued fractions or/and different points of view on
this theory, the reader may consult, for example, a text of Van der Poorten
[462] and the books of Cassels [155], Dajani and Kraaikamp [174], Hardy and
Wright [271], Iosifescu and Kraaikamp [286], Perron [454], Rockett and Szusz
[474], Schmidt [512], and Schweiger [519].
L EMMA 1.1. Any rational number r has exactly two different continued fraction expansions. These are [r ] and [r 1; 1] if r is an integer and, otherwise,
one of them has the form [a0 ; a1 , . . . , an1 , an ] with an 2, and the other one
is [a0 ; a1 , . . . , an1 , an 1, 1].
P ROOF. Let r be a rational number and write r = u/v with v positive and
u and v coprime. We argue by induction on v. If v = 1, then r = u = [u],
and if r = [a0 ; a1 , . . . , an ] with n 1, we have r = a0 + 1/[a1 ; a2 , . . . , an ].
Since a1 1 and r is an integer, we deduce that n = 1 and a1 = 1, thus

r = a0 + 1 = [r 1; 1].
We assume v 2 and that the lemma holds true for any rational of denominator positive and at most equal to v 1. Performing the Euclidean
division of u by v, there exist integers q and c with u = qv + c and
1 c v 1. Thus, u/v = q + c/v and, by our inductive hypothesis, the
rational v/c has exactly two expansions in continued fractions, which we denote by [a1 ; a2 , . . . , an1 , an ] with an 2, and [a1 ; a2 , . . . , an1 , an 1, 1].
Setting a0 equal to q, the desired result follows for u/v.
Unless otherwise explicitly stated, by the continued fraction expansion of a
rational number p/q, we mean [1] if p/q = 1, and if not, the unique expansion
which does not end with 1.
The following algorithm allows us to associate to any irrational real number
an innite sequence of integers. Let us dene the integer a0 and the real
number 1 > 1 by
a0 = [ ]
and 1 = 1/{ }.
We then have = a0 + 1/1 . For any positive integer n, we dene inductively

the integer an and the real number n+1 > 1 by
an = [n ]
and n+1 = 1/{n },
and we observe that n = an + 1/n+1 . We point out that the algorithm does
not stop since is assumed to be irrational. Thus, we have associated to any
irrational real number an innite sequence of integers a0 , a1 , a2 , . . . with an
positive for all n 1.
If is rational, the same algorithm terminates and associates to a nite
sequence of integers. Indeed, the j s are then rational numbers and, if we set
j = u j /v j , with u j , v j positive and gcd(u j , v j ) = 1, an easy induction
shows that we get u j > u j+1 , for any positive integer j with v j = 1. Consequently, there must be some index n for which vn1 = 1 and n is an integer.
Thus, an+1 , an+2 , . . . are not dened. We have = [a0 ; a1 , . . . , an ], and this
corresponds to the Euclidean algorithm.
D EFINITION 1.2. Let be an irrational number (resp. a rational number). Let
a0 , a1 , . . . (resp. a0 , a1 , . . . , a N ) be the sequence of integers associated to
by the algorithm dened above. For any integer n 1 (resp. n = 1, . . . , N ),
the rational number
pn
:= [a0 ; a1 , . . . , an ]
qn
is called the n-th convergent of and an is termed the n-th partial quotient of
. Further, for any integer n 1 (resp. n = 1, . . . , N 1), there exists a real
number n in ]0, 1[ such that
= [a0 ; a1 , . . . , an1 , an + n ].
We observe that the real numbers n occurring in Denition 1.2 are exactly the
real numbers 1/n+1 given by the algorithm.
In all of what follows until the end of Theorem 1.5, unless otherwise explicitly stated, we assume that is a real irrational number and we associate to
the sequences (an )n0 and ( pn /qn )n1 as given by Denition 1.2. However,
the statements below remain true for rational numbers provided that the an s
and the pn /qn s are well-dened.
The integers pn and qn can be easily expressed in terms of an , pn1 , pn2 ,
qn1 , and qn2 .
T HEOREM 1.3. Setting
p1 = 1,
q1 = 0,
p0 = a 0 ,
and
q0 = 1,
we have, for any positive integer n,

pn = an pn1 + pn2
and
qn = an qn1 + qn2 .
P ROOF. We proceed by induction. Since p1 /q1 = a0 + 1/a1 = (a0 a1 +

1)/a1 , the denitions of p1 , q1 , p0 , and q0 show that the theorem is true
for n = 1. Assume that it holds true for a positive integer n and denote by
p0 /q0 , . . . , pn /qn the convergents of the rational number [a1 ; a2 , . . . , an+1 ].
For any integer j with 0 j n + 1 we have
q j1
pj
1
= a0 + ,
= [a0 ; a1 , . . . , a j ] = a0 +
qj
[a1 ; a2 , . . . , a j ]
p j1
thus
p j = a0 p j1 + q j1
and q j = p j1 .
(1.4)
It follows from (1.4) with j = n + 1 and the inductive hypothesis applied to

the rational [a1 ; a2 , . . . , an+1 ] that

pn+1 = a0 (an+1 pn1
+ pn2
) + an+1 qn1
+ qn2

= an+1 (a0 pn1
+ qn1
) + a0 ( pn2
+ qn2
)
and

+ pn2
,
qn+1 = an+1 pn1
whence, by (1.4) with j = n and j = n 1, we get qn+1 = an+1 qn + qn1

and pn+1 = an+1 pn + pn1 , as claimed.
T HEOREM 1.4. For any non-negative integer n, we have
qn pn1 pn qn1 = (1)n
(1.5)
qn pn2 pn qn2 = (1)n1 an .
(1.6)
and, for all n 1,
P ROOF. These equalities are clearly true for n = 0 and n = 1. It then sufces
to argue by induction, using Theorem 1.3.
L EMMA 1.2. For any irrational number and any non-negative integer n, the
difference pn /qn is positive if, and only if, n is even.
P ROOF. We easily check that this is true for n = 0, 1, and 2 and we proceed
by induction. Let n 4 be an even integer. Then = [a0 ; a1 , [a2 ; a3 , . . . , an +
n ]] and the inductive hypothesis implies that [a2 ; a3 , . . . , an + n ] >
[a2 ; a3 , . . . , an ]. Since [a0 ; a1 , u] > [a0 ; a1 , v] holds for all positive real numbers u > v, we get that > [a0 ; a1 , [a2 ; a3 , . . . , an ]] = pn /qn . We deal with
the case n odd in exactly the same way.
As a corollary of Lemma 1.2, we get a result of Vahlen [575].
C OROLLARY 1.3. Let pn /qn and pn+1 /qn+1 be two consecutive convergents
of the continued fraction expansion of an irrational number . Then at least
one of them satises

p < 1 .

q 2q 2
P ROOF. We infer from Lemma 1.2 that is an inner point of the interval bounded by pn /qn and pn+1 /qn+1 . Thus, using (1.5) and the inequality
a 2 + b2 > 2ab, valid for any distinct real numbers a and b, we get

pn
pn
pn+1
1
1
pn+1
1

=
+
,
+ 2 >
=
qn qn+1
qn
qn+1
qn
qn+1
2qn2
2qn+1
and the claimed result follows.
The next two theorems show that the real irrational numbers are in a one-toone correspondence with the set of integer sequences (ai )i0 with ai positive
for i 1.
T HEOREM 1.5. The convergents of even order of any real irrational form
a strictly increasing sequence and those of odd order a strictly decreasing
sequence. The sequence of convergents ( pn /qn )n0 converges to , and we set
= [a0 ; a1 , a2 , . . . ].
Any irrational number has a unique expansion in continued fractions.
P ROOF. It follows from (1.6) that for any integer n with n 2 we have
pn2
pn
(1)n1 an
=
,
qn2
qn
qn qn2
and, since the an s are positive, we deduce that the convergents of even order of
the real irrational form a strictly increasing sequence and those of odd order a
strictly decreasing sequence. To conclude, we observe that, by Lemma 1.2, we
have p2n /q2n < < p2n+1 /q2n+1 for all n 0, and, by (1.5), the difference
p2n /q2n p2n+1 /q2n+1 tends to 0 when n tends to innity. Uniqueness is
clear. Indeed, if (bi )i0 is a sequence of integers with bi positive for i 1
and such that limn+ [b0 ; b1 , b2 , . . . ] exist, then this limit cannot be equal
to limn+ [a0 ; a1 , a2 , . . . ] as soon as there exists a non-negative integer i
with ai = bi .
T HEOREM 1.6. Let a0 , a1 , . . . be integers with a1 , a2 , . . . positive. Then the
sequence of rational numbers [a0 ; a1 , . . . , ai ], i 1, converges to the irrational number whose partial quotients are precisely a0 , a1 , . . .
P ROOF. For any positive integer n, denote by pn /qn the rational number
[a0 ; a1 , . . . , an ]. The recurrence relations obtained in Theorems 1.3 and 1.4
hold true in the present context. As in the proof of Theorem 1.5, we deduce
from (1.5) and (1.6) that the sequences ( p2n /q2n )n1 and ( p2n+1 /q2n+1 )n1
are adjacent. Hence, they converge to the same limit, namely to the irrational
number [a0 ; a1 , a2 , . . . ], whose partial quotients are precisely a0 , a1 , . . . , by
Theorem 1.5.
We observe that for any irrational number the sequences (an )n0 and
(n )n1 given by the algorithm dened below Lemma 1.1 satisfy n =
[an ; an+1 , an+2 , . . . ] for all positive integers n.
T HEOREM 1.7. Let n be a positive integer and = [a0 ; a1 , a2 . . . ] be an
irrational number. We then have
= [a0 ; a1 , . . . , an , n+1 ] =
pn n+1 + pn1
qn n+1 + qn1
10
and
q n pn =
(1)n
(1)n
1
.
=
qn n+1 + qn1
qn
n+1 + [0; an , an1 , . . . , a1 ]
Furthermore, the set of real numbers having a continued fraction expansion

whose n + 1 rst partial quotients are a0 , a1 , . . . , an is precisely the closed
interval bounded by ( pn1 + pn )/(qn1 + qn ) and pn /qn , which are equal to
[a0 ; a1 , . . . , an , 1] and [a0 ; a1 , . . . , an ], respectively.
P ROOF. We proceed by induction, using Theorem 1.3 and noticing that we
have n = an + 1/n+1 and qn /qn1 = [an ; an1 , . . . , a1 ] for all positive
integers n. The last assertion of the theorem follows immediately, since the
admissible values of n+1 run exactly through the interval ]1, +[.
C OROLLARY 1.4. For any irrational number and any non-negative integer
n, we have

1
pn
1
<
<
.

qn (qn + qn+1 )
qn
qn qn+1
P ROOF. Writing = [a0 ; a1 , a2 , . . . ] and n+1 = [an+1 ; an+2 , . . . ], we observe that an+1 < n+1 < an+1 + 1, and we get from Theorem 1.7 that

1
pn
1

<
<
.

q
q
(a
q
qn (an+1 + 1)qn + qn1
n
n n+1 n + qn1 )
The corollary follows then from Theorem 1.3.
The following result of Legendre [359] provides a partial converse to Corollary
1.3.
T HEOREM 1.8. Let be a real number. Any non-zero rational number a/b
with

a < 1

b 2b2
is a convergent of .
P ROOF. We assume that = a/b and we write a/b = /b2 , with = 1
and 0 < < 1/2. By Lemma 1.1, setting an1 = 1 if necessary, we may write
a/b = [a0 ; a1 , . . . , an1 ], with n given by (1)n1 = , and we denote by
p1 /q1 , . . . , pn1 /qn1 the convergents of a/b. Let be such that
=
pn1 + pn2
= [a0 ; a1 , . . . , an1 , ].
qn1 + qn2
11
We check that
pn1
a
1
= =
=
( qn1 pn1 )
2
b
qn1
qn1
b
(1)n1
1
,
=
qn1 qn1 + qn2
whence
=
1 qn2
qn1
and > 1. Denote by [an ; an+1 , an+2 , . . . ] the continued fraction expansion
of . Since a j 1 for all j 1, we have
= [a0 ; a1 , . . . , an1 , ] = [a0 ; a1 , a2 , . . . ],
and Theorem 1.6 yields that a/b = pn1 /qn1 is a convergent of .
A less-known result of Fatou [239] (see Grace [259] for a complete proof) provides a satisfactory converse to Theorem 1.1. It asserts that if the real number
and the rational number a/b satisfy | a/b| < 1/b2 , then there exists an
integer n such that

pn pn+1 + pn pn+2 pn+1
a
belongs to
,
,
.
b
qn qn+1 + qn qn+2 qn+1
D EFINITION 1.3. The real number is said to be badly approximable if there
exists a positive constant c3 ( ) such that

p c3 ( ) for any rational p/q distinct from .

q
q2
The set of badly approximable real numbers is denoted by B.
Theorem 1.1 and Corollary 1.2 show that rational and quadratic real numbers
are badly approximable. However, many other real numbers share this property, as follows from Theorem 1.9.
T HEOREM 1.9. An irrational real number is badly approximable if, and
only if, the sequence of its partial quotients is bounded. Consequently, the set
B is uncountable.
P ROOF. Assume that is badly approximable, and let c4 be a positive real
number such that | p/q| > c4 /q 2 for any rational p/q. Let n be a positive
integer. It follows from Corollary 1.4 that qn qn1 /c4 . Since Theorem 1.3
yields that qn an qn1 , we get an 1/c4 , and the sequence of partial quotients of is bounded.
12
Conversely, if the partial quotients of are not greater than a constant M,

we then have qn+1 (M + 1)qn for any non-negative integer n. Furthermore,
if a/b satises | a/b| < 1/(2b2 ), then, by Theorem 1.8 and Corollary 1.4,
there exists a non-negative integer n such that a/b = pn /qn and

1
1
a >
.

b b(b + qn+1 )
(M + 2)b2
This shows that is badly approximable. The last assertion of the theorem
follows from Theorem 1.6.
1.3 The theorem of Khintchine

In this Section, we aim to prove Theorem 1.10, due to Khintchine [317], by
using the theory of continued fractions, as in [317] and in his book [323]. This
is one of the rst metric results in Diophantine approximation. We denote by
the Lebesgue measure on the real line and, if I is a bounded real interval, we
often simply write |I | = (I ) for its length. A set of Lebesgue measure zero is
called a null set; the complement of a null set is termed a set of full measure,
usually simply called full. As usual, we say that almost no points belong to a
set if this set is null, while a full set contains almost all points.
T HEOREM 1.10. Let : R1 R>0 be a continuous function such that
x x 2 (x) is non-increasing. Then, the set

p

K1 () := R : < (q) for innitely many rational
q

p
numbers
q
+
has Lebesgue measure zero if the sum q=1 q(q) converges and has full
Lebesgue measure otherwise.
The function occurring in Theorem 1.10 is called an approximation function. Throughout this book, we assume for commodity that is continuous
on R1 , although it only requires to be dened for every sufciently large integer. The elements of K1 () are termed -approximable, or approximable at
order .
Since x (x) is non-increasing in Theorem 1.10, for any in K1 ()
there are innitely many reduced rational numbers p/q with | p/q| <
(q). Thus, we may add in the denition of K1 () the extra assumption with
p and q coprime without any change in the conclusion. Furthermore, we point
out that the assumption x x 2 (x) is non-increasing can be removed in
the convergence half of Theorem 1.10.
13
An alternative proof of Theorem 1.10, valid under slightly less restrictive

conditions on the function , is given in Chapter 6. It is based on an improvement of the divergence half of the BorelCantelli Lemma.
An immediate consequence of Theorem 1.10 is that almost all real numbers
have innitely many convergents with | p/q| < q 2 (log q)1 , and almost
all real numbers have only nitely many convergents with | p/q| <
q 2 (log q)2 .
Throughout the proof of Theorem 1.10, we often need the following auxiliary result, referred to as the (convergence half of the) BorelCantelli Lemma.
L EMMA 1.3. Let (E n )n0 be a sequence of Borelian real subsets such that the

sum n0 (E n ) converges. We then have

E n = 0,
N 1 nN
that is, almost all real numbers belong to only a nite number of sets E n .

P ROOF. Let be a positive real number. Since n0(E n ) converges,
there

exists N0 such that, for any integer N N0 , we have nN E n < ; hence,

in particular, N 1 nN E n < .
Before turning to the proof of Theorem 1.10, we state an application of
Lemma 1.2.
D EFINITION 1.4. The real number is said to be very well approximable if
there exists a positive real number such that

p
p < 1
for innitely many rational numbers .

2+
q
q
q
As pointed out in [86], it would seem more appropriate to use the terminology
well approximable instead of very well approximable. To avoid confusion,
we follow current usage.
C OROLLARY 1.5. The Lebesgue measure of the set of very well approximable
numbers is equal to zero.
PROOF OF COROLLARY 1.5 AND OF THE CONVERGENCE HALF OF
THEOREM 1.10. Since K1 () is invariant by translation by 1, it is plainly
enough to prove the conclusion of Theorem 1.10 for the set K1 () [0, 1].
For any positive integer n, denote by E n the union of the intervals [m/n
(n), m/n + (n)] [0, 1] for m = 0, . . . , n. Observe that K1 () [0, 1]
+
is contained in N 1 nN E n . If the sum q=1
q(q) converges, then
n1 (E n ) also converges, and Lemma 1.2 yields that K1 () [0, 1] has
14
Lebesgue measure zero. In particular, taking for the functions x x 21/k ,

where k is a positive integer, we get Corollary 1.5.
The divergence half of Theorem 1.10 is much more difcult to prove. We begin
with some preliminary results on continued fractions. As a standard matter, if
is a real number in [0, 1[, we denote its continued fraction expansion by
[0; a1 , a2 , . . . ] and the sequence of its convergents by ( pn /qn )n0 . Let n 1
be an integer and k = (k1 , . . . , kn ) be a n-tuple of positive integers. We denote
by Fk the set of real numbers in [0, 1] whose partial quotients satisfy ai = ki
for i = 1, . . . , n. By Theorem 1.7, Fk is a closed interval bounded by pn /qn
and ( pn1 + pn )/(qn1 + qn ). We can write Fk as the union

Fk =
F(k,s) ,
s1
where F(k,s) denotes the sub-interval of Fk composed by the real numbers with
an+1 = s. This union is not disjoint but the intersection of two sets F(k,s) and
F(k,s ) with s > s 1 consists of at most one point, which is a rational
number. Since

spn + pn1
(s + 1) pn + pn1
(F(k,s) ) =
sqn + qn1
(s + 1)qn + qn1
=
s2

qn +
qn1 qn1
s
s
,

+ 1 + 1s qn
and
(Fk ) =
1
,
qn (qn + qn1 )
we get
(F(k,s) )
1 1+
= 2
(Fk )
s 1+
qn1
qn
qn1
sqn
1
1+
1
s
qn1
sqn
and thus
(F(k,s) )
2
1
< 2.
<
2
(Fk )
3s
s
(1.7)
Two intermediate results are needed towards the proof of Theorem 1.10. The
rst one is the BorelBernstein theorem [110, 98, 111, 99].
T HEOREM 1.11. Let (u n )n1 be a sequence of positive real numbers. If the

sum n1 u 1
n diverges, then, for almost all = [0; a1 , a2 , . . . ] in [0, 1[,
there exist innitely many integers n such that an u n . Further, if this sum
15
converges, then, for almost all = [0; a1 , a2 , . . . ] in [0, 1[, there exist only a
nite number of integers n such that an u n .
P ROOF. Let m, n, and j be positive integers with n 2 and 1
j n. Let k be a (m + j)-tuple of positive integers, and set k :=
(k1 , . . . , km , km+1 , . . . , km+ j ). Let X > 1 be a real number. We infer from
(1.7) that

1 1
(F(k,s) ) =
(F(k,s) )
(F(k,s) ) 1
(Fk )
3 sX s 2
s<X
sX
s1

1
1
(Fk ).
3(1 + X )
(1.8)
Denote by Fm, j the union of the intervals Fk over all the (m + j)-tuples k with
ki 1 for i = 1, . . . , m and km+i < u m+i for i = 1, . . . , j. By choosing
X = u m+n in (1.8), we get the upper bound

1
(Fm,n ) 1
(Fm,n1 ),
3(1 + u m+n )
whence, by induction,
(Fm,n )
n

i=2
The divergence of the sum

the product
1
n1 u n
n

i=2

1
(Fm,1 ).
3(1 + u m+i )
implies that, for any positive integer m,
1
1
3(1 + u m+i )
and thus (Fm,n ) tend to 0 when n tends to innity. Since the set Bm of real
numbers satisfying am+i < u m+i for all i 1 is contained in every set
Fm,n , its Lebesgue measure is zero. Consequently, if belongs to none of
the sets Bm , and that is indeed the case for almost all , then there exist innitely many integers n such that an u n . This proves the rst part of the
theorem.

The other part is easier. Assume that the sum n1 u 1
n converges. Let n
1 be an integer and denote by E n the set of real numbers satisfying an u n .
The right inequality of (1.7) yields that
2

4 (Fk )
(F(k,s) ) <
(Fk )
.
(1.9)
2
u n+1
su n+1
su n+1 s
16

By (1.9), we get (E n+1 ) < 4/u n+1 , thus the sum n1 (E n ) converges and
we infer from Lemma 1.2 that the Lebesgue measure of the set of real numbers
belonging to innitely many sets E n is zero.
The next statement, due to Borel [110], is a direct consequence of Theorem
1.11.
C OROLLARY 1.6. The Lebesgue measure of the set B of badly approximable
real numbers is equal to zero.
P ROOF. Applying Theorem 1.11 to the divergent sequence (1/n)n1 , we get
that, for almost all real numbers = [a0 ; a1 , a2 , . . . ], there exist innitely
many positive integers n such that an n. Hence, almost all real numbers
have unbounded partial quotients.
The set B is not too small in the sense that its Hausdorff dimension is equal to
1, see Exercise 5.1.
T HEOREM 1.12. There exists a positive real number B and, for almost all real
numbers , an integer n 0 ( ) such that qn < e Bn for all integers n n 0 ( ).
P ROOF. For a real number g 1 and an integer n 2, denote by E n (g) the
set of real numbers in [0, 1[ such that a1 . . . an g. By (1.5), Theorem 1.3,
and Theorem 1.7, this set is a union of intervals, each of which being of length

pn
1
pn + pn1
1

< q 2 < (a . . . a )2 .
q q +q
n
n
n1
n
1
n
Consequently, we get
(E n (g)) <

a1 ...an g
1
.
(a1 . . . an )2
In order to estimate the above summation, we compare it with an integral and

we notice that, for any positive integer a, we have
a+1
dx
1
2
.
a2
x2
a
Thus, if

Jn (g) :=

...
xi 1
x1 ...xn g
dx1
dxn
... 2 ,
2
xn
x1
we get (E n (g)) < 2n Jn (g). For g 1, the integral is on the whole domain
x1 1, . . . , xn 1, thus Jn (g) = 1. For g > 1, an easy induction on n yields
that
n1
(log g)i
1
.
(1.10)
Jn (g) =
g i=0
i!
17
Consequently, if we choose g = e An , where A > 1 is a real number which

will be xed afterwards, we deduce from (1.10) that
n1

(An)i
(An)n
E n (e An ) < 2n eAn
2n eAn n
< 2n eAn n(Ae)n .
i!
n!
i=0

If A is large enough, then the sum n1 (E n (e An )) converges and we infer
from Lemma 1.2 that almost all real numbers in [0, 1] belong to only a nite
number of sets E n (e An ). In other words, for almost all in [0, 1] and for any
sufciently large integer n (in terms of ), we have a1 . . . an < e An and then,
by Theorem 1.3,
qn < 2an qn1 < . . . < 2n e An .

Since for any real number the denominators of the convergents of and +1
are the same, the theorem is proved.
Khintchine
[317] proved that any real number B strictly greater than log 2 +

exp( 2 log 2) satises the conclusion of Theorem 1.12. Actually, a much
stronger statement holds true: there exists a real number such that for al
most all real numbers the sequence ( n qn )n1 converges to . This was established in 1936 by Khintchine [322] and, the same year, Levy [366] proved
that = exp( 2 /(12 log 2)).
We are now ready to complete the proof of Theorem 1.10, the main result
of this Section.
COMPLETION OF THE PROOF OF THEOREM 1.10. We assume that the sum
q1 q(q) diverges and we aim to prove that the set K1 () [0, 1] has full
2Bx
Bx
measure. For any x > 0, set (x) := e
(e ), where B is the real number given by Theorem 1.12. By assumption, the function is non-increasing.
Further, for any positive real numbers a and A with a < A, we have
eB A
A
1
(x)dx =
u(u)du,
B e Ba
a

which diverges when A tends to innity. Consequently, the sum q1 (q)
diverges. Theorem 1.11 asserts then that for almost all real numbers in [0, 1]
we have an+1 1/(n) for innitely many integers n. Thus, for all these
integers n we have

1
1
(n)
pn
<
.

2
q
q q
a q
q2
n
n n+1
n+1 n
However, by Theorem 1.12, there exists a real number B > 0 such that, for
almost all , we have qn < e Bn , that is, n > (log qn )/B, provided that n is
18
large enough in terms of . Since is non-increasing, we get that, for almost

all in [0, 1], there exist innitely many integers n such that

pn < 1 log qn = (qn ),

qn qn2
B
which concludes the proof of the theorem.
1.4 The DufnSchaeffer Conjecture

In the statement of Theorem 1.10, the assumption on the approximation function is quite restrictive, and it would be desirable to weaken it. For instance,
the alternative proof of Theorem 1.10 given in Chapter 6 requires only that
is non-increasing. The result generally conjectured involves the function , the
Euler totient function, dened for all positive integers q by
(q) = Card{1 x q : gcd(x, q) = 1},
and has been proposed by Dufn and Schaeffer at the end of [213].
C ONJECTURE (D UFFIN S CHAEFFER ). Let : R1 R0 be some continuous function. Then the set

p
p

R : < (q) for innitely many rationals
q
q

with gcd( p, q) = 1
has full Lebesgue measure if the sum
+
q=1 (q)(q)
diverges.
When the above sum converges, the above set has Lebesgue measure zero, as
easily follows from Lemma 1.2.
As observed after the statement of Theorem 1.10, Khintchines result holds
also when we demand that the integers p and q are coprime. This is a consequence of the assumption made on and this remark does not apply for
a general function . Thus, we must distinguish between approximation by
rationals not necessarily reduced and approximation by reduced rationals. According to [213], (. . .) the more natural formulation of this problem is in terms
of reduced fractions.
The notorious difculties in the DufnSchaeffer Conjecture are due to the
fact that it is very difcult to control the pairwise intersections of the intervals
] p/q (q), p/q + (q)[, where p and q 1 are integers. However, it has
been proved that the Conjecture holds true under some additional hypotheses.
We choose to quote only two results, a rst one due to Dufn and Schaeffer
1.5 Complementary results on continued fractions
19
[213], and a second one to Erdos [228]; the interested reader can nd an exhaustive survey of that problem (including detailed proofs of Theorems 1.13
to 1.15 below) in Harman [273] (see also his survey [274]).
T HEOREM 1.13. The DufnSchaeffer Conjecture holds true if we assume
that there exists a real number c such that the function x x c (x) is decreasing.
T HEOREM 1.14. The DufnSchaeffer Conjecture holds true if there exists a
positive real number such that, for every positive integer q, we have (q) =
0 or /q 2 . In particular, if (n i )i1 is a strictly increasing sequence of positive
integers, then, for almost all real numbers , it contains innitely many (resp.

a nite number of) denominators of convergents to if the sum i1 (n i )/n i2
diverges (resp. converges).
In the opposite direction, the next result, due to Dufn and Schaeffer [213],

asserts that the divergence of the sum q1 q(q) is certainly not sufcient
to ensure the existence of innitely many solutions to | p/q| < (q) for
almost all .
T HEOREM 1.15. There exists a positive function such that the sum

q1 q(q) diverges and

p < (q)

q
has only a nite number of solutions for almost all .
1.5 Complementary results on continued fractions

In this Section, we list a few complementary results on continued fractions.
The reader is referred to the books quoted at the beginning of Section 1.2 for
deeper statements and proofs (notice that proofs of Theorems 1.16, 1.18, and
1.19 are also given in Exercises 1.2 and 1.3).
Corollary 1.2 asserts that quadratic numbers are badly approximable numbers, thus, by Theorem 1.9, their partial quotients are bounded. However, much
more is known: the sequence of their partial quotients is ultimately periodic.
T HEOREM 1.16. The real irrational number = [a0 ; a1 , a2 , . . . ] has a periodic continued fraction expansion (that is, there exist integers k 0 and
n 1 such that am+n = am for all m k) if, and only if, is a quadratic
irrationality.
The only if part is due to Euler [230], and the if part was established by
Lagrange [347] in 1770.
20
The next result dates back to (at least) 1877 and can be found in Serret [520].
T HEOREM 1.17. Let = [a0 ; a1 , a2 , . . . ] and = [b0 ; b1 , b2 , . . . ] be two
irrational numbers. There exist integers u and v such that au+n = bv+n for
every positive integer n if, and only if, there exist integers a, b, c, and d such
that |ad bc| = 1 and = (a + b)/(c + d).
Two real numbers satisfying the equivalent conditions of Theorem 1.17 are
called equivalent.
Hurwitz [284] improved the second assertion of Theorem 1.1.
T HEOREM 1.18. For any real irrational number , there exist innitely many
rationals p/q satisfying

p < 1 .

q
5q 2
5 cannot be replaced by a larger real number.
We observe that the Golden Section (1 + 5)/2 = [1; 1, 1, . . . , 1, . . . ]

is, up to equivalence, the irrational number which is the most badly approximable by rational numbers. As shown by Hurwitz [284], Theorem 1.18 can
be improved if, besides the rationals, we also exclude the numbers which are
equivalent to the Golden Section.
Further, the constant
T HEOREM
1.19. For any irrational real number which is not equivalent to
(1 + 5)/2, there exist innitely many rationals p/q satisfying

p < 1 .

q
8q 2
The constant
8 cannot be replaced by a larger real number.
In other words, the assumptions of Theorem 1.19 are satised by any irrational
having innitely many partial quotients greater than or equal
to 2. The limiting case is obtained with real numbers equivalent to (1 + 2)/2.
More generally, for a real number we dene
( ) = lim inf qq ,
q+
where denotes the function distance to the nearest integer. Clearly, ( ) =

0 holds for any rational . The set of values taken by the function is called
the Lagrange spectrum. By Theorem 1.18, it is contained in [0, 1/ 5].
T HEOREM 1.20. There exists a sequence of numbers
1
1
1 = > 2 = > 3 > 4 >
5
8
1.6 Exercises
21
tending to 1/3 such that, for all i , there is, up to equivalence, a nite number
of real numbers satisfying ( ) = i .
More results on the Lagrange spectrum can be found in Cusick and Flahive
[173].
1.6 Exercises
E XERCISE 1.1. Use Liouvilles Theorem 1.2 to prove that if the denominators
qn of the convergents of satisfy
lim sup
n+
log log qn
= +,
n
then is transcendental.
E XERCISE 1.2. (Proof of Theorem 1.16). Show that any irrational real number having a periodic continued fraction expansion is a quadratic irrationality.
Prove the converse, following Steinig [543]. Let be a quadratic real number and dene the sequences (an )n0 and (n )n1 by the algorithm given in
Section 1.2. Set 0 := .
1) Show, by induction, that for each non-negative integer n there is an integer polynomial f n (x) := An x 2 + Bn x + Cn , with Bn2 4An Cn positive and
not a square, such that f n (n ) = 0. Prove that An+1 , Bn+1 and Cn+1 are given
by
An+1 = an2 An + an Bn + Cn , Bn+1 = 2an An + Bn , and Cn+1 = An .
(1.11)
2) Observe that
D := B02 4A0 C0 = Bn2 4An Cn ,
for any n 0,
(1.12)
and deduce from (1.11) that there exists an innite set N of distinct positive
integers such that An An1 is negative for any n in N .
3) Deduce from (1.12) that
|Bn | < D, |An | D/4,
and |Cn | D/4,
for any n in N . Conclude.

E XERCISE 1.3. (Proof of Hurwitz Theorems 1.18 and 1.19, following Forder
[248].)
22
Let be an irrational number and assume that

its convergents pn1 /qn1 ,
pn /qn , and pn+1 /qn+1 satisfy | p j /q j | 1/( 5q 2j ) for j = n 1, n, n +1.
1) Prove that we have

1
1
1
1
+ 2 ,
2
qn1 qn
qn
5 qn1
and
deduce that qn /qn1 and qn1 /qn belong to the interval ]( 5 1)/2,
( 5 + 1)/2[. Prove that the same conclusion also holds for qn /qn+1 and
qn+1 /qn . Conclude.
2) With a suitable adaptation of the above proof, show thatif the convergents
pn /qn of an irrational number satisfy | pn /qn | 1/( 8qn2 ) from some
integer n 0 onwards then qn+1 < 2qn + qn1 holds for any sufciently large
integer n. Conclude.
E XERCISE 1.4. Improvement of Dirichlets Theorem 1.1.
Theorem 1.1 asserts that if is a given real number, then, for any positive
integer Q, there exist integers p and q satisfying
1qQ
and |q p| < c/Q,
(1.13)
with c = 1. Show that this result holds true for a real number with a constant
c strictly less than 1 if, and only if, is badly approximable.
A stronger statement is due to Davenport and Schmidt [183]:
Let = [a0 ; a1 , a2 , . . . ] be a real number, and set
c5 ( ) = lim inf [0; an , an1 , . . . , a1 ] [0; an+1 , an+2 , . . . ].
n+
If c > 1/(1 + c5 ( )), then (1.13) has a solution for all Q sufciently large, and
this is not true if c < 1/(1 + c5 ( )).
Use Theorem 1.7
to prove this result and show that the largest possible value
for c5 ( ) is (3 5)/2.
E XERCISE 1.5. A result of Jarnk [292], Satz 6, on approximation order.
Let : R1 R>0 be a non-increasing function such that (x) = o(x 2 )
as x tends to innity. Using the theory of continued fractions, prove that there
exists an uncountable set of real numbers which are approximable at order
, but not at any order c, with 0 < c < 1.
E XERCISE 1.6. In 1962, Erdos [227] showed that every real number (resp.
every non-zero real number) is the sum (resp. product) of two Liouville numbers. He gave two proofs: a rst one is direct and constructive, while a second
1.7 Notes
23
one rests on Baires Theorem. His result has been extended by Rieger [471],
Schwarz [518], and Alniacik and Saias [19], who established the following
statement (recall that a G -set is a countable intersection of open sets):
Let I be a real interval of positive length. Let G be a real G -set dense
in I and ( f n )n0 be a sequence of continuous, nowhere locally constant, real
functions on I . Then, the set n0 f n1 (G) is a G -set dense in I .
Prove Alniacik and Saias assertion. Show that the set of Liouville numbers
is a dense G -set and deduce Erdos result.
1.7 Notes
In 1844 Liouville [368] used a result of Lagrange on continued fractions to
prove that, for any n, the n-th partial quotient of an algebraic number of degree
d 2 is less than some number (independent of n) times the (d 2)-th power
of the denominator of the (n 1)-th convergent. Then, he observed that if is a
real number such that, for any n, its n-th partial quotient an is dened in terms
qn1
of the denominator qn1 of its (n 1)-th convergent by taking an = qn1
n , then the number must be transcendental. These numbers are,
or an = qn1
historically, the rst examples of transcendental numbers. At the end of [368],
Liouville mentioned that analogous results exist for ordinary sums, including

m! for any integer a 2. In 1851, he gave [370] a complete proof
m1 a
of this last assertion. In a second note of 1844, Liouville [369] simplied the
proof given in his rst note [368] by removing the use of Lagranges result. The
reader wishing more information is directed to Chapter XII of Lutzen [374].
Dirichlets Theorem 1.1 has two natural multidimensional extensions (as
also follows from Theorem B.2). Let n 2 be an integer and 1 , . . . , n be real
numbers. On the one hand (approximation of linear forms), for any positive
integer Q, there exist integers p1 , . . . , pn , q not all zero satisfying
| p1 1 + . . . + pn n + q| < Q n ,
max{| p1 |, . . . , | pn |} Q.
On the other hand (simultaneous approximation), for any positive integer Q,

there exist integers p1 , . . . , pn , q not all zero satisfying
max{|q1 p1 |, . . . |qn pn |} < Q 1 ,
|q| Q n .
Davenport and Schmidt [184] (see also [183] and Schmidt [509] for a further
generalization) proved that for almost every n-tuple (1 , . . . , n ) none of these
two forms of Dirichlets Theorem can be improved (in the same sense as in
Exercise 1.4). Raisbeck [466] established a theorem in the opposite direction
(best possible for n = 2), later improved upon by Kaindl [306] for n 3 (see
also Tichy [559] and Langmayr [350] for systems of linear forms).
24
Further extensions of Dirichlets Theorem, where the approximants must

satisfy some restriction, have been considered by Obreskov [441], Rogers
[475], Schmidt [511], and Thurnheer [554, 555, 556, 557, 558].
Khintchine [320] proved that, given any function (Q) which tends to 0 as
Q tends to innity, no matter how rapidly, there exists a pair (1 , 2 ) of real
numbers, with 1, 1 , 2 linearly independent over the rationals, such that the
inequalities
| p1 1 + p2 2 + q| < (Q),
1 max{| p1 |, | p2 |} Q
are soluble in integers for every Q sufciently large (see Theorem XIV of
Chapter V of [155]).
Let be a real irrational number and k be a positive real number. Extending Fatous result [239] quoted after the proof of Theorem 1.8, Worley [601]
has expressed the rational solutions a/b of | a/b| < k/b2 in terms of the
convergents of .
An n-tuple (1 , . . . , n ) of real numbers is said to be badly approximable
if there exists a positive constant c6 such that max1in qi > c6 q 11/n
for any positive integer q. This extends Denition 1.3. The rst proof of the
existence of continuum-many badly approximable pairs of distinct real numbers is due to Davenport [175]. For further references and an exposition of the
theory of - games, see Chapter III of [512]. Schmidt [514] proved the existence of innite sets of real numbers whose nite subsets do not have good
simultaneous rational approximations.
A very general form of Khintchines Theorem 1.10 has been obtained by
Groshev [260], extending earlier work of Khintchine [319] to systems of linear
forms. Let m and n be positive integers and : R1 R>0 be continuous
and such that x x m1 (x)n is non-increasing. Groshev proved that for either almost all or almost no matrices M in Rmn there are innitely many integer
vectors q in Zm such that |q M| < (|q|), depending on whether the series
+ m1
(x)n diverges or converges. Here, for a vector = (1 , . . . , n ),
x=1 x
we set | | = max{|1 |, . . . , |n |} and = (1 , . . . , n ), where lies in
] 1/2, 1/2] and differs from i to within an integer.
Theorem 1.10 asserts in the divergence case the existence of innitely
many rational numbers p/q with | p/q| < (q)/q, but the method of
proof does not give any information regarding the asymptotic behaviour of
their number. An asymptotic formula has been obtained by Erdos [226] and
LeVeque [364], and, with a good error term, by Schmidt [500, 502], see also
Chapter III of [512], Chapter 1 of [540], Chapter 4 of [273], and the survey of
Harman [274].
1.7 Notes
25
Harman [272] established an analogue of Theorem 1.10 where numerators and denominators are restricted to the set of prime numbers. See also
Chapter 6 of his book [273]. An asymptotic formula for the number of solutions in the divergence case, and under some extra hypothesis on the function
, has been established by Jones in the second chapter of his Ph.D. thesis
[302]. Furthermore, Jones [303] extended the results of [272] to simultaneous
approximation.
For inhomogeneous approximation, we refer the reader to Cassels [155],
Hardy and Wright [271], Khintchine [323], and Gruber and Lekkerkerker
[262]. An inhomogeneous analogue of Theorem 1.10, due to Schmidt [502]
(who also obtained an asymptotic formula for the number of solutions in the
divergence case), follows from Exercise 6.2 and Theorem 6.1.
Sullivan [547] established a variant of Khintchines Theorem 1.10 under
a weaker assumption. His proof is of a geometric nature.
Theorems 1.18 and 1.19 are closely related to the works of Korkine and
Zolotareff [337] and Markoff [406], where indenite, binary quadratic forms
are considered. Hurwitz was rst to give these results in the form of a statement
about Diophantine approximation.
Let b 2 be an integer. A real number is said to be simply normal in base
b if each digit 0, 1, . . . , b 1 occurs in its expansion in base b with frequency
1/b. It is said to be normal if it is simply normal in any base b 2. R. C. Baker
(see Montgomery [427], page 203, and Queffelec and Ramare [465]) observed
that there exist badly approximable numbers which are normal. Using a similar
method, that is, combining the results of Bluhm [102, 103] (inspired by a work
of Kaufman [315]) with a Theorem of Davenport, Erdos and LeVeque [177],
Bugeaud [126] proved that there exist Liouville numbers which are normal.
Conversely, an explicit example of a Liouville number which is simply normal
in no base has been given by Martin [408] (see also [126]).
Schmidt [507] established that, for any irrational number which is
not a Liouville number, there exists an irrational number such that /
is a Liouville number. Burger [144] extended Erdos result [227] quoted in
Exercise 1.6, and he proved that there exists a Liouville number such that
log is also a Liouville number. Alniacik [16] established the existence of
Liouville numbers with special properties.
Hall [270] proved that any real number in [0, 1] can be expressed as a sum
of two real numbers having partial quotients less than or equal to 4. See also
related works of Astels [34, 35, 36].
26
From the topological point of view, Gruber [261] proved that the set of
badly approximable real numbers is small (precisely, it is a meager set), while
the set of Liouville numbers is large (its complement is a meager set).
Kargaev and Zhigljavsky [309] established metric results related to the approximation of real numbers by rational numbers with bounded denominators.
A different metric aspect of rational approximation has been investigated
by Kuhnlein [345].
Viola [580] studied rational Diophantine approximation in short intervals.
R. C. Baker [50] proved that real numbers having k distinct rational approximations with given denominators q1 , . . . , qk are quite sparse.
The constant c5 ( ) occurring in Exercise 1.4 allows us to dene the socalled Dirichlet spectrum, studied, for example, by Ivanov [287] and Kopetzky
[335]. Divis [197] and Burger [145] considered closely related problems.
2
In this book, we study various questions related to classications of real numbers and we mainly focus our attention on the approximation of real transcendental numbers by algebraic numbers. In the present Chapter however, we
briey review the most important results which have followed Liouvilles Theorem 1.2 and deal with algebraic approximation to algebraic numbers. Since
a broad literature is available on this topic, we omit most of the proofs and
refer the reader to, for example, the monographs of Mahler [388], Schmidt
[510, 512], A. Baker [44], and Feldman and Nesterenko [244] for further information.
Completeness is not the only reason for making this survey. Indeed, some
results of the present Chapter will be used in subsequent parts of the book. For
instance, Theorem 2.7 (or Theorem 2.6) is crucial for proving the main result
of Chapter 7, namely the existence of T -numbers. These real transcendental
numbers with very specic properties of approximation by algebraic numbers
are dened in Chapter 3.
We divide our exposition into four main Sections, dealing respectively with
rational approximation, effective rational approximation, algebraic approximation to algebraic numbers, and effective algebraic approximation to algebraic
numbers. A broad variety of methods are needed for the proofs of the results
below. In a fth Section, we briey mention various applications to irrationality and transcendence statements.
2.1 Rational approximation

Let be a real algebraic number. We say that is an irrationality measure for
if there exists a positive constant c1 (, ) such that, for any rational number
27
28
p/q distinct from and with q > 0, we have

p > c1 (, ) .

q
q
(2.1)
When the constant c1 (, ) can be explicitly computed, we say that is an

effective irrationality measure for .
Theorem 1.1 asserts that the rational numbers are precisely the numbers for
which 1 is an irrationality measure and that any irrationality measure of any
irrational number is at least equal to 2. This fact can be rephrased as an irrationality criterion, as already mentioned in Chapter 1. In the opposite direction,
Liouvilles Theorem states that the degree of any real algebraic number is an
irrationality measure for . In particular, when is a real algebraic number of
degree at most 2, its smallest irrationality measure is equal to its degree. On
the other hand, when is of degree at least 3, our knowledge was, a century
ago, not very satisfactory. Fortunately, since the early twentieth century, many
mathematicians succeeded in strengthening Liouvilles result.
The rst signicant improvement is due to Thue [553], who proved that, for
any real algebraic number of degree d at least 3, every real number greater
than d/2 + 1 is an irrationality measure for . As an immediate application, it
follows that the Diophantine equation
F(x, y) = m, in integers x, y,
(2.2)
where F(X, Y ) is an irreducible, homogeneous binary form of degree at least

3 with integer coefcients and m is a non-zero integer, has only nitely many
solutions. Equation (2.2) is now called the Thue Equation.
In 1921, Siegel [524] sharpened considerably Thues result by showing that
every real number with

d
+j
> min
1 jd
j +1
is an irrationality measure for . Later, Dyson [223] and independently
Gelfond [256],
slightly rened Siegels estimate by proving that every
greater than 2d is indeed an irrationality measure for . All these results
have been superseded by Roths [477], which can be stated as follows.
T HEOREM 2.1. Let be a real algebraic number of degree d 2. Then, for
any positive real number , there exists a positive constant c2 (, ) such that

p > c2 (, )
(2.3)

q
q 2+
for any rational number p/q with q > 0.
2.2 Effective rational approximation
29
Khintchines Theorem 1.10 may suggest that Theorem 2.1 could be further
rened: a conjecture of Lang claims that the map q q 2+ occurring in (2.3)
could be replaced by q q 2 (log q)1+ .
The common feature of all the results mentioned in the present Section,
except Liouvilles Theorem, is that they all are ineffective. This means that the
methods used by Thue, Siegel, Dyson, Gelfond, and Roth did not allow them
to compute explicitly the constant c1 (, ) occurring in (2.1).
However, as established by Davenport and Roth [179], Theorem 2.1 provides a criterion for proving that a real number with too large partial quotients
cannot be algebraic, which improves the criterion deduced from Liouvilles
Theorem 1.2 (see Exercise 1.1).
T HEOREM 2.2. Let be a real number and, for any integer n 1, denote by
qn the denominator of the n-th convergent in its continued fraction expansion.
If
lim sup
n+
(log log qn ) (log n)1/2

= +,
n
then is transcendental.
In 1957, Ridout [470] obtained an important extension of Theorem 2.1, which
can be formulated as follows.
T HEOREM 2.3. Let be a non-zero algebraic number and let
p1 , . . . , pr , q1 , . . . , qs be distinct rational prime numbers. Let , ,
and c be real numbers with 0 1, 0 1 and c > 0. Let p and q be
restricted to integers of the form p = p p1a1 . . . prar , q = q q1b1 . . . qsbs , where
a1 , . . . , ar , b1 , . . . , bs are non-negative integers and p , q are non-zero
integers satisfying | p | cp and |q | cq . Then, if > + , there are
at most a nite number of solutions of the inequality

1
p

0 < < .
q
q
We observe that, taking = = c = 1 in Theorem 2.3, we recover Roths
Theorem. For more references on this subject, the reader is directed to [388],
pages 73 to 76. We postpone some applications of Theorems 2.2 and 2.3 to
Section 2.5.
2.2 Effective rational approximation

Let be a real algebraic number of degree d greater than or equal to 3. Roths
Theorem 2.1 asserts that every real number strictly larger than 2 is an irrationality measure for , while d is an effective irrationality measure for , by
30
Liouvilles Theorem 1.2. There is a big gap between these results and we may
ask whether admits smaller effective irrationality measures. As an immediate
application, this would imply an effectively computable upper bound for the
size of the Thue equation (2.2) which is polynomial in |m|.
A general result has been obtained by Feldman [241] (see also Chapter 9 of
his book [242]) by means of A. Bakers theory of linear forms in the logarithms
of algebraic numbers [43].
T HEOREM 2.4. Let be a real algebraic number of degree d 3. There exist
effectively computable positive numbers c3 ( ) and ( ) such that

p c3 ( )

q q d ( )
for any rational number p/q.
However, ( ) is very small and, denoting by R the regulator of the number
eld Q( ) (see, for example, [435], page 106, for denition), we may take

1
( ) = 3d+26 d 15d+20 R log max{e, R } ,
as proved by Bugeaud and Gyory [136], who used essentially the same method
as Feldman and A. Baker. Notice that R is smaller than (2d 2 H log(d H ))d1 ,
where H 3 is an upper bound for the absolute values of the coefcients of
the minimal polynomial of over Z (see [136]).
An alternative approach, which yields Theorem 2.4 but does not rest on
the theory of linear forms in logarithms, has been successfully worked out
by Bombieri [104, 105] and developed by Bombieri and several co-authors
[106, 108]. It gives a slightly better value for ( ), as far as the dependence on
the regulator is concerned. As noticed in [120], a combination of both methods
yields the value
1

,
( ) = 1026d d 14d R
which is, at present, the best known general result for d 4. Furthermore, for
cubic irrationalities, the sharpest estimate is due to A. Baker and Stewart [46].
A short and self-contained proof of Theorem 2.4 has been worked out by
Bilu and Bugeaud [101] (see also [591], Corollary 10.18).
Better effective irrationality measures are known for classes of algebraic
numbers, including k-th roots of rational numbers, see [244] (Sections 3.5 and
3.6 of Chapter 1 and Section 4.6 of Chapter 2) for references.
Results like Theorem 2.4 have many applications to the resolution of Diophantine equations.
2.3 Approximation by algebraic numbers
31
2.3 Approximation by algebraic numbers

Up to now, we were only interested in the approximation of a real number by
rational numbers a/b. To this end, we have compared the difference | a/b|
with a natural way to measure the size of a/b, namely with max{|a|, |b|}. We
have now to dene a notion of size to evaluate the complexity of an algebraic
number , which, if possible, coincides with max{|a|, |b|} when is the rational a/b. Many denitions have been proposed (see Appendix A), and we
choose the most natural one, usually termed the naive height, which we simply
call height.
D EFINITION 2.1. The height of a complex polynomial P(X ), denoted by
H(P), is the maximum of the moduli of its coefcients. The height of an
algebraic number , denoted by H(), is the height of its minimal polynomial
over Z.
We point out that, in all questions investigated up to Chapter 7, the degree of
the approximant is xed, and the height is allowed to vary. In this case, and for
these kind of questions, the choice of the size has no particular signicance.
This does not remain true in Chapter 8, where degree and height are allowed to
vary simultaneously, and some of the results presented there are given in terms
of the absolute logarithmic height or in terms of the Mahler measure.
The next theorem, asserted by Roth [477] and proved by LeVeque [363],
extends Theorem 2.1 to number elds.
T HEOREM 2.5. Let K be a number eld and be a real algebraic number
not in K. Then, for any > 0, there exists an (ineffective) positive constant
c4 (, K, ) such that
| | > c4 (, K, ) H()2 ,
(2.4)
for every in K.
An extension of Dirichlets Theorem 1.1 shows that the exponent 2 + in (2.4)
cannot be replaced by a real number strictly less than 2.
Instead of studying approximation of an algebraic number by elements of a
xed number eld, we now consider approximation by algebraic numbers of
degree less than or equal to n. This would generalize Theorem 2.1 in another
direction than Theorem 2.5. Wirsing [599] has given a satisfactory answer to
this problem, by obtaining an exponent depending only on n.
T HEOREM 2.6. Let be a real algebraic number and n 1 be an integer. For
any > 0, there exists a positive (ineffective) constant c5 (, n, ) such that
| | > c5 (, n, ) H()2n ,
for any algebraic number = of degree at most n.
(2.5)
32
Theorem 2.6 reduces to Roths Theorem for n = 1; the proof of Wirsing uses
the main steps of that of Roth. However, a metric argument suggests that it
could be possible to replace the exponent 2n in (2.5) by n 1 .
This conjecture has been proved by Schmidt (see [504] for n = 2 and [506]
for n 3), nearly at the same time as Wirsing published Theorem 2.6, but with
very different arguments.
T HEOREM 2.7. Let be a real algebraic number and n 1 be an integer. Let
be a positive real number. Then there exists a positive (ineffective) constant
c6 (, n, ) such that
|P( )| > c6 (, n, ) H(P)n
for any integer polynomial P(X ) of degree at most n and such that P( ) =
0. Furthermore, there exists a positive (ineffective) constant c7 (, n, ) such
that
| | > c7 (, n, ) H()n1
for any algebraic number = of degree at most n.
The rst assertion of Theorem 2.7 is a corollary to the celebrated Subspace
Theorem of Schmidt, a remarkable result having many important applications
in Diophantine approximation. The proof, very involved, is given with many
details in [512], its main lines being also in [510]. The second assertion of
Theorem 2.7 follows from the rst one, as proved in Theorem 3.5. Notice that
while it is used in Chapter 7 in order to prove that T -numbers (see Denition
3.1) do exist, Theorem 2.6 however is sufcient to get the same conclusion.
Clearly, Theorem 2.7 (and, thus, Theorem 2.8 below) is interesting only for n
at most equal to the degree of minus two, since otherwise it is superseded by
Liouvilles inequality (Corollary A.2).
T HEOREM 2.8. Let n 1 be an integer and be a real algebraic number
of degree d. Then, for any positive real number , there exist an ineffective
constant c8 (, n, ) and innitely many algebraic numbers of degree at most
n such that
| | < c8 (, n, ) H() min{n+1,d}+ .
Theorem 2.8 follows from the rst assertion of Theorem 2.7, as proved in
Theorem 3.5. In the next Section, we show that Theorem 2.8 can be improved
upon when n d 1.
2.4 Effective approximation by algebraic numbers
33

Corollary 1.2 asserts that real quadratic algebraic numbers are badly approximable by rational numbers. This is a particular case of a more general phenomenon: for any integers d 2 and n d 1, real algebraic numbers
of degree d are badly approximable by algebraic numbers of degree n. Theorem 2.9 covers the case n = d 1, while Theorem 2.11 deals with the case
n d.
T HEOREM 2.9. Let be a real algebraic number of degree d 2. There exist
effectively computable positive constants c9 ( ) and c10 ( ) such that
| | c9 ( ) H()d
for any algebraic number of degree

at most d 1,
and
| | c10 ( ) H()d
for innitely many real algebraic numbers

of degree d 1.
The rst part of Theorem 2.9 is a restatement of Corollary A.2, while the second part originates in the work of Wirsing [598]. Bombieri and Mueller [107]
have given an alternative proof, which we reproduce here. They proved Theorem 2.10 below, which implies the second part of Theorem 2.9.
T HEOREM 2.10. Let with | | 1/2 be a real algebraic number of degree
d 2. For every H 2, there exists an algebraic number of degree at most
d 1 such that
| |
d!(d 1)
Hd
and

(d1)2 /d
H() 2d1 d d + 1 H( )
H.
(2.6)
Moreover, is real as soon as H > 2d (d + 1)(3d+5)/2 H( )d1 .

P ROOF. For any real number H 2, let C(H ) denote the convex symmetric
body in Rd dened by
|x0 + x1 + . . . + xd1 d1 | H d+1 ,
|x1 |, . . . , |xd1 | H.
Let i = i (H ), i = 1, . . . , d, be the successive minima of C(H ) (see Ap(i)
(i)
pendix B for the denition). We denote by (x0 , . . . , xd1 ), i = 1, . . . , d,
34
linearly independent points at which the successive minima i are attained and
we set
(i)
(i)
(i)
Pi (X ) = x0 + x1 X + . . . + xd1 X d1 .
We rst show that there exists some integer i for which
|Pi ( )|
i
H
d!
(2.7)
holds. To this end, we set

M = max
1id
|Pi ( )|
,
i
and we observe that 1 , . . . , d are as well the successive minima of the convex
body
C(H, M) := C(H ) {|x1 + 2 x2 + . . . + (d 1) d2 xd1 | M}.
By Theorem B.3, we have the inequalities
2d
1 . . . d vol(C(H, M))
d!
(2.8)
1 . . . d 1,
(2.9)
and
since vol(C(H )) 2d . Further, the volume of C(H, M) is at most equal to the

volume of the convex body dened by the inequalities
|x0 + x1 + . . . + xd1 d1 | H d+1 ,
|x1 + 2x2 + . . . + (d 1)xd1 d2 | M,
|x2 |, . . . , |xd1 | H.
Hence, we get
vol(C(H, M)) 2d M/H.
Combined with (2.8) and (2.9), this implies that M H/d!, thus (2.7) holds
for some index i.
For such an i, we have by construction |Pi ( )| i H d+1 and H(Pi )
i H , since H 2 and | | 1/2. Hence, Lemma A.5 and (2.7) yield that the
polynomial Pi (X ) has a root with
| | (d 1)
d!(d 1)
|Pi ( )|
,
|Pi ( )|
Hd
(2.10)
35
and
H() 2d1 H(Pi ) 2d1 i H,
(2.11)
by Lemma A.3.
It remains for us to bound i . We infer from (2.9) that
i d 1d+1 ,
(2.12)
and we are led to bound 1 from below. By Theorem A.1, we get

|P1 ( )| d d+1 (d + 1)(d1)/2 H(P1 )d+1 H( )d+1 .
(2.13)
Furthermore, by denition, the polynomial P1 (X ) satises |P1 ( )|

1 H d+1 and H(P1 ) 1 H , hence
|P1 ( )| d1 H(P1 )d+1 .
Combining (2.13) and (2.14), we get

(d1)/d
1 d d + 1 H( )
,
(2.14)
(2.15)
and (2.6) follows from (2.10), (2.11), (2.12), and (2.15).

If the algebraic number constructed above is non-real, then, denoting by
P (X ) its minimal polynomial over Z and by its complex conjugate, Lemma
A.6 and (2.10) imply
|P ( )| 2d1 d H() | | | |
(2.16)
2d1 d (d 1)2 (d!)2 H() H 2d .
On the other hand, Theorem A.1 yields
|P ( )| d d+1 (d + 1)(d1)/2 H()d+1 H( )d+1 ,
which, combined with (2.16) and (2.6), gives the upper bound H 2d (d +
1)(3d+5)/2 H( )d1 . This completes the proof of the theorem.
We are now concerned with the approximation of algebraic numbers by
algebraic numbers of same or larger degree. Apparently, Theorem 2.11 did
not appear previously. Its rst part is a restatement of Corollary A.2, while its
second part originates in a work of Davenport and Schmidt [182]. Theorem
2.11 is used in the proof of Theorem 7.5. We recall that an algebraic integer is
an algebraic number whose minimal polynomial over Z is monic.
36
T HEOREM 2.11. Let be a real algebraic number of degree d 2. Let n d

be an integer. There exist positive, effectively computable, constants c11 (, n)
and c12 (, n) such that
| | c11 (, n) H()d
for any algebraic number = of

degree at most n,
and
| | c12 (, n) H()d
for innitely many real algebraic

integers of degree n.
P ROOF. Without loss of generality, we may assume that | | 1/2. Theorem

A.1 implies that for any non-zero integer polynomial P(X ) of degree at most
d 1 we have
|P( )| > (d + 1)3(d1)/2 H( )d+1 H(P)d+1 .
Set = (d + 1)3(d1)
2 /(2d)
H( )(d1)
2 /d
(2.17)
and let H be a real number with
H d > 2n+5 (d + 1)6 (n + 1)5/2 2 .
(2.18)
It follows from (2.17) that the rst minimum 1 of the compact convex C(H )
dened by the inequalities
|xd1 d1 + . . . + x1 + x0 | H d+1 ,
|x1 |, . . . , |xd1 | H,
satises 1 > 1/(d1) . Consequently, by Theorem B.3, the d-th minimum
d of C(H ) is smaller than . Thus, there exist d linearly independent integer
( j)
( j)
( j)
polynomials P j (X ) := xd1 X d1 + . . . + x1 X + x0 , for j = 1, . . . , d, of
degree at most d 1 satisfying
|P j ( )| H d+1 and H(P j ) H,
( j)
(2.19)
for j = 1, . . . , d, and := | det(xi )| d!. Let p be a prime number which

does not divide . Since the product of all prime numbers up to d 2 exceeds d!,
we may assume that p is not larger than d 2 . We argue exactly as Davenport
and Schmidt [182] in order to construct a monic polynomial of degree n, small
when evaluated at .
Since p does not divide , there exists a superscript j such that p does not
( j)
divide x0 . Without loss of generality, we assume that j = 1.
37
We consider the following linear system of d equations in the d real unknowns 1 , . . . , d :

n + p(1 P1 ( ) + . . . + d Pd ( )) = p(d + 1) H d+1

n n1 + p(1 P1 ( ) + . . . + d Pd ( )) = p H + p
|Pi ( )|
(2.20)
1id
1 xm(1) + . . . + d xm(d) = 0 (m = 2, . . . , d 1).

Since the polynomials Pi (X ), for i = 1, . . . , d, are linearly independent,
this system has one and only one solution (1 , . . . , d ). We take a d-tuple
(t1 , . . . , td ) of integers such that |i ti | 1 for i = 1, . . . , d, and we set
xm = t1 xm(1) + . . . + td xm(d) , for m = 0, . . . , d 1.
We consider the polynomial
P(X ) = X n + p(xd1 X d1 + . . . + x1 X + x0 )
= X n + p(t1 P1 (X ) + . . . + td Pd (X )),
which, by a suitable choice of (t1 , . . . , td ), is irreducible. Indeed, by using
Eisensteins Criterion, it is sufcient to check that its constant coefcient,
(1)
(d)
namely p(t1 x0 + . . . + td x0 ), is not divisible by p 2 , since its leading coefcient is congruent to 1 modulo p. We x a (d 1)-tuple (t2 , . . . , td ) and there
remain two possible choices for t1 , which we denote by t1,0 and t1,1 = t1,0 + 1.
(1)
(1)
(d)
Since p does not divide x0 , at least one of the integers t1,0 x0 + . . . + td x0
(1)
(d)
or t1,1 x0 + . . . + td x0 is not divisible by p. This enables us to choose t1
such that the polynomial P(X ) is Eisensteinian with respect to the prime number p, hence, irreducible.
Furthermore, the polynomial P(X ) satises
P( ) = n + p(t1 P1 ( ) + . . . + td Pd ( )),
hence, by (2.18), (2.19), and the rst equation of the system (2.20), we have
0 < p H d+1 P( ) p(2d + 1) H d+1 < 1.
(2.21)
On the other hand, we have

P ( ) = n n1 + p(t1 P1 ( ) + . . . + td Pd ( ));
thus, by (2.19) and the second equation of (2.20), we get
P ( ) p H
(2.22)
38
and
P ( ) p H + 2 p
d
|Pi ( )| p H (1 + 4d(d 1)),
(2.23)
i=1
since | | 1/2 and H(Pi ) (d 1) H for i = 1, . . . , d. Finally, by (2.19)

and the last equations of (2.20), we get
|xm | d H, for m = 2, . . . , d 1,
(2.24)
and we infer from (2.22), (2.23), (2.24), and | | 1/2 that

|x1 | 1 + |P ( )| +
d1
j|x j | (d 3 + 4d 2 p) H.
(2.25)
j=2
It follows from (2.21), (2.24), and (2.25) that

|x0 | 1 + P( ) + |x1 | +
d1
|x j | (d 3 + d 2 + 4d 2 p) H,
j=2
and we nally derive from the upper bound p d 2 that

H(P) 6d 4 H.
(2.26)
By (2.21), (2.22), and the rst assertion of Lemma A.5, the polynomial P(X )
has a root satisfying
P( )
n(2d + 1) H d ,
(2.27)
| | n
P ( )
thus, by (2.26),
| | 6d+1 n (d + 1)(3d
2 +5d+5)/2
H( )(d1) H()d .
2
If were non-real, then its conjugate would also satisfy (2.27) and, by (2.26),
Lemma A.6, and | | 1/2, we would then have
P( ) 2n+5 (n + 1)5/2 (d + 1)6 3 H 12d .
Together with (2.18), this would contradict the lower bound P( ) H d+1
obtained in (2.21). Consequently, is a real algebraic integer of degree n.
For an alternative proof of Theorem 2.9, it sufces to take n = d 1 in the
proof of Theorem 2.11: the Eisenstein Criterion ensures that the polynomial
P(X ) is irreducible.
We could as well state analogues of Theorems 2.9 and 2.11, under the same
hypothesis on , with upper and lower bounds for |P( )|, where P(X ) is an
integer (or a monic integer) polynomial of degree d.
2.5 Remarks on irrationality and transcendence statements
39
2.5 Remarks on irrationality and transcendence statements

It follows from Theorem 1.1 that a real number is irrational if it has innitely many good rational approximants. In a similar way, Roths Theorem
and its generalizations can be applied to establish that is transcendental if it admits innitely many very good rational approximants. Classical results, including the transcendence of the Champernowne number
0.1234567891011121314 . . . , can be obtained by applying Theorems 2.1, 2.2,
and 2.3 (although Mahlers original proof of this statement [380] used a
weaker transcendence criterion), see Mahler [397] and Zhu [616]. For instance,
Ferenczi and Mauduit [246] gave a combinatorial translation of Theorem 2.3
which shows in particular that real numbers whose expansion in a given base is
Sturmian are transcendental. In most of the cases, an interesting question often
remains open: after having shown that a given real number has an irrationality measure strictly larger than 2 to conclude that is transcendental, what can
be said about its smallest irrationality measure?
Another typical problem is the algebraicity of real numbers dened in terms
of their continued fraction expansion. By Corollary 1.6, real transcendental
numbers with bounded partial quotients do exist and it is of interest to construct explicit examples. The rst result of this type is due to Maillet [403] (see
also Section 34 of Perron [454]), who found continued fractions with bounded
partial quotients giving trancendental numbers by using a general form of
Liouvilles inequality (see Corollary A.2) which limits the precision of the
approximation of an algebraic number by quadratic irrationalities. Such an argument is quite natural when we keep in mind that the quadratic irrationalities
coincide with the real numbers having ultimately periodic continued fraction
expansions (Theorem 1.16). Other examples of transcendental numbers with
bounded partial quotients are due to A. Baker [37, 38] who applied Theorem
2.2, asserting that the partial quotients of the continued fraction expansion of
an algebraic number cannot increase too rapidly.
An interesting application of the case n = 2 of Theorem 2.7 has been
worked out by Queffelec [464] (the idea of using Schmidts Theorem in a similar context goes back to Davison [185]), who proved that the real number
= [1; 2, 2, 1, 2, 1, 1, 2, 2, 1, 1, 2, 1, 2, 2, 1, . . . ],
with partial quotients being the ThueMorse word on {1, 2}, that is, the xed
point of the substitution 1 12, 2 21, is transcendental. Again, although
we know that is too well approximable by quadratic numbers to be algebraic,
we do not know precisely how well it is approximable. As we shall see in
the course of the next Chapters, this question appears, in general, to be very
difcult.
40
2.6 Notes
Let be a positive real number. Theorem 2.1 implies that the n-th partial
quotient an of a real algebraic number of degree d satises log an < c13 (1 +
)n for any positive integer n and a suitable ineffective constant c13 , depending
only on and . Using Theorem 2.4, one can get an effective upper bound, but
1 + is then replaced by d 1 , for a (very) small positive number . In
the case of cubic algebraic numbers, this has been considerably
improved by
Wolfskill [600], who obtained the upper estimate log an < c14 ( 3 + )n for
any positive integer n and an effective constant c14 , depending only on and .
Using a method of Dyson [223], Mahler [383] proved that if ( pn /qn )n1
denotes the sequence of convergents of a quadratic or cubic real number, then
the greatest prime factor of qn tends to innity with n. Actually, it follows from
Ridouts Theorem 2.3 that this conclusion holds for any real irrational number.
Shorey [523] established related quantitative results.
Mahler [386] applied Ridouts Theorem 2.3 to get lower bounds for the
distance to the nearest integer of powers of rational numbers.
An interesting complement to Ridouts Theorem 2.3 has been given by
Lagarias [346]. Stewart [544] used a similar construction to investigate a question on divisors of a product of consecutive integers.
Mahler [390] (see also his book [388]) extended Theorem 2.5 and investigated approximation of algebraic numbers by algebraic integers lying in a
given number eld. He also proved an inhomogeneous result, later generalized
by Schmidt, see Section 7.4 of [510].
Allouche, Davison, Queffelec, and Zamboni [6], Davison [186], and Baxa
[55] established extensions of Queffelecs result mentioned in Section 2.5. See
also Liardet and Stambul [367] and Chapter 13 of Allouche and Shallit [7].
3
The set of real numbers splits into algebraic and transcendental numbers, but
these two subsets do not have the same size, the former being countable, while
the latter has the power of continuum. Such a crude classication of real numbers seems to be rather unsatisfactory, and one aims to nd some way to classify the transcendental numbers. First, we have to ask which requirements
should satisfy a good classication. Ideally, for a given real number , we
would like to have a simple criterion to determine the class to which belongs.
Furthermore, two algebraically dependent real numbers should belong to the
same class. The rst classication of transcendental real numbers has been
proposed by Maillet [403, 404], and others were subsequently described by
Perna [453] and by Morduchai-Boltovskoj [430], but none of these has proved
to be relevant. For instance, Maillets classication depends on the size of the
partial quotients of the real numbers and, clearly, does not satisfy the second
requirement.
An attempt towards a reasonable classication was made in 1932 by
Mahler [376], who proposed to subdivide the set of real numbers into four
classes (including the class of algebraic numbers) according to, roughly speaking, their properties of approximation by algebraic numbers. Mahlers classication satises our second requirement: two algebraically dependent real
numbers belong to the same class. However, if a real number is given (that
is, for example, a Cauchy sequence of rational numbers), there is, in general,
very little hope of determining to which class it belongs. Although non-trivial,
Mahlers classication does not seem to be entirely satisfactory since almost all
real numbers (in the sense of the Lebesgue measure) belong to the same class.
Nevertheless, it has been widely studied, as has the closely related Koksmas
classication, which was proposed a few years later [333].
In the present Chapter, we begin by dening Mahlers classication and
we prove some of its properties. We then dene Koksmas classication and
41
42
compare them. Further, we state the Main Problem, which motivates the next
ve Chapters. Finally, we discuss some links between algebraic approximation
and simultaneous rational approximation, and we introduce four other exponents of Diophantine approximation.
3.1 Mahlers classication

Mahlers idea [376] consists in classifying the real numbers according
to the accuracy with which non-zero integer polynomials, evaluated at ,
approach 0. For given positive integer n and real number H 1, we dene the
quantity
wn (, H )
:= min{|P( )| : P(X ) Z[X ], H(P) H, deg(P) n, P( ) = 0},
where H(P) denotes the height of the polynomial P(X ) (see Denition 2.1).
Then, we set
log wn (, H )
wn ( ) = lim sup
log H
H +
and
w( ) = lim sup
n+
wn ( )
.
n
In other words, wn ( ) is the supremum of the real numbers w for which there
exist innitely many integer polynomials P(X ) of degree at most n satisfying
0 < |P( )| H(P)w .
Further, it is an easy exercise (see Exercise 3.1) to check that for any positive
integer n, for any real number and any non-zero rational number a/b, we
have wn ( ) = wn ( + a/b) = wn (a/b). This allows us in most of the proofs
below to assume that belongs to an interval of small length.
For n 1, we have 0 wn ( ) +, since wn (, H ) w1 (, H ) 1
for any H with H | | + 1. Thus, we get 0 w( ) +. Moreover, the
sequence (wn ( ))n1 is clearly non-decreasing. With these notations, Mahler
[376] (actually, he used the Greek letter instead of w; hopefully, the present
notation will not perturb the reader!) divided the set of real numbers into four
disjoint classes.
D EFINITION 3.1. Let be a real number. We say that is an
A-number, if w( ) = 0;
S-number, if 0 < w( ) < +;
3.1 Mahlers classication
43
T -number , if w( ) = + and wn ( ) < + for any n 1;

U -number, if w( ) = + and wn ( ) = + from some n onwards.
Theorem 3.1 below shows that the A-numbers are exactly the algebraic numbers. According to Stolarsky [546], the terminology S-number refers to
Siegel, the letters T and U having then been chosen by alphabetical order.
In another paper of 1932, Mahler [377] proved that almost all (in the
sense of Lebesgue measure) real numbers satisfy wn ( ) 4n for any
positive integer n (actually, his result is slightly stronger) and conjectured
that it should be possible to replace 4 by 1. This has been conrmed by
Sprindzuk [538] in 1965; Chapter 4 is devoted to this problem and related
questions.
Mahler [375] had already introduced the terminology S-number in 1930,
but this, however, does not match with Denition 3.1: the set of S-numbers in
[375] is equal to the union of the sets of A- and S-numbers in [376].
The classes S, T , and U may be further subdivided into innitely many
subclasses, by introducing the notion of type.
D EFINITION 3.2. We dene the type t ( ) of an S-number as the value of
w( ), that is,
t ( ) = lim sup
n+
wn ( )
.
n
Koksma [333] called Index der S-Zahl the quantity

wn ( )
n
n1
sup
(3.1)
and he observed ([333], Satz 4) that any S-number is of index at least 1.

Several authors, including Schneider [517], Kasch and Volkmann [311], A.
Baker [38, 44], and Guting [269], used (3.1) to dene the type of an S-number,
instead of Denition 3.2. As observed by A. Baker and Schmidt [45] in 1970
in a footnote, it was at that time customary to use sup wn ( )/n, but the above
would seem more natural, where they refer to
lim sup
n+
wn ( ) + 1
wn ( ) + 1
and sup
.
n+1
n+1
n1
(3.2)
Indeed, the metric results stated in Chapter 5 show that one of the denitions in
(3.2) should be preferred to (3.1). The rst alternative in (3.2) coincides with
Denition 3.2: in the authors opinion, the lim sup is much more relevant than
44
the supremum. Notice however that the existence of S-numbers with

lim sup
n+
wn ( ) + 1
wn ( ) + 1
< sup
n+1
n+1
n1
remains an open problem. Actually, no matter how one denes the type, Theorem 5.8 asserts that there exist real S-numbers of arbitrary type greater than or
equal to 1.
D EFINITION 3.3. Let be a T -number and, for any n 1, dene the real
number n by wn ( ) = n n . The type of , denoted by t ( ), is the quantity
t ( ) = lim sup n .
n+
The type of a T -number takes its values in [1, +]. Schmidt proved [507, 508]
(see Chapter 7) that there exist T -numbers of arbitrary given type in [3, +],
but the problem of the existence of T -numbers of type strictly less than 3 remains open.
D EFINITION 3.4. Let be a U -number. The type of , denoted by t ( ), is the
smallest positive integer n such that wn ( ) = +.
The U -numbers of type 1 are precisely the Liouville numbers (Denition 1.1).
Furthermore, LeVeque [361] proved that there exist U -numbers of arbitrary
given type (see Section 7.6). There are alternative denitions for the type of a
U -number, see for instance Schmidt [508] and Section 7.1.
We conclude this Section by pointing out that the Lebesgue Density Theorem (see, for example, Riesz and Sz.-Nagy [472], page 13, or Mattila [415],
Corollary 2.14) implies that the function wn , which is invariant by rational
translation, takes the same value at almost all real numbers, that is, for any real
number w, the set of real numbers with wn ( ) = w has either measure zero
or has full measure. Consequently, almost all real numbers belong to the same
class and have the same type. This fact was rst observed by Sprindzuk [530]
(see also [539], page 23).
Furthermore, as soon as the set of real numbers is divided into disjoint
measurable classes with the property that any two real numbers differing by
a rational number belong to the same class (this is obviously true if any two algebraically dependent real numbers belong to the same class), then there exists
a class to which almost all real numbers belong (see Exercise 6.5).
3.2 Some properties of Mahlers classication
45
3.2 Some properties of Mahlers classication

Our rst result gives a lower bound for w( ) when is a real transcendental
number.
P ROPOSITION 3.1. Let n be a positive integer and be a real number not
algebraic of degree at most n. We then have wn ( ) n. In particular, if is
transcendental, then we have w( ) 1.
P ROOF. Without loss of generality, we may assume that 0 < < 1/2. Let
H 2 be real, and consider the system of inequalities
|an n + . . . + a1 + a0 | H n
|a1 |, . . . , |an | H.
By Theorem B.2, it has a non-zero integer solution (a0 , a1 , . . . , an ). Further,
|an n + . . . + a1 + a0 | < H n and 0 < < 1/2 imply that |a0 | H . Since
is not algebraic of degree at most n, this means that there exists a non-zero
integer polynomial PH (X ), of height at most H , such that
0 < |PH ( )| H(PH )n .
By taking arbitrarily large values of H , we get a sequence of distinct integer polynomials (Pk )k1 , whose heights increase to innity, and which satisfy
0 < |Pk ( )| H(Pk )n for k 1. It immediately follows that wn ( ) n.
Consequently, w( ) 1 if is transcendental. We observe that we may alternatively use Dirichlets Schubfachprinzip instead of Theorem B.2, see the
proof of Lemma 8.1.
T HEOREM 3.1. The A-numbers are exactly the real algebraic numbers. Let
be an algebraic real number of degree d and let n be a positive integer. We
then have wn ( ) = min{n, d 1}.
P ROOF. Let be an algebraic number of degree d and let P(X ) be an integer
polynomial of degree n and of height H such that P( ) = 0. We infer from
Theorem A.1 that wn ( ) d 1, hence, we get that w( ) = 0. In view
of the last assertion of Proposition 3.1, this shows that the sets of A-numbers
and of real algebraic numbers coincide. By the rst assertion of Schmidts
Theorem 2.7, the last inequality can be rened to wn ( ) min{n, d 1}.
Since wn ( ) wd1 ( ) holds if n d 1, we then infer from Proposition
3.1 that wn ( ) = min{n, d 1}.
It immediately follows from Theorem 3.1 that real numbers with 0 < w( ) <
1 do not exist.
46
C OROLLARY 3.1. Every real S-number is of type at least 1.

Theorem 3.2, due to Mahler [375, 376], ensures that his classication satises
the second requirement stated at the beginning of the Chapter.
T HEOREM 3.2. Two algebraically dependent real numbers and belong to
the same class.
P ROOF. By Theorem 3.1, we may assume that and are transcendental.
Let n be a positive integer. The constants c1 , c2 , and c3 appearing below are
positive, effectively computable, and depend only on , , and n. Let H 1
be a real number and A(X ) be an integer polynomial of height at most H and
degree at most n with |A( )| = wn (, H ). Let
F(X, Y ) =
M
N
bhk X h Y k =
h=0 k=0
M
Bh (Y )X h
h=0
be a primitive, irreducible, integer polynomial vanishing at the point (, ) and

such that B M (Y ) is not identically zero. For some values of y, the polynomials
A(X ) and F(X, y) have no common zero, hence the resultant (see Denition
A.1)

R(Y ) = Res X A(X ), F(X, Y )
is not the zero polynomial. Moreover,
deg R(Y ) n max deg B j (Y ) n N ,
0 jM
and there exists a constant c1 such that the height of the polynomial R(Y ) is
not greater than c1 H M .
Classical properties of the resultant (see, for example, [349], Chapter IV, or
[586], Chapter 5, or [156], Appendix A) imply that there exist two polynomials
g(X, Y ) and h(X, Y ) satisfying
R(Y ) = A(X )g(X, Y ) + F(X, Y )h(X, Y ),
and such that the height of the polynomial g(X, Y ) is not greater than c2 H M1 .
Hence, we have R() = A( )g(, ) and |g(, )| c3 H M1 . Since |R()|
wn N (, c1 H M ), our choice of the polynomial A(X ) implies that
wn (, H ) = |A( )| c31 H M+1 wn N (, c1 H M ).
(3.3)
Letting H tend to innity, we obtain

wn ( ) M 1 + Mwn N ().
(3.4)
3.3 Koksmas classication
47
It follows that
w( ) lim sup
n+
(M 1)N + M N wn N ()
M N w()
nN
(3.5)
and, inverting the roles played by and , we get

wn () N 1 + N wn M () and w() M N w( ).
Consequently, w( ) and w() are simultaneously nite or innite. Furthermore, wn ( ) is nite for every positive integer n if, and only if, wn () is nite
for every positive integer n. This completes the proof of the theorem.
We may ask whether inequalities (3.4) and (3.5) are optimal: how can the
type of S-numbers and which are roots of a non-zero integer polynomial
P(X, Y ) of bidegree (M, N ) differ? It follows from (3.4) that for any real transcendental number and any positive integers k and n, we have

(3.6)
wn ( ) + 1 k wn ( k ) + 1
and
wn ( k ) wkn ( ).
(3.7)
Inequality (3.6) is sharp for n = 1, since the difference k(w1 ( k ) + 1) 1

w1 ( ) can take any non-negative real value, see Exercise 3.6. Furthermore, it
follows from Theorem 7.7 that for any integer d kn(kn + 2) and for the

j
positive real number with kn = j1 2(d+1) we have wn ( k ) = wkn ( ) =
d. This shows that (3.7) is also sharp.
Moreover, it follows from (3.4) and (3.5) that two algebraically dependent
T -numbers have same type. This is not the case for two algebraically dependent U -numbers, see Theorem 7.4.

Koksmas point of view [333] is close to that of Mahler, but instead of looking
at the approximation of 0 by integer polynomials evaluated at the real number
, Koksma considered the approximation of by algebraic numbers. For given
positive integer n and real number H 1, we dene the quantity
wn (, H )
:= min{| | : real algebraic, deg() n, H() H, = },
where H() denotes the height of the algebraic number (see Denition 2.1).
48
We then set
log(H wn (, H ))
log H
H +
wn ( ) = lim sup
(3.8)
and
w ( ) = lim sup
n+
wn ( )
.
n
wn ( )
In other words,
is the supremum of the real numbers w for which there
exist innitely many real algebraic numbers of degree at most n satisfying
0 < | | H()w1 .
(3.9)
We have wn ( ) = wn ( + a/b) = wn (a/b) for any real number , any

positive integer n, and any non-zero rational number a/b (see Exercise 3.1).
This implies that the function wn takes the same value at almost all real
numbers.
The factor H occurring in (3.8) (corresponding to the exponent 1 of H()
in (3.9)) does not appear in the denition of wn ( ), but it is natural to introduce
it here. Indeed, if an integer polynomial P(X ) satises |P( )| < H(P)w ,
then, by Lemma A.5, it has a root with | | < n H(P)w |P ( )|1 . In
general |P ( )| is likely to be comparable to H(P) (or, equivalently, P(X ) is
likely to have no other root close to ). Then, | | is not much bigger than
H()w1 . Furthermore, we shall see in Chapter 4 that wn ( ) = wn ( ) = n
for almost all real numbers .
For any real number , any positive integer n, and any H | | + 1, we
have wn (, H ) w1 (, H ) (| | + 1)/H , as can be seen by considering the
rational number with denominator [H ] closest to . Thus, we get 0 wn ( )
+. Moreover, the sequence (wn ( ))n1 is clearly non-decreasing, and the
functions w1 and w1 are equal.
D EFINITION 3.5. Let be a real number. We say that is an
A -number, if w ( ) = 0;
S -number, if 0 < w ( ) < +;
T -number, if w ( ) = + and wn ( ) < + for any n 1;
U -number, if w ( ) = + and wn ( ) = + from some n onwards.
Actually, Koksma [333] dened wn (, H ) as the minimum
min{| | : complex algebraic, deg() n, H() H },
in such a way that is transcendental if, and only if, wn (, H ) is positive
for any n 1 and H 1. When Koksma introduced his classication
49
(and also in 1957, when Schneiders book appeared [517]), it was not known
that w ( ) = 0 is a necessary and sufcient condition for to be algebraic
(a result due to Wirsing [598]). For this reason, Koksma only classied the
transcendental numbers in three families (according to [333], an S -number
is a transcendental number with w ( ) < +), and, for consistency, he
called A -numbers the set of algebraic numbers. Thus, our denition of A numbers differs from Koksmas, but this is only apparent. We choose this way
to proceed in order to match with Mahlers and Sprindzuks classications (see
Chapter 8).
In [333] Koksma dened wn ( ) for a complex transcendental number by
taking into account its complex approximants. Since we restrict our attention
to the approximation of real numbers , it is plainly natural to consider only
real algebraic approximants. This denition is also taken by Schmidt in [512],
page 279.
However, it is important to notice that it makes no difference if we also take
into account complex non-real approximants. Indeed, for n 1, H 1, and a
real number , set
wnc (, H )
:= min{| | : algebraic, deg() n, H() H, = }
and
log(H wnc (, H ))
.
log H
H +
wnc ( ) = lim sup

Clearly, we have the inequality
wnc ( ) wn ( ),
which turns out to be an equality, as stated in the next lemma, extracted from
[128].
L EMMA 3.1. For any positive integer n and any real number not algebraic
of degree at most n, we have wnc ( ) = wn ( ).
P ROOF. The case n = 1 is trivial. Let n 2 be an integer, H > 1 be a real
number, and be a real number not algebraic of degree at most n. Let 1 be
an algebraic number of height at most H and of degree n 1 less than or equal
to n such that wnc (, H ) = | 1 |. We may assume that 1 is non-real,
otherwise the lemma is clearly true. Then, the minimal polynomial of 1 over
Z, denoted by P1 (X ), has two distinct roots 1 and 1 lying very close to .
Graces complex version of Rolles Theorem (see, for example, [114], page 25)
asserts that its derivative P1 (X ) has a root 2 in the closed disc centered at
50
(1 + 1 )/2 and of radius |1 1 | cot(/n 1 )/2. Observe that this closed disc
reduces to the point (1 + 1 )/2 if n 1 = 2. Consequently, we have
| 2 | | e 1 | +
|1 1 | n
n | 1 |,
2
deg(2 ) n 1 1, and H(2 ) 2n H(P1 ) 2n n H,

by Lemma A.3. If 2 is non-real, we proceed further in the same way in order to construct an algebraic approximant 3 of whose degree is strictly less
than the degree of 2 . We iterate this process until we end up with a real approximant. This always happens since the degrees of the algebraic numbers we
construct form a strictly decreasing sequence. Consequently, there exists a real
algebraic number with
H() (2n n)n H and | | n n | 1 |,
and = , since is not algebraic of degree at most n. Thus, we have for any
real number H 1

2
wn , 2n n n H n n wnc (, H )
and
wnc ( )

2
log H n n wn , 2n n n H
= wn ( )
lim sup
log H
H +
holds, as asserted.
As for Mahlers classication, the classes S , T , and U may be further subdivided into innitely many subclasses.
D EFINITION 3.6. We dene the -type t ( ) of an S -number as the value
of w ( ), that is,
t ( ) = lim sup
n+
wn ( )
.
n
As for the S-numbers, this denition may not match with the quantity
wn ( )
,
n
n1
sup
which Koksma [333] called Index der S -Zahl . The discussion following
Denition 3.2 can be rephrased with wn and S instead of wn and S. Again,
no matter how the -type is dened, Theorem 5.6 asserts that there are real
S -numbers of arbitrary -type greater than or equal to 1.
51
D EFINITION 3.7. Let be a T -number and, for any positive integer n, dene
the real number n by wn ( ) = n n . The -type of , denoted by t ( ), is the

quantity
t ( ) = lim sup n .
n+
a U -number.
D EFINITION 3.8. Let be

The -type of , denoted by t ( ), is
the smallest positive integer n such that wn ( ) = +.
The U -numbers of -type 1 are precisely the Liouville numbers.
The analogue of Theorem 3.2 holds, namely two algebraically dependent
real numbers belong to the same class. Moreover, they have same -type if
both are T -numbers. The proof is left as Exercise 3.6, where it is also shown
that for any real transcendental number and any positive integers k and n, we
have

wn ( ) + 1 k wn ( k ) + 1
and
( ).
wn ( k ) wkn
(3.10)
As for inequality (3.7), Theorem 7.7 can be used to prove that (3.10) is sharp.
As in Mahlers classication, since the functions wn are invariant by rational
translations, almost all real numbers belong to the same class and have the
same -type. We end this Section with a metric result due to Koksma [333],
Satz 13.
T HEOREM 3.3. Almost all real numbers are S -numbers of -type less than
or equal to 1.
P ROOF. For any positive integers H , n and k, let E(H, n, k) denote the union
of the intervals [ H n11/k , + H n11/k ] over all real algebraic numbers of degree at most n and of height H . Since there are no more than
n(n + 1)(2H + 1)n such algebraic numbers , we get

E(H, n, k) 2n(n + 1)(2H + 1)n H n11/k ,

and the sum H 1 (E(H, n, k)) converges. It then follows from the Borel
Cantelli Lemma 1.2 that (E(n, k)) = 0, where E(n, k) denotes the set of real
numbers such that the inequality
| | < H n11/k
has innitely many solutions in real algebraic numbers of height at most H
and of degree at most n. Since any S -number whose -type is strictly greater
52
than 1, any T -number and any U -number belong to some set E(n, k), we
conclude that almost all real numbers are S -numbers of -type less than or
equal to 1.
3.4 Comparison between both classications

In this Section, we use a result of Wirsing [598] on the approximation of real
numbers by algebraic numbers of bounded degree to show that the classications of Mahler and Koksma are essentially equivalent (rst assertion of Theorem 3.6). Actually, this fact had been established before Wirsings result, but
with a slightly more complicated proof, by Koksma [333] in the paper where
he introduced its classication.
P ROPOSITION 3.2. For any positive integer n and any real number , we have
wn ( ) wn ( ).
P ROOF. We may assume that wn ( ) is positive. Let w be real with 0 < w <
wn ( ). By the denition of wn ( ), there exist innitely many non-zero alge
braic numbers of degree at most n such that | | H()1w . Denoting
by P (X ) the minimal polynomial over Z of such an , Rolles Theorem implies that
|P ( )| | |
max
t[ 1, +1]
|P (t)| c4 H(P ) H()1w ,
for some positive constant c4 depending only on and n. Since H(P ) = H(),
this shows that wn ( ) w . Letting w tend to wn ( ), we get wn ( ) wn ( ),
as claimed.
One may ask whether there exist real numbers such that wn ( ) and wn ( )
are different. The answer is positive, and this will be proved in Chapter 7.
While it is easy to bound wn ( ) from below in terms of wn ( ), it is highly
non-trivial to estimate wn ( ) from below in terms of wn ( ). Such a result
was rst obtained by Wirsing [598] in 1961 (and, later and independently,
by Sprindzuk). He established various lower bounds for wn ( ), when is a
transcendental real number or a real algebraic number of degree at least n + 1.
T HEOREM 3.4. Let n be a positive integer and be a real number which is
not algebraic of degree at most n. Then we have
wn ( ) wn ( ) n + 1,
wn ( ) + 1
,
wn ( )
2
wn ( )
wn ( )
,
wn ( ) n + 1
(3.11)
(3.12)
(3.13)
53
and
n 2 + 16n 8
n
.
(3.14)
+
4
4
Inequality (3.11), under a hidden form, is due to Schneider [517], Hilfssatz 19.
The lower bound (3.13)
and a result slightly weaker than (3.14) (namely, the estimate wn ( ) (n+2+ n 2 + 4n 4)/4, which follows by combining (3.12),
(3.13), and Proposition 3.1) were obtained by Wirsing [598], while (3.14) is a
result of Bernik and Tishchenko [95]. The proof of (3.13) given below is not
the original one of Wirsing and can be found in [122].
We display an immediate consequence of (3.13) and Proposition 3.2.
wn ( )
C OROLLARY 3.2. Let n be a positive integer and be a real number which is

not algebraic of degree at most n. If wn ( ) = n, we then have wn ( ) = n.
We are now able to show how the second assertion of Theorem 2.7 and Theorem 2.8 follow from the rst assertion of Theorem 2.7.
T HEOREM 3.5. The A -numbers are exactly the algebraic numbers. Let be
a real algebraic number of degree d and let n be a positive integer. We then
have wn ( ) = min{n, d 1}.
P ROOF. Inequality (3.12) of Theorem 3.4 combined with Proposition 3.1
shows that w ( ) 1/2 holds for any real transcendental number . Conversely, let be a real algebraic number of degree d and n be a positive integer. We infer from Corollary A.2 that wn ( ) d 1, hence we get that
w ( ) = 0, and the rst assertion of the theorem is proved. Furthermore, we
observe that for n at most equal to d 1, Theorem 3.1 and Corollary 3.2 imply
that wn ( ) = n = min{n, d 1}. In particular, for any n at least equal to d 1,
( ) = d 1, hence w ( ) = d 1, by Corollary A.2.
we have wn ( ) wd1
n
This completes the proof of the theorem.
It easily follows from Proposition 3.2 and Theorem 3.4 that the classications
of Mahler and of Koksma are equivalent.
T HEOREM 3.6. The classications of Mahler and of Koksma coincide, in the
sense that any S-number (resp. T -number, U -number) is an S -number (resp.
T -number, U -number). Further, if is an S-number, we then have t ( )
t ( ) t ( ) + 1. If is either a T -number or a U -number, then its type is
equal to its -type.
P ROOF. For a real transcendental number , we infer from (3.11), (3.12), and
Proposition 3.2 that the inequalities
wn ( ) wn ( ) min{n + wn ( ), 2wn ( )}
(3.15)
54
hold for any positive integer n. Consequently, wn ( ) and wn ( ) are simultaneously nite or innite and w ( ) w( ) min{w ( ) + 1, 2w ( )}. Thus,
we have t ( ) t ( ) t ( ) + 1 if is an S-number. If is a T -number,
then wn ( ) n for innitely many n and we infer from (3.15) that
lim sup
n+
log wn ( )
log wn ( )
log 2wn ( )
lim sup
lim sup
,
log n
log n
log n
n+
n+
which implies that t ( ) = t ( ).

Koksma [333], Satz 11, proved the rst assertion of Theorem 3.6 without using
Theorem 3.4 (see also Schneider [517]). Furthermore, he showed (Satz 10) that
any S-number is an S -number with t ( ) t ( ) t ( ) + 2.
We now turn to the proof of Theorem 3.4. Clearly, the functions w1 and w1
coincide. Let n 2 be an integer and let be a real number which is not algebraic of degree at most n. We rst prove (3.11), and then inequalities (3.12),
(3.14), and (3.13).
Proof of the lower bound (3.11).
Let w be a real number with w < wn ( ). We infer from the denition of wn ( )
and Lemma A.3 that there exist innitely many irreducible integer polynomials
P(X ) of degree at most n satisfying
0 < |P( )| H(P)w .
By Lemma A.8, such a polynomial P(X ) has a root with
| | (2n)n |P( )| H(P)n2 .
This implies the lower bound wnc ( ) w n + 1. We conclude by noticing
that w can be taken arbitrarily close to wn ( ) and by applying Lemma 3.1.
Without any loss of generality, we may assume in the sequel of the proof of
Theorem 3.4 that 0 < < 1/10 holds and, by (3.11), that wn ( ) is nite. In all
what follows (apart from Lemma 3.2 and its proof), the notation a b means
that there exists a real number , depending only on n, such that a b.
Let be a real number with 0 < < 1/2. We infer from the denition of
wn ( ) and Lemma A.3 that there exist innitely many irreducible, primitive,
integer polynomials P(X ) of degree at most n satisfying
0 < |P( )| H(P)wn ( )+ .
Let P(X ) be such a polynomial. If R(X ) is an integer polynomial of degree at
55
most n which is a multiple of P(X ), then, again by Lemma A.3, there exists a
positive constant c5 , depending only on n, with c5 < 1 and H(R) 2c5 H(P).
By Theorem B.2, the system of inequalities
|bn n + . . . + b0 | c5n H(P)n
|b1 |, . . . , |bn | c5 H(P)
has a non-zero integer solution (b0 , . . . , bn ). Set Q(X ) = bn X n + . . . + b1 X +
b0 . If H(P) 2c51 , it follows from the assumption 0 < < 1/10 that H(Q)
is at most equal to c5 H(P). Consequently, by our choice of c5 , the polynomials
P(X ) and Q(X ) have no common factor.
Hence, one can build two sequences (Pk )k1 and (Q k )k1 of non-zero integer polynomials of degree at most n, such that the height of Pk (X ) tends to
innity with k,
|Pk ( )| H(Pk )wn ( )+ , H(Q k )
H(Pk ),
|Q k ( )|
H(Pk )n (k 1),
(3.16)
and
Pk (X ) and Q k (X ) are coprime (k 1).
To conclude, we need an auxiliary lemma, due to Wirsing [598]. Notice that
in Lemma 3.2 below and in its proof, the constant implied in
depends only
on t.
L EMMA 3.2. Let t 2 be an integer and let P(X ) and Q(X ) be coprime
polynomials with integer coefcients of degrees less than or equal to t. Let
be a real number with | | 1 and which is not algebraic of degree less than
or equal to t. Assume that there exist a root of P(X ) and a root of Q(X ) in the
open disc centered at of radius 1. Then, we either have
1
max{P 2 ( ) H(P)t2 H(Q)t , Q 2 ( ) H(Q)t2 H(P)t },
(3.17)
or there exists a real root of the polynomial P Q(X ) such that one of the
following four cases holds:
| |
|P( )| H(P)1 ,
1
| |
|Q( )| H(Q)
| |
P ( ) |Q( )| H(P)
2
t2
| |
|P( )| Q ( ) H(P)
2
t1
(3.18)
(3.19)
H(Q)
t1
H(Q)
t2
(3.20)
(3.21)
56
P ROOF. We denote by 1 , . . . , m the roots of P(X ) and by 1 , . . . , n those

of Q(X ), numbered in such a way that, if pi := |i | and q j := | j | for
i = 1, . . . , m and j = 1, . . . , n, we have p1 . . . pm and q1 . . . qn .
Without loosing any generality, we may assume p1 q1 < 1. Denote by am
the leading coefcient of P(X ) and by bn that of Q(X ). Corollary A.1 applied
with = 1 gives

|P( )|
H(P)
min{1, pi }
|P( )|
(3.22)
1im
and

|Q( )|
H(Q)
min{1, q j }
|Q( )|.
(3.23)
1 jn
If m = 1 or if m > 1 and p2 > 1 (resp. if n = 1 or if n > 1 and q2 > 1), then

1 (resp. 1 ) is real and (3.22) (resp. (3.23)) implies that we are in the case
(3.18) (resp. (3.19)). Thus, we can assume that m 2, n 2, and p2 1,
q2 1. Denoting by R the resultant (see Denition A.1) of the polynomials
P(X ) and Q(X ), we have

1 |R| = |am |n |bn |m
|i j |
|am bn |t
1im
1 jn
max{ pi , q j } =: AB,
1im
1 jn
where
A=
max{ pi , q j }
i: pi 1 j:q j 1
and
B |am bn |t

max{1, pi } max{1, q j }
H(P)t H(Q)t .
1im
1 jn
If p2 q1 , we have p1 p2 q1 1, whence

q 2j
Q 2 ( )H(Q)2 ,
A
j:q j 1
by (3.23). This gives (3.17).
(3.24)
57
Thus, we can assume that q1 < p2 . If moreover q2 is such that p1 q1 <

p2 q2 , then we get

2
pi
qj .
A
q1
i2: pi 1
j2:q j 1
By multiplying (3.24) by p1 q1 , we infer from (3.22) that

2
2
pi
q j H(P)t H(Q)t
p1 p1 q 1
i: pi 1
j:q j 1
|P( )| Q ( ) H(P)t1 H(Q)t2 ,

2
which gives (3.20). In the same way, if q2 is such that p1 q1 q2 < p2 ,

one ends up with (3.21), and the proof of the lemma is complete. Further, 1 is
real since p1 < p2 .
Let k be sufciently large such that the polynomials Pk (X ) and Q k (X ) have
a root in the open disc centered at of radius 1, and apply Lemma 3.2 to the
pairs of polynomials (Pk , Q k ). By (3.16), we get
max{Pk2 ( ) H(Pk )n2 H(Q k )n , Q 2k ( ) H(Q k )n2 H(Pk )n }
H(Pk )1 ,
since wn ( ) n 1/2. Thus, (3.17) cannot hold for k large enough. If
for innitely many k we are in one of the cases (3.18) or (3.19), we then get
wn ( ) n. Consequently, we may assume that for k large enough, we are in
one of the cases (3.20) or (3.21). This implies that there exists a real root k of
the polynomial Pk Q k (X ) such that
| k |2
H(Pk )2wn ( )+2+n3 or | k |2
H(Pk )wn ( )+3 .
Since H(k )
H(Pk ) and can be taken arbitrarily small, we get the lower
bound

n 1 wn ( ) + 1
wn ( ) + 1
,
min n,
,
wn ( ) min n, wn ( )
2
2
2
where we use that wn ( ) n. Consequently, (3.12) holds when wn ( )
2n 1. If wn ( ) > 2n 1, then (3.12) follows from (3.11).
Beginning of the proofs of (3.14) and (3.13).
Let A and H A be positive real numbers with 2 < A < n + 1 and such that, for
any algebraic number of degree at most n and of height at least equal to H A ,
we have
| | H()A .
(3.25)
58
Let > 0 and H 2 be real numbers. By Minkowskis Theorem B.2, there

exists a non-zero integer polynomial P(X ) := an X n + an1 X n1 + . . . +
a1 X + a0 such that
|P( )| H n
|a1 | H 1+
(3.26)
|a2 |, . . . , |an | H.
If |a1 | H , we then have H(P) H , otherwise we get H(P) = |a1 |, since
our assumption on implies that |a0 | < H(P).
The idea of Bernik and Tishchenko [95] is to consider the quantity |P ( )|.
When |P ( )| is not too small, Lemma A.5 asserts that P(X ) has a root very
close to . If H(P) = |a1 |, we infer from 0 < < 1/10 that
|P ( )| = |nan n1 + . . . + 2a2 + a1 |
|a1 | |nan n1 + . . . + 2a2 |
H(P)
|a1 |
=
.
2
2
Recall that is not algebraic of degree at most n. By (3.25), Lemma A.3, and
Lemma A.5, if H is large enough, we then get
|P( )| | | |P ( )| H(P)1A ,
(3.27)
where denotes a root of P(X ) with | | minimal. Further, by (3.26), we

have
|P( )| H(P)(n+)/(1+) ,
which, combined with (3.27), yields
c6
n+1 A
A2
log 3H(P)
for some positive constant c6 depending only on n. Thus, if is sufciently

small, then the height of P(X ) is strictly less than |a1 | and satises H(P) =
max{|a2 |, . . . , |an |}. In other words, under the hypothesis
<
n+1 A
,
A2
(3.28)
and provided that H is large enough, we have constructed a non-zero integer

polynomial P(X ) satisfying H(P) H and |P( )| H n , hence such
that
|P( )| H(P)n .
(3.29)
59
We will exploit this remark in two different ways.

Let be a positive real number satisfying (3.28). Using Theorem B.2 as explained above with arbitrarily large values of H in (3.26), we get that there
exist innitely many integer polynomials P(X ) of degree at most n satisfying (3.29). Furthermore, we infer from Lemma A.3 that there are innitely
many integer polynomials P(X ), irreducible over Q and primitive, such that
|P( )|
H(P)n . Let P(X ) be such a polynomial and consider the system
of inequalities
|bn n + . . . + b0 | c5n H(P)n
|b1 | c5 H(P)1+
|b2 |, . . . , |bn | c5 H(P),
where c5 is the same constant as in the beginning of the proof of (3.12). By
Theorem B.2, it has a non-zero integer solution (b0 , . . . , bn ). Set Q(X ) =
bn X n + . . . + b1 X + b0 . Arguing as above, we infer from the assumption
on that H(Q) c5 H(P) if H(P) is sufciently large. Consequently, our
choice of c5 implies that the polynomials P(X ) and Q(X ) have no common
factor.
Hence, one can build two sequences (Pk )k1 and (Q k )k1 of non-zero integer polynomials of degree at most n, such that the height of Pk (X ) tends to
innity with k,
|Pk ( )| H(Pk )n , H(Q k )
H(Pk ),
|Q k ( )|
H(Pk )n (k 1),
(3.30)
and
Pk (X ) and Q k (X ) are coprime (k 1).
As in the proof of (3.12), we apply Lemma 3.2 to the pairs of polynomials
(Pk , Q k ). We infer from (3.30) that (3.17) cannot hold for k large enough.
If for innitely many k we are in one of the cases (3.18) or (3.19), we then
get wn ( ) n and (3.14) holds. Otherwise, we may assume that for k large
enough, we are in one of the cases (3.20) or (3.21). This means that there exists
a real root k of the polynomial Pk Q k (X ) such that
H(k )2 A
| k |2
H(k )3n3 .
(3.31)
Letting tend to the upper bound (n +1 A)/(A2) given in (3.28), it follows
60
from (3.31) that A satises the inequality

2 A2 (n + 4) A + (3 n) 0,
whence
n+4
n 2 + 16n 8
A
+
.
4
4
The claimed result is proved since wn ( ) A 1.

Since the method allows us to construct innitely many polynomials satisfying
(3.29), we get wn ( ) n + . Letting tend to the upper bound (n + 1
A)/(A 2) given in (3.28), we obtain
n+1 A
wn ( ) n +
,
A2
hence,
2 wn ( ) + 1 n
A
,
wn ( ) + 1 n
and (3.13) follows from Lemma 3.1 and wnc ( ) A 1.
For approximation by real quadratic numbers, Wirsings statement has been
improved upon by Davenport and Schmidt [180], who have obtained the following result, best possible up to the value of the numerical constant.
T HEOREM 3.7. For any real number which is neither rational nor
quadratic, and for any real number c7 greater than 160/9, there exist innitely
many rational or quadratic real numbers satisfying
| | c7 max{1, | |2 } H()3 .
Theorem 3.7 is proved in [180] and also in [512]. It has been extended by
Davenport and Schmidt [181] (up to the value of the numerical constant) as
follows.
T HEOREM 3.8. Let n 2 be an integer and let be a real number which is not
algebraic of degree at most n. Then there exists an effectively computable constant c8 , depending only on and on n, an integer d with 1 d n 1, and
innitely many integer polynomials P(X ) of degree n whose roots 1 , . . . , n
can be numbered in such a way that
|( 1 ) . . . ( d )| c8 H(P)n1 .
Tishchenko [560, 561, 564, 565] obtained slight sharpenings of (3.14) for small
values of n. Furthermore, he proved [563] that there exists a sequence (n )n1
3.5 Some examples
61
of positive real numbers, tending to 3, such that, for any real transcendental
number and any positive integer n, we have wn ( ) n/2 + n . This remains somehow far away from a celebrated conjecture of Wirsing, saying that
wn ( ) n holds for any positive integer n and any real transcendental number
. A heuristic support for this conjecture is the fact that there are integer polynomials P(X ) with arbitrarily large height such that |P( )|
H(P)n . Here
and below, the constant implied by
only depends on and n. By Lemma
A.5, such a polynomial has a root satisfying | |
|P( )/P ( )|. Furthermore, if P(X ) has no other root very close to , the order of magnitude of
|P ( )| is H(P) and we get that | |
H(P)n1 . A difculty arises since
we cannot exclude the case when all but nitely many such polynomials have
two distinct roots very close to .
The plausibility of Wirsings Conjecture and (3.11) motivate the following
problem.
M AIN P ROBLEM . Let (wn )n 1 and (wn )n 1 be two non-decreasing sequences in [1, +] such that
n wn wn wn + n 1, for any n 1.
Then there exists a real transcendental number such that
wn ( ) = wn
and wn ( ) = wn for any n 1.
There is no evidence against (or for!) the Main Problem, which, in other words,
claims that all that is not trivially impossible may occur. It will be the l rouge
of the next Chapters. A summary of the results obtained towards its resolution
is given in Section 7.8.
3.5 Some examples

In this Section, we discuss where some classical numbers, like , e, and log
for a non-zero algebraic number , are located in Mahlers classication.
Popken [463] proved that for any positive integer n there exists a positive
constant c9 (n), depending only on n, such that
|P(e)| H nc9 (n)/(log log H )
(3.32)
for any integer polynomial of degree n and height H sufciently large. It immediately follows that we have wn (e) = n for any positive integer n, thus e
is an S-number of type 1 and, by Corollary 3.2, an S -number of -type 1.
Popkens result was rened by Mahler [376], who showed that we can take
c9 (n) = c10 n 2 log n in (3.32), for some suitable positive, absolute, constant
62
c10 . In the same paper, Mahler proved that e is an S-number for any non-zero
algebraic number , whose type can be explictly bounded (see, for example,
Diaz [187]).
In [384, 385], Mahler showed that and log are not U -numbers, where
= 1 is a non-zero algebraic number. Subsequently, A. Baker [40] (see also
Sorokin [528]) established that for any positive integer n and any positive real
number , we have wn (log(a/b)) n + if a and b are positive integers,
sufciently large in terms of n, , and |a b|.
Further results can be found in the doctoral dissertation of Cijsouw [166] or
in the books of Feldman and Nesterenko [244] (Chapter 2) and of Chudnovsky
[164] (Chapter 2). For example, if and are algebraic numbers with = 0, 1
and irrational, then is either an S-number or a T -number of type at most
2.
3.6 Exponents of Diophantine approximation

After having dened the functions wn and wn in Sections 3.1 and 3.3, we introduce below four other exponents of Diophantine approximation. We survey
known results on these and discuss briey some related questions.
Davenport and Schmidt [182] investigated the approximation of real numbers by algebraic integers of bounded degree. This is an inhomogeneous problem which needs alternative methods, since, for instance, Theorem B.2 is useless in this context. Their approach yields the following result which establishes a link between simultaneous rational approximation and approximation
by algebraic numbers of bounded degree.
P ROPOSITION 3.3. Let n 2 be an integer and let be a real number. Assume
that there exist positive real numbers and w and arbitrarily large values of
X for which the inequalities
|x0 | X,
max |x0 m xm | X 1/w
1m n
(3.33)
have no solution in integers x0 , . . . , xn not all 0. We then have wn ( ) w.

P ROOF. We may assume that < 1. Let X be one of the large numbers specied in the statement of the proposition and set Y = X (w+1)/(wn+w) . Let C(Y )
be the parallelepiped dened by
|x0 | Y n , |x0 m xm | Y 1 , 1 m n.
The rst of its successive minima, denoted by 1 (Y ), satises
1 (Y ) Y (nw)/(w+1) ,
(3.34)
63
since otherwise there would be a non-zero solution to the system (3.33).

The polar body C (Y ) of C(Y ) is dened by
|xn n + . . . + x1 + x0 | Y n , |xm | Y, 1 m n,
and it follows from (3.34) and Theorem B.4 that the (n + 1)-th minimum
n+1 (Y ) of C (Y ) satises
n+1 (Y ) c11 Y (nw)/(w+1) ,
for some positive constant c11 depending only on and n. Thus, there exist
( j)
n + 1 linearly independent integer polynomials P j (X ) := xn X n + . . . +
( j)
( j)
x1 X + x0 , of degree at most n, with
|P j ( )| c11 Y w(n+1)/(w+1)
and H(P j ) c11 Y (n+1)/(w+1) ,

( j)
for any j = 1, . . . , n+1 and, by Theorem B.3, such that | det(xi )| (n+1)!.
Arguing then as in the proof of Theorem 2.11, we use the Eisenstein Criterion
to construct an irreducible integer polynomial P(X ) of degree n by taking a
suitable linear combination of P1 (X ), . . . , Pn+1 (X ) in such a way that P(X )
has a real root with
| | c12 H()1w ,
where c12 only depends on and n. This implies that wn ( ) w, as claimed.
Proposition 3.3 invites us to introduce the functions wn and w n .
D EFINITION 3.9. Let n be a positive integer and let be a real number. We
denote by wn ( ) the inmum of the real numbers w such that there are arbitrarily large values of X for which the inequalities
|x0 | X, 0 < max |x0 m xm | X 1/w
1mn
have a solution in integers x0 , . . . , xn . Further, we denote by w n ( ) the inmum of the real numbers w such that for all sufciently large values of X the
inequalities
|x0 | X, 0 < max |x0 m xm | X 1/w
1m n
have a solution in integers x0 , . . . , xn .
(3.35)
64
It is readily veried that w n ( ) is the supremum of the real numbers w such

that there are arbitrarily large values of X for which the inequalities (3.35) have
no solution in integers x0 , . . . , xn .

Any real number satises w1 ( ) = 1/w1 ( ) and wn ( ) w n ( ). For any
positive integer n and any real algebraic number of degree d 2, we have

wn ( ) = w n ( ) = min{n, d 1}. Furthermore, it follows from Theorem 1.1

and Dirichlets Schubfachprinzip, that max{1, wn ( )} wn ( ) n holds for
any integer n 2 and any irrational real number .
Not surprisingly, wn ( ) and wn ( ) are related to the simultaneous rational
approximation of , 2 , . . . , n .
T HEOREM 3.9. For any positive integer n and any real number not algebraic of degree at most n, we have
n
(n 1)wn ( ) + n
wn ( )
n
wn ( ) n + 1
wn ( )
and wn ( ) w n ( ). In particular, we have wn ( ) > n if, and only if, wn ( ) <
n holds.
The rst assertion of Theorem 3.9 is a direct consequence of Khintchines
Transference Theorem B.5, while the second one follows from Proposition
3.3.
The method of Davenport and Schmidt [182] offers much more exibility
than the approach Wirsing [598] used to prove Theorem 3.4 above, insofar as
we may impose several constraints on the algebraic approximants, for example to be algebraic integers [182, 140] (see also the proof of Theorem 2.11) or
algebraic units [549]. Let n 2 be an integer and let be a real transcendental
number. Applying a result of Davenport and Schmidt [182] which asserts that
the assumptions of Proposition 3.3 are satised with w = [n/2] and a suitable
positive constant , Bugeaud and Teulie [140, 549] deduced from Proposition
3.3 the existence of innitely many real algebraic numbers of degree n satisfying | | c13 H()1[n/2] , with a suitable constant c13 depending only
on and n. Laurent [354] slightly improved upon the result of Davenport and
Schmidt for odd values of n. To state a consequence of his result, we introduce
the following notation: for any positive real number x, set x = x when x is
an integer and x = [x] + 1, otherwise.
T HEOREM 3.10. Let n be a positive integer and let be a real number which
is not an algebraic number of degree less than or equal to n/2. We then have
w n ( ) n/2 and there exists a constant c14 , depending only on and n,
innitely many real algebraic numbers of degree n, and innitely many real
65
algebraic integers of degree n + 1 such that

| | c14 H()1n/2 .
Actually, Laurents result for n = 2 is weaker than the following one, due to
Davenport and Schmidt [182].
rational nor a
T HEOREM 3.11. Let be a real number which is neither
quadratic irrationality. We then have w 2 ( ) (1 + 5)/2 and there exist

a constant c15 , depending only on , innitely many real quadratic numbers ,
and innitely many real algebraic integers of degree 3 such that
| | c15 H()(3+
5)/2
The exponent of H() obtained in Theorem 3.10 yields a slightly weaker lower
bound for wn ( ) than (3.14), but we know the exact degree of the approximants. Furthermore, Theorem 3.7 does not imply the rst assertion of Theorem
3.11.
After having dened wn , wn , wn , and w n , we introduce a fth function w n ,
which appeared for the rst time in [182].
D EFINITION 3.10. Let n be a positive integer and let be a real number.
We denote by w n ( ) the supremum of the real numbers w such that for all
sufciently large values of X the inequalities
0 < |xn n + . . . + x1 + x0 | X w , |xm | X, 1 m n,
(3.36)
have a solution in integers x0 , . . . , xn .

It is readily veried that w n ( ) is the inmum of the real numbers w such that
there are arbitrarily large values of X for which the inequalities (3.36) have no
solution in integers x0 , . . . , xn .
For any positive integer n and any real algebraic number of degree d, we
have w n ( ) = min{n, d 1}, by Theorème 1 of [123]. Assume now that n is
a positive integer and is a real number not algebraic of degree at most n.
It follows from Dirichlets Schubfachprinzip that we have w n ( ) n, and
Davenport and Schmidt [182] (see also Theorem 4.2 of [484]) proved that
w n ( ) 2n 1 holds. Furthermore, the proof of Khintchines Transference
Theorem B.5 shows that the values w n ( ) and w n ( ) are related by
(n 1)w n ( ) + n
n
w n ( )
n.
w n ( ) n + 1
w n ( )
(3.37)
Davenport and Schmidt [183] (for n = 2), R. C. Baker [49] (for n = 3), and
66
Bugeaud [123, 125] (for general n) proved that w n ( ) = n and w n ( ) = n

(this is a direct consequence of (3.37)) hold for almost all real numbers .
If w < n, the volume of the convex body dened by (3.33) tends to zero
as X tends to innity. Thus, it is tempting to conjecture that for any integer
n 2 and any real transcendental number the assumptions of Proposition
3.3 are satised with a suitable positive constant and w = n, or, at least, that
w n ( ) = n holds. This is however surprisingly not true, as was proved by Roy
[479, 480]. Let {a, b} denote the monoid of words on the alphabet {a, b} for
the product given by the concatenation. The Fibonacci sequence in {a, b} is
the sequence of words ( f i )i0 dened recursively by
f 0 = b, f 1 = a, and
f i = f i1 f i2 (i 2).
Since, for every i 1, the word f i is a prex of f i+1 , this sequence converges
to an innite word f = abaabab . . . called the Fibonacci word on {a, b}. For
two positive distinct integers a and b, let a,b = [0; a, b, a, a, b, a, . . . ] be the
real number whose sequence of partial quotients is given by the letters of the
Fibonacci word on {a, b}, and, for any positive integer m, set
m := (m + 1 + m,m+2 )1 = [0; m + 1, m, m + 2, m, m, m + 2, . . . ].
Let denote the Golden Section (1 + 5)/2. Roy showed that m is transcendental and that we have w 2 (m ) = and w 2 (m ) 2 . More precisely, there
exists a positive constant c16 , depending only on m, such that, for any X 1,
the inequalities
0 < x0 X, |x0 m x1 | c16 X 1/
and |x0 m2 x2 | c16 X 1/
have a non-zero integer solution (x0 , x1 , x2 ). If c16 is small enough, the same
conclusion also holds true for the inequalities
|x1 | X, |x2 | X, |x2 m2 + x1 m + x0 | c16 X .
2
Furthermore, Roy [480] studied the approximation of m by real quadratic

numbers and proved the existence of positive constants c17 and c18 , depending
2
only on m, such that |m | c17 H()2 has innitely many solutions
2
in algebraic numbers of degree 2, while |m | c18 H()2 has no
solution in quadratic irrationals . In addition, he established [481, 482] that
there exists a positive constant c19 , depending only on m, such that
|m | c19 H() 1 ,
67
for any algebraic integer of degree at most 3, and

|P(m )| c19 H(P) ,
for any monic, integer polynomial P(X ) of degree at most 3. This proves that
the last assertion of Theorem 3.11 is the best possible (there is a gap in [567])
and disproves the conjecture claiming that, for any real transcendental number
, for any integer n 2, and any positive real number , there exist innitely
many algebraic integers of degree at most n such that | | H()n+ .
This conjecture was supported by a metric statement saying that almost all real
numbers share this approximation property ([140], Corollaire 1). Moreover,
Arbour and Roy [32] proved that w 2 ( ) 2 holds for any real number not
algebraic of degree at most 2. Actually, they showed that if the real number
is such that for any sufciently large positive real number H there exists a
non-zero integer polynomial P(X ) of degree at most 2 and height at most H
2
with |P( )| H /4, then is algebraic of degree at most 2. It is worth
noting that
1+ 5
3+ 5

= w 2 (m ) <
= w 2 (m )
1 = w2 (m ) <
2
2
< w2 (m ) = 2 + 5,
holds, while we know that the functions w2 , w 2 , w 2 , and w2 take the value 2
for almost all real numbers.
To conclude this Section, we dene another function, introduced by
Bugeaud and Laurent [137] and motivated by Dirichlets Theorem 1.1.
D EFINITION 3.11. Let n be a positive integer and let be a real number.
We denote by w n ( ) the supremum of the real numbers w such that for all
sufciently large values of X there exists a real algebraic number of degree
at most n with
0 < | | H()1 X w and H() X.
(3.38)
The discussion at the beginning of Section 3.3 and Dirichlets Theorem 1.1
invite us to consider (3.38) rather than the system of inequalities | |
X w1 , H() X .
For any positive integer n and any real algebraic number of degree d, we
have w n ( ) = min{n, d 1}, by Theorème 1 of [123]. Dirichlets Theorem
1.1 obviously implies that w n ( ) 1 holds for any irrational number and
Corollary A.2 yields the upper bound w n ( ) 2n 1. Furthermore, Corollaire
1 of Bugeaud [123] asserts that we have w n ( ) = n for all positive integers n
and almost all real numbers . Notice that the numbers m dened above satisfy
68
w 2 (m ) = (3 + 5)/2. Other examples of real numbers with w 2 ( ) > 2

and w 2 ( ) < 2 have been given by Bugeaud and Laurent [137]. They further
improved (3.13) by showing that
w n ( )
wn ( )
wn ( ) n + 1
and wn ( )
w n ( )
w n ( ) n + 1
(3.39)
hold for any positive integer n and any real number not algebraic of degree
at most n (see Exercises 3.4 and 3.5).
We conclude by some consideration on inhomogeneous Diophantine approximation. Let n be a positive integer and be a real number. For any real
number , let wn (, ) denote the supremum of the real numbers w for which
there exist innitely many integer polynomials P(X ) of degree at most n satisfying
0 < |P( ) + | H(P)w .
Dene further the n-th exponent of uniform inhomogeneous approximation
wninh by
wninh ( ) = inf wn (, ).
R
Bugeaud and Laurent [138] proved that w n ( ) = wninh ( ) for any positive
integer n and any irrational real number . It would be of interest to study
more closely the exponents wn (, ).
The questions evoked in this Section yield several open problems listed in
Chapter 10.
3.7 Exercises
E XERCISE 3.1. Let a, b, c, and d be rational integers with ad bc = 0
and let be a real number. Let n be a positive integer. Prove that we have
wn ( ) = wn ((a + b)/(c + d)) and wn ( ) = wn ((a + b)/(c + d)). For a
xed real algebraic number , compare wn ( + ) with wn ( ) and wn ( + )
with wn ( ).
E XERCISE 3.2. Let = [a0 ; a1 , a2 , . . . ] be an irrational real number with
convergents p0 /q0 , p1 /q1 , . . . Following Guting [267], use Theorems 1.3 and
1.7 to prove that we have
log an
w1 ( ) = w1 ( ) = 1 + lim sup
,
n+ log qn1
if this lim sup is nite, and w1 ( ) = w1 ( ) = + otherwise.
3.7 Exercises
69
Deduce that w1 ( ) = 1 if the sequence (log n a1 a2 . . . an )n1 converges

(Khintchine [321, 323] proved that this sequence tends to
j1 {log( j ( j +
2) + 1) log( j ( j + 2))}(log j)/ log 2 for almost all real numbers ).
E XERCISE 3.3. Use Bombieri and Muellers method described in Theorem
2.10 to get the lower bound wn ( ) n/(wn ( ) n + 1), for any integer n 1
and any real number which is not algebraic of degree at most n. Prove the
same result by means of the approach of Theorem 2.11.
E XERCISE 3.4. Use the method of Wirsing [598] to prove (3.13) and Theorem
2.10 (up to the numerical constants).
Hint. Let n 2 be an integer and let be a real number which is not
algebraic of degree at most n. Let > 0 and set w = wn ( ) (1 + )2 . Use
Theorem B.2 to prove that there exists a positive constant c20 , depending only
on and n, and, for any positive H , an integer polynomial P(X ) such that
|P( )| H w , |P(1)|, . . . , |P(n 1)| H and
|P(n)| c20 H wn+1 .
Show that H(P) c21 H 1+ for a suitable constant c21 , depending only on
, n, and , and apply Corollary A.1 to get that, when H is large enough,
the polynomial P(X ) has a single (real) root in each small disc centered at
the points , 1, . . . , n 1; in particular, there exists a suitable constant c22 ,
depending only on , n, and , such that P(X ) has a real root with | |
c22 H()1+w/(wn+1) . Conclude and prove the rst inequality of (3.39).
With the same approach, use Theorem A.1 to get Theorem 2.10 (up to the
numerical constants).
E XERCISE 3.5. Show that the proof of (3.13) given in Section 3.4 actually
yields the second inequality of (3.39).
E XERCISE 3.6. Let and be two algebraically dependent transcendental real
numbers. Let F(X, Y ) be an irreducible polynomial with integer coefcients
vanishing at the point (, ). Denote by M (resp. by N ) the degree of F(X, Y )
in X (resp. in Y ). Following Schmidt [508], prove that the inequality

wn ( ) + 1 M wn N () + 1
holds for any positive integer n.
Hint. Use the implicit function theorem to show that there exists a neighbourhood of and a one-to-one differentiable function y dened on this
70
neighbourhood such that y( ) = and F(x, y(x)) is identically zero. Let

H 1 be a real number. Assume that n is an algebraic number of degree
at most n and height at most H such that wn (, H ) = | n | and denote by
(1)
(n)
Q(X ) := an (X n ) . . . (X n ) its minimal polynomial over Z, with
(1)
n = n . Observe that | y(n )|
wn (, H ). Here, and in the sequel of
the proof, the constant implied in
depends only on and on . Since y(n )
is a root of the polynomial
R(Y ) := anM F(n(1) , Y ) . . . F(n(n) , Y ),
which is not identically zero, y(n ) is algebraic of degree less than or equal
to n N . Use Exercise A.1 and Lemma A.3 to prove that H(y(n ))
H M .
Conclude.
E XERCISE 3.7. Show that for any non-negative real number , any positive
integer k, and any real number w k + 1, there exist real numbers such that
w1 ( ) = k(w + 1) 1 and w1 ( k ) = w + .
Hint. Construct inductively two increasing sequences of prime numbers
(qn )n1 , (vn )n1 and two increasing integer sequences ( pn )n1 , (u n )n1 such
that the intervals
k

pn
1
pnk
1
I2n :=
and
k(w+1) q k
k(w+1)
qnk
qn
2qn
n

un
1
un
1
w+1+ ,
w+1+
I2n+1 :=
vn
vn
vn
2vn
satisfy I2 I3 . . . . Let be the positive real number such that { k } is the
intersection of the intervals In , n 2. Using triangular inequalities, estimate
k
the difference | a/b| for any rational number a/b with qnk < b < qn+1
k
k
(distinguish three cases, depending on the size of |a/b pn /qn |, |a/bu n /vn |,
k /q k |). Prove that if (q )
and |a/b pn+1
n n1 and (vn )n1 do not increase too
n+1
k
rapidly, we then have w1 ( ) = w+ and | k a/b| bw1 for any integers
a and b such that b does not belong to the sequence (vn )n1 .
3.8 Notes
Another presentation of Koksmas classication, based on equivalence
classes, can be found in [508].
Let Q be a set of integer polynomials and be a transcendental real
number. At the end of [311], Kasch and Volkmann introduced the quantities
3.8 Notes
71
wn (, H, Q) dened as the minimum of |P( )| over the set of integer polynomials P(X ) of degrees at most n and heights at most H which belong to Q.
A link between Theorem B.3 and the approximation of real numbers by
algebraic numbers has been discussed by Mahler [401]. Notice that his remark
In the case of (3.29), (. . . ) this property on page 159 is incorrect, since the
results of Cassels [154] and Davenport [175] quoted therein do not deal with
n-tuples of the particular form (a, a 2 , . . . , a n ).
Khintchine [318] proved that the set of real numbers for which there exist a positive integer n and a positive constant such that |P( )| H(P)n
for any integer polynomial of degree at most n is a null set.
Galockin [254] proved that values at some algebraic points of Mahler
functions satisfying certain functional equations are S-numbers. Improvements
and extensions to more general Mahler functions are due to Becker-Landeck
[58], Wass [595], Becker [56], Nishioka [438] (see also [439], pages 137 and
138), Nishioka and Topfer [440], and Topfer [569, 570]. Amou [27] gave explicit upper bounds for the type of S-numbers obtained as values at algebraic
points of functions satisfying certain functional equations of Mahlers type.
Lang [348] established that values taken by E-functions at algebraic
points are S-numbers (see Shidlovskii [522] for further references and results).
Chudnowsky [163] established that if the Weierstrass function has algebraic
invariants and complex multiplication, then () is an S-number for any nonzero algebraic number .
In a very original paper, Duverney [222] showed that the sum of the re
n
ciprocals of the Fermat numbers n0 1/(22 + 1) is either an S-number or a
T -number of type at most 2 (actually, with a slight modication of the proof,
it is possible to show that this sum is an S-number of type at most 4, as kindly
communicated to me by Duverney). His proof starts from an irrationality measure and then works by induction.
A. Baker [37] used Theorem 2.5 and a method of Maillet [403] to construct explicitly families of badly approximable real transcendental numbers.
He further gave [38] a condition under which such numbers are U -numbers
of type 2. His criterion, however in a less general form, has been improved by
Baxa [54] and Mkaouar [425].
A. Baker [38] established that the Champernowne number is not a U number. Further, he proved that the set of badly approximable real transcendental numbers contains T -numbers or S-numbers with w1 ( ) arbitrarily
large. His main tool is a renement of Theorem 2.5, providing a criterion
72
which, when satised by a real number , ensures that is transcendental

but not a U -number. This has been used by Bundschuh [142] who dealt with
a class of real numbers having explicit g-adic and continued fraction expansions. A. Bakers criterion has also been used by several authors, including
Oryan [446], Zeren [611], Yilmaz [606], and Gurses [264], to prove that values of specic power series evaluated at algebraic numbers are transcendental
numbers but not U -numbers. Furthermore, Baxa [53] applied it to show that

Cahens constant n0 (1)n /(Sn 1), where S0 = 2 and Sn+1 = Sn2 Sn + 1
for n 0, is transcendental but is not a U -number. Actually, Topfer (Corollary
8 of [569]) proved that Cahens constant is an S-number.
Adhikari, Saradha, Shorey and Tijdeman [3] (see also Saradha and
Tijdeman [493]) applied A. Bakers theory of linear forms in logarithms of
algebraic numbers with algebraic coefcients to prove that numbers such as

1
n
n0 ((3n + 1)(3n + 2)(3n + 3)) ,
n1 (n)/n,
n1 2 Fn /n, where
is any non-principal Dirichlet character and (Fn )n0 the Fibonacci sequence,
are transcendental but are not U -numbers.
Dress, Elkies, and Luca [208] established a characterization of U -numbers
by simultaneous rational approximation. For any given positive integer n, they
prove that a real number is algebraic of degree at most n + 1 or satises
wn ( ) = + if, and only if, for each positive integer d, there exist a positive
constant Md and innitely many integer (d + 1)-tuples (x0 , . . . xd ) such that
1/n
for each k = 1, 2, . . . , d.
|x0 k xk | < Md x0
Under suitable assumptions on the approximation properties of the T number (or the U -number) and the U -number , Caveny [159] proved that
is transcendental. Caveny and Tubbs [160] established a quantitative version of the particular case of Theorem 3.2 saying that if 1 is a U -number and
1 and 2 are algebraically dependent, then 2 is a U -number.
A. Baker [42] established the existence of pairs of distinct U -numbers
which are badly approximable pairs (see the Notes of Chapter 1 for the denition).
A possible generalization of Wirsings problem in dimension 2 is the following question: does there exist a function : Z1 R>0 such that, for
any pair (1 , 2 ) of real transcendental numbers and any integer n 1, there
are innitely many integers H such that the equations
|1 1 | H (n) , |2 2 | H (n)
have a solution in real algebraic numbers 1 and 2 of degree at most n and
3.8 Notes
73
height at most H ? If 1 or 2 is an S-number, it is proved in [131] that the answer is positive with (n) = c23 n, for some positive constant c23 , depending
only on 1 and 2 . See also Tishchenko [568]. In the opposite direction, for
any integer n 1, Roy and Waldschmidt [484] have given explicit examples
of pairs
(1 , 2 ) of Liouville numbers such that max{|1 1 |, |2 2 |} >
H 3 n holds for all algebraic numbers 1 and 2 of degree at most n and

height at most H , when H is sufciently large.
Kopetzky and Schnitzer [336] applied Theorem 3.7 to the approximation
on the unit circle by points whose coordinates are algebraic of degree at most 2.
Diophantine approximation by conjugate algebraic integers has been studied by Roy and Waldschmidt [484], and by Arbour and Roy [32].
Yu [609] dened a multidimensional generalization of Mahlers classication: all points in Cn (or in Rn ) are classied into 3n + 1 disjoint non-empty
classes, denoted by An , Stn , Ttn and Utn with t = 1, . . . , n, in such a way that
any two algebraic equivalent (over Q) points in Cn fall into the same class.
Mahler and Szekeres [402] investigated approximation of real numbers
by roots of integers.
4
In Theorem 3.3, we used the BorelCantelli Lemma 1.2 to prove that almost
all real numbers are S -numbers of -type less than or equal to 1. A similar statement is however much more difcult to establish when we consider
Mahlers classication. The rst important result in this direction is due to
Mahler [377], who showed in 1932 that almost all real numbers satisfy
supn1 (wn ( )/n) 4. At the end of [377], he made the conjecture that the
upper bound 4 could be replaced by 1.
Until Sprindzuk [536, 537, 538] gave in 1965 a complete afrmative answer to that conjecture, there appeared various improvements of Mahlers result. First, Koksma [333] showed that supn1 (wn ( )/n) 3 for almost all
real numbers . This has been strengthened by LeVeque [362], who replaced
the upper bound 3 in Koksmas result by 2. Later on, rening a method introduced by Kasch and Volkmann [311], Schmidt [501] proved the inequality
wn ( ) 2n 7/3 for almost all real numbers and all positive integers n.
Lastly, Volkmann [583, 584] showed that wn ( ) 4n/3 holds for almost all
real numbers and all positive integers n. At the same time, Sprindzuk [533]
obtained a slightly stronger result than Volkmanns, shortly before his resolution of Mahlers Conjecture. Apart from these general results, it had been
established that wn ( ) = n for almost all real numbers and for 1 n 3.
The case n = 1 is nothing but Khintchines Theorem 1.10, while Kubilsus [344] used Vinogradovs method of exponential sums to solve the case
n = 2. A simpler proof is due to Kasch [310], see Exercise 4.1. Furthermore,
Volkmann [582] applied results of Davenport [176] on binary cubic forms
for the case n = 3.
In the present Chapter, we state Sprindzuks result, its renement obtained shortly thereafter by A. Baker [41], and the subsequent improvement
due to Bernik [81], which appears to be the best possible. The proof of
74
4.1 Statements of the theorems
75
Bernik is lengthy and complicated, and we choose not to quote it, since
it is included in Bernik and Dodson [86]. Instead, we give a full proof of
A. Bakers result, following largely Harmans exposition [273]. The end of
Section 4.1 is devoted to results on multiplicative Diophantine approximation obtained by an approach originated in a work of Kleinbock and Margulis
[330].

The purpose of the present Chapter is to give a full proof of the following
result.
T HEOREM 4.1. Almost all real numbers satisfy wn ( ) = wn ( ) = n for
every positive integer n.
C OROLLARY 4.1. Almost all real numbers are S-numbers of type 1 and
S -numbers of -type 1.
Corollary 4.1 improves upon Theorem 3.3. In Chapter 5, it is proved that the
set of S-numbers of type greater than 1 is not too small, in the sense that its
Hausdorff dimension is equal to 1.
In view of Corollary 3.2, which asserts that wn ( ) = n holds if wn ( ) =
n, Theorem 4.1 is an easy consequence of the following result of Sprindzuk
[538, 539].
T HEOREM 4.2. Let n be a positive integer and be a positive real number.
Then, for almost all real numbers , the equation
|P( )| < H n
(4.1)
has only a nite number of solutions in integer polynomials P(X ) of degree at

most n and of height at most H .
Keeping in mind Khintchines Theorem 1.10, it becomes natural to ask whether
the conclusion of Theorem 4.2 remains true if the functions H H n
occurring in (4.1) are replaced by more general (non-increasing) functions.
This has been conrmed by A. Baker [41] in 1966.
T HEOREM 4.3. Let n be a positive integer and : R1 R>0 be a non
increasing continuous function such that the series h1 (h) converges.
|P( )| < n (H )
(4.2)
76

According to A. Baker [44], it seems likely that the function n (H ) [in (4.2)]
can be replaced by H n+1 (H ), and this conjecture is in fact established for
n 3. A heuristic support for A. Bakers conjecture is the observation that
there are about H n integer polynomials P(X ) of height exactly H and degree
at most n. If one of these satises |P( )| < H n+1 (H ), then it could have
(this is not always true!) a root close to , say (see the discussion after Theorem 3.8) with | |
H n (H ), where, as below, the constant implied
by
depends only on and n. Hence, would lie in the union of about H n
intervals centered at the real roots of integer polynomials of degree at most n
and height H . The Lebesgue measure of this union is approximately equal to
(H ). Assuming that the sum of the (H ) converges, it would then follow
from the BorelCantelli Lemma 1.2 that the set of real numbers for which
|P( )| < H n+1 (H ) is satised for innitely many integer polynomials.
P(X ) is a null set.
The main difculty in this approach is that the estimate | |
n
H (H ) does not always hold. To overcome this problem, very precise
information on the distribution of roots of integer polynomials is needed.
Sprindzuk introduced the notion of essential and inessential domains (see Section 14 of Chapter 2 of [540] for the description of the general idea) to solve
it when (H ) = H 1 for some positive real number . His method was
rened by A. Baker [41], and then by Bernik [78, 81], who, in a lengthy and
intricate paper, managed to get rid of the induction lying at the heart of the
proofs of Theorems 4.2 and 4.3 in order to establish the above-mentioned conjecture of A. Baker.
T HEOREM 4.4. Let n be a positive integer and : R1 R>0 be a non
increasing continuous function such that the series h1 (h) converges.
|P( )| < H n+1 (H )
Apart from the monotonicity condition imposed on , which may presumably
be somewhat relaxed (this is however a difcult problem, keeping in mind
the DufnSchaeffer Conjecture mentioned in Section 1.4), Theorem 4.4 is
best possible, as shown by Beresnevich [61], whose results yield the following
strengthening of Theorem 4.1.
77
T HEOREM 4.5. Let : R1 R>0 be a non-increasing continuous func

tion. If the series h1 (h) diverges, then for almost all real numbers and
any positive integer n the equations
|P( )| < H n+1 (H ) and | | < H n (H )
(4.3)
have innitely many solutions in integer polynomials P(X ) and real algebraic
numbers , of degree at most n and of height at most H , respectively. If the

series h1 (h) converges, then, for almost all real numbers , the equations
(4.3) have only nitely many solutions in integer polynomials P(X ) and real
algebraic numbers , of degree at most n and of height at most H , respectively.
The rst part of Theorem 4.5 is Theorem 3 of [61], quoted in Chapter 6 as
Theorem 6.6. The second part derives from Theorem 4.4, using Lemma A.6.
We end this Section by quoting two deep results on multiplicative approximation. Following a suggestion of A. Baker [44], another way to generalize Theorem 4.2 consists in replacing the height H(P) of a polynomial
P(X ) = an X n + . . . + a1 X + a0 by the function + dened by

+ (P) :=
max{1, |ai |}.
1in
A. Baker conjectured that, for almost all real numbers and for any pos1
(P) has only nitely
itive real number , the equation |P( )| +
many solutions in integer polynomials P(X ) of degree at most n. (Actually, he formulated the dual problem, asking whether for any positive real
number and any positive integer n there exist, for almost all real numbers , only nitely many (n + 1)-tuples (q, p1 , . . . , pn ) of integers such
that q 1+ |q p1 | . . . |q n pn | < 1. A classical transference theorem
[594, 516, 604] shows that both questions are equivalent.) This has been established by Sprindzuk [540] for n = 2 (see also Theorem 2 of Yu [608]),
by Bernik and Borbat [82] for n = 3, 4, and for general n by Kleinbock and
Margulis [330], who developed a rather new approach based on the correspondence between approximation properties of numbers and orbit properties of
certain ows on homogeneous spaces (see, for example, Kleinbocks survey
[326], Margulis survey [405], and Starkovs book [542]).
T HEOREM 4.6. Let n be a positive integer and be a positive real number.
(P)
|P( )| < 1
+
most n.
78
Shortly thereafter, Bernik, Kleinbock, and Margulis [89] improved upon Theorem 4.6.
T HEOREM 4.7. Let n be a positive integer and : R1 R>0 be a nonincreasing continuous function such that

(log h)n1 (h) < +.
h1

|P( )| < + (P)
(4.4)

most n.
Observe that the BorelCantelli Lemma 1.2 implies that for a function as in
Theorem 4.7 the set of real n-tuples (1 , . . . , n ) such that there are innitely
many (n +1)-tuples (a0 , a1 , . . . , an ) of integers with |a0 +a1 1 +. . .+an n | <
n
max{|ai |, 1}) has Lebesgue measure zero. A presumably difcult
( i=1
open question is to prove that Theorem 4.7 is best possible in the sense that

(4.4) has innitely many solutions when the sum h1 (log h)n1 (h) diverges.
4.2 An auxiliary result
The lemma below deals with integer polynomials having two zeros close to
each other. Since it is very specic for the problem considered in the present
Chapter, we have not included it in Appendix A.
L EMMA 4.1. Let P(X ) = a(X 1 ) . . . (X n ) and Q(X ) = b(X
1 ) . . . (X n ) be two distinct, irreducible, integer polynomials of degree
n 3. Set
p = |a n (1 3 ) . . . (1 n )|1/2 and
q = |bn (1 3 ) . . . (1 n )|1/2 .
Assume that
|1 2 | |1 j | and |1 2 | |1 j |
(4.5)
for j = 2, . . . , n, and also that

|1 2 | < p and |1 2 | < q .
(4.6)
Then, there exists a positive constant c1 , depending only on n and
4.2 An auxiliary result
79
max1 jn max{| j |, | j |}, such that

|1 1 | c1 min{ p , q }.
P ROOF. Throughout the present proof, the constants implied by
and
only depend (at most) on n and max1 jn max{| j |, | j |}. We set
:=
|1 1 |
min{ p , q }
and we aim to show that 1. The integer polynomial P(X ) being without
multiple roots, its discriminant (see Denition A.1) is a non-zero integer, and
we thus get

|i j | 1.
(4.7)
|a|n1
1i< jn
Let j be an integer with 3 j n. It follows from (4.7) that

|1 2 | |2 j | ( p )2
(4.8)
holds. By (4.5), we have |2 j | 2|1 j |, which, together with (4.6)

and (4.8), implies that p
|1 j |. Further, using the denition of , we
get
| j 1 | |1 1 | + |1 j | p + |1 j |
(1 + )|1 j |.
Taking the product over all the integers j with 3 j n, we obtain

|a n1 P(1 )|
|1 1 | |2 2 | |a n |
(1 + )|1 j |
3 jn
(1 + )
n2
(p )
|1 1 | |2 1 |,
(4.9)
and, reversing the roles of P(X ) and Q(X ), we get

|bn1 Q(1 )|
(1 + )n2 (q )2 |1 1 | |1 2 |.
(4.10)
Since the integer polynomials P(X ) and Q(X ) have no common roots, their
resultant (see Denition A.1) is a non-zero integer and we have

|ab|n
|i j | 1,
1i, jn
hence,
|ab|n1 |P(1 )| |Q(1 )| |2 2 | |1 1 |1 1.
(4.11)
80
Combining (4.9), (4.10), and (4.11) we get

|1 1 | |1 2 | |2 1 | |2 2 | (1 + )2n+4 ( p q )2 .
(4.12)
Further, (4.6) and the denition of yield that
|2 1 | |1 2 | + |1 1 |
(1 + ) p ,
|1 2 | |1 1 | + |1 2 |
(1 + )q ,
and
|2 2 | |1 2 | + |1 2 |
(1 + ) max{ p , q }.
It follows from these three inequalities and (4.12) that
|1 1 | (1 + )2n+1 min{ p , q },
that is,
(1 + )2n+1 .
Thus, we have proved that 1, as wanted.
4.3 Proof of Theorem 4.3
Difculties encountered and general strategy
Let n be a positive integer and : R1 R>0 be a non-increasing continu
ous function such that the series h1 (h) converges. We begin by explaining the general idea underlying the method of Sprindzuk.
Let P(X ) be an integer polynomial of degree at most n and height at most
H , and let E(P) denote the set of real numbers satisfying |P( )| < n (H ).
The precise description of E(P) is very difcult, hence, in most cases we are

unable to show the convergence of the series
(E(P)), where the summation is over all non-zero integer polynomials of degree at most n. This would
imply that

E(P)
0,
(4.13)
H(P)H0
H0 +
which allows us to conclude by the BorelCantelli lemma. However, we may

hope to prove (4.13) even if the series (E(P)) diverges: indeed, (4.13) may
possibly hold if the sets E(P) intersect frequently and a lot. Thus, we are led
to analyze the relative arrangement of the sets E(P) in order to evaluate the
multiple intersections.
81
The basic idea, explained in Section 14 of Chapter 2 of [540] (see also
[538], pp. 220221), consists in considering auxiliary sets E(P)

containing
E(P), and in dividing them into two classes:
The essential intervals, which are the sets E(P)

such that E(P)
has a
small intersection with any interval E(Q),

where Q(X ) is an integer polynomial of same height and same degree as P(X ).
The inessential intervals, which are the sets E(P)

such that E(P)
has a
large intersection with some interval E(Q),

where Q(X ) is an integer polynomial of same height and same degree as P(X ).
Then, the rst case is treated by means of Lemma 4.1 above and the Borel
Cantelli Lemma 1.2, while induction and Lemma 4.1 are used for the second
one.
We proceed by induction on the integer n. For n = 1, the result easily
follows from Lemma 1.2, as in the proof of the easy half of Theorem 1.10. Let
n 2 be a given integer and assume that Theorem 4.3 holds for any positive
integer less than or equal to n 1. We introduce in the sequel various sets of
real numbers (or of polynomials) which depend on n, but we choose not to
indicate this dependence to simplify our notation.
Two reductions of the problem

First, we show that, in order to establish Theorem 4.2, it is plainly enough to
prove that the set of real numbers such that (4.2) has innitely many solutions in integer polynomials P(X ) which are irreducible and leading (this
terminology means that the leading coefcient of P(X ) is equal to its height)
has Lebesgue measure zero.
Let Ar ed be the set of real transcendental numbers for which there exist
innitely many reducible polynomials P(X ) of degree n such that |P( )| <
n (H(P)). These polynomials can be written P(X ) = P1 (X ) P2 (X ), where
the degrees of the polynomials P1 (X ) and P2 (X ) are integers between 1 and
n 1. By Lemma A.3, we have H(P1 ) 2n H(P) and, since is nonincreasing, there exist an integer 1 n 1 and innitely many polynomials P(X ) of degree such that |P( )| < (2n H(P)). For any positive
is non-increasing and the
integer h, set (h)

= (2n h). The function
converges, hence, by our inductive hypothesis applied to the

sum h1 (h)
integer , the set Ar ed is a null set, as expected.
We now deal with the second reduction. By Lagrange interpolation formula,
there exists a positive constant c2 , depending only on n, such that any integer
82
polynomial P(X ) of degree n belongs to a set

P j := {P(X ) : P(X ) Z[X ] of degree n such that |P( j)| c2 H(P)},
for some integer j with 0 j n. Let A denote the set of real transcendental
numbers for which |P( )| < n (H(P)) has innitely many solutions with
P(X ) irreducible and, for j = 0, . . . , n, set
A j = { A : |P( )| < n (H(P)) for innitely many
irreducible polynomials P(X ) in P j }.
Let P(X ) be in P j and irreducible. Set G(X ) = P(X + j) or P(X + j),
in such a way that G(0) is positive. There exists an integer c3 , that we take
greater than c21 , depending only on n, such that H(G) c3 H(P). The leading coefcient G(0) of the polynomial Q(X ) := X n G(X 1 ) then satises
G(0) > c31 H(P). Set c4 = c32 + 1 and R(X ) := Q(c4 X ). The leading coefcient of R(X ) is at least equal to c4n c31 H(P), while its other coefcients
are, in absolute value, bounded by c4n1 c3 H(P), which is strictly less than
c4n c31 H(P). Hence, the polynomial R(X ) is irreducible and leading, and it
satises H(R) = c4n G(0) c5 H(P), for some positive constant c5 depending
only on n.
Thus, for any integer j = 0, . . . , n and for any real number in A j , there
exist innitely many irreducible and leading polynomials R(X ) satisfying

1
R
= | j|n |P( )| < | j|n n H(P)

c4 ( j)

| j|n n c51 H(R) .
Since the function decreases monotonically, for any given positive real num

ber , the sum h1 n (h) diverges if, and only if, the sum h1 n (h)
diverges. Consequently, if we prove that the set of real transcendental numbers
such that (4.2) has innitely many solutions in leading, irreducible, integer
polynomials P(X ) has Lebesgue measure zero, we then get that A j (and, thus,
the set A) is a null set.
Some auxiliary quantities

Since is non-increasing and the series h1 (h) converges, h(h) tends
to 0 when h tends to innity. Thus, there exists a positive integer h 0 such that
h(h) 1/4
holds for h h 0 .
(4.14)
83
Let P(h) denote the set of leading irreducible integer polynomials P(X ) =
an X n + . . . + a1 X + a0 of height h = an . Let P(X ) be in P(h) and denote by
1 , . . . , n its roots. Observe that for any integer j = 1, . . . , n we have

h| j |n |a0 | + . . . + max 1, | j |n1 |an1 |,
hence,
| j |n 1 + | j | + . . . + | j |n1 if | j | 1.
Consequently, we get
| j | 2, for j = 1, . . . , n.
(4.15)
We number the roots of P(X ) in such a way that |1 2 | is the minimum

among the |1 j |s with j = 2, . . . , n. We then set
= |1 2 |,
2n n (h)
,
=
|P (1 )|
= min{, ( )1/2 };
these quantities obviously depend on P(X ), but, to simplify the notation, we
choose not to indicate this dependence. If the real transcendental number
satises |P( )| < n (h) and if 1 is a root of P(X ) with | 1 | minimal,
then we have | 1 | < and | 1 |2 < by Lemma A.5. This implies that
| 1 | < . Consequently, in order to prove Theorem 4.3, it is sufcient to
show that, for almost all real numbers , there exist only nitely many integer
polynomials P(X ) in hh 0 P(h) with a root 1 such that | 1 | < .
Let h h 0 be an integer. For an integer polynomial P(X ) in P(h), we set
I (P) = { R : | 1 | < },
I (P) = { R : | 1 | < 1 (h)}.
By (4.14), these intervals satisfy I (P) I (P) and |I (P)| = (h)|I (P)|.
We introduce the quantity p related to the distribution of the roots of P(X )
dened by p = h 1 if n = 2 and by
p = |h n (1 3 ) . . . (1 n )|1/2 ,
otherwise. We observe that, for any integer n 2, we have
( p )2 |1 2 | = h n1 |P (1 )|,
(4.16)
and we divide P(h) into two classes. The subset P1 (h) of P(h) is composed
of the polynomials satisfying p , while the subset P2 (h) of P(h) contains
84
exactly those with < p . Observe that for n = 2 the set P2 (h) is empty,
since p holds for any quadratic polynomial with integer coefcients.
In the rst step, we show that, for almost all real numbers , there are only
nitely many integer polynomials P(X ) in hh 0 P1 (h) having a root 1 such
that | 1 | < . In the second step, we establish a similar result with P1 (h)
replaced by P2 (h).
The sets P1 (h), essential and inessential intervals

Let h h 0 be an integer. An interval I (P) is said to be inessential if there
exists a polynomial Q(X ) in P1 (h) such that the Lebesgue measure of the
intersection I (P) I (Q) is greater than |I (P)|/2. We observe that the interval
I (Q) is not necessarily inessential since I (P) and I (Q) do not have, a priori,
the same length. We denote by E(h) the set
{I (P) : there exists Q(X ) in P1 (h) such that|I (P) I (Q)| > |I (P)|/2}
of inessential intervals. The set of essential intervals is then

I (P) \ E(h).
E(h) :=
PP1 (h)
Let P(X ) be in P1 (h) and be a real number in I (P). By (4.14), (4.16),

and the hypothesis p , we have
| 1 | < 1 (h) 1 (h) = 2n n1 (h) |P (1 )|1
< h 1n |P (1 )|1 = |1 2 |1 ( p )2 |1 2 |. (4.17)
Our choice of 2 ensures that the assumption of the last statement of Lemma
A.5 is fullled, hence, we get
|P( )| 2n1 |P (1 )| | 1 |
< 2n1 |P (1 )| 1 (h) = 22n1 n1 (h).
In particular, if Q(X ) is in P1 (h) and is such that is in I (Q), we then have
|Q( )| 22n1 n1 (h) and the polynomial R(X ) := P(X ) Q(X ), which
is of degree less than or equal to n 1 since P(X ) and Q(X ) have the same
leading coefcient, satises

|R( )| < 22n n1 (h) 22n n1 H(R)/2 .
(4.18)
By our inductive assumption, for almost all , only a nite number of integer
polynomials R(X ) of degree at most n 1 satisfy (4.18). Consequently, almost
all belong to only a nite number of intersections I (P) I (Q), where P(X )
85
are in some set P1 (h) with h h 0 and I (P) is an inessential interval. This
means that the decreasing intersection

I (P) I (Q)
H h 0 hH
P(X )P1 (h),Q(X )P1 (h),

P(X )= Q(X ),|I (P)I (Q)|>|I (P)|/2
is a null set. Consequently, the Lebesgue measure of

I (P) I (Q)
hH
P(X )P1 (h),Q(X )P1 (h),

P(X )= Q(X ),|I (P)I (Q)|>|I (P)|/2
tends to zero when H goes to innity. However, for polynomials P(X ) and
Q(X ) in P1 (h) with |I (P) I (Q)| > |I (P)|/2, putting I P,Q = I (P) I (Q),
the interval I (P) is contained in the union
I P,Q (I P,Q + P,Q ) (I P,Q P,Q ),
where the plus sign and the minus sign mean translation, and P,Q denotes
the length of I P,Q . Thus, the Lebesgue measure of

I (P)
hH
P(X )P1 (h)

I (P) inessential
tends to zero when H goes to innity, and the set

I (P)
H h 0 hH
P(X )P1 (h)

I (P) inessential
is a null set, as expected. In other words, for almost all real numbers , there
exist only nitely many polynomials P(X ) in hh 0 P1 (h) such that I (P) is
inessential and belongs to I (P), hence, to I (P).
We now consider the essential intervals, still under the assumption that
P(X ) belongs to P1 (h), and we set

I (P).
A(h) =
I (P)E (h)
Since the interval I (P) is essential, a point in I (P) does not belong to an
intersection I (Q) I (R) I (S) of three distinct essential intervals. Indeed,
otherwise, one among the intervals I (P), I (Q), I (R), and I (S) would cover
strictly more than half of the length of another one, which would then be an
inessential interval. Consequently, any point in I (P) belongs to at most three
essential intervals and

|I (P)| 3(h)
I (P) 36(h),
A(h) (h)
I (P)E (h)
I (P)E (h)
86
since, by (4.15) and (4.17), any interval I (P) is included in [6, 6]. It then

follows from Lemma 1.2 and the convergence of h1 (h) that almost all
belong to only a nite number of intervals I (P).
Thus, we have proved that, for almost all , there exist only a nite number
of polynomials P(X ) in hh 0 P1 (h) such that I (P) is essential and belongs
to I (P).
The sets P2 (h)
Since P2 (h) is empty for n = 2, we assume that n 3. We aim to show that,
for almost all real numbers , there are only nitely many polynomials P(X )
in hh 0 P2 (h) with a root 1 such that | 1 | < . Let h h 0 be an integer
and let P(X ) be a polynomial in P2 (h). We have < p , hence, by (4.14) and
(4.16), we get
2 = |P ()|1 n (h)2n |1 2 |
< h 1n (h) |P ()|1 |1 2 | (h)( p )2 .
(4.19)
Using the discriminant as for the proof of (4.8), we infer from (4.15) that there
exists an integer 0 , depending only on n, such that p 20 +1 . For integers
j 1 and 0 with 4 j1 h 0 , set

P2 (h)
P3 ( j) =
4 j1 h<4 j
and
P4 ( j, ) = {P(X ) P3 ( j) : 2 < p 2+1 }.
By (4.15), we have p > 2n( j+1) for any polynomial P(X ) in P3 ( j). Consequently, the set P4 ( j, ) is empty for > n( j + 1).
Let j and be integers such that P4 ( j, ) is non-empty and let P(X ) be
in P4 ( j, ) with I (P) non-empty. Then we have |m 1 | < by the denition of I (P) and |e 1 | 2, by (4.15). Assume that, for some real number
c6 > 1, there are c6 2 or more such irreducible polynomials P(X ) in P4 ( j, ).
They furnish c6 2 distinct complex numbers 1 . Denote by 1 the smallest distance between two of these numbers. Then, a simple covering argument
gives that
2

< 10( + ).
(4.20)
c6 2
2
Further, we infer from (4.14) and (4.19) that p 2+1 . By (4.15) and
< p , we can apply Lemma 4.1. It yields that there exists a positive constant
4.4 Exercise
87
c7 , depending only on n, such that c7 2 . Together with (4.20), this gives

that there exist positive constants c8 and c9 , depending on n but not on and
j, such that
c6 c8 2 2 ( + ) c8 2 1 ( 1 + 1) c9 .
Thus, the real number c6 depends only on n, and we have proved that for at
most c10 polynomials P(X ) in P4 ( j, ) the interval I (P) is non-empty, where
c10 is a constant depending only on n. Recall that, by (4.14) and (4.19), we
have

2+1 (4 j1 ) 2 j+1
for these polynomials. Thus, there exists a positive real number c11 , depending
only on n, such that

P(X )P3 ( j)
|I (P)| =
n(
j+1)
=0
P(X )P4 ( j,)
c10
n(
j+1)
|I (P)|
2+1 c11 j 2 j .
=0
j converges, Lemma 1.2 shows that almost all

Since the series
j1 j2
belong to only a nite number of intervals I (P) with P(X ) in P2 (h). Consequently, we have proved that for almost all real numbers there exist only
a nite number of polynomials P(X ) in hh 0 P2 (h) having a root 1 with
| 1 | < , that is, such that belongs to I (P).
Conclusion
We have proved that for almost all real numbers there exist only a nite number of polynomials P(X ) in hh 0 P(h) having a root 1 with | 1 | < .
Thanks to the observations following the denition of , this implies that
for almost all real numbers there exist only a nite number of integer
polynomials P(X ) with |P( )| < n (H(P)). The proof of Theorem 4.3 is
complete.
4.4 Exercise
E XERCISE 4.1. Proof of Theorem 4.2 for n = 2, following Kasch [310].
1) Let P(X ) = a2 X 2 + a1 X + a0 be an integer polynomial with distinct
roots 1 , 2 and let Disc(P) := a12 4a0 a2 denote its discriminant. Prove that
88
for any complex number we have

2
|P( )|.
min{| 1 |, | 2 |}
|Disc(P)|
2) Let s be a real number with 0 < s < 1. Prove that there exists a positive
constant c12 , depending only on s, such that, for any positive integer H , the
inequality

|Disc(P)|s c12 H 2(1s)
holds, where the summation is taken over all integer polynomials P(X ) of
degree at most 2 and of height H with non-zero discriminant.
3) Combine 1) and 2) with s = 1/2 to conclude using the BorelCantelli
Lemma 1.2.
4.5 Notes
Kubilius result [344] has been rened by Cassels [153] and extended by
Sprindzuk [531] (see also the last Section of [539] and Section 11 of Chapter
2 of [540]).
Bernik and Dombrovski [88] established a metric renement of Theorem
4.2 for general n.
An effective version of Theorem 4.2 has been obtained by Beresnevich
[60] for n = 2 and by Kalosha [307] for n = 3.
Harman [275] established asymptotic formulae for the number of solutions to (4.2) when diverges and satises some extra assumptions.
Using a terminology introduced by Sprindzuk, a curve ( f 1 , . . . , f n ) dened on an open subset U of R is called extremal if, for almost all (in the sense
of the Lebesgue measure) points in U , the supremum of the real numbers w
for which there exist innitely many integer (n + 1)-tuples (q1 , . . . , qn , p)
such that |q1 f 1 ( ) + . . . + qn f n ( ) + p| < max{|q1 |, . . . , |qn |}w is equal
to n. Theorem 4.2 asserts that the curve {(, . . . , n ) : R} is extremal.
Many extensions of this result have been obtained, see, for example, [89]
and the Notes of Chapter 2 of [86]. As for multiplicative approximation, a
curve is called strongly extremal if, for almost all points in U , the supremum of the real numbers w for which there exist innitely many integer
(n + 1)-tuples (q1 , . . . , qn , p) such that |q1 f 1 ( ) + . . . + qn f n ( ) + p| <
n
w/n is equal to n. The rst general results on strong exi=1 max{|qi |, 1}
tremality are due to Kleinbock and Margulis [330] and imply Theorem 4.6.
They have been superseded by Bernik, Kleinbock, and Margulis [89]. Observe
4.5 Notes
89
that these results cover the approximation by lacunary polynomials, since they
include that for any n-tuple (a1 , . . . , an ) of distinct positive integers the curve
{( a1 , . . . , an ) : R} is strongly extremal. Further results are due to Beresnevich [63], Dickinson and Dodson [193], Beresnevich, Bernik, Kleinbock,
and Margulis [68], and Kleinbock [327]. Kleinbock, Lindenstrauss, and Weiss
[329] identied purely geometrical conditions on measures which are sufcient
to guarantee strong extremality, obtaining thus a generalization of the results
from [330].
Theorem 4.3 for n = 2 has been extended by Yu [608] and Mashanov
[409].
Bernik, Dickinson, and Dodson [84] proved the inhomogeneous analogue
of Theorem 4.4, namely that for any positive integer n and any real number
the inequality |P( ) + | < H n+1 (H ) has, for almost all real numbers
, only a nite number of solutions in integer polynomials P(X ) of degree at
most n and height at most H , where is as in Theorem 4.4. The case n = 2
was already established by Beresnevich, Bernik, and Dodson [66], who proved
the inhomogeneous analogue of Theorem 4.5 for n = 2.
The analogue of Theorem 4.4 with the coefcients of the polynomials
being prime numbers has been establish by Bernik, Vasiliev, and Dodson [97].
Bernik [76] conrmed a conjecture of Sprindzuk ([538], Problem C) extending Theorem 4.2. He proved that for any integers k, n with 1 k n and
k
|P(i )| < H n+k1 has
for any positive real number , the inequality i=1
only nitely many solutions in integer polynomials P(X ) of degree at most n
and height at most H for almost all real k-tuples (1 , . . . , k ).
Bernik and Borbat [83] established a two-dimensional analogue of Theorem 4.4.
Slesoratene ([527] and earlier works) established analogues of Theorem
4.2 for polynomials of small degree in two variables. For arbitrary degrees, a
slightly weaker result is due to Vinogradov and Chudnovsky [579] (see also
Bernik [74]).
As for multiplicative approximation, Gallagher [253] proved that for any
positive integer n the inequality q1 . . . qn < q 1 (log q)n (log log q)1
has innitely many solutions in integers q for almost all real n-tuples
(1 , . . . , n ). Asymptotic formulae have been established by Wang and Yu
[593]. Some results from [253] were applied in harmonic analysis by Stokolos
[545].
5
In the preceding Chapters, we have encountered several sets of real numbers of

Lebesgue measure zero, including the set of Liouville numbers, the set of real
numbers with bounded partial quotients, the set of very well approximable
numbers, and the set of S -numbers of -type strictly greater than 1. Some
of them are certainly strictly larger than others: indeed, as it may be seen by
considering continued fraction expansions (see Exercise 1.5), there are very
well approximable numbers other than the Liouville numbers. On the other
hand, the set of S -numbers of -type at least 2 contains the set of S -numbers
of -type at least 3, but the results of Chapters 1 to 4 do not enable us to decide
whether the inclusion is strict or not.
In the present Chapter, we introduce a powerful tool for discriminating between the sets of Lebesgue measure zero, namely the notion of Hausdorff
dimension, developed by Hausdorff in 1919 [276]. Shortly thereafter, Jarnk
[288, 292] and, independently, Besicovitch [100], applied it to number theoretical problems, and they determined the Hausdorff dimension of sets of
real numbers very close to innitely many rational numbers (Theorem 5.2).
Their result has been subsequently generalized in many directions. For instance, A. Baker and Schmidt [45] showed in 1970 that there exist S -numbers
of arbitrarily large but nite -type (Theorem 5.5). In the present Chapter, we
prove both these results and we quote some other extensions of the Jarnk
Besicovitch Theorem. Further renements are stated in Chapter 6.
5.1 Hausdorff measure and Hausdorff dimension

Hausdorffs idea [276] consists in measuring a set by covering it by an innite,
countable family of sets of bounded diameter, and then in looking at what
happens when the maximal diameter of these covering sets tends to 0. The
90
5.1 Hausdorff measure and Hausdorff dimension
91
reader interested in Hausdorff dimension theory is directed, for example, to

Rogers [476], Falconer [234, 236], and Mattila [415].
If U is a non-empty subset of Rn , its diameter, denoted by diam(U ), is by
denition
diam(U ) = sup{x x : x, x U },
where x denotes the maximum of the absolute values of the coordinates of
a point x in Rn . Let J be a nite or innite set of indices. If for some positive

real number the sets E and U j satisfy E jJ U j and 0 < diam(U j )
for any j in J , then {U j } jJ is called a -covering of E.
Let f be a dimension function, that is, a strictly increasing continuous function dened on R>0 (actually, it is enough to assume that f is dened on some
open interval ]0, t[ with t positive) satisfying limx0 f (x) = 0. For any positive real number , set

f
f diam(U j ) ,
H (E) := inf
J
jJ
where the inmum is taken over all the countable -coverings {U j } jJ of E.

f
Clearly, the function H (E) is non-increasing. Consequently,
f
H f (E) := lim H (E) = sup H (E)

0
>0
is well-dened and lies in [0, +]. If E 1 and E 2 are two subsets of Rn with E 1
included in E 2 , we then have H f (E 1 ) H f (E 2 ). Furthermore, H f is subadditive and is a regular outer measure for which the Borelian sets are measurable
(see, for example, Rogers [476] or Mattila [415]), called the H f -measure.
Let f and g be two dimension functions. We say that g corresponds to a
smaller generalized dimension than f and we write g f if
lim
x0
g(x)
= +.
f (x)
Observe that if g f , then g increases faster than f in a neighbourhood of

the origin, and we have Hg (E) H f (E) for any subset E of Rn . Further,
the ordering induced by is not a total ordering since there exist dimension
functions f and g satisfying
lim inf
x0
g(x)
g(x)
= 0 and lim sup
= +.
f (x)
x0 f (x)
L EMMA 5.1. Let f , g, and h be dimension functions satisfying f g h.

If the set E satises 0 Hg (E) < + (resp. 0 < Hg (E) +), we then
have Hh (E) = 0 (resp. H f (E) = +).
92
P ROOF. Assume that E satises 0 Hg (E) < + and let be a positive

real number. There exists > 0 such that h(x) g(x)/(Hg (E) + 1) for any
x with 0 < x < . For any positive real number less than , there exists a
countable -covering {U j } j1 of E such that

g diam(U j ) Hg (E) + 1,
j1
whence

h diam(U j ) .
j1
This means that Hh (E) and yields Hh (E) = 0.

Replacing the functions g and h by f and g, respectively, we get by contraposition the result asserted into brackets.
When f is a power function x x s , with s a positive real number, we write
Hs (E) instead of H f (E) and we call it the s-dimensional Hausdorff measure of the set E. We dene H0 as the counting measure: H0 (E) is equal to
the cardinality of the set E. Furthermore, for any Borelian subset E of Rn , we
have
Hn (E) =
2n (n/2)!
vol(E),
n/2
where vol denotes the Lebesgue measure on Rn . In particular, in the real case,
HId = H1 coincides with the Lebesgue measure on R.
We infer from Lemma 5.1 that Hs (E) Ht (E) when t s 0. Consequently, the function s Hs (E) is non-increasing on R0 .
C OROLLARY 5.1. Let E be a subset of Rn . If there exists s 0 such that
Hs (E) < +, then Hs+ (E) = 0 for any > 0. If there exists s > 0 such
that Hs (E) > 0, then Hs (E) = + for any in ]0, s].
P ROOF. This follows from Lemma 5.1 since we have (x x s ) (x x t )
for any real numbers s and t with t > s > 0.
Corollary 5.1 shows that there is a critical value of s at which Hs (E) jumps
from + to 0. This value is called the Hausdorff dimension of E.
D EFINITION 5.1. Let E be a subset of Rn . The Hausdorff dimension of E,
denoted by dim E, is the unique non-negative real number s0 such that
Hs (E) = 0 if s > s0
5.2 Upper bound for the Hausdorff dimension
93
and
Hs (E) = + if 0 < s < s0 .
In other words, with the notation of Denition 5.1, we have
dim E = inf{s : Hs (E) = 0} = sup{s : Hs (E) = +}.
The main properties of Hausdorff dimension for subsets E, E 1 , E 2 , . . . of Rn
are (see for example [86], p. 65):
(i)
(ii)
(iii)
(iv)
(v)
(vi)
dim E n;
If vol(E) is positive, then dim E = n;
If E 1 E 2 , then dim E 1 dim E 2 ;
dim +
j=1 E j = sup{dim E j : j 1};
The Hausdorff dimension of a nite or countable set of points is 0;
Two sets differing by a countable set of points have the same Hausdorff
dimension.
Observe that there exist sets of Hausdorff dimension zero which are uncountable (for example, the set of Liouville numbers, by Corollaries 1.1 and 5.2
below) as well as uncountable sets of Lebesgue measure zero and with Hausdorff dimension one (for example, the set of badly approximable real numbers,
see Exercise 5.1).
By property (iv), to determine the Hausdorff dimension of a subset E of
Rn , it is enough to know the Hausdorff dimension of intersections of E with
products of bounded intervals.
The choice of another norm in the denition of the diameter does not affect the measure when it is either 0 or +. In particular, the critical exponent is the same. Borelian sets of Hausdorff dimension s with positive nite
s-dimensional Hausdorff measure are called s-sets and enjoy some special
properties (see, for example, [234], Chapters 2 to 4, or [236], Chapter 5).
The introduction of dimension functions is a tool for discriminating between sets of same Hausdorff dimension, as will be made clear in Chapter 6.
In the remainder of the present Chapter, except in the notes at the end, we
study only subsets of R.
5.2 Upper bound for the Hausdorff dimension

The exact calculation of the Hausdorff dimension (and, a fortiori, of the H f measure) of a set E is in most cases a difcult problem. It is however often
possible to bound it from above by applying the following analogue of the
BorelCantelli Lemma 1.2.
94
L EMMA 5.2. Let E be a Borelian subset of R and {U j } j1 be a countable

family of subsets of R such that
E { R : U j for innitely many j 1}.
If f is a dimension function such that

f diam(U j ) < +,
j1
then H f (E) = 0. In particular, if there exists a positive real number s such

that

s
diam(U j ) < +,
j1
then Hs (E) = 0 and dim E s.

P ROOF. By assumption, for any positive integer N , the family {U j } jN
is a covering of E. Let and be positive real numbers. Since the sum

j1 f (diam(U j )) converges, there exists a positive integer N such that
diam(U j ) < for any j N and

f diam(U j ) < .
jN
f
This means that H (E) < , hence, we get H f (E) which, by letting
tend to 0, implies H f (E) = 0.
We give at once an application of Lemma 5.2 to Mahlers classication of
numbers. Corollary 5.2 is due to Kasch and Volkmann [311], but their proof,
slightly more complicated, does not involve Theorem 3.6. A further result is
given in Section 7.5.
C OROLLARY 5.2. The set of T -numbers and the set of U -numbers have
Hausdorff dimension zero.
P ROOF. For any real number w > 1 and any positive integer H0 , the set of
T -numbers and the set of U -numbers are contained in

E(H0 , w) :=
] H nw , + H nw [,
n1 H H0
where the last union is taken over all real algebraic numbers of degree n and
height H . Since there are no more than n(n + 1)(2H + 1)n such algebraic
5.3 The mass distribution principle
95
numbers, we have for any real number s with 1/w < s 2/w the estimate

2H0 + 1 n n(ws1)
Hs E(H0 , w)
4n(n + 1) 2n
H
.
2H0
n1 H H
0
The above double series converges if H0 is greater than 41/(ws1) . Thus, by

Lemma 5.2, the Hausdorff dimension of the set E(H0 , w) is at most 1/w.
Hence, the set of T -numbers and the set of U -numbers have Hausdorff dimension zero. We conclude by Theorem 3.6.
Lemma 5.2, called the HausdorffCantelli Lemma by Bernik and Dodson
([86], Lemma 3.10), provides in general, but not always (see Dodson [199]),
an upper bound which is the exact value of the dimension. Furthermore, a covering argument as Lemma 5.2 is in some cases unfortunately useless, as it can
be seen, for example, in Section 5.7 below. An alternative approach resting on
the study of exponential sums is discussed in [86], Section 4.4.2.

Different methods, more or less sophisticated, allow one to bound from below
the Hausdorff dimension of a real set.
Since the sets occurring in Diophantine approximation are often of this
form, we consider a set K included in a bounded interval E, and dened as
follows. We set E0 = E and we assume that, for any positive integer k, there
exists a nite family Ek of disjoint compact intervals in E such that any interval U belonging to Ek is contained in exactly one of the intervals of Ek1
and contains at least two intervals belonging to Ek+1 . We also suppose that the
maximum of the lengths of the intervals in Ek tends to 0 when k tends to innity. For k 0, we denote by E k the union of the intervals belonging to the
family Ek . Then, we dene a mass distribution on the set
K :=
Ek
k=1
by successive subdivisions. We set (E) = 1, and we distribute this mass

(E)
(E)
between the intervals U1 , . . . , Um 1 composing E1 in such a way that
m 1
(E)
i=1 (Ui ) = 1. Likewise, if V belongs to the family Ek , we dene, for
(V )
(V )
the intervals U1 , . . . , Um V composing Ek+1 and included in V , the masses
m V
(V )
(V )
(V )
(U1 ), . . . , (Um V ) in such a way that i=1
(Ui ) = (V ). Further,
for k 0, we set (E \ E k ) = 0.
Let E be the family consisting of the sets E \ E k for k 0, and of all the
intervals belonging to the families Ek with k 0. We have dened (A) for
96
any set A belonging to E. As claimed in the following result, this is enough in

order to dene (A) for any Borelian set A included in E.
P ROPOSITION 5.1. If K and are as above, then can be extended to all the
subsets of R in such a way that it is a Borelian measure. Moreover, the support
of is included in K.
The proof is involved. We just briey mention that, if A is a Borelian set in E,
we set

(U j ) : A
U j and U j E for j 1 .
(A) = inf
j1
j1
The main difculties occur in showing that dened in this way is indeed a
measure. See, for example, [236], Proposition 1.7.
After having dened a measure on the set K, it is often possible to bound
its Hausdorff dimension from below by using the following lemma, called the
mass distribution principle, or the (easy half of the) Frostman Lemma [251].
L EMMA 5.3. Let be a probability measure with support in a bounded real
set K. Let f be a dimension function. Assume that there exist positive real
numbers and such that
(J ) f (|J |)
for any interval J with length |J | . We then have H f (K) 1/. Furthermore, if f (x) = x s for some real number s in ]0, 1], then dim K s.
P ROOF. If {U j } j1 is a countable -covering of K, with 0 < , then,
denoting by U j the smallest interval containing U j , we have diam(U j ) = |U j |
and

Uj
(U j )
(U j )
1 = (K) =
j1

f diam(U j ) .
j1
j1
j1
We deduce that H (K) 1 for any , and the rst assertion follows
by letting tend to 0. Furthermore, if f (x) = x s for some s > 0, then Hs (K)
is positive, which implies that the Hausdorff dimension of K is at least equal
to s.
f
An illustration of the strength of Lemma 5.3 is provided by the following result, which is Example 4.6 in [236].
97
P ROPOSITION 5.2. Keep the same notation as above. Assume further that
there exists a positive integer k0 such that, for any k k0 , each interval of
E k1 contains at least m k 2 intervals of E k , these being separated by at
least k , where 0 < k+1 < k . We then have
dim K lim inf
k+
log(m 1 . . . m k1 )
.
log(m k k )
(5.1)
P ROOF. Replacing K by a smaller set if needed, one can assume that, for
k k0 , each interval of E k1 exactly contains m k intervals of E k , and
that there exists a positive integer m such that the set E k is composed of
mm k0 . . . m k disjoint intervals U (k) , to which we uniformly give the mass
(U (k) ) = (mm k0 . . . m k )1 . Let U be an interval of length smaller than k0 ,
and let k k0 + 1 be such that
k |U | < k1 .
(5.2)
Observe that the number of intervals of E k which intersect U is, on the one
hand, at most equal to m k (since, by (5.2), U intersects a single interval of
E k1 ) and, on the other hand, at most equal to 1 + |U |/k , since the intervals
of E k are distant by at least k . Consequently, we get

2|U | s
1
2|U |
1
(U )
min m k ,
mm k0 . . . m k
k
k
mm k0 . . . m k m s1
k
2|U |s (m k k )s
,
mm k0 . . . m k1
for any real number s in [0, 1]. This implies that (U )/|U |s is bounded provided that
s < lim inf
k+
log(mm k0 . . . m k1 )
.
log(m k k )
Since m k k tends to 0 when k grows to innity, we conclude by applying

Lemma 5.3.
As pointed out by Falconer [236], if for any k 1 the components of E k have
the same length k and if each interval of E k1 contains exactly m k intervals of
E k , which are evenly spaced in the sense that m k k ck1 for some positive
constant c, then (5.1) becomes
dim K lim inf
k+
log(m 1 . . . m k1 )
.
log(k1 )
(5.3)
Conversely, under the same assumption, the set K can be covered by m 1 . . . m k
98
intervals of length k , thus, by the denition of the Hausdorff dimension, we

have equality in (5.3). Such examples occur frequently in number theory, as it
will be seen throughout this and the next Chapter.
It follows from Proposition 5.2 that the Hausdorff dimension of the triadic
Cantor set is log 2/ log 3. With some additional effort, it is possible to prove
that its measure at the critical exponent is equal to 1 (see [234], Theorem 1.14),
hence, it is an s-set.
There are other techniques for establishing lower bounds for the Hausdorff
dimension. Potential theoretic and Fourier transform methods are briey discussed in Chapter 4 of [236], see also Sections 3.5.7 and 4.1.3 of [86]. A full
treatment can be found Mattila [415].
5.4 Regular systems

The notion of regular system was introduced in 1970 by A. Baker and Schmidt
[45] to compute the Hausdorff dimension of sets KS of real numbers close to
innitely many points of a given countable set S (namely, in [45], sets of algebraic real numbers of bounded degree). Here, we slightly change their definition, in order to point out more accurately which properties of S are really
required to get the Hausdorff dimension of KS .
D EFINITION 5.2. Let E be a bounded open real interval. Let S = ( j ) j1 be
a sequence of distinct real numbers. Let : Z1 R1 and : Z1 R1
be increasing functions such that (n) n and (n) n for any n 1.
The triple (S, , ) is called a regular system of points in E if there exist a
positive constant c1 = c1 (S, , ) and, for any interval I in E, a number
K 0 = K 0 (S, , , I ) such that, for any K K 0 , there exist integers
1 i 1 < . . . < i t ((K ))
with i h in I for any h = 1, . . . , t,
|i h i |
1
(1 h = t),
(K )
and
c1 |I |(K ) t |I |(K ).
In the original work of A. Baker and Schmidt [45], the set S is not indexed
and the function is the identity. The introduction of rests on an idea of
Rynne [487], who dened the concept of weakly regular system, used, for
example, in [130] (see Exercise 5.7). We emphasize that we do not assume
5.4 Regular systems
99
that every point of S belongs to E. Furthermore, we have supposed that E is

bounded, although this was not assumed in [45]. This does not involve any loss
of generality as far as calculating Hausdorff dimension is concerned since any
unbounded set can be covered by a countable collection of bounded, open sets
to which the results may be applied.
In the applications, the real number K 0 (S, , , I ) need not be effectively
computable.
Ideally, when the points of S are very well distributed, then we may hope
that (S, x 1 x, x 2 x) is a regular system for suitable positive real
numbers 1 and 2 . If this occurs, S is said to be optimal and we can get
sharper results on sets of real numbers close to innitely many points of S, see
Chapter 6.
To illustrate the notion of a regular system, we start with an example.
P ROPOSITION 5.3. We order the rational numbers as follows. First, we divide
Q into classes containing rational numbers of the same height, starting with
height 0, 1, 2, and so on. Then within each class, the rationals are ordered in
the usual way (as real numbers). Then, the triple (Q, x 4x, x x) is a
regular system in ]0, 1[.
P ROOF. For any integer B 2, the number of pairs of integers (a, b) with
1 a, b B and gcd(a, b) = 1 is at least equal to

2
B/2 2

B
B2
2
2
2
1
.
B B
B
d
6
3
d=2
On the other hand, strictly fewer than 2B 2 rational numbers have their height
bounded by B. Consequently, for any positive integer j, we have

j
H( j ) 3( j + 1) + 1.
(5.4)
2
Let I be a bounded real interval in ]0, 1[ and Q > 100 be a real number. By
Dirichlets Theorem 1.1, for any in I , there exists a rational number p/q
such that

p < 1
and 1 q Q.
(5.5)

q qQ
The Lebesgue measure of the set E of points in I for which (5.5) is satised
by a rational number p/q with q Q/10 is less than or equal to
[Q/10]+1

q=1
2
(q|I | + 1),
qQ
100
hence, to |I |/4, provided that Q 100|I |1 log(100|I |1 ). Further, for in

I \ E, there exists a rational number p/q such that

p < 10 and Q/10 q Q.

q
Q2
Let { p j /q j }1 jt be a maximal family of rational numbers in I with Q/10
q j Q and | p j /q j p /q | 10/Q 2 for 1 j = t. The union of the
intervals

pj
20 p j
20
2,
+ 2 , 1 j t,
qj
Q qj
Q
covers I \ E, thus its Lebesgue measure is at least 3|I |/4. Consequently, we
get
3
Q 2 |I |.
t
160
Furthermore, it follows from (5.4) that for any j = 1, . . . , t we have p j /q j =
i j with Q 2 /400 i j 4Q 2 . We have proved that for any K K 0
with

K 0 104 |I |2 log2 100|I |1
(5.6)
there exist integers 1 i 1 < . . . < i t 4K such that |i j i | 1/K
for 1 j =
t and t 3K |I |/160. This shows that the triple (Q, x
4x, x x) is a regular system in ]0, 1[.
Proposition 5.4 shows that the set of algebraic numbers of bounded degree
forms a regular system. Further examples are given at the end of this Section
and in Section 6.1.
Thanks to Proposition 5.2, we are able to compute a lower bound for the
Hausdorff dimension of sets of real numbers close to innitely many points in
a given regular system.
T HEOREM 5.1. Let E be a bounded open real interval. Let S = ( j ) j1 be
a sequence of distinct real numbers. Let : Z1 R1 and : Z1
R1 be increasing and assume that the triple (S, , ) is a regular system of
points in E. Let : R1 R>0 be continuous, non-increasing and such
that
lim sup x((x)) < +.
x
Then, the Hausdorff dimension of the set
KS
() = lim sup { E : | j | < ( j)}
j+
5.4 Regular systems
101
satises
() lim inf

dim KS
x+
log x
.
log ((x))
() also depends on E, but we choose not to indicate this

Obviously, the set KS
dependence in the notation; this should not cause any trouble in the sequel.
Although the approximation function needs only to be dened for every
sufciently large integer, we choose for commodity (as in the statement of
Theorem 1.10 and in Chapters 4 and 6) to assume that it is dened on R1 and
continuous.
() obtained in Theorem 5.1
We point out that the lower bound for dim KS
does not depend on the function , as it was observed by Rynne [487].
As will be apparent below, Theorem 5.1 shows that the points in S do not
need to be very well distributed in order that we get the exact dimension of
KS ().
P ROOF. We proceed as follows: we construct a Cantor set K of the form considered in Section 5.3 and contained in KS (); we then apply Proposition
5.2 to estimate its Hausdorff dimension from below. Set E 0 = E. Let c2 and
c3 be positive real numbers such that c2 < 1/3 and x((x)) < (3c2 )1
for any x c3 . By Denition 5.2, there exist a positive constant c4 , depending only on S, , and , an integer t1 , a real number K 1 c3 with
(1)
(1)
c4 |E 0 |K 1 t1 |E 0 |K 1 , and a set A1 = {i 1 , . . . , i t1 +2 } included in
{1, . . . , (K 1 )} such that h is in E 0 and |h | 1/K 1 for any distinct integers h, in A1 . We may further assume that K 1 is a value taken by
the function . Dene

E1 =
j c2 ((K 1 )), j + c2 ((K 1 )) ,
jA1
where A1 is a subset of A1 such that E 1 is contained in E 0 and has maximal

cardinality. Clearly, E 1 is the union of at least t1 intervals, which are pairwise
disjoint since we have assumed K 1 c3 .
We proceed inductively and we let k 1 be an integer such that we have
constructed a set E k contained in E of the form

j c2 ((K k )), j + c2 ((K k )) ,
Ek =
jAk
where K k is in (Z1 ) and Ak is a set of tk integers in {1, . . . , (K k )}.

For j in Ak , denote by U j the interval [ j c2 ((K k )), j +
c2 ((K k ))] and by K U j = K 0 (S, , , U j ) the constant given by Denition
5.2. We construct a new set EU j as a nite union of closed intervals contained
102
in U j . Let K k+1 be a value taken by and greater than (K U j ) for any j in
Ak . To simplify the notation, we omit the subscript j and denote by U one of
the U j s. By the denition of a regular system, there exist a positive integer tU
= {i (U ) , . . . , i (U ) } included
with c4 |U |K k+1 tU |U |K k+1 and a set AU
1
tU +2
in {1, . . . , (K k+1 )} such that h is in U and |h | 1/K k+1 for any
. Dene
distinct integers h, in AU

EU =
h c2 ((K k+1 )), h + c2 ((K k+1 )) ,
hAU
such that E is contained in U and has maximal
where AU is a subset of AU
U
cardinality. Clearly, EU is the union of at least tU intervals, which are pairwise
disjoint since K k+1 c3 .
Set

E k+1 =
EU and K :=
Ek .
jAk
k1
Each interval U j of E k contains at least

tU j c4 |U j |K k+1 c2 c4 ((K k )) K k+1 =: m k+1
1
intervals of E k+1 , these being separated by at least c2 K k+1
=: k+1 , since
c2 < 1/3.
All the assumptions of Proposition 5.2 are fullled, hence, the Hausdorff
dimension of K satises
k1
dim K lim inf

k+

log c2 c4 (K ) K +1 1
=1
.

log c2 c4 (K k ) K k+1 k+1
(5.7)
For any positive real number and any k 2, we may select K k large enough
in such a way that the numerator in (5.7) is larger than (log K k )/(1 + ). Consequently, we have
dim K lim inf
k+
log K k

.
(1 + ) log c22 c4 ((K k ))
Since can be taken arbitrarily small, we get

dim K lim inf
x+
log x
.
log ((x))
Further, we observe that KS () contains K, since is non-increasing and

(n) n for any n 1. Hence, the result follows.
5.5 The theorem of JarnkBesicovitch
103
A closer look at the proof of Theorem 5.1 suggests that we are able to consider
simultaneously several regular systems, and even a countable set of regular
systems. Indeed, at each step k, we use a regular system in order to construct
the set E k+1 from the set E k . If we have at our disposal several regular systems,
we may use any one of them at each step and, provided that we use all of
them innitely many times, the resulting set, that is, K := k1 E k , will be
composed of points close to innitely many elements of each regular system. It
remains for us to bound the Hausdorff dimension of K from below. To this end,
assume that, for any positive integer n, we have a regular system (Sn , n , n )
of points in K, and a non-increasing continuous function n : R1 R>0
with
lim inf xn (n (x)) < +.
x+
Then, following the proof of Theorem 5.1, we get

log x

.
dim K lim inf
x log minn1 n (n (x))
(5.8)
This observation is the key point for the proof of Theorem 5.4.
As observed by Bernik and Dodson [86], page 104, points with an asymptotic distribution in ]0, 1[ with a suitable error term also form a regular system. Indeed, let S = ( j ) j1 be a sequence of real numbers in ]0, 1[. Assume that its discrepancy is controlled in the sense that there exists a positive constant < 1 such that, for any interval I in ]0, 1[ and any real number
Q 1, the number of points 1 , . . . , Q contained in I is asymptotically
equal to Q|I | + O(Q ), where the numerical constant implied in O is independent on I and Q. Dene the function on the set of positive integers by
(n) = [n 1/(1) ]. Then (see [86], page 104) the triple (S, , ) forms a
regular system for some suitable linear function .
As an application to inhomogeneous approximation, Bernik and Dodson
[86] proved that, for almost all real numbers and for any real number > 1,
the set of real numbers such that |q p| < q holds for innitely
many pairs of integers ( p, q) with q positive has Hausdorff dimension 1/ .
This result holds in fact for all irrational numbers (see Exercise 5.7 for a
proof of this assertion).
5.5 The theorem of JarnkBesicovitch

It follows from Corollary 1.5 that, for any > 1, the set

p
p
1
K1 ( ) := R : < 2 for innitely many rationals
q
q
q
104
has Lebesgue measure zero. It is uncountable since it contains the Liouville

numbers. Its Hausdorff dimension has been calculated by Jarnk [289], and,
later and independently, by Besicovitch [100].
T HEOREM 5.2. For any 1, the Hausdorff dimension of the set K1 ( ) is
equal to 1/ .
C OROLLARY 5.3. For any real number 1, we have
1
dim{ R : w1 ( ) 2 1} = .
P ROOF. This is a straigthforward consequence of Theorem 5.2. Indeed, for

any real numbers and with > 1 and 0 < < 1, the set introduced in
the corollary is contained in K1 ( ) and contains K1 ( + ).
Corollary 5.3 does not imply that the Hausdorff dimension of the set
W1 ( ) := { R : w1 ( ) = 2 1}
is equal to 1/ . This result, proved in Exercise 5.3, is a straightforward consequence of Satz 4 of Jarnk [292]. It was rediscovered by Guting [266], thanks
to a renement of Besicovitchs proof [100] of Theorem 5.2.
The set of very well approximable numbers (Denition 1.4) has Lebesgue
measure zero (Corollary 1.5) and Hausdorff dimension 1, by Theorem 5.2. This
is also the case for the set of badly approximable numbers (Denition 1.3), by
Corollary 1.6 and Exercise 5.1.
We deduce Theorem 5.2 from Proposition 5.3 and Theorem 5.1 above. Another proof, which directly depends on Proposition 5.2, is given in Section 5
of Falconers book [236].
PROOF OF T HEOREM 5.2. By Theorem 1.1, K1 (1) is the set of irrational
numbers, hence, its Hausdorff dimension is 1. Let > 1 be a real number.
Our aim is to prove that dim(K1 ( )]0, 1[) = 1/ , which implies Theorem 5.2. For any positive integer q, denote by E(q) the set of in ]0, 1[
for which there exists an integer p such that | p/q| < q 2 . Each point
of the set K1 ( )]0, 1[ is contained in innitely many sets E(q). Since the
series

(q + 1)(2q 2 )s
q1
converges for any real number s greater than 1/ , it follows from Lemma 5.2
that the Hausdorff dimension of K1 ( ) is less than or equal to 1/ .
In order to prove the reverse inequality, denote by S = ( j ) j1 the set of
non-zero rational numbers, ordered as in the statement of Proposition 5.3. For
5.6 Hausdorff dimension of sets of S -numbers
105
any integer j 1, set

( j) = H( j )2 .
By Proposition 5.3, the triple (( j ) j1 , x 4x, x x) is a regular system
and we get from (5.4) that x(4x) 1 holds when x is large enough. Since
() (dened with E =]0, 1[) and K ( )]0, 1[ coincide, it follows
the sets KS
1
from Theorem 5.1 and (5.4) that

1
log x
log x
lim inf
= ,
dim K1 ( ) ]0, 1[ lim inf
x log (4x)
x 2 log H(4x )
as claimed.
Theorem 3.3 asserts that almost all real numbers are S -numbers of -type less
than or equal to 1 (see also Corollary 4.1). However, we have not proved up to
now that S -numbers of -type strictly greater than 1 do exist. We will achieve
this in the present Section.
Recall that An denotes the set of real algebraic numbers of degree bounded
by n. A. Baker and Schmidt [45] have extended Theorem 5.2 as follows.
T HEOREM 5.3. For any integer n 1 and any real number 1, the Hausdorff dimension of the set
Kn ( ) := { R : | | < H() (n+1) for innitely many An }
is equal to 1/ .
Exactly as we derived Corollary 5.3 from Theorem 5.2, we have the following
consequence of Theorem 5.3.
C OROLLARY 5.4. For any integer n 1 and any real number 1, we have
1
dim{ R : wn ( ) (n + 1) 1} = .
In the proof of Theorem 5.3 detailed below we have simplied the approach of
A. Baker and Schmidt by using the weaker form of Frostmans Lemma given
by Proposition 5.2. However, a slight modication of our argument and the use
of Lemma 5.3 instead of Proposition 5.2 yields a sharper result than Corollary
5.4.
Wn ( ) := { R : wn ( ) = (n + 1) 1}
is equal to 1/ .
106
P ROOF. See Exercise 5.3.

Theorem 5.4, due to A. Baker and Schmidt [45], obviously implies that for
any positive integer n, the set of values taken by the function wn includes the
interval [n, +[. It should be pointed out that no other method is known for
proving this last statement.
We now turn to the proof of Theorem 5.3, which proceeds in several steps.
First, we use results from Chapter 4 to show that real algebraic numbers of
bounded degree are well distributed (Proposition 5.4). Then we suitably order
the sets An (Lemma 5.4), and we conclude by applying Theorem 5.1.
P ROPOSITION 5.4. Let n 2 be an integer. For any bounded real interval I
there exist a real number K (I ) and, for any K K (I ), algebraic numbers
1 , . . . , t in An I such that
H(h ) K 1/(n+1) (log K )n , |h k | K 1 (1 h < k t)
(5.9)
and
t
|I |K
.
8
Using Theorem 4.3, A. Baker and Schmidt established Proposition 5.4 (actually under a slightly weaker form). We could apply Theorem 4.4 to get a
sharper statement (that is, with a smaller exponent for (log K ) in (5.9)), as in
[86], page 101, but the result obtained would not yield more applications than
Proposition 5.4. In particular, it would not allow us to get the results established
in Chapter 6.
Proposition 5.4 does not look best possible because of the factor (log K )n in
(5.9): if the real algebraic numbers of degree at most n are evenly distributed,
then Proposition 5.4 should hold with H( j ) cK 1/(n+1) in (5.9) for some
constant c. This is indeed the case, as was proved by Beresnevich [61], see
Section 6.1.
PROOF OF PROPOSITION 5.4. For any integer H 3, denote by R I (H ) the
set of in I for which there exists a real algebraic number of degree at most
n and of height at most H such that
| | < H n1 (log H )n(n+1) .
(5.10)
Let be a real transcendental number in I . Denote by M 1 a real number

2
such that I is contained in [M, M] and let H > 34Mn be a real number.
By Theorem B.2, there exists a non-zero integer polynomial P(X ) = an X n +
107
. . . + a1 X + a0 , of degree at most n, such that

0 < |P( )| (4M)n H n (log H )n
2 1
|a1 | (4M)n H
|a j | H (log H )n1 (2 j n).
(5.11)
Since the height of P(X ) is bounded by 2n H , the height of any root of P(X )
is at most equal to H , by Lemma A.3. We have several cases to distinguish. In
the sequel, we adopt the convention that the constants implied by
depend
only on n and M.
Assume rst that |P ( )| n (4M)n H (log H )n1 . We infer from (5.11)
that we have |a1 |
H (log H )n1 , hence, H(P)
H (log H )n1 . Consequently, we get
|P( )|
H(P)n (log H(P))n1 < H(P)n (log H(P))n1/2 ,
(5.12)
when H is sufciently large, since H(P) tends to innity with H . However,

since the series h2 h 1 (log h)11/2n converges, Theorem 4.3 asserts that
for almost all real numbers , the equation
|Q()| < H n (log H )n1/2
has only a nite number of solutions in integer polynomials Q(X ) of degree
at most n and of height at most H . This shows that the Lebesgue measure
of the set of real numbers satisfying (5.12) tends to zero when H tends to
innity.
Assume now that |P ( )| > n (4M)n H (log H )n1 . Lemma A.5 ensures
that there exists a zero of P(X ) satisfying
| | n
|P( )|
< H n1 (log H )n(n+1) ,
|P ( )|
whence belongs to R I (H ).
Consequently, the Lebesgue measure of the set R I (H ) tends to |I | when
H tends to innity. In particular, there exists H0 such that, for any H H0 ,
we have (R I (H )) |I |/2. Let H H0 be large enough such that K :=
H n+1 (log H )n(n+1) is greater than H . Let 1 , . . . , t be a maximal subset of
An I composed of algebraic numbers satisfying
H(i ) H K 1/(n+1) (log K )n
and
|i j | H n1 (log H )n(n+1) = K 1 .
108
By maximality, any real algebraic number in An such that H( ) H is

separated by at most K 1 from some i . Thus, by (5.10), any real number
belonging to R I (H ) is separated by at most 2K 1 from some i , and the
Lebesgue measure of the union of the intervals centered at the points i for
i = 1, . . . , t and of radius 2K 1 is greater than or equal to (R I (H )). Hence,
we get |I |/2 4t K 1 , as claimed.
In order to apply Theorem 5.1, we have to dene an ordering on the set An of
real algebraic numbers of degree at most n.
L EMMA 5.4. Let n 1 be an integer. We order the set An := ( j ) j1 as
follows. First, we put its elements into classes containing algebraic numbers
of the same height, starting with height 0, 1, 2, and so on. Then within each
class, the elements of An are ordered in the usual way (as real numbers). Then,
there exist two positive constants c5 and c6 , depending only on n, such that, for
any j 1, we have
c5 j 1/(n+1) H( j ) c6 j 1/(n+1) .
(5.13)
P ROOF. The left-hand side inequality in (5.13) is clear, since, for any positive
integer H , there are at most n(2H + 1)n+1 algebraic numbers of height at most
H and degree at most n. As for the right-hand side, let h 5 be an odd integer.
Consider an integer polynomial
P(X ) := h X n an1 X n1 . . . a1 X a0 ,
where a0 is congruent to 2 modulo 4 and, for 0 j n 1, the integer
a j is even and belongs to {0, 2, . . . , 2[h/2]}. By Eisensteins Criterion, the
polynomial P(X ) is irreducible. Furthermore, it has (at least) one real root.
Consequently, there are at least c7 h n real algebraic numbers of height h and
degree n, for some positive constant c7 depending only on n. Hence, there
exists a positive constant c8 , depending only on n, such that, for any positive integer H , there are at least c8 H n+1 real algebraic numbers of height at
most H and degree at most n. This proves the right-hand side inequality of
(5.13).
We now are able to complete the proof of Theorem 5.3.
PROOF OF THEOREM 5.3. Let > 1 be a real number and n 1 be an integer.
By Theorem 5.2, we assume n 2. For any H 1, denote by E(H, n) the set
of real numbers in ]0, 1[ for which there exists an algebraic number in An
with H() = H and
| | < H() (n+1) .
109
This is the union of at most n(n + 1)(2H + 1)n intervals, each of which with
length at most 2H (n+1) . Furthermore, the set H 1 E(H, n) is a countable
covering of Kn ( ). Since for any real number s > 1/ the series

s
n(n + 1)(2H + 1)n 2H (n+1)
H 1
converges, it follows from Lemma 5.2 that the Hausdorff dimension of Kn ( )

is less than or equal to 1/ .
In order to prove the reverse inequality, denote by ( j ) j1 the set An ordered as in Lemma 5.4. For any integer j 1, set
( j) = H( j ) (n+1) .
By Proposition 5.4 and Lemma 5.4, the triple (An , n , x x) is a regular
system, where

(5.14)
n (x) = c9 x (log 3x)n(n+1) (x 1),
for a suitable positive constant c9 , depending only on n. Further, by (5.13), we
have x(n (x)) 1 when x is large enough. Consequently, we infer from
Theorem 5.1 that
log x
log (n (x))
log x
1

= ,
lim inf
x log x c10 log log x
dim Kn ( ) lim inf

x
where c10 is a positive constant depending only on n. This completes the proof
that dim Kn ( ) = 1/ , and this implies that dim Kn (1) = 1.
As announced at the end of Section 5.4, we may use simultaneously innitely
many regular systems.
T HEOREM 5.5. Let (n )n1 be a sequence of real numbers greater than or
equal to 1. We then have
1
dim
Kn (n ) =
.
supn1 n
n1
In particular, for any real number 1, the Hausdorff dimension of the set
Kn ( )
n1
is equal to 1/ . Thus, there exist S -numbers of arbitrarily large -type.
110
P ROOF. The upper bound immediately follows from Theorem 5.3. For the
lower bound, we consider the regular systems (An , n , x x), with n as in
(5.14), and the functions
n : j H( j ) (n+1) .
We conclude by (5.8).
Theorem 5.5 asserts that the Hausdorff dimension of a countable intersection
of sets Kn (n ) is the inmum of the dimensions of the sets Kn (n ). In the next
Chapter, we rene this result by using the notion of intersective sets, introduced
by Falconer.
With a slight modication of the proof of Theorem 5.5, similar to that
suggested for getting Theorem 5.4, it is possible to prove that there exist S numbers with any prescribed -type greater than 1.
T HEOREM 5.6. For any real number 1, the Hausdorff dimension of the
set of real S -numbers of -type is equal to 1/ .
P ROOF. See Exercise 5.4.
Theorems 5.4 and 5.6 also follow from deeper results obtained in Chapter 6
(Theorem 6.3). It seems that Theorem 5.6 appeared for the rst time in [132].
However, it can be relatively easily deduced from the results of A. Baker and
Schmidt [45]. Theorems 5.4 and 5.6 remain true if the -type of an S -number
is dened as the supremum of the sequence ((wn ( ) + 1)/(n + 1))n1 .
5.7 Hausdorff dimension of sets of S-numbers

In the previous Section, we have calculated the Hausdorff dimension of sets of
real numbers related to Koksmas classication. A parallel with the classication of Mahler leads us to consider the sets
Kn ( ) = { R : |P( )| < H(P) (n+1)+1 for innitely many
integer polynomials P(X) of degree n},
for any integer n 1 and any real number 1. Obviously, the sets K1 ( )
and K1 ( ) coincide. By Theorem 4.1, the sets Kn ( ) have Lebesgue measure
zero when > 1, thus we would like to discriminate them using the Hausdorff
dimension. It follows from Lemma A.6 that the Hausdorff dimension of Kn ( )
is at least equal to that of Kn ( ); unfortunately, we cannot apply the Hausdorff
Cantelli Lemma 5.2 to bound it from above, for the same reasons as explained
in Section 4.1.
5.7 Hausdorff dimension of sets of S-numbers
111
A. Baker and Schmidt [45] combined Theorem 3.4 and Lemma 5.2 to show
that
1
2
dim Kn ( ) < ,
(5.15)
which, together with the upper bound dim K2 ( ) 1/ obtained by Kasch

and Volkmann [311], yield that dim K2 ( ) = 1/ (see Exercise 5.5). They
conjectured that the left inequality in (5.15) is indeed an equality. For n and
large enough, this was proved by R. C. Baker [48], and, a few years later, a
complete proof of the conjecture was established by Bernik [75, 77] (see also
[90]).
T HEOREM 5.7. For any n 1 and any 1, the Hausdorff dimension of the
set Kn ( ) is equal to 1/ .
The proof of the upper bound (unlike for Theorem 5.3, this is the most difcult
half) is involved and uses the method of essential and inessential domains, at a
technical level comparable with (and maybe slightly superior to) the proof of
Theorem 4.4. Without much effort, we deduce from Theorem 5.7 the analogues
of Theorems 5.4 and 5.6.
Wn ( ) := { R : wn ( ) = (n + 1) 1}
is equal to 1/ . Furthermore, for any real number 1, the Hausdorff dimension of the set of real S-numbers of type is equal to 1/ .
P ROOF. It follows from Theorem 5.7 that dim Wn ( ) 1/ . Write
Wn ( ) := { R : wn ( ) (n + 1) 1},
and observe that
Wn ( ) = Wn ( )
+

=1
Wn ( + 1/).
(5.16)
For any real numbers > 1 and with 0 < < 1, the set Wn ( + ) is
included in Kn ( ), hence, its Hausdorff dimension is at most equal to 1/ , by
Theorem 5.7. Consequently, we have H f (Wn ( + 1/)) = 0for any integer
n 1, where f is the dimension function f : x x 1/ exp( log 1/x). Furthermore, we infer from Proposition 3.2 and Exercise 5.3 that H f (Wn ( )) >
0, and we get H f (Wn ( )) > 0 by (5.16). Thus the Hausdorff dimension of
Wn ( ) is at least equal to 1/ , and the rst assertion of Theorem 5.8 is proved.
112
As for the second assertion, we observe that, for any real number 1, we
have
+
n=1
Wn ( )
Wn ( )
+
+

n=1 =1
n=1
Wn ( + 1/).
Arguing as above with the same dimension function f , we infer from Proposition 3.2 and Exercise 5.4 that H f (n1 Wn ( )) > 0. Thus, the Hausdorff
dimension of n1 Wn ( ) is at least equal to 1/ and the proof of Theorem 5.8 is complete, since any element of n1 Wn ( ) is an S-number of
type .
The last assertion of Theorem 5.8 remains true if the type of an S-number is
dened as the supremum of the sequence ((wn ( ) + 1)/(n + 1))n1 .
We end this Section by introducing a new exponent of Diophantine approximation, related to multiplicative approximation. Let > 1 be a real
number and let n 1 be an integer. For an integer polynomial P(X ) =
an X n + . . . + a1 X + a0 we introduced in Chapter 4 the quantity

+ (P) :=
max{1, |ai |}.
1in
wn+ ( )
be the supremum of the real numbers w

For a given real number , let
for which there exist innitely many integer polynomials P(X ) of degree at
most n satisfying
0 < |P( )| w
+ (P).
Observe that we have wn+ ( ) wn ( )/n. By Proposition 3.1, this implies that
wn+ ( ) 1 if is transcendental. Furthermore, we set
+1
(P) for innitely many
Kn+ ( ) = { R : |P( )| < 2
+
integer polynomials P(X) of degree n},
and
Wn+ ( ) := { R : wn+ ( ) = 2 1}.
By Proposition 3.1 and Theorem 4.6, almost all real numbers satisfy
wn+ ( ) = 1, thus the Lebesgue measure of Kn+ ( ) is zero. Clearly, Kn+ ( ) contains the sets K1 ( ), . . . , Kn ( ), therefore its Hausdorff dimension is greater
than or equal to 1/ , by Theorem 5.2. Yu [608] conjectured that this is in fact
the exact value. The conjecture for n = 1 is nothing but Theorem 5.2 and it has
been established by Yu [608] for n = 2 (see Exercise 5.6). This result has been
rediscovered by Beresnevich and Bernik [64], but it remains an open problem
5.8 Restricted Diophantine approximation
113
for n 3. Furthermore, the Hausdorff dimension of W2+ ( ) is equal to 1/ ,

as proved in Exercise 5.6.
5.8 Restricted Diophantine approximation

The sets K1 ( ) are dened without any restriction on the rational approximants
p/q. We could however impose some conditions on them, for instance, we
could ask that q belongs to a given innite subset of the integers, like the
prime numbers or the powers of 2, etc. Such questions have been considered
by Borosh and Fraenkel [112], who extended earlier results of Eggleston [225],
and established the following statement.
T HEOREM 5.9. Let N be an innite set of distinct positive integers, and let v
in [0, 1] be the real dened by the conditions

q v diverges and
q v converges for any > 0.
qN
qN
For any real number (1 + v)/2, the Hausdorff dimension of the set

1
p
p

R : < 2 for innitely many rationals with q N
q
q
q
is equal to (1 + v)/(2 ).
The proof of Theorem 5.9 given in [112] is quite technical and is inspired by
the proof of Theorem 5.3 due to A. Baker and Schmidt [45]. It proceeds by
contradiction and does not use the mass distribution principle (see also Chapter 10 of Harman [273]).
In particular, Theorem 5.9 asserts that, for any real number > 3/2, the
Hausdorff dimension of the set of real numbers such that innitely many rational numbers p/q satisfy | p/q 2 | < q is equal to 3/(2 ). Furthermore,
it allows us to answer a conjecture of Erdos related to the convergents of the
continued fraction expansion of real numbers.
C OROLLARY 5.5. Let N be an innite set of distinct positive integers. The
Hausdorff dimension of the set of real numbers having innitely many convergents whose denominators belong to N is greater than or equal to 1/2.
The proof is immediate since any convergent pn /qn of the continued fraction expansion of the real number satises | pn /qn | < qn2 . Actually,
Egglestons results [225] are sufcient to get Corollary 5.5.
We end this Section by stating a corollary to Theorem 5.9 and a result,
due to Harman ([273], Theorem 10.8), on approximation by rational numbers
whose denominators and numerators are restricted.
114
T HEOREM 5.10. Let 1 be a real number. The Hausdorff dimensions of

the sets

p
1
p

R : < 2 for innitely many rationals with q prime
q
q
q
and

p
1
p
R : < 2 for innitely many rationals with p
q
q
q

and q prime
are equal to 1/ .
Restrictions on numerators introduce new difculties which do not seem to be
easy to overcome.
5.9 Exercises
E XERCISE 5.1. The Hausdorff dimension of the set B of badly approximable
real numbers, following Jarnk [288].
We keep the notation of Chapter 1. Let M 2 be an integer and denote by
B M the set of real numbers in [0, 1] having their partial quotients at most equal
to M. Let s be a real number with 0 < s < 1.
1) To determine Hs (B M ), prove that we need only to consider nite coverings U of B M composed of closed intervals having their endpoints in B M .
2) Let U be an interval belonging to U. Prove that there exist a m-tuple
k = (k1 , . . . , km ) of positive integers not greater than M and two integers h
and with 1 h < M such that
U F(k,h) = and U F(k,) = .
Deduce that (Fk ) 4 M 3 |U |.
3) In order to determine whether Hs (B M ) is positive or not, prove that we
need only consider nite coverings U of B M composed of closed intervals Fk ,
where k is an m-tuple of positive integers not greater than M for some positive
integer m.
4) Assume that s is such that for any integer m 0 and any m-tuple k of
positive integers not greater than M we have
(Fk )s
M

h=1
(F(k,h) )s .
5.9 Exercises
115
Prove that this implies that the Hausdorff dimension of B M is at least equal
to s.
5) Use 4) to prove that dim B2 > 1/4 and that, for any integer M 9, we
have dim B M > 1 4/(M log 2).
6) Prove that we have dim B M < 1 1/(8M log M) for any integer M 9.
E XERCISE 5.2. Denote by [0; a1 , a2 , . . . ] the continued fraction expansion of
a real number in [0, 1[. Let b > 1 and c > 1 be real numbers and set
(b, c) := { : an cb
for innitely many n}
and
n
(b,
c) := { : an cb for all positive integers n}.
1) Following Feng, Wu, Liang and Tsen [245], prove that we have
dim (b, c) = dim (b,

c) 1/(b + 1).
2) Following Ganesa Moorthy [255], prove that dim (b, c) 2/(b + 1)
and deduce that the set
U = { : the sequence (log an )1/n is unbounded}
has Hausdorff dimension zero.
Actually, both sets (b, c) and (b,

c) turn out to have Hausdorff dimension 1/(b + 1), as proved by uczak [372].
Hint. For 1), apply Proposition 5.2 with the sets
n
E k := { [0, 1] : cb an 3cb
for 1 n k + 1}
and use Theorems 1.3 and 1.7.

For 2), let k < b be xed. Arguing by contradiction, prove that for every =
n
[0; a1 , a2 , . . . ] with an cb for innitely many n, there are innitely many n
such that the n-th convergent pn /qn to satises qn+1 > qnk . Deduce that for
every positive integer m, the collection of intervals ]r/qq k1 , r/q+q k1 [,
with 1 r q 1 and q m, forms a cover for the set (b,

c). Conclude.
Prove that dim U = 0.
E XERCISE 5.3. Prove that dim Wn ( ) = 1/ holds for any positive integer n
and any real number > 1.
Hint. Use the regular system (An , n , x x) with n as in (5.14)
to construct inductively a Cantor set contained in Kn ( ) and of the same
form as in the proof of Theorem 5.1. Instead of applying Proposition 5.2,
116
argue as in the proof of Proposition 5.2

to dene a measure on Kn ( )
1/
exp( log 1/|U |) for some absolute conand show that (U ) c|U |
stant c and any sufciently
small
interval
U . Use Lemma 5.3 to prove that

for f : x x 1/ exp( log 1/x), we have H f (Kn ( )) > 0. Observe that
H f (Kn ( +1/k)) = 0 for any positive integer k and prove that H f (Wn ( )) >
0. Conclude.
E XERCISE 5.4. Prove that dim n1 Wn ( ) = 1/ holds for any positive integer n and any real number > 1.
Hint. Use the observation at the end of Section 5.4 and the same general
idea as in Exercise 5.3 to show that, with f as in Exercise 5.3, we have
H f (n1 Wn ( )) > 0.
E XERCISE 5.5. Following Kasch and Volkmann [311], prove that dim K2 ( ) =
1/ .
Hint. Use 1) of Exercise 4.1 to nd a suitable countable covering of K2 ( )
and then use 2) of Exercise 4.1 to prove that dim K2 ( ) 1/ .
E XERCISE 5.6. Following Beresnevich and Bernik [64], prove that
dim K2+ ( ) = 1/ . Proceed as in Exercise 5.3 to establish that dim W2+ ( ) =
1/ .
Hint. It is sufcient to show that dim(K2+ ( ) [1, 2]) = 1/ . Let be in
[1, 2] and P(X ) = a2 X 2 + a1 X + a0 be an integer polynomial with |P( )| <
+1
(P).
2
+
Prove that H(P) 7 max{|a1 |, |a2 |}.
Let J ( ) denote the subset of K2+ ( ) composed of the real numbers in
[1, 2] for which there are innitely many integer polynomials Q(X ) = a2 X 2 +
a1 X + a0 with
|Q( )| < + (Q)2 +1 and |Q ( )| <
H(Q)
.
14
Prove that each of these polynomials satises |a2 | |a1 | 8|a2 |, and derive
from Exercise 5.5 that dim J ( ) 3/(4 1).
Assume now that there are innitely many integer polynomials P(X ) =
a2 X 2 + a1 X + a0 with
|P( )| < + (P)2 +1 and |P ( )|
H(P)
.
14
(5.17)
For a xed polynomial P(X ), denote by (P) the set of in [1, 2] satisfying
(5.17). Prove that (P) is the union of at most three intervals, whose lengths
5.10 Notes
117
are at most 28 + (P)2 +1 H(P)1 (use the Mean Value Theorem). For any
in ]1/, 1[, show that the sum

| i (P)|
P(X ) 1i3
converges, where the rst sum is taken over all the integer polynomials
P(X ) = a2 X 2 + a1 X + a0 (sum up over k and , and then over |a1 | in
[2k , 2k+1 1] and |a2 | in [2 , 2+1 1]). Conclude.
E XERCISE 5.7. Prove that, for any real irrational number and any > 1, the
Hausdorff dimension of the set

1
R : n < holds for innitely many integers n
n
is equal to 1/ , a result due to Bugeaud [127] and, independently, to Schmeling
and Troubetzkoy [497].
Hint. The Three Distance Theorem asserts (see, for example, [529], [525]
or [5] 1 ) that, for any integer N 2, the points {}, {2}, . . . , {N } divide
the interval [0, 1] in N + 1 intervals, whose lengths take at most three distinct
values, one of these being the sum of the two others. Deduce that there exists
a strictly increasing innite sequence of positive integers (Nr )r 1 with the
following property: For any interval I in [0, 1] and any integer r large enough
(in terms of |I |), at least |I |K r /3 numbers among {}, . . . , {Nr } belong to I
and are mutually distant by at least 1/K r , where K r denotes the greatest even
integer smaller than (Nr + 1)/3. Conclude by applying Theorem 5.1 with a
suitably chosen function . As observed by Drmota, it is also possible to use
Theorem 1.7 instead of the Three Distance Theorem to solve this exercise.
5.10 Notes
Besides the Hausdorff dimension, there exist various other fractal dimensions, like the lower and upper box-counting dimensions, the Fourier dimension and the packing dimension introduced by Tricot [571, 572], see Chapter 3
of [236] and Chapters 5 and 12 of [415]. Falconer [237] proved that, for any
1, the set K1 ( ) has packing dimension equal to 1.
Theorem 5.2 has been extended to systems of linear forms by Bovey and
Dodson [116].
1 There is a misprint in the statement of the Three Distance Theorem in [5]: is the smallest
k
length, not the largest one.
118
An inhomogeneous analogue of Theorem 5.2 together with an extension

to systems of linear forms has been obtained by Levesley [365] (see also Exercise 6.3 and [497]). His main theorem implies earlier results of Dickinson
[192] and Dodson [202].
Kaufman [315] (see also Bluhm [102]) gave a new proof of Theorem 5.2.
He established the stronger property that, for any real number > 1, there
exists a positive measure with support in a compact subset of K1 ( ) and
whose Fourier-Stieltjes transform satises

(x) =
exp(2i x y) d (y)
R
= o(log |x|) |x|1/(2 ) , |x| +.

Furthermore, he showed that the Fourier dimension (see [415], page 168, for
denition) of K1 ( ) is equal to its Hausdorff dimension. Sets with this property
are called Salem sets and are very rare as deterministic sets; the rst constructions of such sets, due to Salem, were random.
Another extension of Theorem 5.2 has been obtained by Falconer [238].
Let k 1 be an integer, > (k + 1)/2 be a real number and set

pk
1
p1
(k)
k

. . .
<
K ( ) := (1 , . . . , k ) R :
q
q q 2

for innitely many integers p1 , . . . , pk , q .
Bovey and Dodson [115] established that dim K(k) ( ) = k 1+2/(2 k +1).
This provides a generalization of Theorem 5.2, and has been extended to
systems of linear form by Yu [607]. A function F : Rk R is a distance function if, by denition, F is continuous, non-negative and satises
F(t x) = t F(x) for all t 0. Examples of distance functions are provided by the sup norm, the usual Euclidean norm, and by (x1 , . . . , xk )
k
1/k . Sets of the form {( , . . . , ) Rk : F(q x , . . . , q x ) <
k
k
1
1
j=1 |x j |
(q) for innitely many integers q}, where F is a distance function and
some approximation function, have been studied by Dodson [198, 199], who
obtained generalizations of the above quoted result from [115]. Extensions to
systems of linear forms have been worked out by Dodson [198] and Dodson
and Hasan [204].
In order to obtain lower bounds for the Hausdorff dimension of sets of
very well approximable points in a multidimensional setting, Dodson, Rynne,
and Vickers [206], inspired by geometrical ideas of Besicovitch [100], introduced the notion of ubiquitous systems. Actually, in the one dimension case,
5.10 Notes
119
ubiquitous and regular systems are almost equivalent, see Rynne [487] for an
interesting discussion. Using this powerful tool, Theorem 5.2 has been extended to very well approximable linear forms and to very well approximable systems of linear forms in [206] and by Dickinson [190, 191]. We
refer the reader to the survey of Dodson [201], to Chapter 5 of Bernik and
Dodson [86], and to the survey of Beresnevich, Bernik and Dodson [67]. A
new technique involving a multidimensional analogue of regular systems has
been developed by Beresnevich, Bernik, Kleinbock, and Margulis [68].
Weiss [596, 597] proved that, for any real number > 1, almost no points
on a Cantor set (with respect to the standard measure supported on it) belong to
K1 ( ). This has been extended by Kleinbock, Lindenstrauss, and Weiss [329].
Pollington [458] established that the Hausdorff dimension of the set of
real numbers in Kn ( ) which are simply normal (see the Notes of Chapter 1
for the denition) in no base is equal to 1/ .
Hinokuma and Shiga [281] slightly improved Theorem 5.2. Furthermore,
they [282] asserted an extension of Theorem 5.9, but there is apparently a problem in their proof: the term (log N j1 )3 occurring in the denominator of the
fraction at the bottom of page 371 should be (log N j )3 , and this only gives
that the Hausdorff dimension is non-negative. Presumably, Proposition 5.2 cannot be used to get Theorem 5.9, since the points of the regular system are not
sufciently spaced from each other. However, a proof of the assertions claimed
in [282] can be found in [488].
Theorem 5.9 has been extended to simultaneous approximation and then
generalized in several directions by Rynne [486, 488, 489] and by Rynne and
Dickinson [491]. Simultaneous approximation with differing exponents in the
approximation functions for each coordinate is studied in [489, 491].
As a corollary to Theorem 5.9, R. C. Baker and Harman (see Theorem
10.7 in [273]) proved that, given a non-negative function such that the sum

q1 q(q) diverges, the set of real numbers for which | p/q| < (q)
has innitely many solutions in coprime positive integers p and q has Hausdorff dimension 1. This result means that the DufnSchaeffer Conjecture (see
Section 1.4) cannot fail badly.
Vilchinski [577] and Vilchinski and Dombrovski [578] obtained various
metric results in restricted Diophantine approximation.
It follows from Exercise 5.1 and the fact that H1 coincides with the
Lebesgue measure that the set B of badly approximable real numbers is not
an s-set.
120
The results obtained in Exercise 5.1 have been rened by several authors, including Bumby, Cusick, and Hensley. Among other statements, Hensley [277] established that
dim B M = 1
6
2M
72 log M
+ O(M 2 ),
4 M2
as M tends to innity. Further, he proved [278] that dim B2 = 0.5312805...

and gave the rst six signicant digits for the Hausdorff dimensions of some
other sets of real numbers with nitely many different partial quotients. A more
efcient approach to calculating these dimensions has been developed by Jenkinson and Pollicott [301]. Vallee [576] determined the Hausdorff dimension of
sets of real numbers whose continued fraction expansion obeys a not necessarily nite but periodic set of constraints.
Schmidt [503, 505] extended some results of Jarnk [288] by showing that,
for any positive integers m and n, the Hausdorff dimension of the set of badly
approximable m n real matrices is equal to mn. Continued fractions are useless for this problem; Schmidts proof rests on the method of - games (see
also Chapter III of [512]). An inhomogeneous analogue of Schmidts results
has been obtained by Kleinbock [325], using the methods from [330].
The Ganesa Moorthy theorem, proved in Exercise 5.2, contains as an
immediate corollary a result of Cusick [172] asserting that dim{ R :
log2 log2 log2 an n for all n 1} = 0. In the same vein, Hirst [283] showed
that dim{ R : an a n for all n 1} = 1/2. Good [258] claimed that the
Hausdorff dimension of U is zero, but, as noticed, for example, by Hirst [283],
there is a gap in his proof. For related results, see Jarnk [298] and Ramharter
[467, 468].
Extensions of the result proved in Exercise 5.7 to simultaneous approximation and to approximation of linear forms have been studied by Bugeaud
and Chevallier [134].
As for restricted Diophantine approximation, denoting by pn /qn the convergents of a real number, Erdos and Mahler [229] asserted without proof that
there exist real numbers for which the greatest prime factor of both pn and qn
is bounded for innitely many n. They also conjectured that these numbers,
which are necessarily transcendental (by Ridouts Theorem 2.3), are Liouville numbers. The claim of Erdos and Mahler has been proved by Fraenkel
[249]. However, their conjecture remains open. Fraenkel and Borosh [250]
have determined the Hausdorff dimension of sets of real numbers very close
to innitely many rational numbers, whose numerators and denominators are
5.10 Notes
121
mainly composed of xed prime numbers. Furthermore, they have generalized

their result to simultaneous Diophantine approximation [113].
For Hausdorff dimension of exceptional sets of real n-tuples related to ndimensional extensions of Dirichlets Theorem 1.1 established by Davenport
and Schmidt [184], see R. C. Baker [51] and the references given therein.
Let n 2 be an integer and w1 , w2 be real numbers with w1 w2 1
and w1 + w2 > n + 1. Using planar regular systems, Dombrovski [207]
computed the Hausdorff dimension of the set of vectors (1 , 2 ) in the plane
for which the inequalities |1 1 | < H(P)w1 , |2 2 | < H(P)w2 , hold
for innitely many integer polynomials P(X ) of degree at most n, where 1
and 2 are two zeros of P(X ). Further results are due to Kovalevskaya [338].
The Hausdorff dimension of sets of points (1 , 2 ) in R2 for which the
inequality |P(1 )| |P(2 )| < H(P)w has innitely many solutions in integer
polynomials P(X ) of bounded degree has been studied by Pereverzeva [451,
452].
There are many papers concerned with Diophantine approximation on
classical curves and Hausdorff dimension. The reader is directed to Bernik and
Dodson [86] and to the survey by Beresnevich and Bernik [65]. Recent results
are due to Beresnevich [63], Dickinson and Dodson [193], and Rynne [490].
For applications of results connected with this Chapter to the problem of
small denominators or to the Schrodinger equation, see Chapter 7 of Bernik
and Dodson [86], Section 7 of Dodson and Kristensen [205], the survey by
Dodson [203], and Kristensen [343].
6
Deeper results on the measure
of exceptional sets
Theorem 5.4, due to A. Baker and Schmidt [45], asserts that for any integer n
1 and any real number 1 the Hausdorff dimension of the set W n ( ) of real
numbers with wn ( ) = (n + 1) 1 is equal to 1/ . In the present Chapter,
we are concerned with various renements, including the determination of the
Hausdorff measure of Wn ( ) at the critical exponent (Corollary 6.3 below).
There are essentially two new ingredients. On the one hand, we need an
improvement of Proposition 5.4, which is due to Beresnevich [61] and asserts
that real algebraic numbers of bounded degree are distributed as evenly as they
could be. On the other hand, we present a rened analysis of the Hausdorff
measure of sets of real numbers close to innitely many points in a given real
sequence.
One essential tool, introduced in Section 6.1, is the notion of optimal
regular systems (also termed best possible regular systems by Beresnevich,
Bernik, and Dodson [67]). We state four general results on sets of real numbers
close to innitely many points in an optimal regular system. We establish
the rst one in Section 6.2, which allows us to give an alternative proof of
(a slightly stronger form of) Khintchines Theorem 1.10. The second one,
stated in Section 6.3, provides the Hausdorff dimension of general exceptional
sets. The third one, given in Section 6.4, renes Theorem 5.1 inasmuch as it
yields the Hausdorff measure instead of the Hausdorff dimension. Finally, in
Section 6.5, the fourth one shows that the sets of real numbers considered here
share a stability property. Namely, the Hausdorff dimension of a countable
intersection of a family (K j ) j1 of such sets is equal to the minimum of
the Hausdorff dimensions of the K j s. Section 6.6 is devoted to applications
of the general statements obtained in the previous Sections to Diophantine
approximation.
We stress that (apart from Theorems 6.1 and 6.2) none of the results stated
in this Chapter are fully proved in the present book. Theorems 6.6 to 6.10 rest
122
6.1 Optimal regular systems
123
on Proposition 6.1, a difcult result on the distribution of algebraic numbers of

bounded degree. Furthermore, proofs of Theorems 6.3 to 6.5 are quite technical
and lie beyond the scope of this book: the reader is directed to [132, 130] and
to the impressive work of Beresnevich, Dickinson, and Velani [71].
6.1 Optimal regular systems
Our aim is to give more accurate information on sets KS of real numbers close
to innitely many points belonging to a given countable set S. In Chapter 5, we
have seen that if the points in S are evenly distributed in some sense, then we
are able to compute the Hausdorff dimension of KS . However, we would like
to get more, in particular to obtain Khintchine-type results and to determine
at the critical exponent. It appears that we then
the Hausdorff measure of KS
need rened information on the distribution of the points in S, whence the
introduction of the notion of optimal regular system.
D EFINITION 6.1. Let E be a bounded open real interval. Let S = ( j ) j1
be a sequence of distinct real numbers. Then S is called an optimal regular
system of points in E if there exist positive constants c1 , c2 , and c3 , depending
only on S, and, for any bounded interval I in E, a number K 0 = K 0 (S, I )
such that, for any K K 0 , there exist integers
c1 K i 1 < . . . < i t K
with i h in I for h = 1, . . . , t,
|i h i |
c2
K
(1 h = t)
and
c3 |I |K t |I |K .
As in the denition of a regular system, we index the set S and we do not
assume that all points of S belong to E.
Ideally, when the points of the set S are very well distributed, then S is
likely to be an optimal regular system. This is the case for the set of rational
numbers, as follows from Proposition 5.3, and for every set An of real algebraic
numbers of degree at most n, as shown by Beresnevich ([60] for n 2 and [61]
for general n). Observe that Proposition 5.4 is not sharp enough to yield this
assertion. We do not give the proof of the result from [61], since it lies beyond
the scope of the present book. It rests, among others, on a difcult lemma
due to Bernik [79] (extended by Borbat [109]) and auxiliary results (see, for
example, [86], Section 2.4) used in the proofs of Theorems 4.4 and 5.7. One of
these lemmas asserts that, for any given positive real number and for almost
124
all real numbers , all but nitely many integer polynomials P(X ) of degree
n and with |P( )| H(P)n satisfy |P ( )| > H(P)1 ; a statement which
may be compared with Theorem 4.2. We quote Theorem 3 of [61].
P ROPOSITION 6.1. Let n 1 and M 2 be integers and let I be an interval
contained in ] M +1, M 1[. There exist positive constants c4 , c5 , depending
only on n, a number K 0 = K 0 (n, I ) and, for any K K 0 , there are 1 , . . . , t
in An I such that
c4 M n K H(h ) M n K ,
|h | K
n1
t c5 |I |K
(1 h t),
(1 h < t),
n+1
Actually, the existence of c4 is not proved in [61]. However, it is not difcult

to deduce it by following the proof of Beresnevich (see, for example, [124],
Theorème G). The constants c4 and c5 , but not K 0 (n, I ), are explicitly given in [124]. It follows from (5.6) of Chapter 5 that we can take
K 0 (1, I ) = 100|I |1 log(100|I |1 ) for any interval I in [0, 1]. Furthermore,
Beresnevich [60] showed that Proposition 6.1 holds true with K 0 (2, I ) =
48|I |1 log(24|I |1 ) for any interval I in [0, 1]. When n is greater than
or equal to 3, an effective value for K 0 (n, I ) has not been calculated at
present.
C OROLLARY 6.1. We number the elements of An := ( j ) j1 by increasing
order of their height, and, when the heights are equal, by increasing numerical
order. Then, the set An is an optimal regular system in any bounded open real
interval.
Corollary 6.1 immediately follows from Proposition 6.1 and Lemma 5.6. We
draw the readers attention to the assumption that E has to be bounded. Indeed,
otherwise the constants c2 and c3 occurring in Denition 6.1 would have to
depend on I .
Actually, Beresnevich [61] proved that the set of real algebraic numbers of
xed degree n is an optimal regular system. Consequently, all the following
results also hold with the set An replaced by the set of real algebraic numbers
of degree exactly n.
Other examples of optimal regular systems include the set of real algebraic
integers of any xed degree n 2, the set of real algebraic units of any xed
degree n 3, the set of real algebraic integers in a given (non-totally complex)
number eld, and the sequence ({n})n1 for any badly approximable real
number (see Exercise 6.4).
6.2 A Khintchine-type result
125

The aim of this Section is to show that sets of real numbers close to innitely
many points in an optimal regular system satisfy a Khintchine-type statement.
This result is due to Beresnevich [61, 62].
T HEOREM 6.1. Let E be a bounded open real interval. Let S = ( j ) j1
be a sequence of real numbers which is an optimal regular system in E. Let
: R1 R>0 be a non-increasing continuous function and set
KS
() = lim sup { E : | j | < ( j)}.
j+
() is a null set if the sum

Then the set KS
full measure if this sum diverges.
j1 ( j)
converges, and it has
We need several auxiliary lemmas to prove the divergence half of Theorem 6.1.
Here, we closely follow Beresnevich [61].
L EMMA 6.1. Let E be a measurable real set. If there exists an absolute positive constant such that (E I ) (I ) for any bounded real interval I ,
then E has full measure.
P ROOF. If the set R \ E has positive measure, then, by the Lebesgue Density
Theorem (see, for example, Riesz and Sz.-Nagy [472], page 13, or Mattila
[415], Corollary 2.14), there exists a real number x0 such that, for any > 0,
there is > 0 with

(R \ E) [x0 , x0 + ]) 2(1 .
Taking = /2, this yields (E [x0 , x0 + ]) , a contradiction with
our assumption. Consequently, E has full measure.
Lemma 6.2 provides a converse of the BorelCantelli Lemma 1.2. It originates
in a work of Chung and Erdos [165].
L EMMA 6.2. Let E i , i 1, be measurable real sets contained in a bounded

(E i ) diverges, then we have
interval I . If the sum i=1
n
2
i=1 (E i )
n
.
i=1
j=1 (E i E j )

lim sup E i lim sup n
i+
n+
126
P ROOF. We follow Sprindzuk [539], p. 18, and Harman [274]. Let m, n be

positive integers with m n and set

E mn = nj=m E j , M(m, n) = nj=m (E j ),
and
V (m, n) =
n
n

(E j E k ).
j=m k=m
Let m,n denote the characteristic function of E mn and j denote that of E j for
any positive integer j. Since

n
n

j (x) dx =
m,n (x)
j (x) dx,
M(m, n) =
I j=m
j=m
the CauchySchwarz inequality yields that

n
n
m,n (x) dx
j (x) k (x) dx
M(m, n)2
I
I j=m k=m
(6.1)
(E mn ) V (m, n).
Since

M(m, n) M(m, n) + 2m|I | + m 2 |I |2
M(1, n)2
V (1, n)
V (m, n)

and the sum i=1 (E i ) diverges, we let n tend to innity and keep m xed to
obtain
lim sup
n
M(m, n)2
M(1, n)2
lim sup
.
V (m, n)
V (1, n)
n
Combined with (6.1), this gives

M(1, n)2
,
lim E mn lim sup
n
V (1, n)
n
and we now let m tend to innity to get the lemma.
L EMMA 6.3. Let (an )n1 be a non-increasing sequence of positive numbers

such that the sum n1 an diverges. Let be a positive real number and,
for n 1, set bn := min{an , /n}. Then the sequence (bn )n1 is also non
increasing, and the sum n1 bn diverges.
P ROOF. Since (an )n1 and (/n)n1 are non-increasing, so is (bn )n1 . Thus,
for any integer 2, we have
b 2

n=[/2]
bn .
(6.2)
127

Assume that the sum n1 bn converges. It follows from (6.2) that b tends

to 0 as tends to innity. However, since n1 an diverges, the denition of
(bn )n1 yields that there exist innitely many integers such that b = /,

which is a contradiction. Hence, the sum n1 bn diverges.
L EMMA 6.4. Let : R>0 R>0 be a given non-increasing, continuous

function such that the sum h1 (h) diverges (resp. converges). Let and

M be positive real numbers with M > 1. Then, the sum h1 M h ( M h )
also diverges (resp. converges).
P ROOF. This follows from the equality

M x ( M x )dx =
1
1
log M
(y)dy,
which is nothing but the change of variable y := M x .

We now turn our attention to the proof of Theorem 6.1.
PROOF OF THEOREM 6.1. The convergence half is a straightforward application of Lemma 1.2, thus we omit the details. We assume that the sum

j1 ( j) diverges. Let c6 and c7 denote the second and the third positive constants given by Denition 6.1 applied with S. For any j 1, set

j)
j) := min{( j), c6 /(3 j)}. By Lemmas 6.3 and 6.4, the sums j1 (
(

k
k
and k1 2 (2 ) also diverge. Let I be a bounded interval in E and let

K 0 = K 0 (S, I ) be given by Denition 6.1. For any integer k K 0 , there is a
collection Ak (I ) := {i 1 , . . . , i t } of distinct integers such that
1 i 1 < < i t 2k ,
c6
|i h i | k (1 h < t),
2
c7 |I |2k t |I |2k .
For any j in Ak (I ), set
k )}
E k ( j ) := { I : | j | < (2
and dene
Ek =
E k ( j )
and
E(I ) = lim sup E k .
jAk (I )
k+
k ) (2k ) for any positive integer k and since (by Lemma 6.3)
Since (2
is non-increasing, we get

(6.3)
(E(I )) KS () I ,
and we now establish a lower bound for (E(I )).
128
To this end, we observe that for distinct integers j and h in Ak (I ), the

intervals E k ( j ) and E k (h ) are disjoint, since | j h | c6 2k and c6
It follows that
k ), by the denition of .
2k+1 (2
k ) (E k ) 2k+1 |I |(2
k ),
2k+1 c7 |I |(2
(6.4)
whence we get that

(E k ) = +.
(6.5)
k ) > 1,
2k (2
(6.6)
kK 0
Let N0 be such that we have

N0

k=K 0
and let k, , and N with K 0 k < N and N N0 .

For an integer j in Ak (I ), the number of distinct integers h in A (I ) such
that E (h ) and E k ( j ) have non-empty intersection is at most
2+
k)
2(2
k ),
= 2 + 2+1 c61 (2
c6 2
whence
k )).
)(1 + 2 c1 (2
(E E k ( j )) 4(2
6
(6.7)
Since the cardinality of Ak (I ) is less than or equal to 2k |I |, we infer from (6.7)

that

) + c1 2k+ (2
k )(2
) .
(E E k ) 4|I | 2k (2
(6.8)
6
We are now ready to estimate from above the double summation
N
N

(E E k ) =
=K 0 k=K 0
N
N
(E k ) + 2
k=K 0
1
(E E k ). (6.9)
=K 0 +1 k=K 0
We rst observe that

N
(E k ) 2|I |
k=K 0
N

N
k ) 2|I |
2k (2
k=K 0
k)
2k (2
2
,
by (6.6) and N N0 . Furthermore, we notice that

2
1
N
N
N

k

k k
2 (2 )
2 (2 )
2 (2 ) ,
=K 0 +1 k=K 0
=K 0 +1
(6.10)
k=K 0
k=K 0
(6.11)
6.3 Hausdorff dimension of exceptional sets
129
again by (6.6). Finally, we check that

N
1
k )(2
)
2k+ (2
=K 0 +1 k=K 0

N
k)
2k (2
2
,
(6.12)
k=K 0
and we deduce from (6.4) and (6.8)(6.12) that

N
( k=K
(E k ))2
(2c7 |I |)2
0
=: c8 |I |.
N
N
(10 + 8c61 )|I |
=K 0
k=K 0 (E E k )
(6.13)
By (6.3), (6.5), (6.13), and Lemma 6.2, we get

KS () I (E(I )) c8 |I |,
for any bounded interval I in E. Thus, Lemma 6.1 asserts that KS () has full
measure, as claimed.
We observe that the method of the proof of Theorem 6.1 works as well with a
slightly weaker denition of regular system: indeed, the existence of the constant c1 in Denition 6.1 is not needed. This remark also applies for Theorem
6.2, but it does not seem to hold for Theorem 6.3.
6.3 Hausdorff dimension of exceptional sets

Clearly, with the notation of Theorem 6.1, if 1 decreases more rapidly than
2 , then the set KS (1 ) is smaller than the set KS (2 ). The aim of this Section is to determine their Hausdorff dimensions. An indication of how a function g : R1 R>0 grows near innity is provided by its lower order at
innity (g).
D EFINITION 6.2. The lower order at innity (g) of a function g : R1
R>0 is dened by
(g) = lim inf
x+
log g(x)
.
log x
This notion arises naturally in the theory of Hausdorff dimension, see, for example, Dodson [200] and Dickinson [192]. Clearly, (g) is non-negative when
limx+ g(x) = +.
T HEOREM 6.2. Keep the notation of Theorem 6.1, and assume that the sum

j1 ( j) converges. Let denote the lower order at innity of the function
() is equal to 1/.
1/. Then the Hausdorff dimension of the set KS
Observe that the assumption of Theorem 6.2, whose proof is left as Exercise
6.1, implies that is at least equal to 1.
130
6.4 Hausdorff measure of exceptional sets

In the previous Section, we determined the Hausdorff dimensions of the sets
KS (); we are now interested in calculating their Hausdorff measures.
T HEOREM 6.3. Keep the notation of Theorem 6.1. Let f be a dimension function such that limx0 f (x)/x = + and x f (x)/x decreases in a neighbourhood of the origin. Assume that x x f (2(x)) is non-increasing and
tends to 0 when x goes to innity. Then we have

f (2( j)) diverges
H f (KS ()) = + if
j1
and
H f (KS ()) = 0
if

f 2( j) converges.
j1
We point out that Theorem 6.3 does not imply Theorem 6.2.
Since x f (x)/x decreases in a neighbourhood of the origin,
the sum

f
(
j)
converges
if,
and
only
if,
the
sum
f
2(
j)
converges.
j1
j1
The latter sum arises naturally in the proof of Theorem 6.3, that is the reason
why it occurs in its statement, as in Satz 4 of Jarnk [292].
Theorem 6.3, proved in [71] and in [130], renes Theorem 6.1. A straightforward application of the HausdorffCantelli Lemma 5.5 yields the convergence half, while the divergence half is quite technical and much more difcult
to establish. The general idea is however easy to explain and originates in
Jarnks work [292]: we construct inductively a Cantor set as large as possi () and to which we apply the full power of the Frostman
ble contained in KS
Lemma 5.3.
6.5 Sets with large intersection properties

Several authors have introduced large classes of sets of Hausdorff dimension at
most s which turn out to have the property that countable intersections of such
sets also have dimension at most s. Such sets share a large intersection property
in an s-dimensional sense. Examples include the regular sets of A. Baker and
Schmidt (see Theorem 5.4), the Ms -sequences of Rynne [487], constructions
using the ubiquitous systems of Dodson, Rynne, and Vickers [206] and the
Ms -dense construction of Falconer [235]. These constructions have been
unied by Falconer [237], who gave a more direct denition of classes G s
of subsets of Rn with a large intersection property. His theory can be easily
extended to general dimension functions [132].
6.6 Application to the approximation by algebraic numbers
131
D EFINITION 6.3. Let f be a dimension function concave on a neighbourhood

of the origin. We dene G f (R) to be the class of real G -sets F such that
+

Hg i=1
f i (F) = +
for any dimension function g with g f and any sequence of similarity trans+
.
formation { f i }i=1
T HEOREM 6.4. The class G f (R) is closed under countable intersections and
under bi-Lipschitz transformations on R.
T HEOREM 6.5. We keep the notation and hypotheses of Theorem 6.3. We furthermore assume that S is an optimal regular system in any bounded open real
interval. If the sum

f 2( j)
j1
diverges, then the set

lim sup { R : | j | < ( j)}
j+
belongs to the class G f (R).

Theorem 6.5, proved in [132], does not imply Theorem 6.3, and vice-versa. It
yields some results which seem to be out of reach with the methods used in
the proof of Theorem 6.3. Both the stability by countable intersection and the
stability by bi-Lipschitz transform are important properties of G f (R). We refer
the reader to Falconer [235, 236, 237] for more details, and to the next Section
for applications in the context of Diophantine approximation; most of which
have been pointed out in [235, 236, 237].

Since the set of real algebraic numbers of bounded degree is an optimal regular
system (Proposition 6.1), many important results derive in a straightforward
way from Theorems 6.1 to 6.5. Throughout this Section, n denotes a positive
integer and : R1 R>0 is a continuous, non-increasing function. We set

Kn () := R : | | < (H()) for innitely many An ,
and we aim to give accurate metric results on the set Kn (). Observe that,
for any real number 1, the set Kn (x x (n+1) ) coincides with the set
Kn ( ) dened in Chapter 5.
132
The convergence half of Theorem 6.6 is due to Koksma [333], Satz 12.
Its divergence half, established by Beresnevich [61], implies that Theorem 4.4
from Chapter 4 is best possible.
T HEOREM 6.6. The set Kn () has Lebesgue measure zero if the sum
n
n
x=1 x (x) converges, and it has full measure if the sum
x=1 x (x)
diverges.
P ROOF. Let E be a bounded real open interval. By Corollary 6.1, the set An =
( j ) j1 is an optimal regular system in E. We apply Theorem 6.1 with the
dened by (
j) := (H( j )) and afne on any
non-increasing function
interval [ j, j + 1], with j a positive integer. We only have to check that the

j) and j1 j n ( j) have the same behaviour. To this end, we
sums j1 (
infer from Lemma 5.4 that there exist positive constants c9 and c10 , depending
only on n, such that, for any j 1, we have

j) c10 j 1/(n+1) .
(6.14)
c9 j 1/(n+1) (
It follows from (6.14) and the fact that is non-increasing that there exist
positive constants c11 , . . . , c14 , depending only on n, such that
1/(n+1)
[c11 M
]
k=1
c12 k (k)
n
M

j=1
j)
(
1/(n+1)
[c13 M
]
c14 k n (k)
k=1

j)
holds for any sufciently large integer M. Consequently, the sum j1 (

n ( j) converges. Hence, we inconverges if, and only if, the sum
j
j1
fer from Theorem 6.1 that Kn () E has Lebesgue measure zero if the sum
n
x=1 x (x) converges, and has full measure otherwise. Theorem 6.6 follows immediately.
We emphasize that we have given a complete proof of Theorem 6.6 for n = 1,
which depends on Proposition 5.3 and Theorem 6.1. This slightly improves
upon Theorem 1.10, established by Khintchine [317] by means of the theory of
continued fractions. Indeed, we get that if : R1 R>0 is a non-increasing
continuous function, then the set

p
K1 () = R : < (q)
q

p
for innitely many rational numbers
q

has Lebesgue measure zero if the sum
q=1 q(q) converges and full
Lebesgue measure otherwise. Theorem 1.10 requires a stronger assumption
133
on , namely that the function x x 2 (x) is non-increasing. A slightly different proof of Theorem 1.10 can also be found in Cassels [155], Chapter VII.
It follows from Theorem 5.3 that, for any real numbers and with >
1, the set Kn ( ) is strictly contained in Kn ( ). This motivates the study
of the following question, asked in [124].
P ROBLEM . Let 1 and 2 be given non-increasing, continuous, positive functions dened on R1 and such that
1 (x) < 2 (x) for x sufciently large.
Do there exist real numbers approximable by elements of An at the order 2 ,
but not at the order 1? In other words, is the set Kn (2 ) \ Kn (1 ) empty or
not?
This has been solved by Jarnk ([292], Satz 5 and Satz 6; see also Exercise
1.5) under some restriction when n = 1, and by A. Baker and Schmidt [45]
for general n and functions i , i = 1, 2, of the form x x 2i (see Theorem
5.3).
To obtain further results on the above question, we determine the Hausdorff
dimensions and the generalized Hausdorff measures of the null sets occurring
in Theorem 6.6. For the proofs of Theorems 6.7 and 6.8 below, we argue as in
the proof of Theorem 6.6, except that we apply Theorems 6.2 to 6.5 instead of
Theorem 6.1.

n
T HEOREM 6.7. Assume that the sum
x=1 x (x) converges. Denote by
the lower order at innity of the function 1/. We then have
n+1
.
dim Kn () =
T HEOREM 6.8. Let f be a dimension function such that x f (x)/x is decreasing in a neighbourhood

of the origin and limx0 f (x)/x = +. Assume
that x x n+1 f 2(x) is non-increasing. Then, we have H f (Kn ()) = 0
if the sum
+
x n f (2(x))
x=1
converges. Otherwise, we have H f (Kn ()) = + and the set Kn () belongs
to the class G f (R).
Theorem 6.8 allows us to give a rather satisfactory partial answer to the above
Problem. Namely, given the approximation functions 1 and 2 , we construct a suitable dimension function f for discriminating between Kn (1 ) and
134
Kn (2 ), when these sets have Lebesgue measure zero. The idea goes back to
Jarnk [292], Satz 5, and yields Corollary 6.2.
C OROLLARY 6.2. Let 1 and 2 be continuous, positive, non-increasing
n
x i (x)
functions dened on R1 . For i = 1, 2, assume that the sum i=1
converges and that the function x x n i (x) is non-increasing. If the
function
x
21 1 (x)
x
is non-decreasing and tends to innity with x, then the set

Kn (2 ) \ Kn (1 )
is non-empty and has the same Hausdorff dimension as the set Kn (2 ).
For n = 1, Bugeaud [129] proved a sharper result, namely that, for any positive
real number c < 1, any continuous function : R1 R>0 such that

x x 2 (x) is non-increasing and the sum x1 x(x) converges, the set
K1 () \ K1 (c) has same Hausdorff dimension as the set K1 (). This does
not follow from Theorem 6.8, which does not help to get an analogous result
for general n.
Let c1 < c2 and > 1 be given positive real numbers. For a general n 2,
we do not even know whether there exist real numbers such that innitely
many algebraic numbers of degree n satisfy | | < c2 H ()(n+1) ,
but only nitely many of them satisfy | | < c1 H ()(n+1) . However,
for n = 2, a partial positive answer follows from a nested interval construction based on Beresnevichs effective proof [60] of Proposition 6.1 for
n = 2.
For a given positive integer n, the exact order of approximation of the real
number by algebraic numbers of degree at most n is
sup{ : Kn ( )} =
wn ( ) + 1
.
n+1
Inspired by Beresnevich, Dickinson, and Velani, who gave in [70] sharp consequences of Theorem 1 of [195], Bugeaud [124] introduced a rened notion of
exact order: the exact logarithmic order. Recall that for any positive integer
i we denote by logi the i-fold iterated logarithm. For n 1 and t 1, let
0 1, 1 , . . . , t1 be real numbers. For a non-zero real number , denote by
Kn (0 , . . . , t1 , ) the set of real numbers for which the inequality

| | < H()(n+1)0 log H() 1 . . .

logt1 H() t1 logt H()
135
is satised by innitely many algebraic numbers of degree at most n. For

any real number t , we then set = (0 , . . . , t1 , t ) and we consider the set
Wn () := { R : sup{ : Kn (0 , . . . , t1 , )} = t }.
In particular, when t = 0, we have Wn ((0 )) = Wn (0 ). Theorem 6.8 allows us to describe very precisely the metric structure of the sets Wn (): their
Hausdorff dimension is equal to 1/0 , independently of 1 , . . . , t , but they
have different sizes, as it can be seen by considering dimension functions.
Let t 0 and = (0 , . . . , t ) with 0 1. For any non-negative real
number , dene the dimension function f , for u > 0 (and small) by
f , (u) := u +1/0 if t = 0,
and
f , (u) := u 1/0
t1

i=1
logi
1
u
1+i /0
logt
1
u
1+t /0
if t 1.
With the above notation, Theorem 6.9 is an almost straigthforward consequence of Theorem 6.8.
T HEOREM 6.9. Let = (0 , . . . , t ) be a real (t +1)-tuple with 0 > 1. Then,
the Hausdorff dimension of Wn () is equal to 1/0 and, more precisely,

+ if = 0,
f ,
H (Wn ()) =
0
if > 0.
C OROLLARY 6.3. For any positive integer n and any real number > 1, we
have

H1/ Kn ( ) = H1/ Wn ( ) = +.
Corollary 6.3 implies that the sets Kn ( ) and Wn ( ) are not s-sets. For K1 ( ),
this result goes back to Jarnk [290, 291] (see also [205]) and is actually a
particular case of a more general statement due to Bugeaud, Dodson, and
Kristensen [135]: for any given dimension function f , the H f -measure of any
real set invariant by rational translation is either 0 or + (see Exercise 6.5).
In the case n = 1, Theorem 6.9 is contained in Theorem 1 of Beresnevich,
Dickinson, and Velani [70], whose Theorem 3 deals with an even more general
formalism and is deeper.
A further metric statement is given in [124], where it is pointed out that
we do not have any information on the Hausdorff dimension of the sets
Wn ((1, 1/2)). For n 2, we even do not know whether these sets are empty
or not.
136
Theorem 6.10, established in [132], provides a nice application of the theory

of intersective sets.
T HEOREM 6.10. Let = (0 , . . . , t ) with 0 > 1. Then, the Hausdorff dimension of the set
{ R : k Wn ()}
n1 k1
is equal to 1/0 .
C OROLLARY 6.4. For any real number 1, the Hausdorff dimension of the
set
{ R : wn ( k ) = (n + 1) 1}
n1 k1
is equal to 1/ . Thus, the set of real numbers whose integer powers are all
S -numbers of -type has Hausdorff dimension 1/ .
Corollary 6.4 remains true if the -type of an S -number is dened as the
supremum of the sequence ((wn ( ) + 1)/(n + 1))n1 . Unlike Theorem 5.6, it
can presumably not be proved with the tools developed in Chapter 5. It heavily
depends on the notion of intersective sets.
We point out that there exist real transcendental numbers for which
wn ( ) = wn ( 2 ), see Theorem 7.7 for explicit examples. However, it is not

known whether t ( ) = t ( 2 ) holds for any S -number .
For distinct positive integers n and n and given real numbers and
with > 1, the methods presented here are unable to decide whether
the intersection Wn ( ) Wn ( ) is empty or not. This is a challenging open
problem.
6.7 Exercises
E XERCISE 6.1. Prove Theorem 6.2.
Hint. Use Lemma 5.2 and a similar argument as in the proof of Theorem 5.1.
E XERCISE 6.2. Use Theorem 6.8 to prove Theorems 6.9 and 6.10.
E XERCISE 6.3. Let be a given real number. The aim of this exercise is to
establish an inhomogeneous extension of Theorem 5.2: For any > 1, the
6.8 Notes
Hausdorff dimension of the set

R : q <
1
q 2 1
137
holds for innitely many

positive integers q
is equal to 1/ . To this end, prove that the set S composed of the points

p+
, p, q Z, q 1, 0 p q 1,
q
form an optimal regular system in ]0, 1[ and apply Theorem 5.1.
Hint. Let Y > 1 be a real number and I be an interval in ]0, 1[. Set

Y

p
1 p
1
2, + 2 .
J (Y ) = I
q
4q q
4q
q=Y /2
( p,q)=1
Observe that any in J (Y ) has a convergent ph /qh such that Y /2 qh Y <

qh+1 . Write = ph /qh + /(Y qh ) and = (t/qh ) + /(2qh ), with || < 1/2
and | | 1. Prove that there exist integers x and y such that qh x < 2qh
and x ph yqh = t. Show that we have | (y + )/x| 3/Y 2 and argue as
at the end of the proof of Proposition 5.3 to conclude.
E XERCISE 6.4. Let be a badly approximable real irrational number. Use the
Three Distance Theorem (or use Theorem 1.7) as in Exercise 5.7 to prove that
the sequence ({n})n1 is an optimal regular system.
E XERCISE 6.5. Let f be a dimension function, and let E be a real set, invariant
by rational translation. Prove that H f (E) = 0 or + (see [135]). Taking
f = Id, prove that either E or its complementary set is a null set.
Hint. Observe that any real open interval ]a, b[ can be represented as a

countable union of intervals [u j /v j , (u j + 1)/v j ] such that b a = j 1/v j .
Deduce that H f (]a, b[E [0, 1]) (b a)H f (E [0, 1]). Observe that
if E [0, 1] does not have full Lebesgue measure, then it can be covered by

a countable union of open intervals ]am , bm [ such that m (bm am ) < 1.
Conclude.
6.8 Notes
Hinokuma and Shiga [282] (but see notes in Chapter 5) and Rynne [488]
obtained the Hausdorff dimension of K1 () for any arbitrary non-negative approximation function without any monotonicity assumption. See also Dickinson [192].
138
Theorem 6.8 for n = 1 is due to Jarnk [292] and has been extended to systems of linear forms by Dickinson and Velani [195]. Beresnevich, Dickinson,
and Velani [70] derived from Theorem 1 of [195] sharp results on the exact
order of approximation of systems of linear forms.
Under suitable assumptions, the notion of ubiquitous system may be used
to prove that the Hausdorff measure is innite at the critical exponent, see
Theorem 1 of Dodson, Rynne, and Vickers [206].
A (slightly) weaker version of Theorem 6.8 (apart from the last assertion)
appeared in [124]. Measure theoretic laws for lim sup sets in a very general
framework have been established by Beresnevich, Dickinson, and Velani [71].
Applications of their results include Theorems 6.1, 6.2, 6.3, 6.6, 6.7, and 6.8
(apart from the last assertion).
An inhomogeneous analogue of Theorem 6.8 for n = 1 can be found in
[130]. See also [71] for generalizations to systems of linear forms.
For extensions of the case n = 1 of Corollary 6.2 to simultaneous approx
imation and to systems of linear forms, see Cerny
[161] and Jarnk [300].
It is interesting to compare Theorem 5.2 with the results proved in Exercises 5.7 and 6.3.
7
On T -numbers and U -numbers
As Mahler [376] proposed his classication of real numbers, he observed that

the set of U -numbers is non-empty, for it contains the Liouville numbers (Definition 1.1). He further showed that the set of T -numbers and the set of U numbers have Lebesgue measure zero, a result later improved upon by Kasch
and Volkmann [311], who established that both have Hausdorff dimension zero
(Corollary 5.2). However, Mahler was not able to decide whether the set of T numbers is non-empty.
The rst proof of the existence of T -numbers goes back to 1968, thirty-six
years after Mahlers paper, and is due Schmidt [507], who, shortly thereafter,
gave a simpler proof [508], together with a slightly stronger statement. The
main ideas of Schmidts proof lie at the heart of every paper providing new results on the set of T -numbers. In the present Chapter, we establish the existence
of T -numbers following R. C. Baker [47]. Sections 7.2 and 7.3 are devoted
to the proof, which incidentally allows us to construct real numbers with
wn ( ) = wn ( ) for every n 2, a question further discussed in Section 7.4.
We emphasize that the construction of T -numbers is fundamentally a nested
interval construction. This observation is used in Section 7.5 to get one metric
result of R. C. Baker [47], who proved that the set of T -numbers is not too
small, although of Hausdorff dimension zero.
Schmidts construction also establishes the existence of U -numbers of any
given type (Denition 3.4). The problem, however, had already been solved by
LeVeque [361] who, in 1953, produced explicit examples of Liouville numbers
whose positive real m-th root is a U -number of type exactly m. Section 7.6
is devoted to LeVeques result and includes an alternative construction of U numbers of given type. In Section 7.7, we present an approach due to Guting
[269], who, thanks to explicit constructions of real numbers as roots of sums of
convergent series, produced examples of real numbers with prescribed values
for wn ( ) and wn ( ) for nitely many values of n. Finally, in Section 7.8,
139
140
we summarize the results obtained in the previous Chapters towards the Main
Problem.
7.1 T -numbers do exist

R. C. Bakers theorem asserts the existence of T -numbers with specic properties. This provides an important step towards the resolution of the Main Problem stated at the end of Section 3.4.
T HEOREM 7.1. Let (wn )n1 and (wn )n1 be two non-decreasing sequences
in [1, +] such that
wn wn wn + (n 1)/n,
wn > n 3 + 2n 2 + 4n + 3,
for any n 1.
(7.1)

wn ( ) = wn
and
wn ( ) = wn ,
for any n 1.
In particular, T -numbers do exist.

The fact that the function f : n n 3 +2n 2 +4n +3 occurring in the statement
of Theorem 7.1 is of order of magnitude n 3 is due to technical constraints.
Presumably, the conclusion of Theorem 7.1 still holds if f is replaced by any
non-decreasing function g such that g(n)/n tends to innity with n or, even,
such that lim supn+ g(n)/n is innite. Theorem 7.1 implies that there exist
T -numbers of any type greater than or equal to 3. However, the existence of
T -numbers of type strictly smaller than 3 is an open problem. No particular
signicance should be attached to this value 3, which appears for technical
reasons.
Theorem 3.4 asserts that any real transcendental number satises wn ( )
wn ( )+n 1 for every positive integer n. This leads to conjecture that the term
(n 1)/n occurring in (7.1) could perhaps be replaced by n 1, a question
discussed in Section 7.4.
It should be pointed out that Theorem 7.1 does not imply the existence of
real numbers with w( ) = w ( ). This remains an open problem.
Theorem 7.1 is (very slightly) better than the main result of [47], since
we use (7.24) below instead of setting n = wn n + 2, as R. C. Baker
did. Furthermore, it is possible to slightly decrease the lower bound for wn in
(7.1) if we seek for numbers for which wn ( ) and wn ( ) are equal for every
positive integer n (see Exercise 7.1).
Schmidts construction is rather tedious and complicated. The T -numbers
are obtained as limits of sequences of algebraic numbers in a semi-explicit
7.2 The inductive construction
141
way. Indeed, at the heart of his proof, Schmidt used the fact that algebraic
numbers are not too well approximable by algebraic numbers of bounded
degree, which is a result of an ineffective nature (Theorems 2.6 and 2.7). For
this reason, Schmidts method does not allow us to construct explicit examples
of T -numbers. This remains an open problem.
Incidentally, Theorem 7.1 shows the existence of U -numbers of arbitrary
given type, a result established by LeVeque [361] in 1953 with a constructive
method (Theorem 7.4). Following Schmidt [508], there are, however, alternative ways to split the U -numbers into innitely many classes, including the
following one. Write ( ) for the inmum of the numbers such that for every
positive integer n we have
wn (, H ) cn eH ,
as H +,
for some constant cn depending only on n. Observe that if and are two algebraically dependent U -numbers, we then have ( ) = ( ). Consequently,
we may subdivide the set of U -numbers according to the value of ( ).
Using the same ideas as in his proof of the existence of T -numbers, Schmidt
[508] established that, for any in [0, +], there exist uncountably many
U -numbers with ( ) = .
Schmidt [508] also proposed a possible generalization of the notion of
Liouville numbers. A real number eld K being given, a real number is
called K-Liouville if for every positive real number w there is a real number
in K with | | < H()w . According to this denition, the usual Liouville
numbers are termed Q-Liouville. Assuming that F is a collection of real algebraic number elds such that K1 K2 and K1 in F implies K2 in F, Schmidt
proved (again with the same kind of construction) that there exist real numbers
which are K-Liouville precisely for the elds K in F.

Before proceeding with the construction of sequences of real numbers satisfying various conditions, we recall (part of) Schmidts Theorem 2.7, which
follows from his Subspace Theorem.
T HEOREM 2.7. Let be a real algebraic number and n be a positive integer. Then, for any positive real number , there exists a positive (ineffective)
constant (, n, ) such that
| | > (, n, ) H()n1
for any algebraic number of degree at most n.
(7.2)
142
The exponent of H() in (7.2) does not depend on : this fact is crucial for
establishing the existence of T -numbers. Actually, Schmidt applied Theorem
2.6 (that is, the generalization due to Wirsing of Roths Theorem 2.1) in [507]
and in [508], thus getting a lower bound for wn of order of magnitude 2n 3
instead of n 3 in (7.1).
We follow step by step the arguments of R. C. Baker [47], who deduced
Theorem 7.1 from Proposition 7.1 below, as will be shown in Section 7.3. In
fact, we build a T -number; by Theorem 3.6, this is equivalent to constructing
a T -number.
P ROPOSITION 7.1. Let 1 , 2 , . . . be real numbers > 1 and 1 , 2 , . . . be real
numbers in [0, 1]. Let m 1 , m 2 , . . . be positive integers and 1 , 2 , . . . be real
numbers satisfying n > n 3 + 2n 2 + 4n + 3 for any n 1. Then, there exist
positive real numbers 1 , 2 , . . . , prime numbers g1 , g2 , . . . , and integers c1 ,
c2 , . . . such that the following conditions are satised.
(I j )
(I I1 )
(I I j )
g j |c j j + 2(1)m j +1 v j j , where v j = [g j j ]
( j 1).
1 = (c1 + 21/m 1 v1 )/g1 I0 :=]1, 2[.

j = (c j + 21/m j v j )/g j
belongs to the interval I j1 dened by
1
3
j1 + g j1j1 < x < j1 + g j1j1 ( j 2).
2
4
(I I I j )
| j n | n H(n )n
for any algebraic number n of degree n j which is distinct from

1 , . . . , j ( j 1).
Let us see how one should interpret the different parameters introduced in
Proposition 7.1. Observe that the sequence ( j ) j1 we construct is strictly
increasing and bounded, hence, it converges to a limit . Since c j 2g j ,
m
Lemma A.4 implies that for any j 1 the height of j is at most 4m j +1 g j j .
Thus, conditions (I I j+1 ) show that the order of approximation of by the algebraic number j depends only on j and m j , namely that we have | j |
24 j H( j ) j /m j . Consequently, if for some positive integer n and some real
number > 1 there are innitely many indices j such that m j = n and j = ,
then conditions (I I j ) give us that wn ( ) /n 1. Furthermore, conditions
(I I I j ) show that the algebraic numbers other than the j s are not too close to
and they yield that wn ( ) max{n , /n} 1. With a suitable choice of
the sequences (m j ) j1 , ( j ) j1 , and (n )n1 , the real number obtained is a
143
T -number. However, some technical conditions impose that n has to grow

at least as fast as n 3 , thus, with this method, it does not seem to be possible to
construct T -numbers of type smaller than 3.
The role of the parameter is to measure the gap between wn ( ) and wn ( ),
as it will be clear in Section 7.3. Taking j equal to 1 for any j, we have
v j = g j and
cj
j = 21/m j + .
gj
The condition (I j ) is then satised if g j does not divide c j , and the T -numbers
obtained in this way satisfy wn ( ) = wn ( ) for any n.
In Schmidts original construction, the algebraic numbers 21/m j are replaced by arbitrary real algebraic numbers of degree m j . These play a pivotal role. The particular choice of the numbers 21/m j allows us to build real
numbers for which wn ( ) and wn ( ) are different.
PROOF OF PROPOSITION 7.1. In all what follows, we denote by n a real
algebraic number of degree exactly n. We x a sequence (n )n1 in ]0, 1[ such
that, for any n 1, we have
n > n 3 + 2n 2 + 4n + 3 + 20n 2 n .
(7.3)
We begin by adding four extra conditions (I V j ), . . . , (V I I j ) to be satised by

the numbers j .
Set J0 = I0 . For any positive integer j, let J j denote the subset of I j consisting of the real numbers x in I j such that
|x n | 2n H(n )n
for any algebraic number n of degree n j, distinct from 1 , . . . , j , x and
of height H(n ) sufciently large, that is, satisfying
H(n ) (n g j j )1/n .
The supplementary conditions are the following.
(I V j )
j J j1
( j 2).
(V j )
| j j | 2 j H( j ) j for any j = j
( j 1).
(V I j )
1/(n+1+n )
, then | j n | 1/g j
If n j and H(n ) g j
(V I I j )
(J j ) (I j )/2
( j 1).
( j 1)
144
We construct the numbers 1 , 1 , 2 , 2 , . . . by induction. At the j-th stage,

there are two distinct steps. Step (A j ) consists in building an algebraic number
j =
c j + 21/m j v j
gj
satisfying conditions (I j ) to (V I j ). In step (B j ), we show that the number j

constructed in (A j ) satises (V I I j ) as well, provided that g j is chosen large
enough in terms of
1 , . . . , j , 1 , . . . , j , m 1 , . . . , m j , 1 , . . . , j , 1 , . . . , j ,
1 , . . . , j1 , 1 , . . . , j1 .
(7.4)
The symbols o, and

used throughout steps (A j ) and (B j ) mean that the
numerical implicit constants depend (at most) on the quantities (7.4). Furthermore, the symbol o implies as g j tends to innity.
Step (A1 ) is rather easy. Let P(X ) := X m 1 2 v1m 1 denote the minimal
polynomial of 21/m 1 v1 over Z and observe that (I1 ) is satised if, and only if,
g1 does not divide P(c1 ). Let g be a prime number and assume that g divides
P(d), P(d + 1), . . . , P(d + m 1 ) for some integer d. Then g also divides the
difference P(d + j + 1) P(d + j) = m 1 (d + j)m 1 1 + ... for any integer
j = 0, . . . , m 1 1. If m 1 2, we similarly get that g divides m 1 (m 1 1)(d +
j)m 1 2 + ... for any integer j = 0, . . . , m 1 2. Continuing in this way, it
follows that g divides m 1 !. Consequently, if the prime number g1 is larger than
m 1 , then there are g1 numbers 1 = (c1 + 21/m 1 v1 )/g1 in the interval ]1, 2[
satisfying condition (I1 ). These g1 numbers have mutual distances at least
g11 and, since there are only o(g1 ) rational numbers 1 satisfying H(1 )
1/(2+1 )
g1
, we are able to choose 1 such that (V I1 ) is veried. Moreover, by
Theorem 2.7 with n = 1 (this is Roths Theorem 2.1), there exists 1 in ]0, 1[
such that both (I I I1 ) and (V1 ) hold.
Let j 2 be an integer and assume that 1 , . . . , j1 have been constructed.
Step (A j ) is much harder to verify, since we have no control on the set J j1 .
Thus, it seems difcult to check that the condition (I V j ) holds. To overcome

which contains J j1 .
this problem, we introduce a new set J j1
1/m
j
Set j = (c j + 2
v j )/g j for some positive integers c j and g j with
j1
g j j > 8g j1
,
(7.5)

the set formed by the real numbers x in I j1 satisfying
and denote by J j1
|x n | 2n H(n )n
145
for any algebraic number n of degree n j, distinct from 1 , . . . , j , x, and

whose height H(n ) satises the inequalities
j1 1/n
)
H(n ) (2n g nj
(n g j1
2 +n+1+2n
)1/(n n1n ) .
(7.6)
Since
n n 1 n > n 3 + 2n 2 + 2n + 1 + 5n 2 n
> (n + 1)(n 2 + n + 1 + 2nn ),
(7.7)
the exponent of g j in the right member of (7.6) is strictly less than 1/(n + 1).
Thus, there are o(g j ) algebraic numbers n satisfying (7.6), and we observe

is a nite union of intervals, and, more prethat, unlike J j1 , the set J j1
cisely, a union of o(g j ) intervals. We will prove that for g j large enough
we have g j suitable choices for c j such that conditions (I j ) to (V j ) are
fullled.
Let n be an algebraic number of degree n. By Theorem 2.7, there exists a
positive constant (m j , n, n ) such that

vj
g j n c j
| j n | = 21/m j

gj
vj
n1n
v j 1
(m j , n, n )
H v j (g j n c j )
gj
(n 2 +n+1+2nn )
gj
H(n )n1n ,
(7.8)
if g j satises
g j (m j , n, n )1/(nn ) 2(2n+1)(n+1+n )/(nn ) .
Here, we have used v j 1, max{c j , g j , v j } 2g j , and Lemma A.4. In particular, if g j is large enough, we have
| j n | 2n H(n )n
(7.9)
as soon as
H(n )n n1n 2n g nj
2 +n+1+2n
(7.10)

) 1. Since the set J
By (V I I j1 ) and J j1
J j1 , we have (J j1
j1 is
the union of o(g j ) intervals, if g j is a sufciently large prime number, then,
arguing as in step (A1 ), we get that there exist g j numbers j = (c j +

21/m j v j )/g j in J j1
such that (I j ) is satised. Such j s also belong to J j1 ,
since (7.10) implies (7.9), and condition (I V j ) is veried.
146
Thus, we are left with g j suitable algebraic numbers j , mutually distant

by at least g 1
j . Only o(g j ) algebraic numbers n satisfy
1/(n+1+)
H(n ) g j
(7.11)
thus there are g j algebraic numbers j such that | j n | 1/g j for the
numbers n verifying (7.11). Further, Theorem 2.7 ensures that there exists j
in ]0, 1[ such that (V j ) is satised. Consequently, there are g j algebraic
numbers j satisfying (I j ), (I I j ), (I V j ), (V j ) and (V I j ).
It remains for us to show that such a j also satises (I I I j ). To this end, it
sufces to prove that
| j n | n H(n )n
holds for any algebraic number n of degree n < j, which is different from
1 , . . . , j and whose height H(n ) satises
j1 1/n
)
.
H(n ) < (n g j1
Since the sequence (gtt )t1 is increasing, we either have

gnn < n H(n )n ,
(7.12)
or there exists an integer t with n < t < j such that
gtt < n H(n )n gt1t1 .
(7.13)
In the former case, we infer from (Vn ), (7.5) and (7.12) that
| j n | |n n | | j n | 2n H(n )n gnn > n H(n )n .
In the latter case, (I Vt ), (7.5) and (7.13) yield that
| j n | |t n | | j t | 2n H(n )n gtt > n H(n )n .
Thus, condition (I I I j ) holds and the proof of step (A j ) is completed.
Before going on with step (B j ), we observe that the integer c j is far from
being uniquely determined. Indeed, it follows from the argument used in step
(A1 ) that there exist at least
g j (J j1 )
mj +1
candidates j having the property (I j ). In the course of step (A j ), we excluded
only o(g j ) of them. Thus, if g j is sufciently large, we have at least
g j g j1j1
32m j
(7.14)
147
suitable choices for j . This observation is valid at any step j and shows that
the construction actually gives an uncountable set of T -numbers. It will be
used in Section 7.5.
Let j 1 be an integer. For the proof of step (B j ), we rst establish that if
g j is large enough and if x lies in I j , then we have
|x n | 2n H(n )n
(7.15)
for any algebraic number n = j of degree n j such that
/(n n1n )
(n g j j )1/n H(n ) g j j
(7.16)
Let, then, n = j be an algebraic number satisfying (7.16) and let x be in I j ,

that is, such that
3
1 j
gj < x j < gj j.
2
4
(7.17)
1/(n+1+n )
If j (n + 1 + n ) n n 1 n , then H(n ) g j
follows from (V I j ), (7.16), (7.17), and the assumption j > 1 that
and it
|x n | | j n | | j x|
j
g 1
j gj
2g j
2n H(n )n ,
provided that g j is large enough.

Otherwise, we have
j (n + 1 + n ) > n n 1 n ,
(7.18)
and, by (7.8), we get

|x n | | j n | | j x|
j
(n 2 +n+1+2nn )
H(n )n1n g j
(n 2 +n+1+2nn )
H(n )n1n /2.
gj
gj
(7.19)
To check the last inequality, we have to verify that

j
2g j
(n 2 +n+1+2nn )
gj
H(n )n1n .
(7.20)
In view of (7.16), inequality (7.20) is true as soon as

(n+1+n )/(n n1n )
2g j j
(n 2 +n+1+2nn )
gj j gj
148
which, by (7.18), holds for g j large enough when

n + 1 + n
n + 1 + n
< 1 (n 2 + n + 1 + 2nn )
,
n n 1 n
n n 1 n
(7.21)
and in particular when n satises (7.3). Furthermore, we have

(n 2 +n+1+2nn )
gj
H(n )n1n 4n H(n )n .
(7.22)
Indeed, by (7.16), n < 1, and (7.18), we get
H(n )n n1n (n g j j )(n n1n )/n

(n n1n )/n
n g j j
(n n1n )2 /(n (n+1+n ))
> n g j
4n g nj
2 +n+1+2n
since we infer from (7.3) that

(n n 1 n )2 > n (n + 1 + n )(n 2 + n + 1 + 2nn ).
(7.23)
Combining (7.19) and (7.22), we have checked that

|x n | 2n H(n )n
holds under the assumption (7.18). By (7.18), this implies that (7.15) is true if
n satises (7.16) and is not equal to j . Consequently, for g j large enough,
the complement J jc of J j in I j is contained in the union of the intervals
]n 2n H(n )n , n + 2n H(n )n [,
where n runs over the real algebraic numbers of degree n j and height
/( n1)
greater than g j j n
. The Lebesgue measure of J jc is then
j
n=1
j /(n n1n )
H >g j
H nn = o(g j
) = o((I j )),
since for any positive integers H and n there are at most (8H )n algebraic numbers of height H and degree n (see (8.5) in Chapter 8). Thus, we conclude that
we can nd g j large enough such that (J j ) (I j )/2. This completes step
(B j ) as well as the proof of Proposition 7.1.
At this point, we may summarize where the condition n > n 3 + 2n 2 +
4n + 3 appears. There are three steps where it is needed, namely (7.7), (7.21)
and (7.23). Asymptotically, these three inequalities reduce, respectively, to
7.3 Completion of the proof of Theorem 7.1
149
n > (n + 1)(n 2 + n + 2), n > (n + 1)(n 2 + n + 3), and (n n 1)2 >

n (n + 1)(n 2 + n + 1). The most restricting condition is given by (7.21), hence,
our assumption on n .
7.3 Completion of the proof of Theorem 7.1

The sequence ( j ) j1 obtained in Proposition 7.1 is strictly increasing and
bounded, thus it converges to a real number which we denote by . It is easily
seen that if the sequence (m j ) j1 is chosen in such a way that all the integers
1, 2 . . . appear in it innitely many times, then is a T -number. Now, we explain how to choose the sequence ( j ) j1 such that wn ( ) and wn ( ) are
different.
Let (m j ) j1 be a sequence of positive integers taking innitely many times
+ 1) and we dene
each value 1, 2, . . . . For j 1, we set j = m j (wm
j
j
in [0, 1] by
wm
+ (m j 1)(1 j )/m j = wm j .
j
Moreover, for any integer n 1, we set

n = wn ,
(7.24)
in such a way that n > n 3 + 2n 2 + 4n + 3. Let 1 , 2 , . . . , 1 , 2 , . . . be as

in Proposition 7.1 and denote by the limit of the strictly increasing sequence
( j ) j1 .
We x an integer n 1. Observe that the minimal polynomial of j over Z
is the polynomial
P j (X ) = (g j X c j )m 2v mj ,
which is primitive since, by condition (I j ), its leading coefcient is coprime
m
with its constant term. Thus, recalling that c j 2g j , we get that g j j
H( j ) (2g j )m j . Furthermore, for any j 1 we have
j
j + gj
/2 < < j + g j
and the denition of j implies that

1
wm
H( j )
1
wm
/2 | j | 2 j H( j )
(7.25)
Moreover, if m is a real algebraic number of degree m n which is not equal

to one of the j , then, by (I V j ) we have
| j m | m H(m )m ,
150
hence, as j tends to innity,
| m | m H(m )m m H(m )wm 1 ,
(7.26)
+ 1. As m = n for innitely many integers j, it follows

since m wm
j
)
from (7.25), (7.26) and from the fact that the sequence (wm
m1 is increasing
that
wn ( ) = wn .
It remains for us to prove that wn ( ) = wn . This is clear for n = 1, thus we
assume n 2. Denote by j = j,1 , j,2 , . . . , j,m j the roots of the polynomial
P j (X ). Until the end of this proof, we write A
B when there is a positive
constant c(m j ), depending only on m j , such that |A| c(m j ) |B|, and we
j 1
write A B if both A
B and B
A hold. We have | j j,k | g j
for k = 2, . . . , m j , and, since | j j,k | 2| j |, we get
j 1
| j,k | g j
for k = 2, . . . , m j .
Consequently, since H(P j ) g j j , we have

|P j ( )| =

m
gj j
mj

| j,k |
k=1
m m j (wm j +1) (m j 1)( j 1)

gj
gj j gj
w +((m j 1)/m j )( j 1)
=
H(P j ) m j
(7.27)
wm j
H(P j )
Since m j = n for innitely many j, we infer from (7.27) that wn ( ) wn .

In order to show that we have equality, let P(X ) be an integer polynomial of
degree at most n, which we write under the form
P(X ) = a R1 (X ) . . . R p (X ) Q 1 (X ) . . . Q s (X ),
where a is an integer and the polynomials Ri (X ) and Q j (X ) are primitive
and irreducible. We moreover assume that the Ri (X ) do not have a root equal
to one of the s, but that each Q j (X ) has a root equal to some . If k denotes the degree of the polynomial Ri (X ), then, by Lemma A.8, it has a root
satisfying
|Ri ( )| H(Ri )2k | |
n H(Ri )k k+2 = n H(Ri )wk n H(Ri )wn . (7.28)
If denotes the degree of Q j (X ), then (7.27) shows that
|Q j ( )| H(Q j )w H(Q j )wn .
7.4 On the gap between wn and wn
151
Together with (7.28) and Lemma A.3, this gives

wn
|P( )| H(R1 ) . . . H(R p )H(Q 1 ) . . . H(Q s )
H(P)wn ,
and we get wn ( ) = wn , as claimed.
7.4 On the gap between wn and wn
Many ideas from the construction of T -numbers have been used by Bugeaud
[128] to prove that, for any integer n 3, the set of values taken by the function
wn wn includes the interval [0, n/4].
T HEOREM 7.2. Let n 3 be an integer. Let wn and wn be real numbers such
that
wn wn wn + n/4, wn > 2n 3 + 2n 2 + 3n 1.
Then there exists real numbers such that
wn ( ) = wn and wn ( ) = wn .
Exercise 7.2 is devoted to the proof of Theorem 7.2. The main idea is however easy to explain. As pointed out after the statement of Proposition 7.1, the
algebraic numbers 21/m j occurring there have no particular importance for obtaining a T -number, and they could be replaced by other algebraic numbers of
the same degree. Since we are only interested in approximation by algebraic
numbers of degree at most n, we use a suitable version of Proposition 7.1
where m j = n and j = (c j + j )/g j for every j 1. Here, the c j s and the
g j s are positive integers and the j s are real algebraic numbers of degree n.
The differences | j | are precisely controlled and wn ( ) satises
wn ( ) = lim
j+
log | j |
1.
log H( j )
The idea consists in choosing the j s such that the algebraic numbers j have
a complex conjugate j very close to j in terms of H( j ). It then follows
that | j | is very small, and that |P j ( )| is much smaller than H( j )wn ( ) ,

where P j (X ) denotes the minimal polynomial of j over Z. Consequently,
wn ( ) is larger than wn ( ). In [128], Bugeaud took j to be roots of the integer
polynomials Q a,n (X ) := X n 2(a X 1)2 , which were rst mentioned by
Mignotte [421]. It is readily veried that for any integers n 3 and a 10,
the polynomial Q a,n (X ) is irreducible and has two conjugate roots distant by
at most 2n H(Q a,n )n/41/2 : this estimate yields the upper bound n/4 for the
range of values of wn wn in Theorem 7.2.
152
Furthermore, by considering the family of integer polynomials Pa,d (X ) :=

(X d a X + 1)2 2X 2d2 (a X 1)2 mentioned below Theorem A.3, it is
possible to improve Theorem 7.2. For any integer d 3, it follows either
from the results of Muller [433] or of Laurent and Poulakis [355] that the
polynomial Pa,d (X ) is irreducible for any integer a sufciently large, thus it
has two conjugate roots distant by at most 4a 2d . Taking for j one of these
roots, it follows from the proof of [128] that, for any even integer n 6, the
set of values taken by the function wn wn contains the interval [0, n/2[, see
[133].
Unlike Theorem 7.1, Theorem 7.2 is effective. Indeed, it does not rest on
Theorem 2.7, but on Liouvilles inequality. As explained in Section 8 of [128],
the construction of Theorem 7.2 does not ensure the existence of T -numbers
with wn ( ) wn ( ) > (n 1)/n for some positive integer n.
7.5 Hausdorff dimension and Hausdorff measure

It has been proved (Corollary 5.2) that the Hausdorff dimension of the set
of T -numbers is zero. However, thanks to the multiple possible choices for
the integers c j occurring in the proof of Proposition 7.1, R. C. Baker [47]
showed that this set is not too small. Namely, its H f -measure is innite for
any dimension function f increasing in a neighbourhood of the origin faster
than any power function, that is, for any dimension function f such that
f (x x )
for any > 0.
(7.29)
It is possible to rework R. C. Bakers argument to prove the existence of Snumbers for which wn ( ) and wn ( ) are different for nitely many integers n.
T HEOREM 7.3. Let n 1 < . . . < n k be positive integers. Let w1 . . . wk
and w1 . . . wk be real numbers satisfying
wi wi wi + n i /4, wi > 2n i3 + 2n i2 + 3n i 1 (1 i k).
Then there exists a real S-number such that
wni ( ) = wi and wn i ( ) = wi (1 i k).
(7.30)
As explained in [128], the proof of Theorem 7.2 can easily be adapted to assert
the existence of real numbers satisfying (7.30). To prove that some of them
are S-numbers, it is sufcient, by Corollary 5.2, to show that they form a set of
positive Hausdorff dimension. We use a version of Proposition 7.1 in which the
sequence (m j ) j1 takes only the values n 1 , . . . , n k , and each of these appears
7.6 On U -numbers
153
innitely often. Further, proceeding as in Section 7.4, we take j = n i (wi + 1)

if m j is equal to n i . Denote by the maximum of the j and let be a real
number with 0 < < 1/2. We can choose the sequence (g j ) j1 in such a way
that, besides the requirements of Proposition 7.1, we have
g j > 32n k g j1 ,
for any integer j 2. Then, (7.14) implies that, at each step, there are at
least g 1
suitable choices for the integer c j . We thus have a Cantor-type
j
construction (see Section 5 from [47]), at the j-th stage of which there are
intervals of length g
at least g 1
j
j /4 and distant by at least 1/g j . It follows from Proposition 5.2 that the Hausdorff dimension of the residual set
obtained is positive and even at least equal to (1 )/. Consequently, this
set contains S-numbers, and even S-numbers of type less than + 2, by
Theorem 5.8.
7.6 On U -numbers
As mentioned in Chapter 3 and in Section 7.1, there are several ways to dene
subclasses of the set of U -numbers. We consider in the present book the most
classical one, introduced by LeVeque [361] and recalled below.
D EFINITION 7.1. Let m be a positive integer. The Um -numbers are precisely
the U -numbers of type m, that is, the real numbers such that wm ( ) = +
and wn ( ) < + for any integer n with 1 n < m.
Since for any positive integer n both functions wn and wn take simultaneously
nite values or are simultaneously innite, we may replace the functions wm
and w .
and wn in Denition 7.1 by wm
n
The rst examples of U2 -numbers are due to Maillet [403] (see also Perron
[454], pp. 143148). In the rst paper entirely devoted to U -numbers, LeVeque
[361] proved, among other deeper results, that none of the sets of Um -numbers
is empty.
T HEOREM 7.4. For any positive integer m, there exist uncountably many real
Um -numbers.
P ROOF. For any positive integer j, let a j be an element of {2, 4}. We
prove that, for any integer m 1, the positive real m-th root of (3 +

j! )/4 is a U -number. For k 1, set
m
j1 a j 10

1/m
k

pk
a j 10 j! , qk = 4 10k! , and k =
.
pk = 10k! 3 +
qk
j=1
154
Then, we have H(k ) = 4 10k! and

| k | | m km | 2 10(k+1)! 22k+3 H(k )k1 .
(7.31)
Consequently, is an U -number, and its type does not exceed m.

Let be a non-zero real algebraic number of degree n strictly less than m
and of height greater than H(1 ). Then, there exists a positive integer k such
that
H(k ) H()2m H(k+1 ) H(k )k+1 .
(7.32)
It follows from Corollary A.2 that

| k | 2 (m + 1)5m/2 H()m H(k )n
2 (m + 1)5m/2 H(k )m(k+1)/2 .
(7.33)
By taking H() large enough, the index k satises

H(k )m+(k+1)/2 > 22k+3 (m + 1)5m/2 ,
and it follows from (7.31) and (7.33) that | k | > 2| k |. Thus, except
for nitely many algebraic numbers of degree strictly less than m, we have
| | | k | | k | > | k |/2
(m + 1)5m/2 H()m H(k )n (m + 1)5m/2 H()m2mn ,
by (7.32). We conclude that
wn ( ) 2mn + m 1
(7.34)
for n = 1, . . . , m 1, hence, is not a U -number of type strictly less than m.

Using the idea explained in Section 7.5, R. C. Baker [47] proved that, for
any positive integer m, the set of Um -numbers has innite H f -measure for
any dimension function f satisfying (7.29). In particular, the H f -measure
of the set of Liouville numbers is innite if f satises (7.29), and is zero
if there exists a positive real number such that (x x ) f (see also
Exercise 6.5).
In Theorem 7.4, we constructed explicitly Um -numbers and gave effective upper bounds for wn ( ) for any positive integer n smaller than m. These
are somehow unsatisfactory, since they are quite large (a slight improvement
of (7.34) has been obtained by Guting ([269], Satz 10) who, essentially, replaced 2mn by mn). Alniacik, Avci, and Bugeaud [18] improved upon Theorem 7.4 in the sense that they established the existence of Um -numbers
with sharper upper bounds for wn ( ), where n = 1, . . . , m 1. Like Theorem
7.4, Theorem 7.5 below is of an effective nature, since it ultimately rests on
7.6 On U -numbers
155
Theorem 2.11. The underlying idea is to construct as the limit of a rapidly

converging sequence of real algebraic numbers of degree m, using the fact that
these numbers can be quite well approximated by algebraic integers of same
degree.
T HEOREM 7.5. Let m 2 be an integer. There are uncountably many Um numbers with
wn ( ) m + n 1,
for n = 1, . . . , m 1.
The proof of Theorem 7.5 relies on the following consequence of Theorem

2.11.
L EMMA 7.1. Let n 2 be an integer and be a real algebraic number of
degree n. There exist an effectively computable constant ( ) and innitely
many algebraic integers of degree n such that
0 < | | < ( ) H()n .
Moreover, we use repeatedly an easy consequence of the triangle inequality.
L EMMA 7.2. Let (1 , 2 , ) be a triple of real numbers. Let H , N , and c be
positive real numbers such that
0 < |1 2 | < N 1
and |1 | c1 H 1 .
Then, for any positive real number , we have

|2 | > (c + )1 H 1
as soon as
H c1 (c + )1 N .
P ROOF. Assuming that H c1 (c + )1 N , we have

|2 | |1 | |1 2 |
> c1 H 1 N 1

H 1 c1 c1 (c + )1 = H 1 (c + )1 ,
as claimed.
PROOF OF THEOREM 7.5. Let m 2 be an integer. Thanks to repeated use
of Liouvilles inequality, we construct by induction a converging sequence of
real algebraic integers of degree m, whose limit is a Um -number with the given
property. Throughout the proof, for n = 1, . . . , m 1, we denote by n a real
algebraic number of degree n.
Let 1 be a real algebraic number of degree m. By Corollary A.2, we have
|1 n | > c11 H(1 )n H(n )m c11 H(1 )m H(n )m ,
(7.35)
156
with c1 = (m + 1)2m . Furthermore, Lemma 7.1 implies that there is an effectively computable constant (1 ) > 1 and innitely many real algebraic
integers 2 of degree m with
0 < |1 2 | < (1 ) H(2 )m .
(7.36)
We infer from (7.35), (7.36), and Lemma 7.2 applied to the triple (1 , 2 , n )
with H = H(n )m , N = (1 )1 H(2 )m , c = c1 H(1 )m , and = 1/2 that
|2 n | > (c2 + 1/2)1 H(n )m ,
(7.37)
with c2 = c1 H(1 )m , provided that

H(n ) (c2 (2c2 + 1))1/m (1 )1/m H(2 ) := A.
Furthermore, we deduce from Corollary A.2 that
|2 n | > c11 H(2 )n H(n )m ,
and, if H(n ) > A, we infer from (7.38) that

n/m
|2 n | > c11 c2 (2c2 + 1)(1 )
H(n )mn

1
> H(n )mn log 3H(n )
(7.38)
(7.39)
holds if H(n ) exceeds some effectively computable constant c3 depending

only on 1 . We choose 2 such that we have H(2 ) > (1 )H(1 ) and A > c3 .
Then, by (7.37) and (7.39), we see that

1
|2 n | > (c2 + 1/2)1 H(n )mn log 3H(n )
holds for any algebraic number n .
Let k 2 be an integer. Assume that there are real algebraic integers
2 , . . . , k of degree m and real numbers (1 ), . . . , (k1 ) such that

1
1 1
1
|i+1 n | > c2 + + . . . + i
H(n )mn log 3H(n ) ,
2
2
for any n ,
0 < |i i+1 | < (i ) H(i+1 )m and H(i+1 ) > (i ) H(i )i
for i = 1, . . . , k 1.
By Lemma 7.1, there exists a real number (k ) > 1 and a real algebraic
number k+1 of degree m satisfying
H(k+1 ) > (k ) H(k )k and |k k+1 | < (k ) H(k+1 )m .
(7.40)
7.6 On U -numbers
157
By assumption, we have

1
|k n | c41 H(n )mn log 3H(n ) ,
with c4 = c2 + 21 + . . . + 2k+1 , and we infer from Lemma 7.2 applied to
the triple (k , k+1 , n ) with = 2k that

1
|k+1 n | (c4 + 2k )1 H(n )mn log 3H(n )
(7.41)
holds provided that

H(n )m+n log 3H(n ) c41 (2k c4 + 1)1 (k )1 H(k+1 )m .
(7.42)
By Corollary A.2, we have

|k n | c11 H(k )n H(n )m ,
and we apply Lemma 7.2 to the triple (k , k+1 , n ) with the parameters H =
H(n )m , N = (k )1 H(k+1 )m , c = c1 H(k )n and = H(k )n to get
|k+1 n | > (c1 + 1)1 H(k )n H(n )m
provided that
1/m

H(n ) c11 (c1 + 1)1 (k )1 H(k )n
H(k+1 ) := B.
(7.43)
Hence, we have
|k+1 n | > H(n )m1
(7.44)
(c1 + 1)H(k )n H(n ) B.
(7.45)
as soon as n satises
By choosing k+1 with H(k+1 ) large enough in terms of 1 and k , (7.42) is

satised for all algebraic numbers n with H(n ) < (c1 + 1)H(k )n . Thus, by
(7.41), (7.44), and (7.45), we have

1
|k+1 n | > (c4 + 2k )1 H(n )mn log 3H(n )
when H(n ) B.
(7.46)
158
Finally, assuming that H(n ) > B, we use

|k+1 n | > c11 H(k+1 )n H(n )m ,
given by Corollary A.2, to get from (7.43) that, for H(k+1 ) large enough in
terms of m, 1 , and k , we have

n/m
|k+1 n | > c11 c1 (c1 + 1)(k )H(k )n
H(n )mn
1

> H(n )mn (log B)1 > H(n )mn log 3H(n ) .
(7.47)
The desired conclusion follows from (7.46) and (7.47): for any algebraic number n , we have

1
1 1
1
H(n )mn log 3H(n ) .
|k+1 n | > c2 + + . . . + k
2
2
(7.48)
We deduce from (7.40) that, for any integers i, j with 2 i < j, we have
| j i | 2 H(i )i .
(7.49)
Thus, the sequence ( j ) j1 is a Cauchy sequence, and we denote its limit by

. Letting j tend to innity in (7.49), we get | i | 2 H(i )i , whence
( ) = +. Moreover, by (7.48), we get that w ( ) m + n 1, for any
wm
n
n = 1, . . . , m 1, as claimed. Finally, we observe that, at each step, we have in
fact innitely many choices for the algebraic integer k , thus we can construct
uncountably many real numbers satisfying the required properties.
Unfortunately, Theorem 7.5 does not assert the existence of Um -numbers
with a prescribed value of wn ( ) or wn ( ) for some integer n with 1 n < m.
However, using Lagranges Theorem 1.16 (asserting that the continued fraction
expansion of a real number is ultimately periodic if, and only if, this number
is a quadratic irrationality), we can construct real U2 -numbers with any prescribed value for w1 ( ) (see Exercise 7.3). Moreover, again with the use of
continued fractions, Alniacik [9] gave effective constructions of U -numbers
which are not well approximable by rational numbers.
T HEOREM 7.6. For any integer m 2, there exist Um -numbers with
w1 ( ) = 1.
Exercise 7.4 is devoted to a proof of Theorem 7.6. Actually, the Um -numbers
constructed by Alniacik in [9] satisfy w1 ( ) = 1 and wn ( ) c5 m 6 for any
integer n with 2 n < m and some absolute constant c5 .
7.7 A method of Guting
159
7.7 A method of Guting

The main idea behind LeVeque work [361] mentioned in Section 7.6 can be
phrased as follows: Any real number with sufciently many very good algebraic approximations of given degree cannot have innitely many very, very
good algebraic approximants of smaller or comparable degree.
This can be seen by applying Liouvilles inequality under the form given by
Corollary A.2. Indeed, assume that the algebraic numbers 1 and 2 are very
close to the transcendental real number . Suppose that the algebraic number
satises
H(1 ) < H() < H(2 ).
For i = 1, 2 we have
| | |i | |i |,
and Corollary A.2 gives a lower bound for |i |. Thus, if |i | is sufciently small, say smaller than |i |/2, then | | must be greater than
|i |/2, thus greater than |i |; consequently, cannot be a very good
approximation of . Of course, the above argument is rather rough, since we
have omitted any reference to the heights and to the degrees of the algebraic
numbers involved. However, it lies at heart of Gutings work [269], where the
existence of real numbers with prescribed values for wn ( ) for some integers
n is established in a constructive way.
T HEOREM 7.7. Let n 1 and k 0 be integers. Let d be a real number with
(d + 1 n)(d + 1 n k) > n(n + k)(d + 1). Then, there exist real numbers
with
( ) = d.
wn ( ) = wn ( ) = . . . = wn+k ( ) = wn+k
In particular, the functions wn and wn take any value d > (2n + 1 +
4n 2 + 1)/2.
P ROOF. Let d, n, and k be as in the statement of the theorem. Throughout the
proof, the numerical constants implied by
and depend, at most, on n, k,
and d. Let (n i )i1 be a strictly increasing sequence of positive integers such
that n i+1 /n i tends to d + 1 and gcd(n i , n) = 1 for any i 1. Dene the
positive real number by

2n j .
n =
j1
Let be a real number with 0 < < 1. Let i 0 1 be such that d + 1 <
160
n i+1 /n i < d + 1 + holds for any integer i i 0 . For i i 0 , the polynomial

Pi (X ) := 2n i X n 2n i
i
2n j
j=1
satises H(Pi ) = 2n i and, since n i and n are coprime, a result by Dumas [214]
(see also [585], page 76) asserts that Pi (X ) is irreducible. It follows from

Pi ( ) =
2n i n j =
H(Pi )1n j /ni , Pi ( ) = n n1 H(Pi ),
j>i
j>i
and Lemma A.5 that Pi (X ) has a root i such that H(i ) = H(Pi ) and
H(i )d1
| i |
H(i )d1+ .
(7.50)
It is immediate that, for i i 0 , we have

H(i ) H(i+1 ) H(i )d+1+ .
(7.51)
Let be a real algebraic number of degree n + k. There exists an integer i with

H(i ) H() < H(i+1 ).
(7.52)
We may assume that H() is sufciently large in order to ensure that i i 0 .

Thus, (7.50) and (7.51) are satised. We distinguish two cases and introduce a
real number u > n, which will be specied later on.
First, we assume that = i and
H()n H(i )u .
(7.53)
Corollary A.2 asserts that |i | H()n H(i )nk , whence, by (7.53),

we get
|i | H(i )nku .
(7.54)
By (7.50), (7.52), (7.54), and the triangle inequality | | |i || i |,

we have | | H(i )d1+ , and thus | | H()d1+ , as soon as
u <d +1nk
(7.55)
and H() is large enough.

We assume now that = i+1 and H()n > H(i )u . We then get |i+1
| H()n H(i+1 )nk by Corollary A.2, and we infer from (7.50),
H() < H(i+1 ), and the triangle inequality | | |i+1 | | i+1 |
that | | H()d1+ holds if H()n H(i+1 )nk H()d1+ .
161
This condition is fullled as soon as

H(i+1 )n+k
H(i )(d+1n)u/n ,
thus, by (7.51), as soon as
u>
n(n + k)(d + 1 + )
.
d +1n
(7.56)
We can select a real number u such that (7.55) and (7.56) hold simultaneously
if satises
n(n + k)(d + 1 + )
< d + 1 n k .
d +1n
( ) = d as
Together with (7.50), by letting tend to 0, we obtain that wn+k
soon as
(d + 1 n k)(d + 1 n) > n(n + k)(d + 1),
(7.57)
as claimed. The last statement of the theorem follows by applying the rst
assertion to the pair (n, k) = (1, n 1). Indeed, condition (7.57)
then becomes
d(d n + 1) > n(d + 1), that is, precisely d > (2n 1 + 4n 2 + 1)/2.
A similar argument yields the result for the function wn , as proved by
Guting [269]. Instead of taking an arbitrary algebraic number with (7.52),
we take a polynomial P(X ) and we consider the index i for which H(Pi )
H(P) < H(Pi+1 ). We then apply Theorem A.1 and Rolles Theorem to get
that wn+k ( ) = d.
Guting ([269], Satz 7 and Satz 8) gave analogous, but slightly weaker, statements for roots of real numbers dened by their continued fraction expansions.
In that case, there is a new difculty, since the polynomials playing the roles
of Pi (X ) may not be irreducible.
It follows from a general result of Nishioka [438, 439] that, for any positive

j
integer d, the number j1 2(d+1) is an S-number.

In Chapters 5, 6 and in Sections 7.1 to 7.7 of the present Chapter, we have
developed different methods yielding to results towards the Main Problem introduced in Chapter 3.
M AIN P ROBLEM . Let (wn )n1 and (wn )n1 be two non-decreasing sequences
in [1, +] such that
n wn wn wn + n 1,
for any n 1.
162

wn ( ) = wn and wn ( ) = wn for any n 1.
We rst summarize the results concerning a single function wn or wn . As emphasized in Chapter 1, the function w1 (which is equal to w1 ) is well understood thanks to the theory of continued fractions. Further, Theorem 7.7
asserts, in an
with w >
explicit way, that wn and wn take any value w
2
(2n + 1 + 4n + 1)/2. In the range n w (2n + 1 + 4n 2 + 1)/2,
using Hausdorff dimension theory, it is proved in Chapter 5 that there exist
real numbers and with wn ( ) = w and wn ( ) = w. Moreover, by means
of a Cantor-type construction based on the effective, explicit result of Beresnevich [60] (see below Proposition 6.1), it is possible to construct real numbers
with prescribed values for w2 ( ).
The Main Problem is partially solved when the sequence (wn )n1 increases
sufciently rapidly, that is, faster than n 3 +2n 2 +4n+3 (see Theorem 7.1). This
result follows ultimately from Schmidts Subspace Theorem, hence, it does not
yield explicit examples of transcendental numbers with the required property.
Moreover, this approach is up to now the only one allowing us to conrm the
existence of real numbers with wn ( ) = wn ( ) for some integer n 2. The
real numbers obtained in this way are limits of sequences of algebraic numbers
which converge very rapidly. A similar idea is also used in Section 7.6, but the
construction is simpler.
For given integers n < n and real numbers n wn wn with (wn +
1)(n + 1) = (wn + 1)(n + 1), we can use the Hausdorff dimension theory to
ensure the existence of real numbers with wn ( ) = wn and wn ( ) = wn .
However, this method is ineffective.
Alternatively, fully explicit examples of real numbers with (very) specic
values of wn 1 ( ), . . . , wn k ( ) for some integers n 1 < . . . < n k have been
obtained as n-th roots of suitable convergent series, see Section 7.7.
7.9 Exercises
E XERCISE 7.1. Modify (very) slightly the proof of Theorem 7.1 to show that if
0 < w1 w2 . . . is a sequence in [1, +] such that wn > n 3 +2n 2 +3n+1
for any n 1, then there exists a real number such that wn ( ) = wn and
wn ( ) = wn for any n 1.
Hint. Observe that the exponent of g j in (7.8) is replaced by (n 2 + n +
2nn ) since we have g j = v j . Consequently, (7.7), (7.21), and (7.23) have
to be modied accordingly. Furthermore, (7.24) can be replaced by n =
wn + 1.
7.10 Notes
163
E XERCISE 7.2. Prove Theorems 7.2 and 7.3.
Hint. For in [(n 1)/n, n/4], use for j a root close to [g j ]1 of the
polynomial X n 2([g j ]X 1)2 , where = 2(n n + 1)/(n 2). To get

Theorem 7.2, argue as in the proof of Theorem 7.1, replacing, however, the use
of Theorem 2.7 by that of Corollary A.2 and (7.24) by n = wn n + 2. With
a slight modication, prove Theorem 7.3.
E XERCISE 7.3. Let w 1 be a real number. Apply Theorem 1.16 to prove that
there exist uncountably many U2 -numbers with w1 ( ) = w.
E XERCISE 7.4. Proof of Theorem 7.6.
Let m 2 be an integer and let := [0; a1 , a2 , a3 , . . .] be a real algebraic
number of degree m. For any j 1, denote by q j the denominator of the jth convergent of . Let (rk )k1 be an increasing sequence of positive integers
such that the ratio (log qrk+1 )/(log qrk ) tends to innity with k. For any j 1,
set b j = a j if j does not belong to the sequence (rk )k1 and b j = a j + 1
otherwise. Our aim is to prove that the number := [0; b1 , b2 , b3 , . . .] is a
Um -number with w1 ( ) = 1.
First, use Roths Theorem 2.1 to show that w1 ( ) = 1. Next, for any integer
k 1, set k := [0; b1 , b2 , . . . , brk , ark +1 , ark +2 , . . .], which is an algebraic
number of degree m. Show that (log | k |)/(log H(k )) tends to innity
with k. It remains for us to prove that is not a Un -number for some integer
n < m. Let be a real algebraic number of degree at most m 1 and with large
height. Let k be the integer such that qrk H() < qrk+1 . Bound | | from
below in terms of H() by using the inequality | | | | | |
1/m 3
with = k or k + 1 according as H() qrk+1 or not. Conclude.
7.10 Notes
Guting [269] gave a sufcient condition for a real number to be a T number.
Alniacik [11] used continued fractions to construct real numbers with
specic properties. He claimed that, for any positive real number , there exist T -numbers with w1 ( ) 1 + . However, it seems that his work contains serious gaps (not only because he did not use the correct denition of
T -numbers). His idea was to construct as a limit of algebraic numbers
= [a0 ; a1 , . . . , an , ], where the ai s are integers and the s are algebraic
numbers. Compared with Schmidt who considered a limit of numbers of the
shape + p /q , the advantage of Alniaciks approach is that the continued
164
fraction expansion allows us to control w1 ( ) in a satisfactory way. The remaining problem is to ensure that has good, but not too good, algebraic
approximants. To this end, one needs to establish a precise lower bound for
the height of , essentially in terms of the denominator of the n -th convergent of . This does not seem to be easy, and Alniaciks assertion still remains
unproved.
LeVeque [361] introduced the notion of strong (and weak) Liouville numbers. This has been further studied and rened by Alniacik [8, 10, 15] who
dened semi-strong Um -numbers and irregular semi-strong Um -numbers. For
any positive integers m and k, Alniacik [15] proved that every algebraic number of degree m can be represented by the sum and the product of two Umk numbers. Further, he established [12] that every real number, except possibly
Liouville numbers, can be represented as the sum of two U2 -numbers. His approach, somehow intricate, depends on the theory of continued fractions and
is explicit. Shortly thereafter, Pollington [459] adapted the method used by
Schmidt to conrm the existence of T -numbers to show that, for any positive
integer m, every real number can be expressed as a sum of two Um -numbers.
He also briey outlined the proof that every real number can be written as a
sum of two T -numbers. Petruska [455] showed that the sum or the product of
an arbitrary number of strong Liouville numbers is either rational or a Liouville
number.
Results on the distribution of the sequence (n) for a U2 -number have
been established by Baxa [54].
Burger and Struppeck [150] have investigated the statistical behaviour of
the partial quotients of real U2 -numbers. They proved the existence of U2 numbers with the property that if translated by any non-negative integer and
then squared, the result is a Liouville number (see Burger [148] for rened
statements). They further showed that there exists a real positive Liouville
number whose square root has, in its continued fraction expansion, only the
partial quotients 1 and 2, the partial quotient 2 occurring with probability 0,
for some given probability measure on the set of positive integers.
Other constructions of Um -numbers have been obtained by Oryan [442,
444, 446, 447] and by Yilmaz [606]. Results of Zeren [610, 611] (see also
Mahler [392] and Braune [117]) on values of gap power series at algebraic
points have been extended by Gurses [264, 265].
We have been concerned with the Lebesgue measure and with Hausdorff measures. However, a rather different measure for the size of sets of
real numbers arises in harmonic analysis. A set E of real numbers is called
7.10 Notes
165
an M0 set if it carries a probability measure , the FourierStieltjes transform

of which

exp(2i x y) d(y)
(x)
=
E
vanishes at innity. Aside from the RiemannLebesgue Lemma, no purely

metric property of E can ensure that E is an M0 set. Further, there exist sets
of full Lebesgue measure which do not share this property (see, for example,
[304]). This is also the case for the triadic Cantor set (see, for example, [305]
and [415], p. 168). Kaufman [314, 315] proved that the set B of badly approximable real numbers carries a measure in M0 , and so does any set K1 ( )
for > 1 (see also the notes at the end of Chapter 5). The former result
has been used by Pollington and Velani [460] towards Littlewoods Conjecture
(see Chapter 10). Moran, Pearce, and Pollington [428], following Schmidts
approach, proved that the set of T -numbers is an M0 set and claimed that, for
any positive integer m, the set of Um -numbers is an M0 set.
Haseo Ki [324] studied the sets of A-, S-, T - and U -numbers from the
point of view of Descriptive Set Theory. He established their possible locations in the Borel hierarchy (see, for example, the book of Kechris [316] for
an introduction to that topic). He followed the main steps of Schmidts construction and proved that the set of T -numbers is 03 -hard, while the set of
U -numbers is 30 -complete.
Let w be a positive real number and m 2 be an integer. Feldman [240]
constructed real numbers with w < wn ( ) m(m + 2)w for any integer
n = 1, . . . , m 1.
Amou [25] determined the values of wn ( ) and wn ( ) for the Champernowne number and any integer n = 1, . . . , 4. He used the fact, observed by
Mahler [380], that has a sequence of very good rational approximants whose
heights do not increase too rapidly. Lemma 1 of [25] rests on an idea of Guting
[269].
8
Other classications of real and
complex numbers
In 1932, Mahler [376] introduced the rst relevant classication of complex

numbers into several classes. To this end, for given positive integers n and
H and for any complex number , he considered the minimum of the real
numbers |P( )|, where P(X ) runs through the (nite) set of integer polynomials of degree at most n and height at most H , which do not vanish at .
Then, he let rst H tend to innity, and then n. This order is arbitrary, and
we may as well do the converse, or let tend to innity some given function
of the height and the degree. The former suggestion has been proposed by
Sprindzuk [532] in 1962, and the latter one by Mahler [393] in 1971. Both
yield new classications of complex numbers, to which Sections 8.1 and 8.2
are devoted, respectively. In Section 8.3, we present further results on the approximation by algebraic numbers, which, to some extend, rene Wirsings
Theorem 3.4.
Unlike in the previous chapters, we approximate complex numbers, and not
only real numbers. The main reason for doing this is that the results obtained
here are not sharp enough to ensure that, when we start with a real number, the
approximants we construct are also real numbers.
We warn the reader that in all the metric statements below the expression
almost all refers to the Lebesgue measure on the line (resp. on the plane) if
the set under consideration is a real set (resp. a complex set). A similar remark
applies for the Hausdorff dimension.
8.1 Sprindzuks classication

According to Sprindzuk [532], for a complex number and for positive
integers n and H , we set
166
167
wn (, H ) := min{|P( )| : P(X ) Z[X ], H(P) H, deg(P) n,

P( ) = 0},
w(,
H ) = lim sup
n
log log(1/wn (, H ))
,
log n
and
H ).
w(
) = sup w(,
H 1
In case w(
) = + and if there exists a positive integer H with w(,
H) =
+, we denote by H0 ( ) the smallest integer with this property and, otherwise, we put H0 ( ) = +. Further, we set
(,
H ) = lim sup
n
log wn (, H )
)
n w(
and
(
) = lim sup
H
(,
H)
.
log H
The quantities w(
) and (
) are called the -order and the -type of , respectively.
D EFINITION 8.1. Let be a complex number. We say that is an
A-number,
if 0 w(
) < 1 or if w(
) = 1 and (
) = 0;
S-number,
if 1 < w(
) < + or if w(
) = 1 and (
) > 0;
T -number, if w(
) = + and H0 ( ) = +;
U -number, if w(
) = + and H0 ( ) < +.
Two algebraically dependent complex numbers belong to the same class (see
Exercise 8.1), thus this classication satises the second requirement stated in
the Introduction to Chapter 3.
Sprindzuk [532] established that the A-numbers

are exactly the algebraic
numbers (see Exercise 8.1) and that the -type of any real (resp. complex)
transcendental number of -order 1 is at least 1 (resp. at least 1/2).
Let (w H ) H 1 be a sequence of numbers in [0, +] such that there is a
positive integer H0 with
w H = 0 for H < H0 and 1 w H w H +1 + for H H0 .
(8.1)
If is a complex transcendental number, then the sequence (w(,
H )) H 1 satises (8.1), and, conversely, we may ask whether, for any sequence (w H ) H 1
168
with (8.1), there exists a complex number such that w(,
H ) = w H for any
H 1. This is the analogue of the Main Problem stated in Chapter 3. Amou
[28] gave in 1996 an afrmative answer to that question.
T HEOREM 8.1. Let (w H ) H 1 be a sequence of numbers in [0, +] satisfying
(8.1). Then, there exist real numbers such that w(,
H ) = w H for any H 1.
The proof is involved and will not be reproduced here. We merely give some of
the main ideas. While we were previously interested in the distribution of algebraic numbers of bounded degree, the proof of Theorem 8.1 requires information on the distribution of algebraic numbers of xed height. Using an effective
version of a theorem of Kornblum on the polynomial analogue of Dirichlets
theorem on primes in arithmetic progressions, Amou obtained the following
auxiliary result. Let H be a positive integer and let be a real number with
| | = 1 and (H + 1)1 < | | < H + 1. Then there exists a positive constant
c(, H ), depending only on and H , and an innite set N (, H ) of positive
integers such that, for each n in N (, H ), there are at least (2H + 1)n/9 real algebraic numbers of degree n, height H , and with | | < exp{c(, H ) n}.
The lower estimate for the number of real algebraic numbers with these
properties is sharp and appears to be crucial in the remaining part of the proof,
which essentially rests on the method developed by Schmidt for constructing
T -numbers (see Chapter 7).
with
C OROLLARY 8.1. For any real number w 1, there exist S-numbers
w(
) = w.
There exist T -numbers. For any positive integer H , there exist
U -numbers with H0 ( ) = H .
Corollary 8.1 shows that Sprindzuks classication is non-trivial. As for metric results, Sprindzuk [532] established that almost all complex numbers are
S-numbers
of -order at most 2 and conjectured that 2 may be replaced by 1.
This was claimed by Chudnovsky [164], who only sketched the proof. Amou
[28] supplied a complete proof and gave some improvements in a subsequent
work [29].
of T HEOREM 8.2. Almost all complex (resp. real) numbers are S-numbers
order 1.
Theorem 8.2 is a straightforward consequence of Theorem 8.3, due to Amou
and Bugeaud [31].
T HEOREM 8.3. Let be a positive real number. Then, for almost all complex
(resp. real) numbers , there exists a positive constant c1 (, ), depending only
on and on , such that

|P( )| > exp (3 + )n log(n H ) n log n
(8.2)
169
for all integer polynomials P(X ) of degree n and height H satisfying

max{n, H } c1 (, ).
P ROOF. We only give a proof for the real case, since the complex case follows
exactly the same lines. Let I be a real interval of length 1. Let n 0 4 be
an integer which will be chosen later. In view of Sprindzuks Theorem 4.2, for
almost all real numbers in I , there are only nitely many integer polynomials
P(X ) of degree n less than n 0 and height H with

|P( )| exp (1 + )n log H .
Consequently, in order to prove the theorem, it is sufcient to show that the set
E of real numbers in I for which there exist innitely many integer polynomials P(X ) of degree n at least n 0 and height H such that

|P( )| exp (3 + )n log(n H ) n log n
is a null set. To this end, for any positive integers n, s, and H with 1 s n
and n n 0 , we consider the set A(H, n, s) of complex algebraic numbers
satisfying P() = 0 with multiplicity s for some integer polynomial P(X ) of
degree at most n and height H . Let E(H, n, s) denote the set of all real numbers
for which there exists an algebraic number in A(H, n, s) such that

3
3+
3+
2
n
log
H
+
n
log
n
, (8.3)
| | exp
s
s
s2
2s 2
and set E(H, n) := E(H, n, 1) . . . E(H, n, n). By Lemma A.7, each in
E belongs to innitely many sets E(H, n), thus
E
+
+

N n 0 n=N H =1
E(H, n)
E(H, n).
(8.4)
H0 1 n=n 0 H =H0
Observe that the cardinality of A(H, n, s) is bounded by n times the number

of integer polynomials of degree at most n and of height H , hence, we get

Card A(H, n, s) n (2H + 1)n+1 (2(H 1) + 1)n+1
2n(n + 1)(2H + 1)n 23n H n .
(8.5)
Furthermore, for any element in A(H, n, s), the minimal polynomial of

over Z, denoted by P (X ), divides some integer polynomial of degree n and
height H . By Lemma A.3, we have H(P ) 2n/s H 1/s , hence,
n
2 2
2
(8.6)
Card A(H, n, s) (21+n/s H 1/s + 1)1+n/s 27n /s H 2n/s .
s
Let D(H, n, s) be the domain in the complex plane consisting of numbers
whose distance from the interval I is less than the number in the right-hand
170
side of (8.3). It follows from Theorems A.1 and A.2 that

| | 2n/s n 5n/(2s) H 2n/s
holds for any , in A(H, n, s) with = . For s 3, this yields the upper
estimate
Card (A(H, n, s) D(H, n, s)) 23n/s n 5n/(2s) H 2n/s ,
(8.7)
if n is large enough.
To bound the Lebesgue measure of E(H, n, s), we
combine (8.3) with (8.5)
7/ log n s n, if
(resp. with (8.6),
with
(8.7))
if
s
=
1,
2
(resp.
if
n

3 s < n 7/ log n), and, choosing n 0 sufciently large, we get for any
s = 1, . . . , n the upper bound (E(H, n, s)) (n H )(2+) as soon as n n 0 .
Thus, we obtain
(E(H, n)) (n H )(1+)
and the double sum

(E(H, n))
nn 0 H 1
converges. The BorelCantelli Lemma 1.2 and (8.4) then show that the set E
has zero Lebesgue measure.
Maybe, it is possible to replace (8.2) by

|P( )| > exp (1 + )n log H f (n) ,
for some suitable function n f (n). Such a result seems however difcult to
prove.
A renement of the proof of Theorem 8.3 yields a strengthening of Theorem
8.2, whose proof is left as Exercise 8.2.
T HEOREM 8.4. There exists a complex (resp. real) set Ec (resp. Er ) of Hausdorff dimension zero such that, for all complex (resp. real) numbers not in Ec
(resp. Er ), there exists a positive constant , depending only on , such that

|P( )| > exp n log(n H )
for all integer polynomials P(X ) of degree n and height H . In particular, the
complex (resp. real) S-numbers

of -order strictly larger than 1, the complex
(resp. real) T -numbers and the complex (resp. real) U -numbers form sets of
Hausdorff dimension zero.
We may as well dene an analogous classication of the complex numbers in
terms of their approximation properties by algebraic numbers. For a complex
8.2 Another classication proposed by Mahler
171
number and positive integers H and n, we set

w n (, H ) := min{| | : algebraic, deg() n, H() = H, = }.
Observe that w n (, H ) does not coincide with wn (, H ) dened in Chapter 3,
which is equal to the minimum of the w n (, h) for h varying between 1 and
H . We then set
log log(1/w n (, H ))
w (, H ) = lim sup
.
log n
n
The function H w (, H ) describes how well can be approximated by
algebraic numbers of height H .
The method used to prove Theorem 8.1 allowed Amou [28] to obtain a
similar statement for the functions H w (, H ).
T HEOREM 8.5. Let (w H ) H 1 be a sequence of numbers in [0, +] satisfying
w H = 0 for H < H0 , 1 w H + for H H0 ,
for some positive integer H0 . There exist real numbers such that w (, H ) =
w H for any H 1.
We refer to [28] for a proof of Theorem 8.5. We point out that in the above
theorem the sequence (w H ) H 1 is not assumed to be non-decreasing, unlike
the sequence (w H ) H 1 in Theorem 8.1.
In 1971, Mahler [393] introduced a new classication of complex numbers in
terms of their approximation properties by algebraic numbers. Unlike his rst
classication and Sprindzuks one, he allowed degree and height to vary simultaneously. In this Section, we study this classication, and we report on various
results obtained by Mahler [393], Durand [216, 219], Nesterenko [436], and
Amoroso [21].
For an integer polynomial P(X ) = an X n + . . . + a1 X + a0 of degree n, we
set L(P) := |a0 | + . . . + |an | its length and
(P) := 2n L(P) = 2n (|a0 | + . . . + |an |)
(8.8)
its size. We may replace the number 2 in (8.8) by any real number strictly
greater than 1 and L(P) by the naive height H(P) of P(X ) without any notable change in the results below. Although the length is at present rarely used,
we choose to keep the original denition of Mahler. For a non-zero algebraic
number , we dene L() and () as L(P) and (P), respectively, where
P(X ) denotes the minimal polynomial of over Z. Setting (0) = 2, we have
172
() 2 for any algebraic number , and we point out that, for any positive
real number M, there exist only nitely many complex algebraic numbers
with () M.
The idea of Mahler was to associate to a given complex number a nonnegative valued non-decreasing function u O(u | ) of the integer variable
u, called the order function of .
D EFINITION 8.2. Let be a complex number. The order function u O(u |
) of is dened for any integer u 2 by
O(u | ) = sup{ log |P( )| : P(X ) Z[X ], (P) u, P( ) = 0}.
Analogously, we dene the order function u O (u | ) by
O (u | ) = sup{ log | | : algebraic, () u, = }.
Both functions u O(u | ) and u O (u | ) are non-decreasing. We
check that u O(u | ) vanishes identically if, and only if, is an algebraic
integer in an imaginary quadratic eld. Otherwise, O(u | ) is positive as soon
as u is sufciently large.
Mahler [393] dened a partial ordering and an equivalence relation on the
set of positive valued non-decreasing functions. Let a and b be two such functions of the integer variable u. If there exist two positive integers c and u 0 and
a positive real number such that
a(u c ) b(u) for u u 0 ,
then we write a(u) b(u) and b(u)
a(u). The relation denes a partial
ordering. If a(u) b(u) and b(u) a(u) hold simultaneously, then we write
a(u) b(u), which denes the equivalence relation . With respect to this
relation, the order functions can be distributed into disjoint classes, and
denes a partial ordering on these classes, which is not a total ordering. By
denition, two complex numbers and belong to the same class if, and only
if, their order functions u O(u | ) and u O(u | ) are in the same
class.
Mahler [393] (see Exercise 8.3) proved that two algebraically dependent
complex transcendental numbers belong to the same class, thus this classication satises the second requirement stated in the Introduction of Chapter 3 (if
we adopt the convention that all algebraic numbers belong to the same class).
The next lemma is used in the proofs of Theorems 8.6, 8.9, and 8.11.
L EMMA 8.1. Let be a complex number. Let n be an integer with n 2 and
H be a real number. There exist a positive constant c2 , depending only on ,
173
and an absolute positive constant c3 such that, for any H c2 , there is a

non-zero integer polynomial P(X ) with
deg P n,
H(P) H,
and
|P( )| H c3 n .
Moreover, if n 50, suitable values for c2 and c3 are given by (4 + | |)50 and
0.455, respectively.
H 30 and set H = [ H ]2 /2 if [ H ] is even, and

P ROOF. Assume
H = ([ H ] 1)2 /2 otherwise. Put c4 = 1 + | | + . . . + | |n . The

(2H + 1)n+1 points an n + . . . + a1 + a0 , where a0 , . . . , an are integers
with H a0 , . . . , an H , lie in the square centered at the origin and
of sidelength 2H c4 . We divide this square into (2H )n+1 small squares of
sidelength 2 H (2H )(n+1)/2 c4 . By Dirichlets Schubfachprinzip, two among
these (2H + 1)n+1 points lie in the same small square, thus we get a non-zero
integer polynomial P(X ) of degree at most n and height at most 2H such
that
|P( )| 21n/2 H
(n1)/2
c4 ,
and the desired estimate follows from H 4 H 2H H and n 2. The

last assertion is an easy computation and reproduces the numerical values obtained in [356]. We observe that we may alternatively use Theorem B.2 instead
of Dirichlets Schubfachprinzip (see the proof of Proposition 3.1).
Theorem 8.6, the proof of which is left as Exercise 8.4, derives from
Lemma 8.1.
T HEOREM 8.6. Let be a complex number which is not an algebraic integer
in an imaginary quadratic eld. Then O(u | ) log u if is algebraic,
otherwise we have O(u | ) (log u)2 .
We infer from Theorem 8.6 that all algebraic numbers which are not integers
in an imaginary quadratic eld belong to the same class.
At the end of [393], Mahler addressed some (at that time) open problems,
including the following ones. Do there exist uncountably many distinct classes
of real numbers? Do there exist real transcendental numbers and such that
the functions O(u | ) and O(u | ) are not comparable? Given a positive
valued non-decreasing function a, establish necessary and sufcient conditions for the existence of a real number with O(u | ) a(u). Does
there exist such a function a with O(u | ) a(u) for almost all real
numbers ?
In the following text, we answer (at least partially) these questions and we
investigate the relationship between O(u | ) and O (u | ).
174
P ROPOSITION 8.1. For any transcendental complex number and any real
number 2, we have O (u | )
O(u | ) and
O (u | )
O(u | )
lim sup
= + iff lim sup
= +.
(log u)
u
u (log u)
The proof of Proposition 8.1 is left as Exercise 8.5. As noticed by Mahler
[393], the works of Koksma [333] and Wirsing [598] suggest that the results for
O (u | ) should be completely analogous to those for O(u | ). However,
this is surprisingly not the case: Corollary 8.4 below asserts that O (u | )
O(u | ) does not hold for all transcendental real numbers .
D EFINITION 8.3. Let A denote the set composed of the rational integers
and the non-real quadratic integers. For any complex number not in A ,
the transcendence type of is
( ) = sup{ 0 : O(u | ) (log u) }.
Furthermore, for in [0, +], we set A = { C : ( ) = }.
By Theorem 8.6, any complex transcendental number satises ( ) 2 and
A1 consists of all algebraic numbers which are not in A . Further, we observe that the transcendence type of is the inmum of the real numbers
such that log |P( )| > (log (P)) holds for any integer polynomial P(X )
of sufciently large size (see Amoroso [21], Chudnovsky [164], and Waldschmidt [588]; there are some subtle differences between the denitions of
transcendence type).
Theorems 8.7 and 8.8 are concerned with the sets A . The rst assertion of
Theorem 8.7 has been proved independently by Durand [216] and Nesterenko
[436], while the second assertion is due to Amoroso [21, 24].
T HEOREM 8.7. The set A2 has full Lebesgue measure, that is, almost all complex numbers (resp. real numbers) have transcendence type 2. Furthermore,
there exists a complex set Ec (resp. a real set Er ) of Hausdorff dimension zero
such that, for all complex (resp. real) numbers not in Ec (resp. not in Er ), we
have O(u | ) (log u)2 .
P ROOF. Let be a complex transcendental number. If O(u | )
(log u)2 ,
we have O(u | ) (log u)2 by Theorem 8.6. Otherwise, for any positive real
number , there exist innitely many integer polynomials P(X ) such that

2
log |P( )| < log((P)) < (deg P) log(H(P) deg(P)) .

2
Thus, by Theorem 8.4, the set of complex (resp. real) transcendental numbers
for which O(u | )
(log u)2 has Hausdorff dimension zero, and the proof
of Theorem 8.7 is complete.
T HEOREM 8.8. Let q 2 and (3 +

positive integer k, set
k
bk = q
and
k =
175
5)/2 be real numbers. For any

k
2b j .
j=1
The transcendence type of the Liouville number

2bk = lim k
(q) :=
k1
k+
is exactly equal to . Furthermore, for any real number > 2, there are
uncountably many real numbers in A .
(q)
P ROOF. Write instead of . Let k be a positive integer. The minimal polynomial over Z of k is 2bk X 2bk k , thus we have
bk /2 < log (k ) < bk .
(8.9)
Further, since the b j s are pairwise distinct, we get

2bk+1 k 21bk+1 .
(8.10)
Combining (8.9), (8.10), and the inequalities

bk /2 < bk+1 < (2bk ) ,
(8.11)
we deduce that there exist absolute, positive constants c5 and c6 such that

exp c5 (log (k )) k exp c6 (log (k )) .
(8.12)
Thus, we have ( ) by Proposition 8.1.
We now deal with the reverse inequality. Let be an algebraic number
with = j for any integer j 1. By Corollary A.2, we have | k |
exp{2 log () log (k )}, whence, by (8.9), we get
| k | ()2bk .
(8.13)
By (8.10), (8.11), and the triangle inequality | | | k | | k |, we

infer from (8.13) that, under the assumption
bk 1 9 log (),
(8.14)
we have
| |
1
()2bk .
2
(8.15)
176
Let be the smallest positive integer k for which (8.14) is satised. If > 1,
we have
1
< 9 log () b 1 .
b1
Together with (8.11), this yields

/( 1)
b 2 9/( 1) log ()
,
and it follows from (8.15) that there exists a positive constant c7 , depending
only on and q, such that

(8.16)
| | exp c7 (log ())1+/( 1) .
Taking c7 large enough, (8.16) also holds if = 1. Since 1 + /( 1)
, it follows from (8.12) and (8.16) that there exists a positive constant c8 ,
depending only on and q, such that

| | exp c8 (log ())
holds for any algebraic number . Consequently, we have
lim sup
u
O (u | )
< +,
(log u)
hence, ( ) by Proposition 8.1, and ( ) = , by (8.12).

For the last assertion of the theorem, the reader is directed to Amoroso [21]
(notice that Amoroso used the size function max{log H(P), deg(P)} rather
than (P)). For any given real number > 2, he constructed inductively (and
in an effective way) uncountably many real transcendental numbers with prescribed type as the limits of sequences of algebraic numbers with increasing
degrees.
The rst assertion of Theorem 8.8, due to Durand [216], allowed him to answer
Problem 1 of Mahler [393], since any two functions u (log u)t1 and u
(log u)t2 , where t1 and t2 are positive, distinct real numbers, belong to different
classes. Furthermore, as observed by Durand [219], the set A+ of complex
numbers of innite type is uncountable, for it contains the Liouville numbers

given by the series j0 2b j , where (b j ) j1 is any increasing sequence of
distinct integers with
log b j+1
= +.
lim sup
log b j
j
This can be shown by a suitable modication of the beginning of the proof of
Theorem 8.8.
177
C OROLLARY 8.2. There are uncountably many distinct classes of real

numbers.
The construction given in Theorem 8.8 also implies the existence of real numbers whose order functions are not comparable.
C OROLLARY 8.3. Let t 3 be an integer and, for any k 1, set
bk = 2t
2k
and dk = 4t .
Then the Liouville numbers

=
2bk and =
2dk
k1
k1
do not satisfy either one of the following order relations: O(u | )

O(u | ),
O(u | )
O(u | ), O (u | )
O (u | ), O (u | )
O (u | ).
P ROOF. By Theorem 8.8, we have ( ) = t and () = t 2 , thus the relations
O(u | )
O(u | ) and O (u | )
O (u | ) cannot hold. The remaining
part of the proof is left as Exercise 8.6.
Theorems 8.9 and 8.10 are needed to show that the functions u O(u | )
and u O (u | ) behave differently for some complex numbers . They
have been announced by Durand [219] (up to the values of the numerical constants in Theorem 8.9).
T HEOREM 8.9. For any transcendental complex number and for any real
number with (4+| |)1000 , there exists an algebraic number satisfying
() and
log | |
(log )(log ())

.
5000
In particular, we have
lim sup
u
O (u | )
> 0.
(log u)2
We do not give a specic proof of Theorem 8.9 here, since it is an immediate

consequence of Corollary 8.5, established in Section 8.3.
T HEOREM 8.10. Let be a transcendental complex number. Let and be
positive real numbers with
lim inf
u
O (u | )
>0
(log u)
(8.17)
178
and
lim sup
u
O(u | )
= +.
(log u)
(8.18)
We then have ( 1) < .

P ROOF. Since is transcendental, we may assume that 2. By Proposition
8.1, (8.18) is then equivalent to
lim sup
u
O (u | )
= +.
(log u)
Thus, there exists a sequence (k )k1 of algebraic numbers with

10 2(k ) < (k+1 ) and log | k | 3k log (k ) ,
for any k 1. Let k be a positive integer and be a non-zero algebraic number
such that = k and
1

1
k
log (k )
log () k log (k )
(8.19)
2
holds. By Corollary A.2, we have log | k | 2 log () log (k ). Since
| | | k | | k |, we infer from (8.19) that

/( 1)
.
(8.20)
log | | 5 k 1/( 1) log ()
However, by (8.17), there exists an absolute positive constant c9 such that

log | | c9 log () ,
which, combined with (8.20), implies that

/( 1)
.
c9 log () 5 k 1/( 1) log ()
Since k can be taken arbitrarily large, we get ( 1) < , as claimed.
Theorem 8.6 immediately implies that
O(u | )
O(u | )
lim inf
>0
lim sup
2
u (log u)2
u (log u)
holds for any transcendental complex number . The analogous statement, with
the function O(u | ) replaced by O (u | ), is, however, not true.
C OROLLARY 8.4. For any complex transcendental number , we have
lim inf
u
iff
O (u | )
O(u | )
> 0 iff lim sup
< +
2
(log u)2
u (log u)
O (u | )
lim sup
< +.
2
u (log u)
179
Thus, any complex transcendental number of type strictly greater than 2

satises O (u | ) O(u | ). Furthermore, apart from a set of Hausdorff
dimension zero, any complex number satises
O (u | ) O(u | ) (log u)2 .
P ROOF. Let be a complex transcendental number. By Theorem 8.6, the second and the third inequalities of the corollary are equivalent. Furthermore, it
follows directly from Theorem 8.10 that (8.18) does not hold with = 2 if
(8.17) holds for = 2. Hence, it only remains for us to prove that the second
inequality of the corollary implies the rst one. Assume that there exists a positive constant c10 such that log | | c10 (log ())2 for any algebraic
number . By Theorem 8.9, for any sufciently large integer u, there exists an
algebraic number u satisfying
(u ) u
and

log | u | 104 (log u) log (u ) .
Consequently, we get log u 104 c10 log (u ) and log | u |

(108 c10 )1 (log u)2 . This yields O (u | ) (108 c10 )1 (log u)2 for u large
enough, and completes the proof of the rst assertion of the corollary.
As an immediate consequence, if the type of strictly exceeds 2, we then
have
lim sup
u
O(u | )
= +
(log u)2
and
lim inf
u
O (u | )
= 0.
(log u)2
This implies that O (u | ) (log u)2 , and we deduce from Theorem 8.6
that O (u | ) O(u | ) holds. Finally, the last assertion of the corollary
follows from Theorem 8.7.
Many interesting questions on the order functions u O(u | ) and
u O (u | ) have not been investigated up to now. For instance, we may
introduce, for any complex transcendental number , the quantity

O (u | )
=
0
.
i( ) := inf 0 : lim inf
u (log u)
We infer from Corollary 8.4 that i( ) = 2 for almost all complex numbers and from Theorem 8.10 that i( ) /( 1) if ( ) 2, hence,
i( ) = 1 if has innite type. However, it seems to be difcult to construct
explicit examples of complex numbers with prescribed values for ( ) and
i( ).
180
8.3 Transcendence measures and measures of algebraic approximation

Let be a given complex transcendental number. If, for any integer polynomial P(X ), we establish explicit (even up to the numerical constants) lower
bounds for |P( )| in terms of the degree of P(X ) and its height, we then get
immediately information on the classes to which belongs in the different
classications of numbers we have considered. This motivates the introduction
of the notion of transcendence measure.
D EFINITION 8.4. Let be a complex transcendental number. A function :
R1 R1 R>0 is a transcendence measure for if, for any sufciently
large integer n and any sufciently large real number H , we have |P( )|
exp{(n, log H )}, for all non-zero integer polynomials P(X) of degree at
most n and height at most H .
For instance, a transcendence measure for is (see Waldschmidt [589])
240 n (log H + n log n) (1 + log n),
which implies that is either an S-number or a T -number, and that its transcendence type is 2.
It follows from Denitions 8.2 and 8.4 that the transcendence type ( ) of a
given complex number is the inmum of the real numbers for which there
exists a positive constant c(, ) such that c(, ) (n + log H ) is a transcendence measure for .
We may as well adopt Koksmas point of view and introduce an analogue
of Denition 8.4. It becomes in that context more convenient to use Mahlers
measure (dened in Appendix A) instead of the naive height, although this is
not followed by all authors.
D EFINITION 8.5. Let be a complex transcendental number. A function
: R1 R1 R>0 is a measure of algebraic approximation for if
there exists a positive constant with the following property: for any positive
integer n and any real number M with n and log M n, we have
| | exp{(n, log M)} for all algebraic numbers of degree at most
n and Mahlers measure at most M.
For instance, a measure of algebraic approximation for is (see [589])
3 238 n (log M + n log n) (1 + log n).
From the knowledge of a measure of algebraic approximation for a complex
number we immediately get information on its location in Koksmas classication.
8.3 Transcendence measures and algebraic approximation
181
Connection between transcendence measures and measures of algebraic approximation are discussed in [589] and in Chapter 15 of [591]. A useful tool
is Lemma A.7 [189]. Examples of transcendence measures and measures of
algebraic approximation can be found in [164], pp. 4147, and in [244], pp.
163178.
Up to the present Section, we have used a function of the nave height and
the degree in order to measure the size of an algebraic number. However, a
rened notion of height, called the absolute height, is also frequently used in
transcendental number theory. It has its own advantages and disadvantages and
yields, for various kinds of problems, very precise results. The absolute height
of an algebraic number is denoted by h() and is dened as
h() =
1
log M(),
deg()
where M() is the Mahler measure of .

The remaining part of this Section is devoted to some results by Diaz [188],
Laurent and Roy [357], and Roy and Waldschmidt [483], in which both the degrees and the heights of the approximants are allowed to vary simultaneously.
The most interesting feature in Theorem 8.11, due to Diaz [188] (see also
Philippon [457]) and inspired by ideas of Laurent and Roy [356], is the assumption M n +1, which advantageously replaces the condition log M n
occurring in [356]. This is a direct consequence of the good dependence on the
2
degree n (essentially n n instead of 2n ) in Lemma A.8.
T HEOREM 8.11. Let be a complex number. Let n be an integer and M a real
number with n 50, M n + 1, and M (4 + | |)100 . Then, there exists an
algebraic number with
deg() n, M() M, and

6
(n log M() + deg() log M) .
| | exp
1000
P ROOF. We follow step by step the proof of Diaz [188]. By Lemma 8.1 applied with , n, and H := M(n + 1)1/2 , there exists a non-zero integer polynomial P(X ) of degree at most n and height at most H satisfying |P( )|
exp{0.455 n log H }. By Lemma A.3, we have M(P) (n+1)1/2 H(P) M
and
|P( )| exp{0.227 n log M},
(8.21)
since M n + 1. Write P(X ) as a product P(X ) = a P1 (X ) . . . Pk (X ) of

irreducible integer polynomials, where a is a non-zero integer and P j (X ) has
182
degree n j 1 for j = 1, . . . , k. By the additivity of the degrees and the

multiplicativity of the Mahler measure, we have
n log M = (n 1 + . . . + n k ) log M
and
n log M n(log M(P1 ) + . . . + log M(Pk )),

and (8.21) yields that

|P j ( )| |P( )| exp 0.113
(n j log M
1 jk
1 jk

+ n log M(P j )) .
Consequently, there exists an index j with 1 j k such that

|P j ( )| exp 0.113 (n j log M + n log M(P j )) .
(8.22)
Let be a root of P j (X ) such that | | is minimal. For any positive integer

, Lemma A.8 and (8.22) yield that

| |(+1)/2 exp 0.113 (n j log M + n log M(P j ))
1/2

+ n j log 2 + n j log n j M(P j ) .
Since n j n M and M(P j ) M, we get

| |(+1)/2 exp (0.113 0.5)n j log M

+ (0.113 + 1)n log M(P j ) + n j log 2 .
We take = 18 and bound log 2 by (log M)/200 in order to obtain

| |171 exp 1.034 deg() log M + n log M() ,
since deg() = n j and M() = M(P j ). This proves the theorem.
C OROLLARY 8.5. Let be a complex number. Let n be an integer and be
a real number with n 50 and 215n (4 + | |)100 . Then there exists an
algebraic number satisfying
deg() n, () , and

3
(n log () + deg() log ) .
| | exp
1000
P ROOF. Set M := 23n . We infer from Theorem 8.10 that there exists an
algebraic number with deg() n, M() M, and

6
(n log M() + deg() log M) .
(8.23)
| | exp
1000
8.3 Transcendence measures and algebraic approximation
183
Lemma A.2 implies that () 23 deg() M(), hence, we get

n log M() + deg() log M n log () + deg() log 5n deg()

n log () + deg() log /2,
(8.24)
by the assumption log 10n. The corollary follows from (8.23) and (8.24),
since () 23n M .
Theorem 8.9 follows from Corollary 8.5 applied with n and satisfying
(4 + | |)1000 and n = [(log )/14]. Actually, Durand [221], pp. 9596,
asserted a stronger result than Theorem 8.9: namely that, for any transcendental complex number and any positive real number , there exist innitely
many algebraic numbers such that

(log ())2
.
| | exp
48 log 2 +
Conversely, he claimed that, for almost all transcendental complex numbers
and any positive real number , there exist only nitely many algebraic numbers such that

(log ())2
.
| | exp
8 log 2
Theorem 8.12 below, due (up to the numerical constants) to Laurent and Roy
[357], provides a more general statement than Corollary 8.5.
T HEOREM 8.12. Let be a transcendental complex number. Let (n k )k0 be a
non-decreasing sequence of positive integers and let (tk )k0 be an unbounded
non-decreasing sequence of positive real numbers. Assume that, for any k 0,
we have
n k 50, tk 2n k , n k+1 2n k , tk+1 2tk .
Then, there exist innitely many integers k 0 with the following property:
for each of these k, there exists an algebraic number satisfying
n k tk
nk
deg() n k , log M() tk , and log | |
.
4000
22000
The proof of Theorem 8.12 requires many steps, which are displayed in Exercise 8.7. An interesting feature of Theorem 8.12 is the lower bound for the
degree of the approximants.
Corollary 8.6 should be compared with Theorem 3.4, which however provides no lower bound for the degrees of the algebraic approximants. Although
superseded by Theorem 3.12 (which depends on a statement not proved in the
present book), we quote it.
184
C OROLLARY 8.6. Let be a transcendental complex number. For any integer

n 50, there exist innitely many algebraic numbers with
n
deg() n and | | M()n/22000 .
4000
P ROOF. Set n k = n and tk = k for each integer k 2n. Since is transcendental, the set of algebraic numbers given by Theorem 8.12 applied with these
two sequences is necessarily innite.
Up to the numerical constants, Corollary 8.7 is due to Roy and Waldschmidt
[483] (see also Corollaire 2 from [357]).
C OROLLARY 8.7. Let be a transcendental complex number and 8000
be a real number. There exist innitely many algebraic numbers with
h() and log | | 108 deg()2 .
P ROOF. Set n k = k and tk = k/4000 for any k 1. Since 8000, we
have tk 2n k for any k 1. Applying Theorem 8.12, we get for innitely
many positive integers k an algebraic number k with
h(k ) =
tk
log M(k )
=
deg(k )
n k /4000
and
log | k |
k 2
n k tk
108 deg(k )2 .
=
22000
88 106
Innitely many of the k s are distinct since is trancendental.

Theorem 8.12 asserts the existence of innitely many integers k with a given
property, but it gives no information regarding the distribution of these integers.
In fact, U -numbers show that we cannot hope for a much more precise result
than Theorem 8.12 (Brownawell [118], Bugeaud [123], and Laurent [353]),
although a rened statement holds for almost all complex numbers [123].
8.4 Exercises
are
E XERCISE 8.1. Use Theorems A.1 and B.2 to prove that the A-numbers
exactly the algebraic numbers. Use estimate (3.3) of Chapter 3 to show that
two algebraically dependent complex numbers belong to the same class in
Sprindzuks classication.
E XERCISE 8.2. Let > 5 be a real number and denote by E() the set of real
numbers for which the equation |P( )| < exp{ n log(n H )} has innitely
8.4 Exercises
185
many solutions in integer polynomials P(X ) of degree at most n and height at

most H . Prove that the Hausdorff dimension of E() is less than or equal to
5/( 3) and establish Theorem 8.4.
E XERCISE 8.3. Prove that the order functions of two algebraically dependent
transcendental complex numbers are equivalent.
E XERCISE 8.4. Use Theorem A.1, Dirichlets Schubfachprinzip, and Lemma
8.1 to prove Theorem 8.6.
E XERCISE 8.5. Use Lemmas A.6 and A.8 to prove Proposition 8.1.
E XERCISE 8.6. Completion of the proof of Corollary 8.3.
We keep the notation of Corollary 8.3. For any integer k 1, set k =
k
b j and u = ( ). Show that there exists a positive constant c ,
k
2k
11
j=1 2
depending only on t, such that O (u k | ) c11 b2k+1 . Let r be a positive
integer and let be an algebraic number with () u rk . Argue as in the
(1+t 2 /(t 2 1))/t
proof of Theorem 8.8 to show that O (u rk | ) c12 b2k+1

holds for
a constant c12 , depending only on t and r , and for k large enough. Prove that
O (u | )
O (u | ) does not hold and make the suitable adaptations to
establish that O(u | )
O(u | ) does not hold either.
E XERCISE 8.7. Proof of Theorem 8.12. Let be a transcendental complex
number. Proceed as in the beginning of the proof of Theorem 8.11 to show that
for any sufciently large integer k there exists a non-zero integer polynomial
Pk (X ) such that
deg(Pk ) n k , log M(Pk ) tk /2, and log |Pk ( )| n k tk /10.
Then, prove that (at least) one of the following two statements holds.
(i) There exists a factor Q k (X ) of Pk (X ) in Z[X ] which is a power of an
irreducible, integer polynomial and satises
deg(Q k) n k , log M(Q k) tk /2, and log |Q k ( )| n k tk /15;
(ii) There exist two factors Fk (X ) and G k (X ) of Pk (X ) in Z[X ] which are
coprime and satisfy
deg(Fk ) + deg(G k ) n k , log M(Fk ) + log M(G k ) tk /2
and
log max{|Fk ( )|, |G k ( )|} n k tk /30.
186
We further need a rst auxiliary result (Lemme 3.4 and Proposition 3.5 from
[483]).
Let be a complex number and r be a positive real number. Let F(X ) =
m

(X i ) and G(X ) = bn nj=1 (X j ) be non-constant complex
am i=1
polynomials of degree m and n, respectively. Let f (resp. g) denote the number
of roots of F(X ) (resp. of G(X )) in the closed disc of radius r centered at
and set
:= min{| | : root of F G(X )}.
We then have
f g |Res(F, G)| 2mn M(F)ng M(G)m f |F( )|g |G( )| f .
(8.25)
If furthermore |am | 1 and |bn | 1 hold, then, for any non-negative integer
s, we have
min{1, }s
2 /4
|Res(F, G)| 2mn M(F)n M(G)m max{|F( )|, |G( )|}s .

(8.26)
Hint. For inequality (8.25), set pi := | i | and q j := | j |,

in such a way that we have p1 . . . pm and q1 . . .

qn . Bound 1i f,g< jn |i j | from above in terms of the q j s and

f <im,1 jg |i j | from above in terms of the pi s. Further, show that
we have

g
f

fg
|i j |
2 pi
2q j ,
(2)
1i f
1i f
1 jg
1 jg
and

|i j | 2(m f )(ng)
f <im
g< jn

f <im
ng
max{1, |i |}
m f
max{1, | j |}
g< jn
Prove (8.25) by combining the above inequalities.

If s m + n and all the | | are distinct, where runs over the roots of
FG(X ), then observe that a suitable choice of r yields (8.26). Use a continuity
argument to treat the case when all the | | are not distinct. Finally, deal
with the case s > m + n.
8.4 Exercises
187
The second auxiliary result, asserted below, follows from (8.26) with a suitable
choice of s.
Let n be an integer and t be a real number with t n 1. Let be a real
number. Assume that there exist coprime integer polynomials F(X ) and G(X )
such that
deg(F) n, log M(F) t and
log |F( )| nt/62,
deg(G) n, log M(G) t and log |G( )| nt/62.

Then, there exists an algebraic number such that
(F G)() = 0 and log | | nt/11000.
The third auxiliary result, asserted below, is a consequence of the second one.
Let n be an integer and t be a real number with t n 7 and tn 650.
Let be a real number. Assume that there exist coprime integer polynomials
F(X ) and G(X ) such that
deg(F) n, log M(F) t, and log |F( )| nt/30
deg(G) n, log M(G) t, and log |G( )| nt/30.
Then, there exists a root of the polynomial F G(F + G)(X ) which satises
n/4000 deg() n, log M() 2t,
and log | | nt/11000.
Hint. The second auxiliary result applied to the pair of polynomials (F, G)
yields an algebraic approximant 1 of . Without any loss of generality, we
may assume that F( ) = 0. Use Lemma A.3 to check that log M(F + G) 2t
and apply the second auxiliary result with 2t in place of t to the pair of polynomials (G, F +G). We then get another algebraic approximant 2 of . Observe
that 1 and 2 are distinct. Show that any non-zero algebraic number satises log || deg()h(). Set := max{deg(1 ), deg(2 )} and derive that
log |1 2 | 3.7t. Give an upper bound for log |1 2 | using the triangle
inequality and conclude.
Finish the proof of Theorem 8.12.
Hint. You may proceed as follows. If condition (ii) is satied for innitely many integers k, conclude by applying the third auxiliary result with
n = n k and t = tk /2 to the polynomials Fk (X ) and G k (X ). If condition (i)
is satised for any k sufciently large, then observe that there are innitely
many integers k such that the polynomials Q k1 (X ) and Q k (X ) are coprime.
188
Conclude by applying the third auxiliary result with n = n k and t = tk /2 to

these polynomials.
8.5 Notes
Clearly, we may dene other classications of transcendental complex numbers as well, by simultaneously varying the upper bounds for the degree and
height of the approximants, denoted by n and H , respectively. For many reasons, it turns out to be convenient to regard n and log H as equivalent. Another
example of size has been proposed by Mahler [395], where, for an integer polynomial P(X ) = an X n + . . . + a1 X + a0 , the quantity (P) is replaced by the
product (2 + |a0 |) . . . (2 + |an |).
Philippon [456] investigated classications of complex numbers by using
non-standard analysis. He dened the equivalence relation on the eld of
complex numbers by saying that if, and only if, is algebraic over the
eld Q() and is algebraic over the eld Q(). The quotient set C/ is the
set of algebraically closed subelds of C of transcendence degree 1; the special point in this set is the class of algebraic numbers. Using Diophantine approximation properties of complex numbers, Philippon dened a non-standard
local distance called dam (acronym of distance dapproximation mutuelle,
that is, mutual approximation distance translated into French) on C/ which
endows this space with a non-discrete Hausdorff topology. He recovered the
classication dened by Mahler [393] in 1971 by considering the dam to the
special point.
Chudnovsky ([164], Theorem 2.8, page 45) proved that e is a S-number
of -order 1. Other examples of S-numbers

of -order 1 have been given by
Amou [26].
Nesterenko [436] and Durand [217] introduced a multidimensional generalization of the classication dened by Mahler in [393]. They dened a
function (u | 1 , . . . , n ) extending to Cn the order function O(u | ). Durand showed some properties of which appropriately generalize those of
O(u | ). A conjecture of Chudnovsky ([163], Problem 1.3, page 178) has
been conrmed by Amoroso [24]. As for the real case, the metric statement
of Nesterenko [436], Theorem 2, gives a weaker result than expected and the
conjecture proposed on page 237 of [436] is still open.
Gelfonds criterion ([257], Chapter III, Section 4, Lemma VII) asserts
that the existence of a, in some sense, gap-free innite sequence of non-zero
integer polynomials taking small values at a given complex number implies
8.5 Notes
189
that is algebraic. It has been subsequently rened by Brownawell [118] and

Waldschmidt [587] (see also Chapitre 5 of [588]). A Gelfondtype criterion
with multiplicities has been established by Laurent and Roy [356]. Theorem
2b of Davenport and Schmidt [182] (see also [32]) and Theorem 4.2 of Roy
and Waldschmidt [484] are further variants of Gelfonds criterion.
Proposition 3.1 and Theorem 3.1 yield a necessary and sufcient condition for a real number to be transcendental. Mahler [394, 396] obtained a
simplied criterion, depending only on the approximation behaviour of a single sequence of integer polynomials of arbitrary degrees. Mahlers result can
be compared to transcendence criteria of Durand [215, 218].
Cijsouw [166], Waldschmidt [589], Chudnovsky [164], Diaz [187],
Nesterenko and Waldschmidt [437], and others obtained transcendence measures for some families of numbers, including log , e and , for non-zero
algebraic numbers , with log = 0 and irrational. Galockin [254], Miller
[424], Becker-Landeck [58], Wass [595], Becker [56], Nishioka [438, 439],
Nishioka and Topfer [440], Topfer [569, 570], and others established transcendence measures for values of Mahler functions at algebraic points. Lang
[348], Brownawell [119], and others (for further references and results, see,
for example, Shidlovskii [522] and Section 5.2 of Chapter 5 of [244]) obtained
transcendence measures for values of E-functions at algebraic points.
Let n and t be two real numbers with t n 1 and set
Q(n, t) := { Q : [Q() : Q] n, log M() t}.
The distribution of the sets Q(n, t) in the complex plane has been studied by
Waldschmidt [590], who dened specic sets of pairs (n, t). His results are
in some respect not entirely satisfactory and Bugeaud [123] was led to consider subsets of R (or C) rather than sets of pairs (n, t). For any > 0, he
introduced
F :=

<
n 0 1
h 0 1
nn 0
th 0 n

nt
nt
]
e
,
+
e
[
Q(n,t)R
and
F :=

>
n 0 1
Q(n,t)R
h 0 1
nn 0
nt
] e
th 0 n
nt c
, + e
[ .
Further, for a real number , he dened

0 ( ) := max{ > 0 : F } and 0 ( ) := min{ > 0 : F },
190
with the convention that 0 ( ) = 0 (resp. 0 ( ) = +) if belongs to none of

the sets F (resp. F ). The relationship between Waldschmidts and Bugeauds
denitions are explained in [123]. It is proved in [123] that there exist two
real numbers and with 1/850 1, such that almost all real
numbers satisfy 0 ( ) = and 0 ( ) = . This statement expresses that the
quality of the algebraic approximation is the same for almost all real numbers.
Presumably, we have = = 1.
Laurent [353] studied the location of the algebraic approximants of the

Liouville number j1 2 j! . Simultaneous algebraic approximation of Liouville numbers has been considered by Roy [478].
Results in a similar spirit as Theorem 8.12, but without a lower bound
for the degree of the approximants have been obtained by Laurent and Roy
[356] (see also [358] for deep generalizations to, for example, approximation
by hypersurfaces) and Laurent [352] (see his Theorem 4 for a weaker condition
between tk and n k ).
9
In Chapters 1 to 7, we have exclusively considered approximation of real numbers. However, Mahler [376] and Koksma [333] dened their classications
for complex numbers as well, and Mahler [378] also introduced an analogous
classication for the transcendental numbers in the eld Q p , the completion of
Q with respect to the prime number p. Furthermore, approximation in the eld
of formal power series has also been investigated, for example, by Sprindzuk
[534, 539]. In the present Chapter, we consider each of these settings, and we
briey describe the state of the art for the problems corresponding to those
studied in Chapters 1 to 7. Roughly speaking, it is believed (and it often turns
out to be true) that Diophantine approximation results in the real case have got
their complex and p-adic analogues, the proofs of which are a (more or less)
straightforward adaptation of those in the real case. This however does not hold
true anymore for Diophantine approximation in elds of power series. For instance, the analogue of Roths Theorem 2.1 does not exist when the ground
eld has positive characteristic, see, for example, the surveys by Lasjaunias
[351] and by Schmidt [515] for additional information.
9.1 Approximation in the eld of complex numbers

Let be a complex non-real number and let n be a positive integer. Following
Mahler [376] and Koksma [333], we dene the quantities wn ( ) and wn ( ),
and the classes A, S, T , U , A , S , T , and U , exactly as in Chapter 3.
Notice (see, for example, (16) of [598]) that we have w1 ( ) = w1 ( ) = 0,
w2 ( ) = w2 ( ), and w3 ( ) = w3 ( ). Both classications turn out to be equivalent, see, for example, Schneider [517]. Using Dirichlets Schubfachprinzip it
is easy to show that wn ( ) (n 1)/2 holds for any complex non-real number
which is not algebraic of degree at most n. Further, by a suitable modication of the proof of Theorem A.1 (see the Notes at the end of Appendix A),
191
192
we have wn ( ) (d 2)/2 if is algebraic of degree d, regardless of the positive integer n. This implies that wn ( ) = (n 1)/2 holds true for any complex
non-real algebraic number of degree n + 1. Moreover, proceeding exactly as
in the proof of Proposition 3.2, we get wn ( ) wn ( ) for any complex number and any positive integer n. Wirsing [598] proved the complex analogue of
Theorem 3.4.
T HEOREM 9.1. Let n 2 be an integer and be a complex non-real number
which is not algebraic of degree at most n. Then, we have
n1
,
2
wn ( )
wn ( )
,
2
wn ( )
wn ( )
,
2wn ( ) n + 2
wn ( ) wn ( )
(9.1)
(9.2)
and
wn ( )
n
.
4
(9.3)
Inequality (9.3) is a consequence of (9.1), (9.2), and wn ( ) (n 1)/2. It has

been slightly improved by Tishchenko [562].
It follows from (9.2) that wn ( ) = (n 1)/2 holds as soon as we have
wn ( ) = (n 1)/2. Combined with the observation above Theorem 9.1, this
implies that wn ( ) = (n 1)/2 holds for any complex non-real algebraic
number of degree n + 1. A partial result concerning the approximation of
complex non-real algebraic numbers follows from Schmidts Subspace Theorem. Namely, Evertse [233] showed that, for a complex non-real algebraic
number of degree d and a positive integer n, we have wn ( ) = wn ( ) =
min{(n 1)/2, (d 2)/2} either if [Q( ) : Q( ) R] (n + 3)/2 or
if [Q( ) R : Q] [(n + 1)/2]. It is likely that this assumption could be
removed.
At the end of [377], Mahler conjectured that almost all complex numbers
satisfy wn ( ) n/2 for all positive integers n. The fact that the equality
wn ( ) = (n 1)/2 should hold almost everywhere has been conjectured by
Kasch [310] and established for n = 2 in [310]. Shortly thereafter, Volkmann
[581] conrmed Kaschs conjecture for n = 3. Using his powerful method of
essential and inessential domains, Sprindzuk [538, 539] established the complex analogues of Theorems 4.1 and 4.2.
T HEOREM 9.2. Almost all complex numbers satisfy wn ( ) = wn ( ) =
(n 1)/2 for every positive integer n.
9.2 Approximation in the eld of Gaussian integers
193
The complex analogue of Theorem 4.3 is due to A. Baker [41]. Bernik and
Vasiliev [96] proved the complex analogue of Theorem 4.4 and deduced from
their result that complex algebraic numbers of bounded degree form a regular
system with parameters of regularity slightly better than in [94]. Kleinbock
[328] established the complex analogue of Theorem 4.6.
Bernik and Sakovich [94] used regular systems of complex algebraic numbers to establish the complex analogue of Theorem 5.3. They obtained partial
results towards the complex analogue of Theorem 5.7 (see also R. C. Baker
[48] and Sakovich [492]).
R. C. Baker [47] proved the existence of complex non-real numbers for
which wn ( ) and wn ( ) differ from pre-assigned values for every even integer
n 4, by constructing T -numbers with this property. Bugeaud [133] showed
that, for any given integer n 4, there exist complex non-real numbers for
which wn ( ) differs from wn ( ).
9.2 Approximation in the eld of Gaussian integers

The most natural complex extension of approximation to real numbers by rational integers is approximation to complex numbers by ratios of Gaussian
integers. This has been considered by Hermite and Hurwitz in the nineteenth
century (see [332], Chapter IV, Section 4). Unlike in the real case, a continued
fraction approach did not give the analogue of Hurwitz Theorem 1.18. This
was established by Ford [247], who used additional geometrical ideas based
on the Picard group SL (2, Z). Theorem 4.5 of Dodson and Kristensen [205]
provides an analogue of Theorem 1.1, but is presumably not best possible. The
problem of generalizing the theory of continued fractions to the complex eld
has been studied by A. L. Schmidt [499].
LeVeque [360] established the analogue in Z[i] of Khintchines Theorem
1.10. Another proof (yielding a slightly sharper result) has been given by Sullivan [547]. For extensions to other imaginary quadratic elds, see [547] and
Nakada [434].
Melian and Pestana [416] proved that balls in C of radius 1/|w|2 and centered at z/w, where z, w are Gaussian integers with w = 0 form a welldistributed system, a notion close to that of regular system. They obtained
Theorem 5.2 and its analogue in imaginary quadratic elds. Diophantine approximation in hyperbolic space has been investigated by many authors, see
[205] and Section 7.7 of Bernik and Dodson [86] for references. A full proof
of the analogue in Z[i] of Theorem 5.2 (as well as other results, including the
existence of badly approximable complex numbers), independent of the hyperbolic space framework, has been given by Dodson and Kristensen [205].
194
9.3 Approximation in the p-adic elds

In this Section, p denotes a given prime number. Every non-zero rational number a can be expressed uniquely under the form a = p m a , where m is an
integer and a is a rational number whose numerator and denominator are
prime to p. By denition, the p-adic valuation v p (a) is equal to m and we
set v p (0) = +. It denes a p-adic metric | | p on Q, which we normalize
by setting | p| p = p 1 . Since v p (a + b) min{v p (a), v p (b)} for any rational
numbers a and b, the metric | | p is ultrametric. The completion of Q with
respect to | | p is the p-adic eld Q p of p-adic numbers. Each element in
Q p has a unique representation
=

r =m
cr pr ,
where m and the coefcients cr are integers with cm non-zero and 0 cr

p 1 for any r m. We then have | | p = p m . In order to simplify the
notation, we write in this Section | | instead of | | p for the p-adic absolute
value. Furthermore, we denote by Q p a xed algebraic closure of Q p and we
recall that | | extends uniquely to Q p . For more information and results on
p-adic elds, see, for example, the books of Amice [20], Mahler [398], Robert
[473], and Schickhof [494].
In analogy with his classication of complex numbers, Mahler [378] proposed a classication of p-adic numbers. Let be an element of Q p . For given
n 1 and H 1, dene the quantity
wn (, H ) := min{|P( )| : P(X ) Z[X ], H(P) H, deg(P) n, P( ) = 0}
and set
wn ( ) = lim sup
H
log(H wn (, H ))
wn ( )
and w( ) = lim sup
.
log H
n
n
(9.4)
In analogy with Koksmas classication of complex numbers, we dene the

quantity
wn (, H ) := min{| | : algebraic in Q p , deg() n,
H() H, = },
(9.5)
where H() (resp. deg()) denotes the height (resp. the degree) of the p-adic
number , that is, by denition, the height (resp. the degree) of its minimal
195
polynomial over Z. Then, we set

wn ( ) = lim sup
H
log(H wn (, H ))
w ( )
and w ( ) = lim sup n . (9.6)
log H
n
n
In other words, wn ( ) (resp. wn ( )) is the upper limit of the real numbers w

for which there exist innitely many integer polynomials P(X ) (resp. algebraic
numbers in Q p ) of degree at most n satisfying
0 < |P( )| H(P)w1 (resp. 0 < | | H()w1 ).
Exactly as in Chapter 3, we call an A-, S-, T - or U -number and an A -, S -,
T - or U -number according to the behaviour of the sequences (wn ( ))n1
and (wn ( ))n1 , respectively. As in the real and complex cases, both classications turn out to be equivalent.
Mahler [378] proved that two algebraically dependent elements and in
Q p belong to the same class. Furthermore, if there is an integer polynomial
F(X, Y ) of degree M in X and degree N in Y such that F(, ) = 0, then, for
any integer n 1, we have
wn ( ) + 1 M(wn N () + 1) and wn () + 1 N (wn M ( ) + 1).
Actually, the denitions of the quantities wn ( ) and wn ( ) given here
differ from those used by Mahler [378], Sprindzuk [539], and Schlickewei
[496]. Indeed, for these authors, the numerator of the rst fraction in (9.4)
is log wn (, H ) instead of log(H wn (, H )) (the same applies to (9.6)).
This means that there is a shift by 1 in the value of the critical exponent, which
however does not imply any change regarding the class of a given p-adic number. We have adopted this choice in order to have, as will be stated below,
wn ( ) = wn ( ) = n for almost all p-adic numbers , with respect to the Haar
measure on Q p (see, for example, [539] or [86] for denition).
The shift by 1 is however not the only difference between our denition of
wn ( ) and the previous denitions. Indeed, in the previous literature, wn ( )

is dened as in (9.6), with however wn (, H ) replaced by the minimum of
| | over all numbers = which are zero of an integer polynomial of
degree at most n and height at most H . The point is that not every such is in
Q p . However, as can be seen by using the p-adic analogue of Rolles Theorem,
both denitions coincide (compare with the discussion before Lemma 3.1). To
prove this claim, let n 1 be an integer, H > 1 be a real number and
be a p-adic number not algebraic of degree at most n. Let 1 be an algebraic
196
number in Q p of height at most H and degree n 1 at most n, such that

| 1 | = min{| | : algebraic in Q p , deg() n, H() H, = }.
We may assume that 1 is not in Q p , otherwise there is nothing to prove.
(1)
(2)
(n )
Denote by 1 := 1 , 1 , . . . , 1 1 the conjugates of 1 numbered in such
a way that
(2)
(n 1 )
| 1 | | 1 | . . . | 1
|.
(2)
If | 1 | < | 1 |, then Krasners Lemma (see, for example, [473],

page 130) implies that 1 lies in Q p , that we have excluded. Consequently,
(2)
the minimal polynomial P1 (X ) of 1 over Z has two roots 1 and 1 with
(2)
| 1 | = | 1 |. Let be the largest integer such that
(2)
| p p 1 | = | p p 1 | < p 1/( p1) .

The polynomial P1 ( p X + ) has then two roots in the open disc of radius
p 1/( p1) . By the p-adic version of Rolles Theorem (see, for example, [473],
page 316), we deduce that the polynomial P1 ( p X + ) has a root in the open
unit disc. Thus, the integer polynomial P1 (X ) has a root 2 with H(2 )
2n 1 H(P1 ) 2n n H and
| 2 | < p p 2 | 1 |,
by our choice of . We do not know whether 2 is a p-adic number, but, if this is
not the case, we iterate this process as soon as we end up with an approximant
lying in Q p . This always happen since the degrees of the algebraic numbers
we construct from a strictly decreasing sequence. Consequently, there exists a
p-adic number with
2
H() 2n n n H and | | p 2n | 1 |.
By the same argument used to conclude the proof of Lemma 3.1, this shows
that the value of wn ( ) does not increase if the minimum in (9.5) is taken over
all algebraic numbers in Q p of degree at most n and height at most H . This
proves our claim.
Proposition 9.1, due to Mahler [378], provides the p-adic analogue of
Proposition 3.1.
P ROPOSITION 9.1. Let n 1 be an integer and let be a p-adic number
which is not algebraic of degree at most n. Then, we have wn ( ) n and, if
is transcendental, w( ) 1.
197
Proceeding exactly as in the proof of Proposition 3.2, we also get wn ( )

wn ( ) for any p-adic number and any positive integer n. Conversely, we
have the following analogue of Theorem 3.4.
T HEOREM 9.3. Let n 1 be an integer and be a p-adic number which is
not algebraic of degree at most n. Then, we have
wn ( ) wn ( ) n + 1,
wn ( ) + 1
,
2
n
wn ( )
wn ( ) n + 1
wn ( )
and
wn ( )
n
+
4
n 2 + 8n
.
4
(9.7)
(9.8)
(9.9)
(9.10)
Theorem 9.3 slightly improves Theorem 1 of Morrison [432], who got (9.7)
and wn ( ) min{n 1, (wn ( ) + 1)/2} instead of (9.8). Throughout this
paragraph, the numerical constants implied by
depend only on and on n.
Morrisons proof follows exactly the same lines as that of (3.12) of Theorem
3.4, and splits into several cases. In one case, using the notation of the bottom
of page 342 of [432], he obtained that there exist innitely many primitive,
integer polynomials P(X ) = at X t + . . . + a1 X + a0 = at (X 1 ) . . . (X
t ) of degree t at most n such that |P( )|
H(P)n1 , where the roots
of P(X ) are numbered in such a way that | 1 | | 2 | . . .
| t | and | 1 | < 1, | 2 | > 1. According to [432], this implies that
wn ( ) n 1, but this yields actually the sharper estimate wn ( ) n, as
we show now. Without loss of generality, we may assume that | | = 1. Hence,
we get |1 | = 1 and | j | = | j | > 1 for j = 2, . . . , t. Consequently,
| 1 1 | = | | and | 1 1
j | = 1 holds for j = 2, . . . , t. Setting
Q(X ) := X t P(1/ X ), we have |Q( 1 )| = |a0 ( 1 11 ) . . . ( 1 t1 )|
H(P)n1 , thus |a0 ( )|

H(P)n1 . Furthermore, relations between
coefcients and roots of Q(X ) yield that v p (a j ) v p (a0 ) for j = 0, . . . ,
t 1. Since P(X ) is primitive, we get |a0 | = 1, hence | |
H(P)n1 ,
as expected. Consequently, we obtain wn ( ) min{n, (wn ( ) + 1)/2}, which,
combined with (9.7), gives (9.8).
Estimate (9.9) is due to Bugeaud and Teulie [548], page 62, and rests essentially on [550], where Teulie carried the approach of Davenport and Schmidt
[182] to the eld of p-adic numbers. Combined with (9.8) and Proposition 9.1,
this gives (9.10). Tishchenko [566] obtained a slight renement of (9.10).
198
The denitive result on the approximation of p-adic algebraic numbers by

p-adic algebraic numbers of bounded degree is a consequence of (9.9) and the
p-adic version of Schmidts Subspace Theorem, due to Schlickewei [495].
T HEOREM 9.4. Let be an algebraic p-adic number of degree d and let n 1
be an integer. Then we have
wn ( ) = wn ( ) = min{n, d 1}.
Teulie [551] improved an earlier result of Morrison [432] by showing that
w2 ( ) 2 holds for every p-adic number not algebraic of degree at most 2,
thus establishing the p-adic analogue of Theorem 3.7 (this has been previously
claimed by Guntermann, page 347 of [263]).
Metric results over the p-adic elds are discussed in [539] and in Chapter 6
of [86]. The Haar measure replaces the Lebesgue measure, while the Hausdorff
dimension is dened exactly as the Hausdorff dimension on R.
T HEOREM 9.5. Almost all p-adic numbers satisfy wn ( ) = wn ( ) = n for
every positive integer n.
Theorem 9.5 follows from (9.9) and the p-adic analogue of Theorem 4.2 due to
Sprindzuk [539]. Previously, Turkstra [573] proved in his dissertation that almost all p-adic numbers are S-numbers. Further, partial results towards the
resolution of the p-adic version of Mahlers conjecture were due to Lock
[371], Kasch and Volkmann [312, 313] (however, there is a gap in [313]), and
Sprindzuk [535].
The p-adic analogue of Theorem 5.2 has been established by Melnicuk
[417] and that of Theorem 5.7 by Morotskaya [431].
T HEOREM 9.6. For any integer n 1 and any real number 1, the Hausdorff dimension of any of the sets
{ Q p : wn ( ) (n + 1) 1}, { Q p : wn ( ) (n + 1) 1},
{ Q p : wn ( ) = (n + 1) 1}, { Q p : wn ( ) = (n + 1) 1}
is equal to 1/ .
Theorem 9.6 is a restatement of Theorem 6.19 of [86]. Partial results were
obtained by Bernik and Morotskaya [91].
Beresnevich, Bernik and Kovalevskaya [69] proved that algebraic numbers
of bounded degree in Q p form an optimal regular system and they established
a complete p-adic analogue of Theorem 4.5 (see also Kovalevskaya [339, 340]
for the convergence case and the p-adic analogue of Theorem 4.4, and Beresnevich and Kovalevskaya [72] for the p-adic analogue of [62]). The p-adic
analogue of Theorem 4.6 is due to Kleinbock and Tomanov [331].
9.4 Approximation in elds of formal power series
199
Schlickewei [496] (see also [279]) proved that p-adic T -numbers exist by
adapting to the p-adic case the proof of Schmidt [507]. Alniacik [14] tried to
carry the proof of Schmidt [508] to the p-adic case, but the denition of T numbers he used is not the correct one. Hernandez [279] adapted [128] to the
p-adic case and proved that, for any integer n 2, there exist p-adic numbers
for which wn ( ) differs from wn ( ).
The existence of p-adic Um -numbers has been proved by Alniacik [8, 13],
who carried some of LeVeques results [361] to the p-adic case. Zeren [610],
Oryan [445], Xin [602], and Yilmaz [606] have constructed p-adic Um numbers.
9.4 Approximation in elds of formal power series

Let k be a (nite or innite) eld of arbitrary characteristic and denote by k[x]
the ring of polynomials with coefcients in k and by k(x) the quotient eld
of k[x]. We dene a non-Archimedean absolute value | | on k(x) by setting
|0| = 0 and
| f /g| := exp{degx ( f ) degx (g)}, for any non-zero
polynomials f , g in k[x],
where degx denotes the degree of a polynomial. The completion of k(x) for
this absolute value is the eld of formal power series K with coefcients in k.
an algebraic closure of k(x) contained in the algebraic closure
We denote by K
of K.
Let n be a positive integer and be in k. In analogy with Chapter 3, we
dene wn ( ) as the supremum of the real numbers w for which there exist
innitely many polynomials P(X ) in k[x][X ] of degree at most n satisfying
0 < |P( )| H(P)w ,
and wn ( ) as the supremum of the real numbers w for which there exist in of degree at most n and height arbinitely many algebraic numbers in K
trarily large satisfying
0 < | | H()w1 .
Here, the height of P(X ) is the maximum of the absolute values of its coefcients (these are elements of k[x]) and the height of is the height of its
minimal polynomial over k[x]. We require the height of the approximants to
be arbitrarily large to avoid trivialities: when the eld k is innite, there are
innitely many algebraic elements of bounded degree and bounded height in
unlike in the complex and p-adic cases.
K,
200
With the above dentions of wn ( ) and wn ( ), we dene the classes A, S,

T , U , A , S , T , and U exactly as in Chapter 3. For k a nite eld, this
classication has been introduced by Bundschuh [141].
Mahler [382] worked out an analogue in elds of formal power series to
Minkowskis theory of geometry of numbers. It follows from the analogue of
Theorem B.2 that wn ( ) n holds for any in K which is not algebraic of
degree at most n (when k is nite, see Amou [30] for a best possible statement
obtained via Dirichlets Schubfachprinzip).
When is algebraic of degree d and k has characteristic zero, Ratliff [469]
(see also Dubois [211]) proved that wn ( ) = min{n, d 1}. This does not hold
any more when k is a nite eld, as shown by de Mathan [412].
Guntermann [263] investigated the analogue of Theorem 3.4 in elds of
power series and proved Theorem 9.7 below.
T HEOREM 9.7. Let n be a positive integer. For any in K which is not algebraic of degree at most n, we have wn ( ) (n + 1)/2 and w2 ( ) 2 if
n = 2.
When k is a eld of characteristic zero, Theorem 9.7 and the inequality
wn ( ) wn ( ) n + 1 were previously obtained by Sprindzuk [534].
Sprindzuk [539] proved the analogue of Theorem 4.1 when k is a nite eld.
Like in the p-adic case, the Haar measure replaces the Lebesgue measure (see
[539], Part II, Chapter I).
T HEOREM 9.8. When k is a nite eld, almost all in K satisfy wn ( ) = n
for every positive integer n.
At the end of [263], Guntermann stated without proof that one can deduce
from Theorem 9.8 that almost all in K satisfy wn ( ) = n for every integer
n 1. However, this presumably needs an estimate like wn ( ) n/(wn ( )
n + 1), which does not seem to be in the literature; most likely, results of [382]
combined with the method of proof of Theorem 2.10 (see Exercise 3.3) should
yield such an estimate.
As noticed by Sprindzuk [534] (see also [539], page 150), taking for k the
eld of complex numbers, we get a classication of analytic functions in terms
of their behaviour with respect to approximation by algebraic functions.
When k is a nite eld, Kristensen [341] established the analogues of Theorems 1.10 and 5.2, as well as multidimensional extensions.
Examples of U2 -numbers for k arbitrary have been given by Burger and
Dubois [149] and Dubois [212]. Oryan [443] gave explicit constructions of
Um -numbers when k is a nite eld.
9.5 Notes
201
9.5 Notes
As for multiplicative approximation, metric results in the complex domain
involving the function + introduced in Chapter 4 have been established by
Yu [608].
Bernik and Morozova [92, 93] studied approximation of complex numbers by lacunary polynomials. They established [92] the exact value of the
Hausdorff dimension of the set of complex numbers at which the lacunary
polynomials of the form a3 X m + a2 X n + a1 X + a0 approximate 0 with a
given error term.
Generalizations of Dirichlets Theorem 1.1 and Khintchines Transference
Theorem B.5 to S-integer approximation in a number eld have been worked
out by Burger [143].
For simultaneous Diophantine approximation and improvements of
Dirichlets Theorem 1.1 in real quadratic elds, see Burger [146, 147].
The analogues in Z[i] of Theorems 6.6 and 6.8 have been established in
[71].
Hightower [280] proved the analogue of Theorem 3.7 for the approximation of complex numbers by algebraic numbers of degree at most 2 over a
given imaginary quadratic eld.
Markovich [407] proved the case n = 2 of Theorem 4.2 for polynomials
with coefcients in a given real number eld.
A. L. Schmidt [498] studied the approximation of quaternions.
Teulie [552] established the p-adic analogue of a result of Peck [450].
Abercrombie [1] established the existence of badly approximable p-adic
integers and thus provided a p-adic analogue of Jarnks results [289]. He used
an approximation scheme of Mahler [381], which is a substitute for the continued fraction algorithm. His work can be viewed as the p-adic analogue of
Schmidts paper [503].
Abercrombie [2] generalized the case n = 1 of Theorem 9.6 to systems
of linear forms and established the p-adic analogue of a result of Bovey and
Dodson [116]. This has been further extended by Dickinson, Dodson, and Yuan
[194].
Diophantine approximation by conjugate algebraic integers in p-adic
elds has been studied by Roy and Waldschmidt [484].
202
Jarnk [297] proved a p-adic generalization of Khintchines Theorem

1.10. It has been extended to systems of linear forms by Lutz [373].
A p-adic inhomogeneous analogue of Theorem 4.2 has been established
by Bernik, Dickinson, and Yuan [85]. For n = 1 (that is, for rational inhomogeneous approximation), the result is due to Lutz [373].
Zheludevich [614] proved Sprindzuks Conjecture H3 [541] on simultaneous approximation in the real, complex, and p-adic elds. This generalizes
an earlier result of Bernik [76] (see Notes to Chapter 4) and allowed him to extend [612, 613, 615] Theorem 4.2 to simultaneous approximation by algebraic
numbers. See the survey of Bernik [80] for further references. Zheludevichs
results have been extended by Kleinbock and Tomanov [331], in particular to
multiplicative approximation. See also Kalosha [308].
The p-adic analogues of Theorems 6.6 and 6.8 have been established by
Beresnevich, Dickinson, and Velani [71].
Jarnk [296] established a p-adic transference theorem.
Slesoratene [526] worked out an analogue of Sprindzuks theorem 4.2
for polynomials of degree two in two p-adic variables. For arbitrary degrees, a
weaker result is due to Yanchenko [605].
Menken [418] studied semi-strong p-adic U -numbers. He proved the padic analogue of Erdos result [227], that is, that any p-adic number is the sum
of two p-adic Liouville numbers.
Vaa nanen [574] established transcendence measures for the values of the
p-adic exponential function at algebraic points. Xu [603] proved that the values
of p-adic E-functions at algebraic points are S-numbers and he established a
trancendence measure. Molchanov [426] showed that values at some p-adic
algebraic points of Mahler functions satisfying certain functional equations
are p-adic S-numbers.
Diophantine approximation over the ring of adeles has been studied by
D. G. Cantor [151], who established an analogue of Khintchines Theorem
1.10.
Zong [617] worked out analogues of Mahlers results [401] in p-adic
elds and in elds of formal power series.
Zeroinnity laws for Diophantine approximation over p-adic elds and
elds of formal power series over a nite eld have been obtained by Bugeaud,
Dodson, and Kristensen [135].
9.5 Notes
203
Sprindzuk [539], page 160, claimed (and this has been later established
by Kleinbock [328]) that Theorem 4.3 holds in the complex case, with the
exponent n replaced by (n 1)/2. According to him, probably a similar result
is true for locally compact elds with a non-Archimedean valuation.
Burger [144] investigated decompositions of elements from an arbitrary
local eld into Liouville numbers.
An analogue of Khintchines Theorem 1.10 for formal power series over
a nite eld has been established by de Mathan [410] (see also Fuchs [252],
Kristensen [341], and Inoue and Nakada [285]).
Bundschuh [141] established transcendence measures for classical elements of elds of formal power series over a nite eld. His method has been
applied by Ozdemir
[448, 449], who showed that some explicitly given elements of K are not U -numbers.
Amou [30] established a formal power series analogue of Theorem 8.3
when k is a nite eld.
When k is nite, Becker [57] established transcendence measures for the
values of generalized Mahler functions and gave the rst explicit examples of
S-numbers.
Badly approximable linear forms over a eld of formal series have been
considered by Kristensen [342], who provided the analogue of Schmidts generalization [505] of Jarnks result [288].
10
We begin this Chapter with a short survey on the celebrated Littlewood Conjecture. We then gather open problems encountered in the preceding Chapters
with several new questions.
The reader interested in open questions in Diophantine approximation is
also directed to survey papers by Schmidt [513], Waldschmidt [592], Beresnevich and Bernik [65], and to the Appendix of Montgomerys book [427].
10.1 The Littlewood Conjecture

A famous open problem in Diophantine approximation is Littlewoods Conjecture which claims that, for any given pair (, ) of real numbers and for any
positive real number , there exist integers q, r , and s with q > 0 such that
q |q r | |q s| .
Denoting by the distance to the nearest integer, this statement is equivalent to
inf q q q = 0.
q1
(10.1)
Obviously, the conjecture holds true if or have unbounded partial quotients

in their continued fraction expansions. This is also the case if the numbers
1, , and are linearly dependent over the rational integers, by Dirichlets
Theorem 1.1.
Apart from these easy remarks, very little is known towards a proof or a
disproof of Littlewoods Conjecture. The rst important contribution is due to
Cassels and Swinnerton-Dyer [158] who showed that (10.1) holds when 1, ,
and belong to the same cubic eld. To this end, they applied Minkowskis
Theorem B.2 to get a dual formulation of the conjecture (actually, they only
used and proved the if part of Lemma 10.1).
204
10.1 The Littlewood Conjecture
205
L EMMA 10.1. Let and be real numbers such that 1, , and are linearly
independent over the rational integers. Then
inf q q q = 0
q1
holds if, and only if, we have

inf |x + y + z| max{|y|, 1} max{|z|, 1} : x, y, z Z,

(y, z) = (0, 0) = 0.
An open problem is to decide whether Littlewoods Conjecture holds for pairs

(, 1/ ), when the real number has bounded partial quotients. It follows
from Lemma 10.1 that the answer is positive if is not badly approximable by
quadratic numbers, that is, when we have
inf{|x + y + 2 z| max{|x|, |y|, |z|}2 : x, y, z Z,
(x, y, z) = (0, 0, 0)} = 0. (10.2)
However, the existence of transcendental real numbers not satisfying (10.2)
is still unproved, while it follows from Theorem A.1 that real cubic numbers
do not verify (10.2). Thus, Cassels and Swinnerton-Dyer [158] provide examples of real numbers such that Littlewoods Conjecture is true for the pair
(, 1/ ), although (10.2) does not hold. Since it remains unknown whether or
not cubic real numbers have bounded partial quotients, their result does not
yield examples of pairs of badly approximable real numbers for which Littlewoods Conjecture holds. However, Pollington and Velani [460] conrmed
that such pairs do exist.
T HEOREM 10.1. Let be a real number with bounded partial quotients. Then
the Hausdorff dimension of the set of real numbers with bounded partial
quotients such that the pair (, ) satises the Littlewood Conjecture is equal
to 1.

Chowla and DeLeon [162] observed that ( 2, 3) satises
the
Littlewood
Conjecture provided 0 is a limit point of the sequence (( 6/4)( 2 + 1)n )n1

modulo 1. The rst explicit non-trivial examples of pairs of real numbers (, )
such that (10.1) holds have been given by de Mathan [413].
De Mathan and Teulie [414] have proposed a very interesting mixed Littlewood Conjecture, claiming that, for any real number and any prime number
p, we have
inf q q |q| p = 0.
q1
(10.3)
206
They proved that (10.3) holds (actually, their result is much more general)
when is a quadratic irrationality, which can be viewed as the p-adic analogue
of the above quoted result from [158].
10.2 Open questions

This Section is devoted to open problems. Instead of formulating them in terms
of questions, we merely prefer to propose statements whose validity is open.
In most of the cases there is no evidence for, or against, the assertion claimed.
We do not recall the partial results obtained towards these problems, since they
can be easily found in the present book.
We begin by assertions on Mahlers and Koksmas classications of numbers, we continue with metric number theory, and we end up with rational
approximation and questions related to polynomials.
Problem 1 has been pointed out at the end of Section 3.4 of Chapter 3.
No argument can reasonably be put forward against it, although there is no
evidence in favour of it.
P ROBLEM 1. (Main Problem) Let (wn )n1 and (wn )n1 be two nondecreasing sequences in [1, +] such that
wn ( ) = wn and wn ( ) = wn for any n 1.
The rst conjecture, due to Wirsing [598], deals with the approximation of real
transcendental numbers by real algebraic numbers of bounded degree.
P ROBLEM 2. (Wirsings Conjecture) For any integer n 1 and for any real
transcendental number , we have wn ( ) n.
We propose a slightly weaker problem than Wirsings Conjecture.
P ROBLEM 3. The -type of any S -number is at least equal to 1, that is, we
have
lim sup
n
for any
S -number
wn ( ) + 1
1
n+1
As we discussed in Chapter 3, there are several possibilities to dene the type

of an S-number, but two of them may coincide.
10.2 Open questions
207
P ROBLEM 4. There exist real S-numbers with

lim sup
n
wn ( ) + 1
wn ( ) + 1
< sup
n+1
n+1
n1
and/or lim sup

n
wn ( ) + 1
w ( ) + 1
< sup n
.
n+1
n+1
n1
Corollary 3.2 asserts that wn ( ) = n holds if wn ( ) = n, but the converse is

an open question.
P ROBLEM 5. For any positive integer n, we have wn ( ) = n if wn ( ) = n.
Corollary 3.2 also suggests the following claim.
P ROBLEM 6. Any S-number of type 1 is an S -number of -type 1.
It is known (see Chapter 7) that there exist real numbers with wn ( ) = wn ( )
for every integer n 2, but it is not known if there exist S-numbers with this
property.
P ROBLEM 7. There exist real numbers such that w( ) = w ( ).
The existence of T -numbers has been proved by Schmidt (see Chapter 7).
However, it is at present not known whether or not there exist T -numbers with
some specic properties.
P ROBLEM 8. There exist T -numbers such that
lim
n+
wn ( )
= +.
n
P ROBLEM 9. Let be a real number with 1 < 3. There exist real T numbers of type .
P ROBLEM 10. Give an effective proof of the existence of T -numbers.
Let be a transcendental real number and let n, k be positive integers with
k 2. At the end of Section 3.2 (resp. in Exercise 3.6), we established a relation between wn ( ) and wn ( k ) (resp. between wn ( ) and wn ( k )). A natural
question (answered when n = 1 in Exercise 3.7) asks whether or not it is
possible to rene these estimates.
P ROBLEM 11. For any positive integers n, k, and any non-negative real number , there exist real transcendental numbers such that

k wn ( k ) + 1 = wn ( ) + 1 +
208
and

k wn ( k ) + 1 = wn ( ) + 1 + .
A similar question concerns the relations between wkn ( ) and wn ( k ).
P ROBLEM 12. For any positive integers n, k and any non-negative real number , there exist real transcendental numbers such that
wkn ( ) = wn ( k ) +
and
wkn
( ) = wn ( k ) + .
Problems 11 and 12 invite us to ask whether all integer powers of an S-number

have same type.
P ROBLEM 13. For any S-number and any non-zero integer k we have
t ( ) = t ( k ) and t ( ) = t ( k ).
The formulation of Problem 14, closely related to Problem 4, is rather vague.
There is no known result in this direction, and Theorem 7.7 suggests that Problem 14 could be very difcult.
P ROBLEM 14. Let be a transcendental real number and let n, m be positive
integers with n > m. Give a lower estimate of wn ( ) (resp. wn ( )) in terms of
( )).
n, m, and wm ( ) (resp. wm
In Section 3.6, we introduced the functions wn , w n , w n , and w n . Problem 15
is motivated by Theorem 3.9 establishing a link between wn and simultaneous
rational approximation.
P ROBLEM 15. Let n 2 be an integer and let wn n and wn n be two
real numbers satisfying
n
(n 1)wn + n
wn
.
wn n + 1
wn
There exist real numbers such that wn ( ) = wn and wn ( ) = wn .
Jarnks results [293, 295] on Khintchines Transference Theorem B.5 allows
us to think that the answer of Problem 15 could be afrmative.
P ROBLEM 16. For any integer n 2, nd relations between the six functions
wn , wn , wn , w n , w n , and w n . Find the sets of values taken by each of these
functions and the sets of their limit points.
10.2 Open questions
209
P ROBLEM 17. For any integer n 3 and any real transcendental number ,
we have w n ( ) = w n ( ) = n.
By Proposition 3.3 and the proof of Theorem 2.11, a positive answer to Problem 17 implies a positive answer to Problem 18.
P ROBLEM 18. For any integer n 4, any positive real number , and any
real transcendental number , there exist a constant c1 (, n, ) and innitely
many real algebraic integers of degree less than or equal to n such that
| | c1 (, n, ) H()n+ .
The assumption n 3 in Problem 17 (resp. n 4 in Problem 18) is
needed because of Roys results [480, 481] asserting
the existence of real
transcendental numbers with w 2 ( ) = (1 + 5)/2 and such that |
| c2 H()(3+ 5)/2 holds for any algebraic integer of degree at most

three, for a suitable positive constant c2 depending only on . In view of
this result, it may be doubtful that the answers of Problems 17 and 18 are
afrmative.
Recall that an algebraic unit is an algebraic integer whose minimal polynomial over Z has constant coefcient equal to 1. Approximation by algebraic
units has been investigated by Teulie [549].
P ROBLEM 19. For any integer n 3, any positive real number and any real
transcendental number , there exist a constant c3 (, n, ) and innitely many
real algebraic units of degree less than or equal to n such that
| | c3 (, n, ) H()n+1+ .
It is known that the conclusions of Problems 17, 18, and 19 hold true for almost
all real numbers .
Theorème 6 of [140] and Theorem 2 of [125] give links between approximation by algebraic integers of degree at most n and algebraic numbers of
degree at most n 1.
P ROBLEM 20. For any integer n with n 3, compare the quality of approximation by algebraic numbers of degree at most n 1 with the quality of approximation by algebraic integers of degree at most n, and with the quality of
approximation by algebraic units of degree at most n + 1.
We now state more precise questions on the approximation by algebraic numbers. For a given positive integer n and a given real number , we observe that
the quantities wn ( ) and wn ( ) have been dened as the inma of two sets,
hence, they provide no information on the following problems:
210
(P.1) Do there exist a positive constant c(, n) and innitely many integer polynomials P(X ) of degree at most n such that
|P( )| c(, n) H(P)wn ( ) ?
(P.2) Do there exist a positive constant c (, n) and innitely many algebraic
numbers of degree at most n such that
| | c (, n) H()wn ( )1 ?
After having determined the exact value of wn ( ) and that of wn ( ), it is natural to ask whether (P.1) and (P.2) have an afrmative answer. This corresponds
to the notion of eigentlicher Index introduced by Koksma [333]. In order to
take into account these two problems, we introduce some new notation.
D EFINITION 10.1. Let n be a positive integer and be a real number. Let w
.
be a positive real number. We write wn ( ) = w if wn ( ) = w holds and the
.
answer to (P.1) is positive. Likewise, we write wn ( ) = w if wn ( ) = w holds
and the answer to (P.2) is positive.
For instance, it follows from Theorem 4.5 that almost all real numbers satisfy
.
.
wn ( ) = n and wn ( ) = n for any positive integer n. We propose a renement
of Problem 1.
P ROBLEM 21. Let (wn )n1 and (wn )n1 be two non-decreasing sequences in
[1, +] such that
.
.
wn ( ) = wn and wn ( ) = wn .
Some of the results of Chapter 7 show, or can be adapted to show, that the
answer to Problem 21 is positive for certain pairs (n, w).
Schmidt [512], p. 258, proposed a slightly stronger conjecture than Wirsings; however, he expressed in Summer 2003 serious doubts on the validity of
his conjecture.
P ROBLEM 22. (Schmidts Conjecture) For any positive integer n and any real
transcendental number , there exist a constant c4 (, n) and innitely many
real algebraic numbers of degree less than or equal to n such that
| | c4 (, n) H()n1 .
10.2 Open questions
211
Using Denition 10.1, Schmidts Conjecture can be rephrased as follows: For

any integer n 1 and any real transcendental number , we have either
.
wn ( ) > n or wn ( ) = n.
Davenport and Schmidt [180] gave a positive answer to Problem 22 in the
case n = 2, but Problem 23, where we x the exact degree of the approximants
instead of an upper bound for it, remains unsolved, even for n = 2.
P ROBLEM 23. For any integer n 2 and any real transcendental number ,
there exist a constant c5 (, n) and innitely many real algebraic numbers of
degree n such that
| | c5 (, n) H()n1 .
Results from Roy [480, 481] would speak in favour of the existence of transcendental numbers which do not satisfy the conclusion of Problem 23, even
for n = 2.
The next problem deals with transcendental numbers badly approximable
by algebraic numbers.
P ROBLEM 24. Let n be a positive integer. There exist a real transcendental
number and positive constants c6 (, n) and c7 (, n) such that
| | c6 (, n) H()n1
for any real algebraic number

of degree n
and
| | c7 (, n) H()n1
for innitely many real

algebraic numbers of degree n.
Theorem 1.9 answers positively Problem 24 for n = 1, however, there is no

contribution for n 2. Furthermore, Theorem 2.9 asserts that for any integer
d 2, real algebraic numbers of degree d are badly approximable by real
algebraic numbers of degree at most d 1.
Problem 24 can be formulated in terms of polynomials, as well.
P ROBLEM 25. Let n be a positive integer. There exist a real transcendental
number and a positive constant c8 (, n) such that
|P( )| c8 (, n) H(P)n for any integer
polynomial P(X ) of degree n.
(10.4)
212
Khintchine [318] proved that the set of real numbers for which (10.4) holds
for some positive integer n and some positive constant c8 (, n) has Lebesgue
measure zero.
Very little is known regarding transcendental numbers which are badly approximable by algebraic numbers of bounded degree.
P ROBLEM 26. Let n 2 be an integer. There exist transcendental real numbers such that, for some positive constant c9 (, n), we have

1
| | c9 (, n) H()n1 log 3H() ,
for any non-zero real algebraic number of degree at most n.
Of course, if Problem 26 could be solved afrmatively, it then would be desirable to determine the Hausdorff dimension of such exceptional sets. Problem
26 can be formulated in terms of polynomials, as well.
The next problem extends a question posed by Beresnevich, Dickinson, and
Velani [70] in the case of (simultaneous) rational approximation.
P ROBLEM 27. Let n be a positive integer and let > 1 be real. Is the set of
real numbers for which there exists a positive constant c10 ( ) such that
| | H() (n+1) for innitely many in An
and
| | c10 ( ) H() (n+1) for every in An
non-empty? If yes, determine its Hausdorff dimension.
Problem 27 has been solved by Bugeaud [129] when n = 1. One may also
replace the approximation functions x x (n+1) by more general nonincreasing functions .
Problem 28 is likely to be difcult, especially for n 3.
P ROBLEM 28. For any integer n 2 and any real number > 1, determine
the Hausdorff measure at the critical exponent 1/ of the set Kn ( ).
Problems 29 and 30 deal with the exponents of approximation introduced in
Section 3.6.
P ROBLEM 29. Let n 2 be an integer. Let wn be a real number with 1
wn < n. Determine the Hausdorff dimension of the set of real numbers such
that wn ( ) = wn .
P ROBLEM 30. Determine the Hausdorff dimension of the set of real numbers
such that w 2 ( ) > 2 (resp. w 2 ( ) > 2).
There are very few results concerning the existence of real numbers with prescribed order of approximation by algebraic numbers of different degrees.
10.2 Open questions
213
P ROBLEM 31. Let n and n be positive distinct integers. Let 1 and 1

be real numbers. Then the set Wn ( )Wn ( ) is non-empty, and its Hausdorff
dimension is equal to min{1/, 1/ }.
P ROBLEM 32. Let n 2 be an integer and 1 be a real number. The sets
Wn ( ) and Wn ( ) contain badly approximable real numbers.
The existence of normal (resp. non-normal) numbers with specic approximation properties by rational numbers has been investigated by several authors
(see Notes in Chapters 1 and 5).
P ROBLEM 33. Let n be a positive integer and > 1 be a real number. The
sets Wn ( ) and Wn ( ) contain normal numbers and numbers which are simply
normal in no base.
P ROBLEM 34. There exist normal T -numbers and T -numbers which are simply normal in no base.
The next two problems concern Diophantine approximation on sets of
Lebesgue measure zero. They were posed by Mahler [394, 399, 400].
P ROBLEM 35. There exist very well approximable numbers, other than Liouville numbers, in the triadic Cantor set.
P ROBLEM 36. Every point in the triadic Cantor set is either rational or transcendental.
It is known that the triadic Cantor set contains badly approximable real numbers (see, for example, Theorem 10.3 of [329]).
We propose three problems on multiplicative approximation. With the functions wn+ introduced in Section 5.7, we set
w+ ( ) = lim sup wn+ ( ),
n+
and we say that is an

A+ -number, if w+ ( ) = 0;
S + -number, if 0 < w + ( ) < +;
T + -number, if w+ ( ) = + and wn+ ( ) < + for any n 1;
U + -number, if w + ( ) = + and wn+ ( ) = + from some n onwards.
P ROBLEM 37. What can be said about the sets of A+ -, S + -, T + -, and U + numbers?
Problem 37 requires presumably new methods and seems to be very difcult.
214
P ROBLEM 38. Let n 1 be an integer and : R1 R>0 be a nonincreasing continuous function such that

(log h)n1 (h) = +.
h1

|P( )| < + (P)
(10.5)
has innitely many solutions in integer polynomials P(X ) of degree at most n.

Theorem 4.7 asserts that if the above sum converges, then (10.5) has only a
nite number of solutions in integer polynomials P(X ) of degree at most n.
P ROBLEM 39. Let be a positive real number and let n be a positive integer.
Then the Hausdorff dimension of the set Kn+ ( ) of real numbers such that the
inequality
|P( )| < + (P)2 +1
has innitely many solutions in integer polynomials P(X ) of degree at most n
is equal to 1/ .
This is Theorem 5.2 for n = 1 and has been proved for n = 2 (see Exercise 5.6).
We recall a celebrated problem on rational approximation of algebraic numbers, which is perhaps due to Khintchine (see [521], p. 156).
P ROBLEM 40. Any real algebraic number of degree at least 3 has unbounded partial quotients in its continued fraction expansion, that is, for
any positive real number , there exists a rational number p/q such that
| p/q| < /q 2 .
As observed in the Notes in Chapter 2, known upper bounds for the partial
quotients of real algebraic numbers of degree at least 3 are presumably much
larger than the truth.
P ROBLEM 41. Let be an algebraic number of degree at least 3. Give sharp
upper bounds for its partial quotients.
We mentioned in Section 1.4 the DufnSchaeffer Conjecture.
P ROBLEM 42. (DufnSchaeffers Conjecture) Let : R1 R0 be some
continuous function. Then the set

p
p
R : < (q) for innitely many rationals
q
q

with gcd( p, q) = 1
has full Lebesgue measure if the sum
+
q=1 (q)(q)
diverges.
10.2 Open questions
215
There are many open problems concerning continued fractions. One of them is
related to Ridouts Theorem 2.3, and has been rstly formulated by Erdos and
Mahler [229].
P ROBLEM 43. Let be a real transcendental number for which there exist a
positive number M and innitely many convergents pn /qn such that the greatest prime factor of pn qn is less than M. Then is a Liouville number.
The fact that such numbers do exist has been asserted by Erdos and Mahler
[229] and proved by Fraenkel and Borosh [250], who also showed that they
form a set of Hausdorff dimension zero.
Maillet [403] proved that if is a Liouville number and f a rational function
with rational coefcients, then f ( ) is also a Liouville number. This motivated
the following question posed by Mahler [400].
P ROBLEM 44. Which analytic functions f have the property that if is any
Liouville number, then so is f ( )? In particular, are there entire transcendental
functions with this property?
Bernik and Dombrovski [87] and Alniacik [17] obtained some results related
to Problem 44.
For two positive real numbers i and j with i + j = 1, let us denote by
B(i, j) the set of pairs (, ) of real numbers for which there exists a positive
constant c11 (, ) such that
max{q1/i , q1/j }
c11 (, )
q
holds for any positive integer q. Schmidt [513] proposed the following conjecture, investigated by Pollington and Velani [461].
P ROBLEM 45. (Schmidts Conjecture) For any pairs (i, j) and (i , j ) of positive real numbers with i + j = i + j = 1, the sets B(i, j) and B(i , j ) have
non-empty intersection.
As noted by Schmidt [513], a counter-example to his conjecture would imply
Littlewoods Conjecture.
P ROBLEM 46. (Littlewoods Conjecture) Let and be real numbers. For
any positive , there exist integers q, r , and s with q > 0 and
q |q r | |q s| .
We display a particular case of the Littlewood Conjecture, and a related
question.
216
P ROBLEM 47. Let be a badly approximable real number. For any positive
, there exist integers q, r , and s with q > 0 and
q |q r | |q 1 s| .
By Lemma 10.1, there is a connection between Problems 25 and 47.
P ROBLEM 48. Let i be a real number with 0 i 1. There exist transcendental real numbers and and a positive constant c12 (, ) such that
c12 (, )
,
| p + q + r |
max{| p|, |q|, 1}1+i min{| p|, |q|, 1}1i
for any non-zero integer triple ( p, q, r ).
Davenport [175] proved that Problem 48 is true for i = 1. By Lemma 10.1,
Problem 48 for i = 0 corresponds to the Littlewood Conjecture.
We recall the mixed Littlewood Conjecture, proposed by de Mathan &
Teulie [414].
P ROBLEM 49. Let be a real number and p be a prime number. For any
positive , there exist integers q and r with q > 0 and
q |q r | |q| p .
As for restricted simultaneous Diophantine approximation, Schmidt [511] (see
also Thurnheer [554, 555, 556, 557, 558]) investigated the following question.
P ROBLEM 50. Let and be real numbers. There exist a constant c13 (, )
and innitely many triples ( p, q, r ) of integers with p and q positive and
| p + q + r | < c13 (, ) (max{ p, q})2 .
(10.6)
An easy metric argument shows that Problem 50 holds true for almost all pairs
(, ) of real numbers. One may ask for the Hausdorff dimension of the set of
pairs (, ) for which (10.6) holds with an exponent of max{ p, q} greater than
2, for innitely many triples ( p, q, r ) of integers with p and q positive.
We recall a question posed in Chapter 8.
P ROBLEM 51. Let be a positive real number. Then, for almost all real numbers , there exist a constant c14 (, ), depending only on and on , and a
constant c15 (n) depending only on n such that

|P( )| > exp (1 + )n log H c15 (n)
for all integer polynomials P(X ) of degree n and height H satisfying
max{n, H } c14 (, ).
There are many connections between approximation by algebraic numbers and
separation of roots of integer polynomials.
10.3 Notes
217
P ROBLEM 52. Let n 2 and H be positive integers. Denote by L(n, H )

(resp. L (n, H ), L 0 (n, H ), L 0 (n, H )) the minimum of the minimal distance
between two of the roots of a (resp. an irreducible, a monic, an irreducible
and monic) separable integer polynomial P(X ) of degree n and height at most
H . Give sharp upper bounds for the upper limits as H tends to innity of the
quantities
log L(n, H ) log L (n, H ) log L 0 (n, H )
,
,
,
log H
log H
log H
log L 0 (n, H )
.
and
log H
In particular, we may ask whether the dependence on H(P) in Theorem A.3 is
best possible.
The point of view taken in Problem 52 is that of Wirsing, since we x an
upper bound for the degree and let then the height tend to innity. We may as
well do the converse.
P ROBLEM 53. With the notation of Problem 52, for a xed positive integer
H , determine the asymptotic behaviour of the functions n L(n, H ), n
L (n, H ), n L 0 (n, H ), and n L 0 (n, H ) as n tends to innity.
To conclude, we state a conjecture of Schmidt [513] on approximation in a
given number eld.
P ROBLEM 54. (Schmidts Conjecture) For any number eld K and any positive real number , we have

2
,
| | > c16 (K, ) max{H(), H()}
for any distinct , in K, where c16 (K, ) is some constant depending only on
K and on .
Schmidts Conjecture, if established, would provide a considerable improvement on Theorem 2.5. For results towards this problem, see the survey of Evertse [232].
Most of the questions posed in this Section can be addressed in the context
of Diophantine approximation in p-adic elds or in elds of formal power
series, as well.
10.3 Notes
Davenport and Lewis [178] have proved that the analogue of the Littlewood
Conjecture does not hold in the elds of formal power series over innite elds.
218
Specic examples have been given by A. Baker [39] and Cusick [170, 171].
Armitage [33] showed that the analogue of Littlewoods Conjecture does not
hold in the elds of formal power series over elds of characteristic at least
equal to 5. De Mathan [411] established that, in any algebraic extension of
the eld with two elements, the Littlewood Conjecture holds true for every
pair of quadratic elements. De Mathan and Teulie [414] have investigated the
analogue of Problem 49 in elds of power series and they established that this
conjecture does not hold if the ground eld is innite.
Some of Cassels and Swinnerton-Dyers results [158] have been extended
by Peck [450].
Problem 40 can be addressed also for p-adic algebraic numbers. An interesting contribution is due to Lagarias [346]. In a eld or formal power series,
the situation is different: there exist algebraic elements of degree at least three
which have bounded partial quotients (see Baum and Sweet [52] and the surveys [351] and [515] for additional references).
M. Einsiedler, A. Katok and E. Lindenstrauss have established that the set
of exceptions to the Littlewood Conjecture has Hausdorff dimension zero.
Appendix A
Lemmas on polynomials
In the present Appendix, we gather several lemmas on polynomials, which

have been used throughout the preceding Chapters. They provide us with estimates, for example, for the distance between two algebraic numbers, or between a complex number and the set of zeros of an integer polynomial. The
rst systematic study of such questions is due to Guting [268] and no signicant progress has been made since that time, at least concerning the most
useful lemmas. Apart from in Chapter 8, we have always used the most natural
height, that is, the naive height (Denition 2.1), in order to measure the size
of an algebraic number. In particular, we did not work with either the logarithmic height, or with the Mahler measure, since the problems investigated in
Chapters 1 to 7 do not require such precise and sophisticated tools. Proofs of
several classical auxiliary results can nevertheless be slightly simplied by the
use of the Mahler measure, dened in Section A.1. Nevertheless, we express
most of the statements in this Appendix only in terms of the naive height. Furthermore, we mention totally explicit examples, due to Bugeaud and Mignotte
[139], showing that some of the classical results gathered here are near to best
possible.
A.1 Denitions and useful lemmas

The height (or naive height) of a polynomial with complex coefcients
P(X ) = an X n + . . . + a1 X + a0 = an (X 1 ) . . . (X n ),
denoted by H(P), is the maximum of the moduli of its coefcients. If P(X ) is
non-zero, its Mahler measure is, by denition, the quantity
M(P) := |an |
n

i=1
219
max{1, |i |}.
220
The height (resp. the Mahler measure) of an algebraic number , denoted by

H() (resp. by M()), is the height (resp. the Mahler measure) of its minimal polynomial over Z (that is, the integer polynomial of lowest positive
degree, with coprime coefcients and positive leading coefcient, which vanishes at ).
The Mahler measure of a polynomial appeared for the rst time in a paper
by Mahler [387], which contains the following lemma.
L EMMA A.1. For any non-zero complex polynomial P(X ), we have

1
log |P(e2i t )| dt .
M(P) = exp
(A.1)
P ROOF. This is Jensens formula, see, for example, Ahlfors [4], page 205, or
Lemma 1.9 of Everest and Ward [231].
Unlike the naive height, the Mahler measure is a multiplicative function; for
this reason, it is, in many cases, much easier to handle. It turns out, however,
that these two notions of size are comparable.
L EMMA A.2. Let P(X ) be a non-zero complex polynomial of degree n. We
have the inequalities

1
n
H(P) M(P) n + 1 H(P).
(A.2)
[n/2]
P ROOF. Write P(X ) = an X n + . . . + a1 X + a0 = an (X 1 ) . . . (X n ).
For any integer j = 0, . . . , n 1, the relations between coefcients and roots
of P(X ) give

s1 . . . sn j .
a j = (1)n j an
1s1 <...<sn j n
n
The above sum is composed of at most [n/2]
terms, each of which has modulus at most equal to M(P)/|an |. We thus get

n
M(P).
|a j |
[n/2]
Since |an | M(P), this proves the left-hand side of (A.2).
As for the right-hand side, the convexity of the exponential function together with the CauchySchwarz inequality yield
1
1/2
1
2it
2i t 2
|P(e
)| dt
|P(e
)| dt
M(P)
0

1/2
= |an |2 + . . . + |a0 |2
n + 1 H(P),
as claimed.
A.1 Denitions and useful lemmas
221
Lemma A.2 allows us to recover estimate (12) of Wirsing [598].

C OROLLARY A.1. Let P(X ) = an (X 1 ) . . . (X n ) be a non-zero complex polynomial of degree n with leading coefcient an in R. Let be a complex
number and be a positive real number. We then have
2n1 (n + 1)1/2 max{1, | |}n

2
n+1

|P( )|
| i |
H(P)
i:| |<
i

n
|P( )|
1 n
max{, } max{1, | |}n
.
[n/2]
H(P)
P ROOF. The Mahler measure of the polynomial P (X ) := P(X + ) is

M(P ) = |an |
n

max{1, | i |}.
i=1
Since |P( )| = M(P )
n
min{1, | i |}, Lemma A.2 implies

n
1
|P( )|
|P( )|
n
.
min{1, | i |}
H(P
)
H(P )
[n/2]
n+1
i=1
i=1
The inequalities
2n1 max{1, | |}n H(P) H(P ) 2n+1 max{1, | |}n H(P)
then yield
2n1 (n + 1)1/2 max{1, | |}n

2n+1

|P( )|
| i |
H(P)
i:| |<1

n
|P( )|
,
max{1, | |}n
[n/2]
H(P)
and the corollary is established for = 1. The general statement now follows
straightforwardly.
The next lemma relates the height of a product of polynomials to the product of
the heights of these polynomials. Koksma and Popkens original result [334],
Hilfssatz 13, has been considerably rened by Gelfond [257], and Lemma A.3
is often referred to as Gelfonds Lemma.
L EMMA A.3. Let P1 (X ), . . . , Pr (X ) be non-zero complex polynomials of degree n 1 , . . . , nr , respectively, and set n = n 1 + . . . + nr . We then have
2n H(P1 ) . . . H(Pr ) H(P1 . . . Pr ) 2n H(P1 ) . . . H(Pr ).
(A.3)
222
P ROOF. We follow an idea of Mahler [387]. Lemma A.2 and the multiplicativity property of the Mahler measure imply

1
1

nr
n1
...
H(P1 ) . . . H(Pr ) M(P1 ) . . . M(Pr )
[n 1 /2]
[nr /2]
= M(P1 . . . Pr ) n + 1 H(P1 . . . Pr ).
Further, we check that

nr
n1
n + 1 2n ,
...
[n 1 /2]
[nr /2]
since

2 / + 1 for any integers 0, k 0
k
Thus, we get the left inequality of (A.3). Moreover, we have

r
r
r

H(P1 . . . Pr )
(n j + 1)H(P j )
2n j H(P j ) = 2n
H(P j ),
j=1
j=1
j=1
which concludes the proof of the lemma.

L EMMA A.4. Let be a non-zero algebraic number of degree n. Let k, a, b,
and c be integers with c = 0. We then have
H( k ) 2n (n + 1)|k|/2 H()|k|
and

H
a + b
c

2n+1 H() max{|a|, |b|, |c|}n .
P ROOF. The rst assertion easily follows from Lemma A.2 and the fact that
M( k ) M()|k| . As for the second assertion, denoting by P(X ) the minimal polynomial of over Z, we see that Q(X ) := a n P(cX/a b/a) is
the one of (a + b)/c. Since the height of Q(X ) is bounded from above by
2n+1 H() max{|a|, |b|, |c|}n , the proof is complete.
A.2 Liouvilles inequality

The easiest version of Liouvilles inequality states that the absolute value of
any non-zero integer is at least equal to one. More generally, the results of
this Section give non-trivial lower bounds for the distance between two distinct algebraic numbers and for the value of an integer polynomial evaluated
at an algebraic number. Liouvilles inequality is a generic name for similar
223
estimates. Standard tools to investigate these questions are the notions of discriminant and resultant. Recall that a separable polynomial is a polynomial
with distinct roots.
D EFINITION A.1. Let P(X ) = an (X 1 ) . . . (X n ) and Q(X ) = bm (X
1 ) . . . (X m ) be non-constant integer polynomials. The resultant of P(X )
and Q(X ), dened as

n
Res(P, Q) = anm bm
(i j )
1in 1 jm
anm
1in
n
Q(i ) = (1)mn bm
P( j ),
1 jm
is an integer and is non-zero if, and only if, P(X ) and Q(X ) have no common
root. The discriminant of P(X ), dened as

(i j )2 = (1)n(n1)/2 an1 Res(P, P ),
Disc(P) = an2n2
1i< jn
is an integer and is non-zero if, and only if, P(X ) is separable.

Further results on resultants and discriminants can be found in most of the
textbooks on algebra, for example, in [349], Chapter IV, in [586], Chapter 5,
or in [156], Appendix A.
For any integer n 2 and any non-zero integer a we have
Res(X n a X + 1, a X 1) = 1
and
Res(X n a X + 1, (a + 1)X n X n1 a X + 1) = 1,
hence, it seems to be difcult to improve the trivial lower bound for the absolute value of the resultant of integer polynomials with no common root.
Theorem A.1 is of comparable strength as Theorem 6 of Guting [268]. An
earlier (and less precise) version goes back to Cohn [167], who used it to construct transcendental numbers.
T HEOREM A.1. Let P(X ) and Q(X ) be non-constant integer polynomials of
degree n and m, respectively. Denote by a zero of P(X ) of order s and by
a zero of Q(X ) of order t. Assuming that P() = 0, we have
|P()| (n + 1)1m/t (m + 1)n/(2t) H(P)1m/t H(Q)n/t

n
max{1, ||}
(A.4)
224
and
| | 21n/s (n + 1)1/(2s)m/(st) (m + 1)n/(2st)

H(P)m/(st) H(Q)n/(st) max{1, ||} max{1, ||} .
P ROOF. Set P(X ) = an (X 1 )s1 . . . (X p )s p and Q(X ) = bm (X
1 )t1 . . . (X q )tq , where = 1 , = 1 , s = s1 , t = t1 , and the i s (resp.
the j s) are pairwise distinct. Denote by Q 1 (X ) = b(X 1 ) . . . (X q1 )
the minimal polynomial of over Z. Since the resultant of P(X ) and Q 1 (X )
is a non-zero integer, we get

|P(i )|,
1 |Res(P, Q 1 )| = |b|n
1iq1
and, using that |P(i )| (n + 1) H(P)(max{1, |i |})n for i = 2, . . . , q1 , we

obtain

n
M(Q 1 )
1 |b|n |P()|(n + 1)q1 1 H(P)q1 1
.
(A.5)
|b| max{1, ||}
Then,
(A.4) follows from (A.5), q1 m/t, M(Q 1 ) M(Q)1/t , and M(Q)
m + 1 H(Q). Furthermore, by combining (A.4) with the estimates

|P()| = | |s |an | | 2 |s2 . . . | p |s p

ns
2ns | |s |an | max{1, ||}

s

s
max{1, |2 |} 2 . . . max{1, | p |} p

ns
M(P)
2ns | |s max{1, ||}
max{1, ||}s
and M(P) n + 1 H(P), we get the second assertion of the theorem.

For any integer n 2 and any non-zero integer a, set
P(X ) = X n a X + 1,
Q(X ) = a X 1, and
R(X ) = (a + 1)X X n1 a X + 1.
n
Let (resp. ) be the root of P(X ) (resp. of R(X )) closest to 1/a, and set
= 1/a. We observe that
| | 2 |a|n1 = 2 H(P)1 H(Q)n
and
| | 2 |a|2n 3 H(P)n H(R)n
hold provided that |a| is large enough in terms of n. Consequently, Theorem
225
A.1 is best possible regarding the dependence on the heights of the polynomials.
A lower bound for the distance between two distinct roots of the same polynomial is slightly less easy to obtain, especially when there are multiplicities. Theorem A.2, due to Amou and Bugeaud [31], strengthens Theorem 7
2
of Guting [268] in terms of n (a factor 2n /st is removed). A similar improvement has been obtained previously by Amou [28], but in his result the
dependence on H(P) is less sharp than in [268].
T HEOREM A.2. Let P(X ) be an integer polynomial of degree n 2. Denote
by a zero of P(X ) of order s and by a zero of P(X ) of order t. Assuming
that = , we have
| | 2n/t n n/t3n/(2st) H(P)2n/(st) max{1, ||} max{1, ||}
if t > s, while we have
| | 2n/s n n(2s+3)/(4s ) H(P)n/s
2
2 +1/(2s)
max{1, ||}3/2
max{1, ||}
3/2
if s = t.
P ROOF. Let Q(X ) = a(X 1 ) . . . (X d ) be a separable integer polynomial with = 1 and such that Q(X ) divides P(X ) in Z[X ] and the polynomials Q(X ) and P(X )/Q s (X ) are coprime. Since the resultant of Q(X ) and
P (s) (X )/s! is a non-zero integer, we get
|P (s) (i )|
.
(A.6)
1 |a|ns
s!
1id
For any integer i = 2, . . . , d, we obtain
n

|P (s) (i )|
k
max{1, |i |}nk
H(P)
s
s!
k=s

n+1
H(P) max{1, |i |}ns .

s+1
(A.7)
Denoting by an the leading coefcient of P(X ), we have

|P (s) ()|
= |an |
| |
s!
=
P( )=0
2nst |an | | |t max{1, ||}nst

max{1, | |},
=, =
P( )=0
(A.8)
226
where the roots of P(X ) are counted with their multiplicities in the above products. Now, the combination of (A.6), (A.7), (A.8), d n/s with the inequality
M(Q) M(P)1/s yields

1n/s
n+1
H(P)
| |t 2n+s+t M(P)n/s
s+1
t
max{1, ||} max{1, ||}t ,
and the rst assertion of the theorem follows from Lemma A.2 since n 2
and

n+1
(n + 1)(n/2)s .
s+1
If s = t, we may assume that 2 = . We have the analogue of (A.8) for
|P (s) ()|/s!, namely the upper bound

|P (s) ()|
2n2s |an | | |s max{1, ||}n2s
max{1, | |},
s!
=, =
P( )=0
(A.9)
where, again, the roots of P(X ) are counted with their multiplicities in the
above product. Combining (A.9) with (A.6), (A.8), and (A.7) for i = 3, . . . , d,
we get

2n/s
n+1
H(P)
| |2s 22n+4s M(P)1n/s
s+1
max{1, ||}3s max{1, ||}3s ,
which yields the last assertion of the theorem.
In case of separable polynomials, Theorem A.2 is superseded by the following
result of Mahler [391].
T HEOREM A.3. Let P(X ) be a separable, integer polynomial of degree n 2.
For any two distinct zeros and of P(X ) we have
| | > 3 (n + 1)(2n+1)/2 max{1, ||, ||} H(P)n+1 .

P ROOF. Denote by an the leading coefcient of P(X ) and by 1 , . . . , n its
roots, numbered in such a way that 1 = and 2 = . Without loss of
generality, we may assume that || ||. Recalling that
Disc(P) = an2n2 (det M)2
with
M = (i )
1in
0 jn1
A.3 Zeros of polynomials
227
we substract the second line of M from the rst one and we apply Hadamards
inequality to get the upper bound

n1
|Disc(P)| |an |2n2
| j j |2
j=1

1 + |i |2 + . . . + |i |2n2
2in
|an |2n2 | |2

n1

| j1 + j2 + . . . + j1 |2
j=1
n n1
max{1, |i |}2n2
2in
< | |2
n3
3
max{1, ||}2 n n1 M(P)2n2 .
(A.10)
Since the polynomial P(X ) has distinct roots, Disc(P) is a non-zero integer
and (A.10) and Lemma A.2 yield the theorem.
As observed in [139], for any integer d 2 and any non-zero integer a sufciently large in terms of d, the polynomial Pa,d (X ) := (X d a X + 1)2
2X 2d2 (a X 1)2 has two real roots distant by at most 4a 2d . This example
shows that, when n is even, the exponent of H(P) in Theorem A.3 cannot be
replaced by any number greater than n/2.
We deduce from Theorems A.1 and A.3 the following useful lower estimate
for the distance between two distinct algebraic numbers.
C OROLLARY A.2. Let and be two distinct non-zero algebraic numbers of
degree n and m, respectively. Then we have
| | 2 (n + 1)m/2 (m + 1)n/2
max{2n (n + 1)(m1)/2 , 2m (m + 1)(n1)/2 } H()m H()n .
It is an interesting open question to decide whether the dependence on the
degrees m and n in Theorems A.1 to A.3 is optimal or not. There is apparently
no contribution to this problem.

This Section is concerned with the following problem: if P(X ) is a complex
polynomial which is small but does not vanish at a complex number , what
can be said on the distance between and the set of zeros of P(X )?
228
L EMMA A.5. Let P(X ) = an (X 1 ) . . . (X n ) be a non-constant complex

polynomial. Let be a complex number with P( ) = 0. Assume that 1 is such
that | 1 | is minimal. If P ( ) = 0, then we have
| 1 | n
|P( )|
.
|P ( )|
(A.11)
If P (1 ) = 0, then we have
|P( )|
|P (1 )|
(A.12)
|P( )| |1 2 |
,
|P (1 )|
(A.13)
| 1 | 2n1
and, if n 2,
| 1 |2 2n2
for any root 2 = 1 of P(X ). If, further, | 1 | |1 j | for j = 2, . . . , n,

then we have
| 1 | 21n
|P( )|
.
|P (1 )|
(A.14)
P ROOF. To prove (A.11), it is enough to note that the rational function

P (X )/P(X ) can be written as

P (X )
1
=
.
P(X )
X
i
1in
(A.15)
Taking X = in (A.15), we obtain

|P ( )|
n
,
|P( )|
| 1 |
and the proof is complete. To get (A.12), we combine
| j |
|P( )|
= | 1 |

|P (1 )|
|1 j |
2 jn
(A.16)
with the fact that, for any integer j = 2, . . . , n, we have

|1 j | | 1 | + | j | 2| j |.
(A.17)
Moreover, for any integer j = 2, . . . , n, the triangle inequality ensures that

| j | 2|1 j | holds as soon as | 1 | |1 j |. Thus, (A.14)
follows from (A.16).
229
Finally, we get (A.13) by combining

| 1 | | 2 | | j |
|P( )|
=
|P (1 )|
|1 2 |
|1 j |
3 jn
| 1 |2 | j |
|1 2 | 3 jn |1 j |
and (A.17).
L EMMA A.6. Let P(X ) = an (X 1 ) . . . (X n ) be a non-constant integer
polynomial of degree n. Let be a complex number which is not a root of
P(X ). Then, for any subset J of {1, . . . , n}, we have

n

n + 1 H(P),
(A.18)
|an | jJ | j | 2n max{1, | |}
and, in particular,

n
|P( )| 2n max{1, | |}
n + 1 H(P) min1 jn | j |.
P ROOF. We have

|an |
| j | |an |
2 max{| |, | j |}
jJ
jJ

n
max{1, | j |}
2 max{1, | |} |an |
n

n
jJ
2 max{1, | |} M(P),
n
and we apply Lemma A.2 to get (A.18). The last assertion follows by taking
J = {1, . . . , n} \ {k} where k is such that | k | = min1 jn | j |.
Lemma A.7, due to Diaz and Mignotte [189], slightly improves upon an earlier
result of Chudnovsky [164]. To take multiplicities into account, Chudnovsky
introduced the notion of semi-discriminant, which generalizes that of discriminant to polynomials with multiple roots. Let P(X ) = an (X 1 )s1 . . . (X
p )s p be an integer polynomial of degree n = s1 + . . . + s p . Assume that
i = j if 1 i = j p. Then the semi-discriminant of P(X ), denoted by
SD(P), is the quantity
SD(P) = ann2
p

P (si ) (i )
.
si !
i=1
(A.19)
If P(X ) is separable, then SD(P) reduces to the discriminant of P(X ). Actually, Chudnovsky [164] dened the semi-discriminant with ann1 instead of
ann2 in (A.19). He proved that an SD(P) is a non-zero rational integer and he
used this observation to get Lemma 1.12 of [164], which is slightly weaker
230
than Lemma A.7 below. Amou [28] used the same idea to obtain a weaker
version of Theorem A.2. Further, Diaz and Mignotte [189] pointed out that
the use of the semi-discriminant can be replaced by the consideration of a
suitable resultant. Their method, which is also applied to get Theorem A.2
above, enabled them to considerably simplify Chudnovskys proof. Lemma
A.7 follows from [189], except for the last assertion, kindly communicated by
Amou.
L EMMA A.7. Let P(X ) be a non-constant integer polynomial of degree n. Let
be a complex number and be a root of P(X ) such that | | is minimal.
Then, denoting by s the multiplicity of as root of P(X ), we have
| |s n n+3n/(2s) H(P)2(n/s1) |P( )|.
Further, if P(X ) is the s-th power of a separable polynomial of degree at least
2, we get
| |s 2n/2 n n/2+3n/(4s) H(P)n/s3/2 |P( )|.
P ROOF. Without any restriction, we may assume that n = s, hence, n/s 2.
Let Q(X ) = a(X 1 ) . . . (X d ) be as in the proof of Theorem A.2,
with = 1 . For any root of P(X ) different from , we have | |
| | + | | 2| |. Thus, we get

|P (s) ()|
= |an |
| | 2ns |P( )| | |s ,
s!
=
(A.20)
P( )=0
where an denotes the leading coefcient of P(X ) and the roots of P(X ) are
counted with their multiplicities in the above product. We combine (A.6),
(A.7), and (A.20) to obtain
d1

n+1
s
ns
M(Q)ns |P( )|.
(A.21)
| | 2
s+1 H(P)
Since

n+1
(n + 1)(n/2)s ,
s+1
d n/s, and M(Q) M(P)1/s ,
the rst assertion follows from (A.21) and Lemma A.2.

For the last statement, we take Q(X ) such that Q(X )s = P(X ). We may
assume that = 2 satises
| | | | | i |, for i = 3, . . . , d,
231
hence, we get | | 2| | for any root of P(X ) different from .

Consequently, we have
|P (s) ()|
2n2s | |s |P( )| | |s | |s
s!
2ns |P( )| | |s ,
(A.22)
since | | 2| |. Combining (A.6), (A.20), (A.22), and (A.7) for

i = 3, . . . , d, we get

1+d/2
n+1
s
ns
| | 2
H(P)
M(Q)(ns)/2 |P( )|,
s+1
which, since d = n/s, completes the proof of the lemma.
For the case of a separable polynomial, Lemma A.7 has been improved upon
by Feldman [243] (Chapter 7, Lemma 1.7) and Diaz [188] (for 2 in Lemma
A.8, which strengthens Lemma 5 of Laurent and Roy [356]).
L EMMA A.8. Let P(X ) be a non-constant, separable, integer polynomial of
degree n. Let be a complex number and be a root of P(X ) such that | |
is minimal. Then, for any positive integer with < n we have
(2| |)(+1)/2 2n (n )(n)/2 M(P)n1 |P( )| .
(A.23)
In particular, we have
| |
2 (2n)n3/2 H(P)n2 |P( )|.
(A.24)
If, moreover, |P( )| 1, then, for any positive integer , we have

| |(+1)/2 2n n n/2 M(P)n |P( )| .
(A.25)
P ROOF. We may assume that = . We order the roots 1 , . . . , n of P(X )

in such a way that = 1 and | 1 | . . . | n |. Set Q (X ) =
an (X +1 ) . . . (X n ), where an denotes the leading coefcient of P(X ).
Putting

|i j |,
:=
1i< jn
i
the discriminants Disc(P) and Disc(Q ) of P(X ) and Q (X ) satisfy

|Disc(P)|1/2 = |an | |Disc(Q )|1/2 .
(A.26)
The quantity |Disc(Q )|1/2 equals |an |n1 times the absolute value of the
232
determinant of the matrix (i )+1in,0 jn1 . As in the proof of Theorem A.3, we infer from Hadamards inequality that

max{1, |i |}n1 ,
|| (n )(n)/2
+1in
hence, we get
|Disc(Q )|1/2 (n )(n)/2 M(Q )n1 ,
(A.27)
and, since Disc(P) is a non-zero integer, it follows from (A.26) and (A.27) that
1 |an | (n )(n)/2 M(P)n1 .
(A.28)
For positive integers i and j with i < j n, we have |i j | 2| j |,

thus

| j | j1
| j | . (A.29)
|an | |an | 2n(+1)/2
< jn
1 j
Since
|an |

< jn
| j | = |P( )|
| j |
1 j
and | 1 | . . . | |, we infer from (A.29) that

|an | 2n |P( )| (2| 1 |)(+1)/2 ,
which, together with (A.28), gives (A.23). Choosing = 1 in (A.23), we get
(A.24). Further, (A.25) is clear when n = 1 and it follows from (A.23) if
= 1, . . . , n 1. If n 2, we infer from (A.23) with = n 1 that
| | 2 |P( )|2/n .
(A.30)
For n, we raise both sides of (A.30) to the power n/2 and, noticing that
| | 1, we get an inequality in fact stronger than (A.25).
Theorems A.2 and A.3 should be compared with Lemmas A.7 and A.8, respectively. In both cases, the result obtained for separable polynomials is considerably better than that for arbitrary polynomials: roughly speaking, we gain a
factor 2 in the exponent of H(P). Furthermore, the polynomials Pa,d (X ) dened below the proof of Theorem A.3 show that the dependence on the height
in (A.24) is close to be best possible: indeed, when n is even, the exponent
of H(P) in (A.24) cannot be replaced by any number smaller than 1 + n/2.
However, the dependence on the degree may possibly be improved.
A.5 Notes
233
A.4 Exercises
E XERCISE A.1. Let P(X ) := an (X 1 ) . . . (X n ) be an irreducible, integer
polynomial. Let i 1 , . . . , i t be positive integers with 1 i 1 < . . . < i t n.
Prove that the number an i1 . . . it is an algebraic integer
(see, for example,
[591], pp. 7172) with absolute value at most equal to n + 1 H(P).
E XERCISE A.2. Study how sharp Theorems A.1, A.2, and Lemma A.7 are in
terms of s and t (see [31]).
E XERCISE A.3. Let P(X ) be a separable, integer polynomial of degree n 2
and let 1 , . . . , k be distinct zeros of P(X ), with 2 k n. With a suitable
modication of the proof of Theorem A.3, show that there exists a positive,
effective constant c(n) such that

|i j | c(n) H(P)n+1 .
(A.31)
1i< jk
By considering the polynomials Pa,d,k (X ) := (X d a X +1)k 2X dkk (a X

1)k , where a and d are integers with d 2, prove that (A.31) is near to best
possible (see [139]).
E XERCISE A.4. Prove that Theorem A.3 is, up to the numerical constant, best
possible for cubic, reducible, integer polynomials.
Hint. Following an idea of Deshouillers (see [419]), consider the polynomials Pk (X ) := (qk X pk )(X 2 2), where
pk /qk denotes the k-th convergent
in the continued fraction expansion of 2.
A.5 Notes
The formula (A.1) can be generalized to dene the Mahler measure of a polynomial in nitely many variables, see Mahler [389] and Chapter 3 of Everest
and Ward [231].
For inequalities relating the naive height of a polynomial to other notions
of size, see papers by Durand [220] and the Annex to Chapter 3 in [591].
Theorem A.1 can be considerably improved if is non-real, see, for example, Guting [268]. Roughly speaking, one can take the square roots of the
lower bounds obtained. The interested reader may easily write down the details of the proof and improve upon several other results from this Appendix in
similar particular cases.
234
There are more general versions of Liouvilles inequality providing

non-trivial lower bounds for the non-zero quantity |P(1 , . . . , n )|v , where
1 , . . . , n are elements of an algebraic number eld K, the polynomial
P(X 1 , . . . , X n ) has integer coefcients, and v is some place on K. We refer
to [591] for estimates involving the Mahler measure and the Weils height (see
especially Exercise 3.5). When and are real algebraic numbers with the
Mahler measure of relatively small compared with its degree, improvements
upon Liouvilles inequality have been obtained for | | and | 1|, see
Mignotte [420], Amoroso [22, 23], Dubickas [210], and the last chapter of
[512]. A p-adic analogue has been worked out by Bugeaud [121].
Dubickas [209] and Mignotte [422] have slightly improved Theorem A.3.
Their method also yields lower bounds for products of differences of roots of
a separable polynomial (like the statement given in Exercise A.3). Polynomial
root separation is an important problem in computer algebra, see, for example,
Rump [485] for more references. Following Collins and Horowitz [169], for
any positive integers n 2 and H , we denote by L(n, H ) the minimum of the
minimal distance between two of the roots of a separable, integer polynomial
P(X ) of degree n and height at most H . Mignotte [419] and Mignotte and
Payafar [423] studied L(n, H ) and introduced the related quantities L (n, H ),
L 0 (n, H ), and L 0 (n, H ), where we restrict the consideration to irreducible,
monic, and irreducible and monic polynomials, respectively. The determination of the upper limits when H tends to innity of log L(n, H )/ log H and
of the related quantities is apparently a very difcult problem.
Supported by computational evidence, Collins [168] conjectured that the
minimal distance between two real roots of an integer polynomial P(X ) of
degree n and height H is at least equal to n n/4 H n/2 . This is compatible
with the examples given below the proof of Theorem A.3.
Appendix B
Geometry of numbers
Geometry of numbers turns out to be a very useful tool in Diophantine approximation. For instance, it allows us to construct non-zero integer polynomials
taking small values at prescribed points. In the course of the book, we applied
several times the rst Theorem of Minkowski and the second Theorem of
Minkowski, which are Theorems B.2 and B.3 below, respectively. We give
a full proof of Theorem B.2, but not of Theorem B.3, which is much deeper.
Throughout this Appendix, n denotes a positive integer. A set C in Rn having
inner points and contained in the closure of its open kernel is called a body (or
a domain).
T HEOREM B.1. Let C be a bounded convex body in Rn , symmetric about the
origin and of volume vol(C). If vol(C) > 2n or if vol(C) = 2n and C is compact,
then C contains a point with integer coordinates, other than the origin.
P ROOF. This proof is due to Mordell [429]. By classical arguments from elementary topology, it is enough to treat the case where vol(C) > 2n . For any
positive integer m, denote by Cm the set of points of C having rational coordinates with denominator m. As m tends to innity, the cardinality of Cm becomes equivalent to vol(C)m n , and is thus strictly larger than (2m)n when m is
large enough. Dirichlets Schubfachprinzip asserts that there exist two distinct
points A = (a1 /m, . . . , an /m) and B = (b1 /m, . . . , bn /m) in Cm with ai , bi
integers and ai bi (mod 2m) for i = 1, . . . , n. By symmetry and convexity, the point (A B)/2, which is not the origin and has integer coordinates,
belongs to C.
T HEOREM B.2. Let (u i j )1i, jn be a matrix with real coefcients and
with determinant 1. Let A1 , . . . , An be positive real numbers satisfying
A1 . . . An = 1. Then there exists a non-zero integer point (x1 , . . . , xn ) such
235
236
Geometry of numbers
that
|u i1 x1 + . . . + u in xn | < Ai ,
1 i n 1,
and
|u n1 x1 + . . . + u nn xn | An .
P ROOF. Since the absolute value of the determinant of the matrix (u i j )1i, jn
is equal to the product of the Ai s, we may assume that A1 = = An = 1.
For i = 1, . . . , n and x = (x1 , . . . , xn ), we set
L i (x) := u i1 x1 + . . . + u in xn .
The linear system |L i (x)| 1, 1 i n, denes a symmetric, bounded,
compact convex body of volume 2n . By Theorem B.1, it contains a non-zero
integer point. Classical topological arguments then allow us to replace n 1
of the large inequalities by strict inequalities.
We now state a deep generalization of Theorem B.2. Let C be a convex,
compact body in Rn , symmetric about the origin and of volume vol(C). Let
1 := 1 (C) be the inmum of the real numbers such that C contains an
integer point other than the origin. Since C is compact, this inmum is a minimum. Further, we have 0 < 1 < +. Setting := 2 vol(C)1/n , the volume
of the compact convex set C is equal to 2n , and Theorem B.2 implies the
upper bounds 1 and
n1 vol(C) 2n .
(B.1)
Theorem B.3 allows us to strengthen (B.1). For any integer j = 1, . . . , n,

denote by j := j (C) the inmum of the positive real numbers for which C
contains j linearly independent integer points. Each of the j s is a minimum
and we have
0 < 1 2 . . . n < +.
The real numbers 1 , . . . , n are called the successive minima of C.
T HEOREM B.3. For a set C as above, we have
2n
1 . . . n vol(C) 2n .
n!
Moreover, if the points x 1 , . . . , x n are linearly independent and realize the
successive minima of C, then we have
det(x 1 , . . . , x n ) n!.
Geometry of numbers
237
P ROOF. This is Theorem V and Corollary, pages 218 and 219, of Cassels
[157]. The proof of the upper bound for the product of the i is difcult, while
the lower bound is easier to establish.
Any hyperplane not containing the origin can be put in the form y1 x1 + . . . +
yn xn = 1, and so it may be represented as a point y = (y1 , . . . , yn ) in Rn . It
turns out that the points y corresponding to hyperplanes having empty intersection with C form a convex set C , called the polar body of C.
T HEOREM B.4. Let 1 , . . . , n be the successive minima of a convex body C
and let 1 , . . . , n be the successive minima of the polar body C . Then, for
any integer j = 1, . . . , n, we have
1 j n+1 j n!.
P ROOF. This is Theorem VI, page 219, of Cassels [157].
To conclude this Appendix, we state and prove Khintchines Transference Theorem, which relates small values of linear forms to simultaneous Diophantine
approximation. Let x = (x1 , . . . , xn ) be a real n-tuple. We dene w(x) and
w (x) to be the suprema of the real numbers w and w for which the inequalities

nw
|u 0 + u 1 x1 + . . . + u n xn | max |u i |
1in
and
max |v0 xi + vi | |v0 |(1+w )/n
1in
have innitely many integer solutions (u 0 , . . . , u n ) and (v0 , . . . , vn ), with

v0 = 0. It follows from Theorem B.2 that w(x) and w (x) are non-negative.
Theorem B.5 shows how they are related.
T HEOREM B.5. With the above notation, we have
w(x)
w (x) w(x).
2
n + (n 1)w(x)
(B.2)
P ROOF. We may assume that n 2. We begin by treating some particular

cases. If w(x) = 0 (resp. w (x) = 0), then the left (resp. the right) inequality
of (B.2) holds trivially. If |u 0 + u 1 x1 + . . . + u n xn | = 0 for some non-zero
integer (n + 1)-tuple (u 0 , . . . , u n ), then 1, x1 , . . . , xn are linearly dependent
over Z and there exists a positive constant c such that max1in |v0 xi + vi |
c|v0 |1/(n1) has innitely many integer solutions (v0 , . . . , vn ) with v0 = 0.
Consequently, we have w(x) = + and w (x) 1/(n 1), hence, (B.2)
238
Geometry of numbers
holds. Furthermore, if max1in |v0 xi + vi | = 0 for some non-zero integer

(n + 1)-tuple (v0 , . . . , vn ), then x1 , . . . , xn are rational numbers and w(x) =
w (x) = +, hence, (B.2) holds.
We now assume that we are in none of these particular cases. Let w > 0 be
a real number and let u 0 , . . . , u n be integers, not all zero, such that
0 < := |u 0 + u 1 x1 + . . . + u n xn | U nw ,
where U = max{|u 0 |, . . . , |u n |}.
(B.3)
By Theorem B.2, there exist v0 , . . . , vn integers, not all zero, such that
|v0 xi vi | 1/n
(i = 1, . . . n),
(B.4)
|v0 u 0 + v1 u 1 + . . . + vn u n | < 1,
since the system of linear forms appearing in the left-hand members of these
inequalities has determinant . The u i s and the vi s are integers, thus we get
v0 u 0 + v1 u 1 + . . . + vn u n = 0.
(B.5)
Combining (B.3) and (B.5), we obtain

|v0 | = |(v0 x1 v1 )u 1 + . . . + (v0 xn vn )u n | n U 1/n .
(B.6)
It follows from (B.3) that U 1/(n+w) , thus (B.6) gives

|v0 | n 1/n11/(n+w) = n (n
2 +(n1)w)/(n 2 +nw)
which, combined with (B.4), yields the left inequality of (B.2).

As for the right inequality, let w > 0 be a real number and let v0 , . . . , vn
be integers with v0 = 0 and
0 < := max |v0 xi + vi | |v0 |(1+w )/n .

1in
Without any loss of generality, we assume that = |v0 x1 + v1 |. By Theorem

B.2, there exist u 0 , . . . , u n integers, not all zero, such that
| u 0 + u 1 x1 + . . . + u n xn | (w +n)/(w +1) ,
|v0 u 0 + . . . + vn u n | < 1,
|u i |
1/(w +1)
(B.7)
(i = 2, . . . , n),
since the system of linear forms appearing in the left-hand members of these
inequalities has determinant . Arguing as previously, we get (B.5) and we

deduce that |u 1 | n 1/(w +1) , which, combined with (B.7), completes the
proof of Theorem B.5.
Geometry of numbers
239
Theorem B.5 is due to Khintchine [318, 320], and its proof has been simplied
by Mahler [379] (see, for example, Gruber and Lekkerkerker [262], Section
45.3, Cassels [155], Chapter V, Theorem IV, or Schmidt [512], Chapter IV,
Section 5). Inequalities (B.2) turn out to be best possible, as proved by Jarnk
[293, 295], who got some related results [294]. The generalization of Theorem
B.5 to systems of linear forms is due to Dyson [224] and has been proved to be
best possible by Jarnk [299]. For other transference theorems, see Wang, Yu,
and Zhu [594], Schmidt and Wang [516], and the survey of Xu [604].
References
[1] Abercrombie, A. G. (1995). Badly approximable p-adic integers. Proc. Indian

Acad. Sci. Math. Sci., 105, 123134. (Cited in Chapter 9.)
[2]
(1995). The Hausdorff dimension of some exceptional sets of p-adic integer
matrices. J. Number Theory, 53, 311341. (Cited in Chapter 9.)
[3] Adhikari, S. D., Saradha, N., Shorey, T. N. & Tijdeman, R. (2001). Transcendental innite sums. Indag. Math. (N. S.), 12, 114. (Cited in Chapter 3.)
[4] Ahlfors, L. V. (1966). Complex Analysis: An Introduction of the Theory of Analytic Functions of One Complex Variable. Second edition. McGraw-Hill Book
Co., New YorkTorontoLondon. (Cited in Appendix A.)
[5] Alessandri, P. & Berthe, V. (1998). Three distance theorems and combinatorics
on words. Enseignement Math., 44, 103132. (Cited in Chapter 5.)
[6] Allouche, J.-P., Davison, J. L., Queffelec, M. & Zamboni, L. Q. (2001). Transcendence of Sturmian or morphic continued fractions. J. Number Theory, 91,
3966. (Cited in Chapter 2.)
[7] Allouche, J.-P. & Shallit, J. (2003). Automatic Sequences: Theory, Applications, Generalizations. Cambridge University Press, Cambridge. (Cited in
Chapter 2.)
[8] Alniacik, K. (1979). On the subclasses Um in Mahlers classication of the tran
scendental numbers. Istanbul Univ.
Fen Fak. Mecm. Ser., 44, 3982. (Cited in
Chapters 7 & 9.)
[9]
(1982). On Um -numbers. Proc. Amer. Math. Soc., 85, 499505. (Cited in
Chapter 7.)
[10]
(1983). On Mahlers U-numbers. Amer. J. Math., 105, 13471356. (Cited in
Chapter 7.)
[11]
(1986). On T-numbers. Glasnik Math., 21, 271282. (Cited in Chapter 7.)
[12]
(1990). Representation of real numbers as sums of U2 -numbers. Acta Arith.,
55, 301310. (Cited in Chapter 7.)
[13]
(1991). On p-adic Um -numbers. Istanbul Univ.
Fen Fak. Mat. Der., 50, 17.
(Cited in Chapter 9.)
[14]
(1992). On p-adic T-numbers. Doga Math., 16, 119128. (Cited in
Chapter 9.)
[15]
(1992). On semi-strong U -numbers. Acta Arith., 60, 349358. (Cited in
Chapter 7.)
240
References
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
241
(1994). A note on some U-numbers. Indian J. Pure Appl. Math., 25, 689691.
(1998). The points on curves whose coordinates are U -numbers. Rend. Mat.
Roma, Serie VII, 18, 649653. (Cited in Chapter 10.)
Alniacik, K., Avci, Y. & Bugeaud, Y. (2003). On Um -numbers with small transcendence measures. Acta. Math. Hungar., 99, 271277. (Cited in Chapter 7.)
(1994). Une remarque sur les G -denses. Arch. Math.,
Alniacik, K. & Saias, E.
Amice, Y. (1975). Les nombres p-adiques. Collection SUP: Le Mathematicien,
No. 14, Presses Universitaires de France, Paris. (Cited in Chapter 9.)
Amoroso, F. (1988). On the distribution of complex numbers according to their
transcendence types. Ann. Mat. Pura Appl., 151, 359368. (Cited in Chapter 8.)
(1996). Algebraic numbers close to 1 and variants of Mahlers measure. J.
Number Theory, 60, 8096. (Cited in Appendix A.)
(1998). Algebraic numbers close to 1: results and methods. In Number Theory (Tiruchirapalli, 1996), pp. 305316; Amer. Math. Soc., Providence, RI. (Cited
in Appendix A.)
(2001). Some metric results in transcendental numbers theory. In Introduction to algebraic independence theory. Lecture Notes in Math. 1752, Springer,
Berlin, 227237. (Cited in Chapter 8.)
Amou, M. (1991). Approximation to certain transcendental decimal fractions by
algebraic numbers. J. Number Theory, 37, 231241. (Cited in Chapter 7.)
(1991). Algebraic independence of the values of certain functions at a transcendental number. Acta Arith., 59, 7182. (Cited in Chapter 8.)
(1992). An improvement of a transcendence measure of Galochkin and
Mahlers S-numbers. J. Austral. Math. Soc. Ser., 52, 130140. (Cited in Chapter
3.)
(1996). On Sprindzuks classication of transcendental numbers. J. reine
angew. Math., 470, 2750. (Cited in Chapter 8 & Appendix A.)
(1996). Transcendence measures for almost all numbers. In Analytic Number Theory (Kyoto, 1995). Surikaisekikenkyusho Kokyuroku, 961, 112116 (in
Japanese). (Cited in Chapter 8.)
(1996). A metrical result on transcendence measures in certain elds.
J. Number Theory, 59, 389397. (Cited in Chapter 9.)
Amou, M. & Bugeaud, Y. (Preprint). Sur la separation des racines des
polynomes et une question de Sprindzuk. (Cited in Chapter 8 & Appendix A.)
Arbour, B. & Roy, D. (2004). A Gel fond type criterion in degree two. Acta
Arith., 111, 97103. (Cited in Chapters 3 & 8.)
Armitage, J. V. (1969). An analogue of a problem of Littlewood. Mathematika,
16, 101105; Corrigendum and addendum (1970) in Mathematika, 17, 173 178.
Astels, S. (2000). Cantor sets and numbers with restricted partial quotients.
Trans. Amer. Math. Soc., 352, 133170. (Cited in Chapter 1.)
(2001). Sums of numbers with small partial quotients II. J. Number Theory,
(2002). Sums of numbers with small partial quotients II. Proc. Amer. Math.
Soc., 130, 637642. (Cited in Chapter 1.)
242
References
[37] Baker, A. (1962). Continued fractions of transcendental numbers. Mathematika, 9, 18. (Cited in Chapters 2 & 3.)
[38]
(1964). On Mahlers classication of transcendental numbers. Acta Math.,
111, 97120. (Cited in Chapters 2 & 3.)
[39]
(1964). On an analogue of Littlewoods Diophantine approximation problem.
Michigan Math. J., 11, 247250. (Cited in Chapter 10.)
[40]
(1964). Approximations to the logarithms of certain rational numbers. Acta
Arith., 10, 315323. (Cited in Chapter 3.)
[41]
(1966). On a theorem of Sprindzuk. Proc. Roy. Soc. London, 292, 92104.
(Cited in Chapters 4 & 9.)
[42]
(1966/67). On Mahlers classication of transcendental numbers II:
Simultaneous Diophantine approximation. Acta Arith., 12, 281288. (Cited in
Chapter 3.)
[43]
(1966/68). Linear forms in the logarithms of algebraic numbers IIV. Mathematika, 13 (1966), 204216; 14 (1967), 102107 & 220224; 15 (1968), 204
[44]
(1974). Transcendental Number Theory, Cambridge University Press. (Cited
in Introduction, and Chapters 2, 3 & 4.)
[45] Baker, A. & Schmidt, W. M. (1970). Diophantine approximation and Hausdorff
dimension. Proc. London Math. Soc., 21, 111. (Cited in Chapters 3, 5 & 6.)
[46] Baker, A. & Stewart, C. L. (1988). On effective approximations to cubic irrationals. In New Advances in Transcendence Theory. A. Baker (ed.), pp. 124,
Cambridge University Press, Cambridge. (Cited in Chapter 2.)
[47] Baker, R. C. (1976). On approximation with algebraic numbers of bounded degree. Mathematika, 23, 1831. (Cited in Chapters 7 & 9.)
[48]
(1976). Sprindzuks theorem and Hausdorff dimension. Mathematika, 23,
184197. (Cited in Chapters 5 & 9.)
[49]
(1976). Metric Diophantine approximation on manifolds. J. London Math.
[50]
(1979). On numbers with many rational approximations. Math. Proc. Cambridge Philos. Soc., 86, 2527. (Cited in Chapter 1.)
[51]
(1992). Singular n-tuples and Hausdorff dimension II. Math. Proc. Cambridge Phil. Soc., 111, 577584. (Cited in Chapter 5.)
[52] Baum, L. E. & Sweet, M. M. (1976). Continued fractions of algebraic power
series in characteristic 2. Ann. of Math., 103, 593610. (Cited in Chapter 10.)
[53] Baxa, C. (1994). Fast growing sequences of partial denominators. Acta Math.
Inform. Univ. Ostraviensis, 2, 8184. (Cited in Chapter 3.)
[54]
(1997). On the distribution of the sequence (n) with transcendental . Acta
[55]
(To appear). Extremal values of continuants and transcendence of certain continued fractions. Adv. in Appl. Math. (Cited in Chapter 2.)
[56] Becker, P.-G. (1991). Effective measures for algebraic independence of the
values of Mahler type functions. Acta Arith., 58, 239250. (Cited in Chapters 3 & 8.)
[57]
(1994). Transcendence measures for the values of functions satisfying generalized Mahler type functional equations in arbitrary characteristic. Publ. Math.
Debrecen, 45, 269282. (Cited in Chapter 9.)
References
243
[58] Becker-Landeck, P.-G. (1986). Transcendence measures by Mahlers transcendence method. Bull. Austral. Math. Soc., 33, 5965. (Cited in Chapters 3 & 8.)
[59] Beckett, S. (1953). LInnommable, Les e ditions de Minuit.
[60] Beresnevich, V. (1996). Effective measure estimates for sets of real numbers
with a given error of approximation by quadratic irrationalities. Vests Akad.
Navuk Belarus Ser. Fz. Mat. Navuk, 4, 1015 (in Russian). (Cited in Chapters
4, 6 & 7.)
[61]
(1999). On approximation of real numbers by real algebraic numbers. Acta
Arith., 90, 97112. (Cited in Chapters 4, 5 & 6.)
[62]
(2000). Application of the concept of regular system of points in metric number theory. Vests Nats. Akad. Navuk Belarus Ser. Fz. Mat. Navuk, 1, 3539 (in
Russian). (Cited in Chapters 6 & 9.)
[63]
(2002). A Groshev type theorem for convergence on manifolds. Acta Math.
Hungar., 94, 99130. (Cited in Chapters 4 & 5.)
[64] Beresnevich, V. & Bernik, V. (2000). A. Bakers conjecture and Hausdorff dimension. Publ. Math. Debrecen, 56, 263269. (Cited in Chapter 5.)
[65]
(2002). Diophantine approximation on classical curves and Hausdorff dimension. In Analytic and Probabilistic Methods in Number Theory (Palanga, 2001),
2027, TEV, Vilnius. (Cited in Chapters 5 & 10.)
[66] Beresnevich, V., Bernik, V. I. & Dodson, M. M. (1997). Inhomogeneous nonlinear Diophantine approximation. In Papers in Honour of V. G. Sprindzuks
60th. Birthday, Inst. Math., Belarus Acad. Sci., 1320. (Cited in Chapter 4.)
[67]
(2002). Regular systems, ubiquity and Diophantine approximation. In
A Panorama of Number Theory or The View from Bakers Garden. G.
Wustholz (ed.), Cambridge University Press, pp. 260279. (Cited in Chapters 5
& 6.)
[68] Beresnevich, V., Bernik, V. I., Kleinbock, D. Y. & Margulis, G. A. (2002).
Metric Diophantine approximation: the KhintchineGroshev theorem for nondegenerate manifolds. Moscow Math. J., 2, 203225. (Cited in Chapters 4
& 5.)
[69] Beresnevich, V. V., Bernik, V. I. & Kovalevskaya, E. I. (Preprint). On approximation of p-adic numbers by p-adic algebraic numbers. (Cited in Chapter 9.)
[70] Beresnevich, V. V., Dickinson, H. & Velani, S. L. (2001). Sets of exact logarithmic order in the theory of Diophantine approximation. Math. Ann., 321,
[71]
(Preprint). Measure theoretic laws for lim sup sets. (Cited in Chapters 6
& 9.)
[72] Beresnevich, V. V. & Kovalevskaya, E. I. (2003). On Diophantine approximation
of dependent quantities in the p-adic case. Mat. Zametki. 73, 2237 (in Russian).
English transl. in Math. Notes, 73 (2003), 2135. (Cited in Chapter 9.)
[73] Bernhard, T. (1973). Das Kalkwerk, Suhrkamp Verlag.
[74] Bernik, V. I. (1977). Induced extremal surfaces. Mat. Sb. (N. S.), 103, 480 489,
630 (in Russian). (Cited in Chapter 4.)
[75]
(1979). On the BakerSchmidt hypothesis. Dokl. Akad. Nauk BSSR, 23, 392
[76]
(1980). A metric theorem on the simultaneous approximation of zero by
the values of integral polynomials. Izv. Akad. Nauk SSSR Ser. Mat. 44, 2445
244
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
References
(in Russian). English transl. in Math. USSR Izv., 16 (1981), 2140. (Cited in
Chapters 4 & 9.)
(1983). Application of the Hausdorff dimension in the theory of Diophantine
approximations. Acta Arith., 42, 219253 (in Russian). English transl. in Amer.
Math. Soc. Transl., 140 (1988), 1544. (Cited in Chapter 5.)
(1984). A proof of Bakers conjecture in the metric theory of transcendental numbers. Dokl. Akad. Nauk. SSSR, 277, 10361039 (in Russian). English transl. in Soviet Math. Dokl., 30 (1984), 186189. (Cited in
Chapter 4.)
(1986). A property of integer polynomials that realize Minkowskis theorem
on linear forms. Dokl. Akad. Nauk BSSR, 30, 403405, 477 (in Russian). (Cited
in Chapter 6.)
(1988). Applications of measure theory and Hausdorff dimension to the theory of Diophantine approximation. In New Advances in Transcendence Theory (Durham, 1986), 2536, Cambridge University Press, Cambridge. (Cited in
Chapter 9.)
(1989). On the best approximation of zero by values of integral polynomials.
Acta Arith., 53, 1728 (in Russian). (Cited in Chapter 4.)
Bernik, V. I. & Borbat, V. N. (1997). Polynomials with overfalls in values of
coefcients and a conjecture of A. Baker. Vests Nats. Akad. Navuk Belarus Ser.
Fz. Mat. Navuk, 3, 58 (in Russian). (Cited in Chapter 4.)
(1997). Simultaneous approximation of zero by values of integer-valued polynomials. Tr. Mat. Inst. Steklova, 218, 5873 (in Russian). English transl. in Proc.
Steklov Inst. Math., 218 (1997), 5368. ( Cited in Chapter 4.)
Bernik, V. I., Dickinson, H. & Dodson, M. M. (1998). Approximation of real
numbers by values of integer polynomials. Dokl. Nats. Akad. Nauk Belarusi, 42,
No.4, 5154, 123 (in Russian). (Cited in Chapter 4.)
Bernik, V. I., Dickinson, H. & Yuan, J. (1999). Inhomogeneous Diophantine approximation on polynomials in Q p . Acta Arith., 90, 3748. (Cited in Chapter 9.)
Bernik, V. I. & Dodson, M. M. (1999). Metric Diophantine approximation on
manifolds. Cambridge Tracts in Mathematics 137, Cambridge University Press.
(Cited in Introduction, and Chapters 1, 4, 5, 6 & 9.)
Bernik, V. I. & Dombrovski, I. V. (1992). U3 -numbers on curves in R2 .
Vests Akad. Navuk Belarus Ser. Fz. Mat. Navuk, 3/4, 37, 123. (Cited in
Chapter 10.)
(1994). Effective estimates for the measure of sets dened by Diophantine
conditions. Trudy Mat. Inst. Steklov, 207, 3541 (in Russian). English transl. in
Proc. Steklov Inst. Math., 207 (1995), 3540. (Cited in Chapter 4.)
Bernik, V., Kleinbock, D. & Margulis, G. A. (2001). Khintchine-type theorems
on manifolds: the convergence case for standard and multiplicative versions.
Internat. Math. Res. Notices, 9, 453486. (Cited in Chapter 4.)
Bernik, V. I. & Melnichuk, Y. I. (1988). Diophantine approximation and Hausdorff dimension. Akad. Nauk. BSSR, 1988 (in Russian). (Cited in Introduction
and Chapter 5.)
Bernik V. I. & Morotskaya, I. L. (1986). Diophantine approximations in Q p and
Hausdorff dimension. Vests Nats. Akad. Navuk Belarus Ser. Fz. Mat. Navuk,
3, 39, 123 (in Russian). (Cited in Chapter 9.)
References
245
[92] Bernik, V. I. & Morozova, I. M. (1998). Diophantine approximations by

lacunary polynomials and the Hausdorff dimension. Vests Nats. Akad.
Navuk Belarus Ser. Fz. Mat. Navuk, 4, 4751, 142 (in Russian). (Cited in
Chapter 9.)
[93]
(1998). Diophantine approximations and zeros of lacunary polynomials.
Dokl. Nats. Akad. Nauk Belarus, 42, No.6, 58, 122 (in Russian). (Cited in
Chapter 9.)
[94] Bernik, V. I. & Sakovich, N. V. (1994). Regular systems of complex algebraic
numbers. Dokl. Akad. Nauk Belarus, 38, 1013, 122 (in Russian). (Cited in
Chapter 9.)
[95] Bernik, V. I. & Tishchenko, K. (1993). Integral polynomials with an overfall of
the coefcient values and Wirsings problem. Dokl. Akad. Nauk Belarus, 37,
No.5, 911 (in Russian). (Cited in Chapter 3.)
[96] Bernik, V. I. & Vasiliev, D. V. (1999). A Khintchine-type theorem for integervalued polynomials of a complex variable. Proceedings of the Institute of Mathematics, 3, 1020. Tr. Inst. Mat. (Minsk), 3, Natl. Akad. Nauk Belarus, Inst.
Mat. Minsk (in Russian). (Cited in Chapter 9.)
[97] Bernik, V. I., Vasiliev, D. V. & Dodson, M. M. (2002). Metric theorems on approximation of real numbers by special real numbers. Dokl. Nats. Akad. Nauk
Belarus, 46, No.3, 6063 (in Russian). (Cited in Chapter 4.)
[98] Bernstein, F. (1912). Uber

eine Anwendung der Mengenlehre auf ein aus der
Theorie der sakularen Storungen herruhrendes Problem. Math. Ann., 71, 417
[99]
(1912). Uber
geometrische Wahrscheinlichkeit und u ber das Axiom der
beschrankten Arithmetisierbarkeit der Beobachtungen. Math. Ann., 72, 585
[100] Besicovitch, A. S. (1934). Sets of fractional dimension (IV); on rational approximation to real numbers. J. London Math. Soc., 9, 126131. (Cited in
Chapter 5.)
[101] Bilu, Yu. & Bugeaud, Y. (2000). Demonstration du theorème de Baker
Feldman via les formes lineaires en deux logarithmes. J. Theor. Nombres Bordeaux, 12, 1323. (Cited in Chapter 2.)
[102] Bluhm, C. (1998). On a theorem of Kaufman: Cantor-type construction of linear fractal Salem sets. Ark. Mat., 36, 307316. (Cited in Chapters 1 & 5.)
[103]
(2000). Liouville numbers, Rajchman measures, and small Cantor sets.
Proc. Amer. Math. Soc., 128, 26372640. (Cited in Chapter 1.)
[104] Bombieri, E. (1993). Effective Diophantine approximation on Gm . Ann.
Scuola Norm. Sup. Pisa Cl. Sci., 20, 6189. (Cited in Chapter 2.)
[105]
(1997). The equivariant ThueSiegel method. In Arithmetic Geometry (Cortona, 1994), 7086, Sympos. Math., XXXVII, Cambridge University Press, Cambridge. (Cited in Chapter 2.)
[106] Bombieri, E. & Cohen, P. B. (1997). Effective Diophantine approximation on
Gm , II. Ann. Scuola Norm. Sup. Pisa Cl. Sci., 24, 205225. (Cited in Chapter 2.)
[107] Bombieri, E. & Mueller, J. (1986). Remarks on the approximation to an algebraic number by algebraic numbers. Mich. Math. J., 33, 8393. (Cited in
Chapter 2.)
246
References
[108] Bombieri, E., van der Poorten, A. J. & Vaaler, J. D. (1996). Effective measures
of irrationality for cubic extensions of number elds. Ann. Scuola Norm. Sup.
Pisa Cl. Sci., 23, 211248. (Cited in Chapter 2.)
[109] Borbat, V. N. (1995). Joint zero approximation by the values of integral polynomials and their derivatives. Vests Akad. Navuk BSSR Ser. Fz. Mat. Navuk,
1, 916, 123 (in Russian). (Cited in Chapter 6.)
(1909). Les probabilites denombrables et leurs applications
[110] Borel, E.
arithmetiques. Rend. Circ. Math. Palermo, 27, 247271. (Cited in Chapter 1.)
[111]
(1912). Sur un problème de probabilites relatif aux fractions continues.
Math. Ann., 72, 578584. (Cited in Chapter 1.)
[112] Borosh, I. & Fraenkel, A. S. (1972). A generalization of Jarnks theorem on Diophantine approximations. Indag. Math., 34, 193201. (Cited in
Chapter 5.)
[113]
(1975). A generalization of Jarnks theorem on Diophantine approximations
to Ridout type numbers. Trans. Amer. Math. Soc., 211, 2338. (Cited in Chapter 5.)
[114] Borwein, P. & Erdelyi, T. (1995). Polynomials and polynomial inequalities.
Graduate Texts in Mathematics 161, Springer-Verlag, New York. (Cited in
Chapter 3.)
[115] Bovey, J. D. & Dodson, M. M. (1978). The fractional dimension of sets whose
simultaneous rational approximations have errors with a small product. Bull.
London Math. Soc., 10, 213218. (Cited in Chapter 5.)
[116]
(1986). The Hausdorff dimension of systems of linear forms. Acta Arith.,
[117] Braune, E. (1977). Uber

arithmetische Eigenschaften von Luckenreihen mit algebraischen Koefzienten und algebraischem Argument. Monatsh. Math., 84,
[118] Brownawell, D. (1974). Sequences of Diophantine approximations. J. Number
Theory, 6, 1121. (Cited in Chapter 8.)
[119]
(1985). Effectivity in independence measures for values of E-functions. J.
Austral. Math. Soc., 39, 227240. (Cited in Chapter 8.)
[120] Bugeaud, Y. (1998). Bornes effectives pour les solutions des e quations en
S-unites et des e quations de ThueMahler. J. Number Theory, 71, 227244.
[121]
(1998). Algebraic numbers close to 1 in non-Archimedean metrics. Ramanujan Math. J., 2, 449457. (Cited in Appendix A.)
[122]
(2000). On the approximation by algebraic numbers with bounded degree.
Algebraic Number Theory and Diophantine Analysis (Graz, 1998), 4753, de
Gruyter, Berlin. (Cited in Chapter 3.)
[123]
(2000). Approximation par des nombres algebriques. J. Number Theory, 84,
[124]
(2002). Approximation par des nombres algebriques de degre borne
et dimension de Hausdorff. J. Number Theory, 96, 174200. (Cited in
Chapter 6.)
[125]
(2002). Approximation by algebraic integers and Hausdorff dimension. J.
London Math. Soc., 65, 547559. (Cited in Chapters 3 & 10.)
[126]
(2002). Nombres de Liouville et nombres normaux. C. R. Acad. Sci. Paris,
References
[127]
[128]
[129]
[130]
[131]
[132]
[133]
[134]
[135]
[136]
[137]
[138]
[139]
[140]
[141]
[142]
[143]
[144]
[145]
[146]
[147]
[148]
247
(2003). A note on inhomogeneous Diophantine approximation. Glasgow

Math. J., 45, 105110. (Cited in Chapter 5.)
(2003). Mahlers classication of numbers compared with Koksmas. Acta
Arith., 110, 89105. (Cited in Chapters 3, 7 & 9.)
(2003). Sets of exact approximation order by rational numbers. Math. Ann.,
(To appear). An inhomogeneous Jarnk theorem. J. Analyse Math. (Cited in
Chapters 5 & 6.)
(2003). Approximation simultanee par des nombres algebriques. J. Theor.
Nombres Bordeaux, 15, 665672. (Cited in Chapter 3.)
(To appear). Intersective sets and Diophantine approximation. Mich. Math.
J. (Cited in Chapters 5 & 6.)
(Preprint). Mahlers classication of numbers compared with Koksmas, II.
Bugeaud, Y. & Chevallier, N. (Preprint). On simultaneous inhomogeneous Diophantine approximation. (Cited in Chapter 5.)
Bugeaud, Y., Dodson, M. M. & Kristensen, S. (Preprint). Zero-innity laws in
Diophantine approximation. (Cited in Chapters 6 & 9.)

Bugeaud, Y. & Gyory, K. (1996). Bounds for the solutions of ThueMahler
equations and norm form equation. Acta Arith., 74, 273292. (Cited in
Chapter 2.)
Bugeaud, Y. & Laurent, M. (Preprint). Exponents of Diophantine Approximation and Sturmian Continued Fractions. (Cited in Chapter 3.)
(Predprint). Exponents of Inhomogeneous Diophantine Approximation.
Bugeaud, Y & Mignotte, M. (To appear). On the distance between roots of
integer polynomials. Proc. Edinburgh Math. Soc. (Cited in Appendix A.)
Bugeaud, Y. & Teulie, O. (2000). Approximation dun nombre reel par des
nombres algebriques de degre donne. Acta Arith., 93, 7786. (Cited in Chapters
3 & 10.)
Bundschuh, P. (1978). Transzendenzmasse in Korpern formaler Laurentreihen. J. reine angew. Math. 299/300, 411432. (Cited in Chapter 9.)
(1980) Uber
eine Klasse reeller transzendenter Zahlen mit explicit angebbarer g-adischer und Kettenbruch-Entwicklung. J. reine angew. Math., 318,
Burger, E. B. (1992). Homogeneous Diophantine approximation in S-integers.
Pacic J. Math., 152, 211253. (Cited in Chapter 9.)
(1996). On Liouville decompositions in local elds. Proc. Amer. Math.
Soc., 124, 33053310. (Cited in Chapters 1 & 9.)
(1996). Uniformly approximable numbers and the uniform approximation
spectrum. J. Number Theory, 61, 194208. (Cited in Chapter 1.)
(1999). On real quadratic elds and simultaneous Diophantine approximation. Monatsh. Math., 128, 201209. (Cited in Chapter 9.)
(2000). On simultaneous Diophantine approximation in the vector space Q+
Q. J. Number Theory, 82, 1224. (Cited in Chapter 9.)
(2001). Diophantine inequalities and irrationality measures for certain transcendental numbers. Indian J. Pure Appl. Math., 32, 15911599. (Cited in
Chapter 7.)
248
References
[149] Burger, E. B. & Dubois, E. (1994). Sur les quotients partiels de U-nombres
dans un corps de series formelles. C. R. Acad. Sci. Paris, 319, 421426. (Cited
in Chapter 9.)
[150] Burger, E. B. & Struppeck, T. (1993). On frequency distributions of partial quotients of U-numbers. Mathematika, 40, 215225. (Cited in Chapter 7.)
[151] Cantor, D. G. (1965). On the elementary theory of Diophantine approximation over the ring of adeles. I. Illinois J. Math., 9, 677700. (Cited in
Chapter 9.)
[152] Cantor, G. (1874). Uber

eine Eigenschaft des Inbegriffs aller reellen algebraischen Zahlen. J. reine angew. Math., 77, 258262. (Cited in Chapter 1.)
[153] Cassels, J. W. S. (1951). Some metrical theorems in Diophantine approximation, V: On a conjecture of Mahler. Proc. Cambridge Philos. Soc., 47, 1821.
[154]
(1955). Simultaneous Diophantine approximation, II. Proc. London Math.
[155]
(1957). An introduction to Diophantine approximation. Cambridge Tracts in
Math. and Math. Phys., vol. 99, Cambridge University Press. (Cited in Chapters
1 & 6, and Appendix B.)
[156]
(1986). Local elds. London Mathematical Society Student Texts 3,
Cambridge University Press, Cambridge. (Cited in Chapter 3 and Appendix A.)
[157]
(1997). An Introduction to the Geometry of Numbers. Springer Verlag. (Cited
in Appendix B.)
[158] Cassels, J. W. S. & Swinnerton-Dyer, H. P. F. (1955). On the product of three
homogeneous linear forms and indenite ternary quadratic forms. Philos.
Trans. Roy. Soc. London, Ser. A., 248, 7396. (Cited in Chapter 10.)
[159] Caveny, D. (1993). U-numbers and T-numbers: Some elementary transcendence and algebraic independence results. Number Theory with an Emphasis
on the Markoff Spectrum (Provo, UT, 1991), 4352, Dekker, New York. (Cited
in Chapter 3.)
[160] Caveny, D. & Tubbs, R. (1993). The arithmetic of well-approximated numbers. Number Theory with an Emphasis on the Markoff Spectrum (Provo, UT,
1991), 5359, Dekker, New York. (Cited in Chapter 3.)
[161] Cerny,
K. (1952). Sur les approximations diophantiennes. Cehoslovack.
Mat.
Z.,
[162] Chowla, S. & DeLeon, M. J. (1968). On a conjecture of Littlewood. Norske
Vid. Selsk. Forh. (Trondheim), 41, 4547. (Cited in Chapter 10.)
[163] Chudnovsky, G. V. (1980). Algebraic independence of the values of elliptic
functions at algebraic points. Elliptic analogue of the LindemannWeierstrass
theorem. Invent. Math., 61, 267290. (Cited in Chapters 3 & 8.)
[164]
(1984). Contributions to the theory of transcendental numbers. (Translated from the Russian by J. A. Kandall.) Math. Surveys Monographs,
19, Amer. Math. Soc., Providence, R. I. (Cited in Chapters 3 & 8, and
Appendix A.)

[165] Chung, K. L. & Erdos, P. (1952). On the application of the BorelCantelli
lemma. Trans. Amer. Math. Soc., 72, 179186. (Cited in Chapter 6.)
[166] Cijsouw, P. L. (1972). Transcendence Measures. Doctoral dissertation, University of Amsterdam. (Cited in Chapters 3 & 8.)
References
249
[167] Cohn, H. (1946). Note on almost-algebraic numbers. Bull. Amer. Math. Soc.,
52, 10421045. (Cited in Appendix A.)
[168] Collins, G. E. (2001). Polynomial minimum root separation. J. Symbol.
Comp., 32, 467473. (Cited in Appendix A.)
[169] Collins, G. E. & Horowitz, E. (1974). The minimum root separation of a polynomial. Math. Comp., 28, 589597. (Cited in Appendix A.)
[170] Cusick, T. W. (1967). Littlewoods Diophantine approximation problem for series. Proc. Amer. Math. Soc., 18, 920924. (Cited in Chapter 10.)
[171]
(1971). Lower bound for a Diophantine approximation function. Monatsh.
Math., 75, 398401. (Cited in Chapter 10.)
[172]
(1990). Hausdorff dimensions of sets of continued fractions. Quart. J. Math.
Oxford, 41, 277286. (Cited in Chapter 5.)
[173] Cusick, T. W. & Flahive, M. E. (1989). The Markoff and Lagrange spectra.
Mathematical Surveys and Monographs, 30. American Mathematical Society,
Providence, RI. (Cited in Chapter 1.)
[174] Dajani, K. & Kraaikamp, C. (2002). Ergodic theory of numbers. Carus Mathematical Monographs, 29. Mathematical Association of America, Washington,
DC. (Cited in Chapter 1.)
[175] Davenport, H. (1954). Simultaneous Diophantine approximation. Mathematika, 1, 5172. (Cited in Chapters 1, 3 & 10.)
[176]
(1961). A note on binary cubic forms. Mathematika, 8, 5862. (Cited in
Chapter 4.)

[177] Davenport, H., Erdos, P. & LeVeque, W. J. (1963). On Weyls criterion for uniform distribution. Michigan Math. J., 10, 311314. (Cited in
Chapter 1.)
[178] Davenport, H. & Lewis, D. J. (1963). An analogue of a problem of Littlewood.
Michigan Math. J., 10, 157160. (Cited in Chapter 10.)
[179] Davenport, H. & Roth, K. F. (1955). Rational approximations to algebraic numbers. Mathematika, 2, 160167. (Cited in Chapter 2.)
[180] Davenport, H. & Schmidt, W. M. (1967). Approximation to real numbers by
quadratic irrationals. Acta Arith., 13, 169176. (Cited in Chapters 3 & 10.)
[181]
(1967/68). A theorem on linear forms. Acta Arith., 14, 209223. (Cited in
Chapter 3.)
[182]
(1969). Approximation to real numbers by algebraic integers. Acta Arith.,
15, 393416. (Cited in Chapters 2, 3, 8 & 9.)
[183]
(1970). Dirichlets theorem on Diophantine approximation. Symposia
Mathematica, Vol. IV (INDAM, Rome, 1968/69), pp. 113132. Academic
Press, London. (Cited in Chapters 1 & 3.)
[184]
(1970). Dirichlets theorem on Diophantine approximation, II. Acta Arith.,
[185] Davison, J. L. (1989). A class of transcendental numbers with bounded partial
quotients. In Number Theory and Applications, R. A. Mollin, (ed.) pp. 365
371. Kluwer Academic Publishers. (Cited in Chapter 2.)
[186]
(2002). Continued fractions with bounded partial quotients. Proc. Edinburgh Math. Soc., 45, 653671. (Cited in Chapter 2.)
[187] Diaz, G. (1993). Une nouvelle minoration de | log |, | exp |, et
algebriques. Acta Arith., 64, 4357. (Cited in Chapters 3 & 8.)
250
[188]
[189]
[190]
[191]
[192]
[193]
[194]
[195]
[196]
[197]
[198]
[199]
[200]
[201]
[202]
[203]
[204]
[205]
[206]
References
(1997). Une nouvelle propriete dapproximation diophantienne. C. R. Acad.

Sci. Paris, 324, 969972. (Cited in Chapter 8 and Appendix A.)
Diaz, G. & Mignotte, M. (1991). Passage dune mesure dapproximation a` une
mesure de transcendence. C. R. Math. Rep. Acad. Sci. Canada, 13, 131134.
(Cited in Chapter 8 and Appendix A.)
Dickinson, H. (1993). The Hausdorff dimension of systems of simultaneously
small linear forms. Mathematika, 40, 367374. (Cited in Chapter 5.)
(1994). The Hausdorff dimension of sets arising in metric Diophantine approximation. Acta Arith., 68, 133140. (Cited in Chapter 5.)
(1997). A remark on the JarnkBesicovitch theorem. Glasgow Math. J., 39,
Dickinson, H. & Dodson, M. M. (2001). Diophantine approximation and Hausdorff dimension on the circle. Math. Proc. Cambridge Phil. Soc., 130, 515522.
Dickinson, H., Dodson, M. M. & Yuan, J. (1999). Hausdorff dimension and padic Diophantine approximation. Indag. Math. (N. S.), 10, 337347. (Cited in
Chapter 9.)
Dickinson, H. & Velani, S. L. (1997). Hausdorff measure and linear forms. J.
reine angew. Math., 490, 136. (Cited in Chapter 6.)
Dirichlet, L. G. P. (1842). Verallgemeinerung eines Satzes aus der Lehre von
den Kettenbruchen nebst einige Anwendungen auf die Theorie der Zahlen. S.B. Preuss. Akad. Wiss., 9395. (Cited in Chapter 1.)
Divis, B. (1972). An analog to the Lagrange numbers. J. Number Theory, 4,
Dodson, M. M. (1984). A note on the HausdorffBesicovitch dimension of systems of linear forms. Acta Arith., 44, 8798. (Cited in Chapter 5.)
(1991). Star bodies and Diophantine approximation. J. London Math. Soc.,
(1992). Hausdorff dimension, lower order and Khintchines theorem in metric Diophantine approximation. J. reine angew. Math., 432, 6976. (Cited in
Chapter 6.)
(1993). Geometric and probabilistic ideas in the metrical theory of Diophantine approximation. Tsp. Mat. Nauk, 48, 77106 (in Russian). English transl.
in Russian Math. Surveys, 48 (1993), 73102. (Cited in Chapter 5.)
(1997). A note on metric inhomogeneous Diophantine approximation. J.
Austral. Math. Soc., 62, 175185. (Cited in Chapter 5.)
(2002). Exceptional sets in dynamical systems and Diophantine approximation. Rigidity in dynamics and geometry (Cambridge, 2000), 7798, Springer,
Berlin. (Cited in Chapter 5.)
Dodson, M. M. & Hasan, S. (1992). Systems of linear forms and covers for star
bodies. Acta Arith., 61, 119127. (Cited in Chapter 5.)
Dodson, M. M. & Kristensen, S. (To appear). Hausdorff dimension and Diophantine approximation. Fractal Geometry and Applications: A Jubilee of
Benoit Mandelbrot, Proceedings of Symposia in Pure Mathematics, American
Mathematical Society. (Cited in Chapters 5, 6 & 9.)
Dodson, M. M., Rynne, B. P. & Vickers, J. A. G. (1990). Diophantine approximation and a lower bound for Hausdorff dimension. Mathematika, 37, 5973.
References
251
[207] Dombrovski, R. I. (1989). Simultaneous approximations of real numbers by

algebraic numbers of bounded degree. Dokl. Akad. Nauk BSSR, 33, 205208,
[208] Dress, A., Elkies, N. & Luca, F. (Preprint). A characterization of Mahlers generalized Liouville numbers by simultaneous rational approximation. (Cited in
Chapter 3.)
[209] Dubickas, A. (1992). An estimation of the difference between two zeros of a
polynomial. New Trends in Probability and Statistics, Vol. 2 (Palanga, 1991),
1721, VSP, Utrecht. (Cited in Appendix A.)
[210]
(1998). On algebraic numbers close to 1. Bull. Austral. Math. Soc., 58, 423
434. (Cited in Appendix A.)
[211] Dubois, E. (1977). Application de la methode de W. M. Schmidt a`
lapproximation de nombres algebriques dans un corps de fonctions de
caracteristique zero. C. R. Acad. Sci. Paris, 284, 15271530. (Cited in
Chapter 9.)
[212]
(1996). On Mahlers classication in Laurent series elds. Rocky Mountain
J., 26, 10031016. (Cited in Chapter 9.)
[213] Dufn, R. J. & Schaeffer, A. C. (1941). Khintchines problem in metric Diophantine approximation. Duke J., 8, 243255. (Cited in Chapter 1.)
[214] Dumas, G. (1906). Sur quelques cas dirreductibilite des polynomes a` coefcients rationnels. J. Math. pures appl., (6) 2, 191258. (Cited in Chapter 7.)
[215] Durand, A. (1973/74). Un critère de transcendence. Seminaire Delange
PisotPoitou: Groupe detude de theorie des nombres, 15ème annee, numero
G11. (Cited in Chapter 8.)
[216]
(1974). Quatre problèmes de Mahler sur la fonction ordre dun nombre transcendant. Bull. Soc. Math. France, 102, 365377. (Cited in Chapter 8.)
[217]
(1975). Fonctions -ordre et classication de C p . C. R. Acad. Sci. Paris,
280, 10851088. (Cited in Chapter 8.)
[218]
(1977). Independance algebrique de nombres complexes et critères de transcendance. Compositio Math., 35, 259267. (Cited in Chapter 8.)
[219]
(1978). Approximations algebriques dun nombre transcendant. C. R. Acad.
Sci. Paris, 287, 595597. (Cited in Chapter 8.)
[220]
(1990). Quelques aspects de la theorie analytique des polynomes I, II. Cinquante ans de polynomes (Paris, 1988), pp. 142, 4385, Lecture Notes in
Math., 1415, Springer, Berlin. (Cited in Appendix A.)
[221]
(1990). Approximations algebriques dun nombre transcendant. Cinquante
ans de polynomes (Paris, 1988), pp. 9496, Lecture Notes in Math., 1415,
Springer, Berlin. (Cited in Chapter 8.)
[222] Duverney, D. (2001). Transcendence of a fast converging series of rational
numbers. Math. Proc. Cambridge Phil. Soc., 130, 193207. (Cited in Chapter 3.)
[223] Dyson, F. J. (1947). The approximation to algebraic numbers by rationals.
Acta Math., 79, 225240. (Cited in Chapter 2.)
[224]
(1947). On simultaneous Diophantine approximations. Proc. London Math.
Soc., 49, 409420. (Cited in Appendix B.)
[225] Eggleston, H. G. (1951). Sets of fractional dimension which occur in some
problems in number theory. Proc. London Math. Soc., 54, 4293. (Cited in
Chapter 5.)
252
References
[226] Erdos, P. (1959). Some results on Diophantine approximation. Acta Arith., 5,

[227]
(1962). Representations of real numbers as sums and products of
Liouville numbers. Michigan Math. J., 9, 5960. (Cited in Chapters
1 & 9.)
[228]
(1970). On the distribution of convergents of almost all real numbers. J.
Number Theory, 2, 425441. (Cited in Chapter 1.)

[229] Erdos, P. & Mahler, K. (1939). Some arithmetical properties of the convergents
of a continued fraction. J. London Math. Soc., 14, 1218. (Cited in Chapters 5
& 10.)
[230] Euler, L. (1737). De fractionibus continuis. Commentarii Acad. Sci. Imperiali
Petropolitanae, 9. (Cited in Chapter 1.)
[231] Everest, G. & Ward, T. (1999). Heights of polynomials and entropy in algebraic dynamics. Universitext, Springer-Verlag London, Ltd, London. (Cited in
Appenidx A.)
[232] Evertse, J-H. (2000). Symmetric improvements of Liouvilles inequality: A
survey. In Algebraic Number Theory and Diophantine Analysis (Graz, 1998),
129141, de Gruyter, Berlin. (Cited in Chapter 10.)
[233]
Approximation of complex algebraic numbers by algebraic numbers of
bounded degree. www.math.leidenuniv.nl/evertse/publicaties.shtml (Cited in
Chapter 9.)
[234] Falconer, K. (1985). The geometry of fractal sets. Cambridge Tracts in Mathematics, 85, Cambridge University Press. (Cited in Chapter 5.)
[235]
(1985). Classes of sets with large intersections. Mathematika, 32, 191 205.
[236]
(1990). Fractal Geometry: Mathematical Foundations and Applications. John
Wiley & Sons. (Cited in Chapters 5 & 6.)
[237]
(1994). Sets with large intersection properties. J. London Math. Soc., 49,
[238]
(2000). Representation of families of sets by measures, multifractal analysis and Diophantine approximation. Math. Proc. Cambridge Philos. Soc., 128,
[239] Fatou, P. (1904). Sur lapproximation des incommensurables et des series
trigonometriques. C. R. Acad. Sci. Paris, 139, 10191021. (Cited in Chapter 1.)
[240] Feldman, N. I. (1962). On transcendental numbers having an approximation of
given type. Uspehi mat. Nauk, 17, No.5 (107), 145151 (in Russian). (Cited in
Chapter 7.)
[241]
(1971). An effective renement of the exponent in Liouvilles theorem. Izv.
Akad. Nauk, 35, 973990 (in Russian). English transl. in Math. USSR Izv., 5
(1971), 9851002. (Cited in Chapter 2.)
[242]
(1981). Approximation of algebraic numbers. Moskov. Gos. Univ., Moscow
(in Russian). (Cited in Chapter 2.)
[243]
(1982). Hilberts seventh problem. Moskov. Gos. Univ., Moscow (in Russian). (Cited in Appenidx A.)
[244] Feldman, N. I. & Nesterenko, Yu. V. (1998). Number Theory IV: Transcendental
Numbers. Encyclopaedia of Mathematical Sciences 44. Springer-Verlag, Berlin.
(Cited in Chapters 2, 3 & 8.)
References
253
[245] Feng, D.-J., Wu, J., Liang, J.-C. & Tseng, S. (1997). Appendix to the paper by
T. /Luczak a simple proof of the lower bound. Mathematika, 44, 5455. (Cited
in Chapter 5.)
[246] Ferenczi, S. & Mauduit, C. (1997). Transcendence of numbers with a low complexity expansion. J. Number Theory, 67, 146161. (Cited in Chapter 2.)
[247] Ford, L. R. (1925). On the closeness of approach of complex rational fractions
to a complex irrational number. Trans. Amer. Math. Soc., 27, 146154. (Cited
in Chapter 9.)
[248] Forder, H. G. (1963). A simple proof of a result on Diophantine approximation.
Math. Gaz., 47, 237238. (Cited in Chapter 1.)

[249] Fraenkel, A. S. (1964). Transcendental numbers and a conjecture of Erdos and
Mahler. J. London Math. Soc., 39, 405416. (Cited in Chapter 5.)
[250] Fraenkel, A. S. & Borosh, I. (1965). Fractional dimension of a set of transcendental numbers. Proc. London Math. Soc., 15, 458470; corrigendum: Proc.
London Math. Soc., 16 (1966), 192. (Cited in Chapters 5 & 10.)
[251] Frostman, O. (1935). Potentiel dequilibre et capacite des ensembles avec
quelques applications a` la theorie des fonctions. Meddel. Lunds Univ. Math.
Sem., 3, 1118. (Cited in Chapter 5.)
[252] Fuchs, M. (2002). On metric Diophantine approximation in the eld of formal
Laurent series. Finite Fields Appl., 8, 343368. (Cited in Chapter 9.)
[253] Gallagher, P. (1962). Metric simultaneous Diophantine approximation. J. London Math. Soc., 37, 387390. (Cited in Chapter 4.)
[254] Galockin, A. I. (1980). A transcendence measure for the values of functions
satisfying certain functional equations. Mat. Zametki., 27, 175183, 317 (in
Russian). English transl. in Math. Notes, 27, 8388. (Cited in Chapters 3 & 8.)
[255] Ganesa Moorthy, C. (1992). A problem of Good on Hausdorff dimension.
Mathematika, 39, 244246. (Cited in Chapter 5.)
[256] Gelfond, A. O. (1948). Approximation of algebraic irrationalities and their logarithms. Vest. Mosk. Univ., 9, 325 (in Russian). (Cited in Chapter 2.)
[257]
(1952). Transcendental and algebraic numbers. Gosudartsv. Izdat. Tehn.Teor. Lit., Moscow (in Russian). Translated from the rst Russian edition
by Leo F. Boron, Dover publ., New York, 1960. (Cited in Chapter 8 and
Appenidx A.)
[258] Good, I. J. (1941). The fractional dimensional theory of continued fractions.
Proc. Cambridge Phil. Soc., 37, 199228. (Cited in Chapter 5.)
[259] Grace, J. H. (1918). The classication of rational approximations. Proc. London Math. Soc., 17, 247258. (Cited in Chapter 1.)
[260] Groshev, A. V. (1938). A theorem on systems of linear forms. Dokl. Akad.
Nauk SSSR, 19, 151152 (in Russian). (Cited in Chapter 1.)
[261] Gruber, P. M. (1983). In most cases approximation is irregular. Rend. Sem.
Mat. Univers. Politecn. Torino, 41, 1933. (Cited in Chapter 1.)
[262] Gruber, P. M. & Lekkerkerker, C. G. (1987). Geometry of numbers. Series Bibliotheca Mathematica, 8, North-Holland, Amsterdam. (Cited in Chapter 1 and
Appendix B.)
[263] Guntermann, N. (1996). Approximation durch algebraische Zahlen
beschrankten Grades im Korper der formalen Laurentreihen. Monatsh.
254
References
[264] Gurses, H. (2000). Uber

verallgemeinerte Luckenreihen. Istanbul Univ.
Fen
Fak. Mat. Derg., 59, 5988. (Cited in Chapters 3 & 7.)
[265]
(2000). Uber
eine Klasse verallgemeinerter Luckenreihen mit algebraischen
Koefzienten. Istanbul Univ.

Fen Fak. Mat. Derg., 59, 89102. (Cited in Chapter 7.)
[266] Guting, R. (1963). On Mahlers function 1 . Mich. Math. J., 10, 161179.
[267]
(1965). Uber
den Zusammenhang zwischen rationalen Approximationen und Kettenbruchentwicklungen. Math. Z., 90, 382387. (Cited in
Chapter 3.)
[268]
(1967). Polynomials with multiple zeros. Mathematika, 14, 181196. (Cited
in Appendix A.)
[269]
(1968). Zur Berechnung der Mahlerschen Funktionen wn . J. reine angew.
Math., 232, 122135. (Cited in Chapters 3 & 7.)
[270] Hall, M. (1947). On the sum and product of continued fractions. Ann. of Math.,
[271] Hardy, G. H. & Wright, E. M. (1979). An Introduction to the Theory of Numbers,
5th. Edition. Clarendon Press. (Cited in Chapter 1.)
[272] Harman, G. (1988). Metric Diophantine approximation with two restricted
variables, III. Two prime numbers. J. Number Theory, 29, 364375. (Cited in
Chapter 1.)
[273]
(1998). Metric Number Theory. LMS Monographs New Series, vol. 18.
Clarendon Press. (Cited in Introduction, and Chapters 1, 4 & 5.)
[274]
(2000). Variants of the second BorelCantelli lemma and their applications
in metric number theory. Number Theory, 121140, Trends Math., Birkhauser,
Basel. (Cited in Chapters 1 & 6.)
[275]
(2003). Simultaneous Diophantine approximation and asymptotic formulae
on manifolds. Acta Arith., 108, 379389. (Cited in Chapter 4.)
[276] Hausdorff, F. (1919). Dimension und a usseres Ma. Math. Ann., 79, 157179.
[277] Hensley, D. (1992). Continued fractions Cantor sets, Hausdorff dimension and
functional analysis. J. Number Theory, 40, 336358. (Cited in Chapter 5.)
[278]
(1996). A polynomial time algorithm for the Hausdorff dimension of continued fractions Cantor sets. J. Number Theory, 58, 945; erratum J. Number
Theory, 59 (1996), 419. (Cited in Chapter 5.)
[279] Hernandez, S. (Preprint). Comparing classications of p-adic transcendental
numbers. (Cited in Chapter 9.)
[280] Hightower, C. J. (1975). Approximation to complex numbers by certain biquadratic numbers. J. Number Theory, 7, 293309. (Cited in Chapter 9.)
[281] Hinokuma, T. & Shiga, H. (1992). A remark on theorem of Jarnk. Ryukyu
Math. J., 5, 16. (Cited in Chapter 5.)
[282]
(1996). Hausdorff dimension of sets arising in Diophantine approximation.
Kodai Math. J., 19, 365377. (Cited in Chapters 5 & 6.)
[283] Hirst, K. E. (1970). A problem in the fractional dimension theory of continued
fractions. Quart. J. Math. Oxford, 21, 2935. (Cited in Chapter 5.)
[284] Hurwitz, A. (1891). Uber

die angenaherte Darstellung der Irrationalzahlen
durch rationale Bruche. Math. Ann., 39, 279284. (Cited in Chapter 1.)
References
255
[285] Inoue, K. & Nakada, H. (2003). On metric Diophantine approximation in positive characteristic. Acta Arith., 110, 205218. (Cited in Chapter 9.)
[286] Iosifescu, M. & Kraaikamp, C. (2002). Metrical Theory of Continued Fractions. Mathematics and its Applications 547. Kluwer Academic Publishers,
Dordrecht. (Cited in Chapter 1.)
[287] Ivanov, V. A. (1978). A theorem of Dirichlet in the theory of Diophantine approximations. Mat. Zametki, 24, 459474, 589 (in Russian). English transl. in
Math. Notes, 24 (1978), 747755. (Cited in Chapter 1.)
[288] Jarnk, V. (1928/29). Zur metrischen Theorie der diophantischen Approximationen. Prace Mat. Fz., 36, 91106. (Cited in Chapters 5 & 9.)
[289]
(1929). Diophantischen Approximationen und Hausdorffsches Mass. Mat.
Sbornik, 36, 371382. (Cited in Chapters 5 & 9.)
e Akad.,
[290]
(1930). Nekolik poznamek o Hausdorffove mre. Rozpravy Tr. Cesk
40, c. 9, 8 p. (Cited in Chapter 6.)
[291]
(1930). Quelques remarques sur la mesure de M. Hausdorff. Bull. Int. Acad.
Sci. Boheme, 16. (Cited in Chapter 6.)
[292]
(1931). Uber
die simultanen Diophantische Approximationen. Math. Z., 33,
505543. (Cited in Chapters 1, 5 & 6.)
[293]
(1935). Uber
einen Satz von A. Khintchine. Prace Mat. Fz., 43, 116.
(Cited in Chapter 10 and Appendix B.)
e
[294]
(1936). O simultannich diofantickych approximacich. Rozpravy Tr. Cesk
Akad., 45, c. 19, 16 p. (Cited in Appendix B.)
[295]
(1936). Uber
einen Satz von A. Khintchine, 2. Mitteilung. Acta Arith., 2,
122. (Cited in Chapter 10 and Appendix B.)
[296]
(1939). Uber
einen p-adischen Ubertragungssatz.
Monatsh. Math. Phys.,
[297]
(1945). Sur les approximations diophantiennes des nombres p-adiques. Revista Ci. Lima, 47, 489505. (Cited in Chapter 9.)
[298]
(1954). Contribution to the metric theory of continued fractions. Czechoslovak Math. J., 4 (79), 318329 (in Russian). (Cited in Chapter 5.)
[299]
(1959). Eine Bemerkung zum Ubertragungssatz.
Bulgar. Akad. Nauk Izv.
Mat. Inst., 3, 169175. (Cited in Appendix B.)
[300]
(1969). Un theorème dexistence pour les approximations diophantiennes.
Enseignement Math., 15, 171175. (Cited in Chapter 6.)
[301] Jenkinson, O. & Pollicott, M. (2001). Computing the dimension of dynamically
dened sets: E 2 and bounded continued fractions. Ergod. Th. Dynam. Sys., 21,
[302] Jones, H. (2001). Contributions to Metric Number Theory. Ph.D. thesis, Cardiff.
[303]
(2001). Khintchines theorem in k dimensions with prime numerator and
denominator. Acta Arith., 99, 205225. (Cited in Chapter 1.)
[304] Kahane, J.-P. & Katznelson, Y. (1973). Sur les ensembles dunicite U() de
Zygmund. C. R. Acad. Sci. Paris, 227, 893895. (Cited in Chapter 7.)
[305] Kahane, J.-P. & Salem, R. (1963). Ensembles Parfaits et Series
Trigonometriques. Herrmann. (Cited in Chapter 7.)
[306] Kaindl, G. (1976). Simultane Diophantische Approximationen. Osterreich.

Akad. Wiss. Math. Naturwiss. Kl. S.-B. II, 185, 387404. (Cited in Chapter 1.)
256
References
[307] Kalosha, N. I. (2001). On the approximation properties of algebraic numbers

of the third degree. Vests Nats. Akad. Navuk Belarus Ser. Fz. Mat. Navuk, 3,
1217, 140 (in Russian). (Cited in Chapter 4.)
[308]
(2003). Simultaneous approximations of zero by values of integral polynomials in R Q p . Dokl. Nats. Akad. Nauk Belarus, 47, No.2, 1922 (in
Russian). (Cited in Chapter 9.)
[309] Kargaev, P. & Zhigljavsky, A. (1996). Approximation of real numbers by rationals: some metric theorems. J. Number Theory, 61, 209225. (Cited in Chapter 1.)
[310] Kasch, F. (1958). Uber

eine metrische Eigenschaft der S-Zahlen. Math. Z., 70,
[311] Kasch, F. & Volkmann, B. (1958). Zur Mahlerschen Vermutung u ber SZahlen. Math. Ann., 136, 442453. (Cited in Chapters 3, 4, 5 & 7.)
[312]
(1959/60). Metrische Satze u ber transzendente Zahlen in P-adischen
Korpern. Math. Z., 72, 367378. (Cited in Chapter 9.)
[313]
(1962). Metrische Satze u ber transzendente Zahlen in P-adischen Korpern,
II. Math. Z., 78, 171174. (Cited in Chapter 9.)
[314] Kaufman, R. (1980). Continued fractions and Fourier transforms. Mathematika, 27, 262267. (Cited in Chapter 7.)
[315]
(1981). On the theorem of Jarnk and Besicovitch. Acta Arith., 39, 265267.
[316] Kechris, A. S. (1995). Classical Descriptive Set Theory. Springer-Verlag, Berlin
and New York. (Cited in Chapter 7.)
[317] Khintchine, A. Ya. (1924). Einige Satze u ber Kettenbruche, mit Anwendungen
auf die Theorie der diophantischen Approximationen. Math. Ann., 92, 115
[318]
(1925). Zwei Bemerkungen zu einer Arbeit des Herrn Perron. Math. Z., 22,
274284. (Cited in Chapters 3 & 10, and Appendix B.)
[319]
(1926). Zur metrischen Theorie der diophantischen Approximationen.
Math. Z., 24, 706714. (Cited in Chapter 1.)
[320]
(1926). Uber
eine Klasse linearer diophantischer Approximationen.
Rendiconti Circ. Mat. Palermo, 50, 170195. (Cited in Chapter 1 and
Appendix B.)
[321]
(1935). Metrische Kettenbruchprobleme. Compositio Math., 1, 361382.
[322]
(1936). Zur metrischen Kettenbruchetheorie. Compositio Math., 3, 275
[323]
(1964). Continued Fractions. The University of Chicago Press. (Cited in
Chapters 1 & 3.)
[324] Haseo Ki. (1995). The Borel classes of Mahlers A, S, T and U numbers. Proc.
Amer. Math. Soc., 123, 31973204. (Cited in Chapter 7.)
[325] Kleinbock, D. (1999). Badly approximable systems of afne forms. J. Number
Theory, 79, 83102. (Cited in Chapter 5.)
[326]
(2001). Some applications of homogeneous dynamics to number theory.
Smooth Ergodic Theory and its Applications (Seattle, WA, 1999). Proceedings
of Symposia in Pure Mathematics, 69, Amer. Math. Soc., Providence. (Cited in
Chapter 4.)
References
[327]
[328]
[329]
[330]
[331]
[332]
[333]
[334]
[335]
[336]
[337]
[338]
[339]
[340]
[341]
[342]
[343]
[344]
[345]
257
(2003). Extremal subspaces and their submanifolds. Geom. Funct. Anal.,

(To appear). BakerSprindzuk conjectures for complex analytic manifolds.
Algebraic groups and Arithmetic. T.I.F.R., India. (Cited in Chapter 9.)
Kleinbock, D., Lindenstrauss, E. & Weiss, B. (To appear). On fractal measures
and Diophantine approximation. Selecta Math. (Cited in Chapters 4, 5 & 10.)
Kleinbock, D. & Margulis, G. A. (1998). Flows on homogeneous spaces and
Diophantine approximation on manifolds. Ann. of Math., 148, 339360. (Cited
in Chapters 4 & 5.)
Kleinbock, D. & Tomanov, G. (Preprint). Flows on S-arithmetic homogeneous
spaces and applications to metric Diophantine approximation. (Cited in Chapter 9.)
Koksma, J. F. (1936). Diophantische Approximationen. Ergebnisse d. Math. u.
ihrer Grenzgebiete, 4, Springer. (Cited in Introduction and Chapter 9.)
(1939). Uber
die Mahlersche Klasseneinteilung der transzendenten Zahlen
und die Approximation komplexer Zahlen durch algebraische Zahlen.
Monatsh. Math. Phys., 48, 176189. (Cited in Chapters 3, 4, 6, 8, 9 & 10.)
Koksma, J. F. & Popken, J. (1932). Zur Transzendenz von e. J. reine angew.
Math., 168, 211230. (Cited in Appendix A.)
Kopetzky, H. G. (1989). Some results on Diophantine approximation related to
Dirichlets theorem. Number Theory (Ulm, 1987), 137149, Lecture Notes in
Math., 1380, Springer, New York. (Cited in Chapter 1.)
Kopetzky, H. G. & Schnitzer, F. J. (1986). Quadratische Approximationen am
Einheitskreis. Arch. Math., 46, 144147. (Cited in Chapter 3.)
Korkine, A. & Zolotareff, G. (1873). Sur les formes quadratiques. Math. Ann.,
Kovalevskaya, E. I. (1999). Simultaneous approximations of real numbers by
algebraic numbers of a given type. Vests Nats. Akad. Navuk Belarus Ser. Fz.
Mat. Navuk, 3, 59, 138 (in Russian). (Cited in Chapter 5.)
(1999). A metric theorem on the exact order of approximation of zero by
values of integer polynomials in Q p . Dokl. Nats. Akad. Nauk Belarus, 43,
No.5, 3436, 123 (in Russian). (Cited in Chapter 9.)
(2000). The convergence Khintchine theorem for polynomials and planar
p-adic curves. Number theory (Liptovsky Jan, 1999), Tatra Mt. Math. Publ.
Kristensen, S. (2003). On well-approximable matrices over a eld of formal series. Math. Proc. Cambridge Philos. Soc., 135, 255268. (Cited in
Chapter 9.)
(Preprint). Badly approximable systems of linear forms over a eld of formal
series. (Cited in Chapter 9.)
(2003). Diophantine approximation and the solubility of the Schrodinger
equation. Phys. Lett. A, 314, 1518. (Cited in Chapter 5.)
Kubilius, I. (1949). On the application of I. M. Vinogradovs method to the
solution of a problem of the metric theory of numbers. Doklady Akad. Nauk
SSSR (N.S.), 67, 783786 (in Russian). (Cited in Chapter 4.)
Kuhnlein, S. (1996). On a measure theoretic aspect of Diophantine approximation. J. reine angew. Math., 477, 117127. (Cited in Chapter 1.)
258
References
[346] Lagarias, J. C. (1981). A complement to Ridouts p-adic generalization of the

ThueSiegelRoth theorem. Analytic Number Theory (Philadelphia, 1980),
264275, Lecture Notes in Math., 899, Springer, BerlinNew York. (Cited in
Chapters 2 & 10.)
[347] Lagrange, J. L. (1770). Additions au memoire sur la resolution des e quations
numeriques. Mem. Berl., 24. (Cited in Chapter 1.)
[348] Lang, S. (1962). A transcendence measure for E-functions. Mathematika, 9,
[349]
(2002). Algebra, revised third edition. Graduate Texts in Mathematics, 211.
Springer-Verlag, New York. (Cited in Chapter 3 and Appendix A.)
[350] Langmayr, F. (1980). On Dirichlets approximation theorem. Monatsh. Math.,
[351] Lasjaunias, A. (2000). A survey of Diophantine approximation in elds of
power series. Monatsh. Math., 130, 211229. (Cited in Chapters 9 & 10.)
[352] Laurent, M. (1999). Diophantine approximation and algebraic independence. Colloque FrancoJaponais: Theorie des Nombres Transcendants
(Tokyo, 1998), 7589. Sem. Math. Sci., 27, Keio Univ. Yokohama. (Cited in
Chapter 8.)
[353]
(1999). Some remarks on the approximation of complex numbers by algebraic numbers. Proceedings of the 2nd Panhellenic Conference in Algebra and
Number Theory (Thessaloniki, 1998). Bull. Greek Math. Soc., 42, 4957. (Cited
in Chapter 8.)
[354]
(2003). Simultaneous rational approximation to the successive powers of a
real number. Indag. Math. (N.S.), 11, 4553. (Cited in Chapter 3.)
[355] Laurent, M. & Poulakis, D. (2004). On the global distance between two
algebraic points on a curve. J. Number Theory, 104, 210254. (Cited in
Chapter 7.)
[356] Laurent, M. & Roy, D. (1999). Criteria of algebraic independence with multiplicities and interpolation determinants. Trans. Amer. Math. Soc., 351, 1845
1870. (Cited in Chapter 8 and Appendix A.)
[357]
(1999). Sur lapproximation algebrique en degre de transcendance un. Ann.
Inst. Fourier (Grenoble), 49, 2755. (Cited in Chapter 8.)
[358]
(2001). Criteria of algebraic independence with multiplicities and approximation by hypersurfaces. J. reine angew. Math., 536, 65114. (Cited in
Chapter 8.)
[359] Legendre, A.-M. (1830). Theorie des Nombres, troisième e dition, Tome 1. Paris.
[360] LeVeque, W. J. (1952). Continued fractions and approximations in k(i), I and
II. Indag. Math., 14, 526545. (Cited in Chapter 9.)
[361]
(1953). On Mahlers U-numbers. J. London Math. Soc., 28, 220 229.
[362]
(1953). Note on S-numbers. Proc. Amer. Math. Soc., 4, 189190. (Cited in
Chapter 4.)
[363]
(1956). Topics in Number Theory, Vols. 1 & 2. Addison-Wesley Publishing
Co., Inc., Reading, MA. (Cited in Chapter 2.)
[364]
(1960). On the frequency of small fractional parts in certain real sequences,
II. Trans. Amer. Math. Soc., 94, 130149. (Cited in Chapter 1.)
References
259
[365] Levesley, J. (1998). A general inhomogeneous JarnkBesicovitch theorem.

[366] Levy, P. (1936). Sur le developpement en fraction continue dun nombre choisi
au hasard. Compositio Math., 3, 286303. (Cited in Chapter 1.)
[367] Liardet, P. & Stambul, P. (2000). Series de Engel et fractions continuees.
J. Theor. Nombres Bordeaux, 12, 3768. (Cited in Chaprer 2.)
[368] Liouville, J. (1844). Remarques relatives a` des classes très-etendues de quantites dont la valeur nest ni algebrique, ni meme reductible a` des irrationnelles
algebriques. C. R. Acad. Sci. Paris, 18, 883885. (Cited in Chapter 1.)
[369]
(1844). Nouvelle demonstration dun theorème sur les irrationnelles
algebriques. C. R. Acad. Sci. Paris, 18, 910911. (Cited in Chapter 1.)
[370]
(1851). Sur des classes très-etendues de quantites dont la valeur nest ni
algebrique, ni meme reductible a` des irrationnelles algebriques. J. Math. pures
appl., (1) 16, 133142. (Cited in Chapter 1.)
[371] Lock, D. J. (1947). Metrisch-Diophantische Onderzoekingen in K(P) en
K n (P). Dissertation, Vrije Univ., Amsterdam. (Cited in Chapter 9.)
[372] /Luczak, T. (1997). On the fractional dimension of sets of continued fractions.
Mathematika, 44, 5053. (Cited in Chapter 5.)
(1955). Sur les Approximations Diophantiennes Lineaires et p-adiques.
[373] Lutz, E.
Hermann. (Cited in Chapter 9.)
[374] Lutzen, J. (1990). Joseph Liouville 18091882: Master of Pure and Applied
Mathematics. Studies in the History of Mathematics and Physical Sciences, 15,
Springer-Verlag, New York. (Cited in Chapter 1.)
[375] Mahler, K. (1930). Uber

Beziehungen zwischen der Zahl e und den Liouvilleschen Zahlen. Math. Z., 31, 729732. (Cited in Chapter 3.)
[376]
(1932). Zur Approximation der Exponentialfunktionen und des Logarithmus, I, II. J. reine angew. Math., 166, 118150. (Cited in Chapters 3, 7, 8 & 9.)
[377]
(1932). Uber
das Mass der Menge aller S-Zahlen. Math. Ann., 106, 131
139. (Cited in Chapters 3, 4 & 9.)
[378]
(1934/35). Uber
eine Klasseneinteilung der p-adischen Zahlen. Mathematica Leiden, 3, 177185. (Cited in Chapter 9.)
[379]
(1936). Neuer Beweis einer Satz von A. Khintchine. Mat. Sbornik, 43, 961
962. (Cited in Appendix B.)
[380]
(1937). Arithmetische Eigenschaften einer Klasse von Dezimalbruchen.
Proc. Akad. v. Wetensch., Amsterdam, 40, 421428. (Cited in Chapters 2 & 7.)
[381]
(1940). On a geometrical representation of p-adic numbers. Ann. of Math.,
[382]
(1941). An analogue to Minkowskis geometry of numbers in a eld of series. Ann. of Math., 42, 488522. (Cited in Chapter 9.)
[383]
(1949). On the continued fraction of quadratic and cubic irrationals. Ann.
Mat. Pura Appl., (4) 30, 147172. (Cited in Chapter 2.)
[384]
(1953). On the approximation of . Indag. Math., 15, 3042. (Cited in
Chapter 3.)
[385]
(1953). On the approximation of logarithms of algebraic numbers. Philos.
Trans. Roy. Soc. London. Ser. A, 245, 371398. (Cited in Chapter 3.)
[386]
(1957). On the fractional parts of the powers of a rational number, II. Mathematika, 4, 122124. (Cited in Chapter 2.)
260
[387]
[388]
[389]
[390]
[391]
[392]
[393]
[394]
[395]
[396]
[397]
[398]
[399]
[400]
[401]
[402]
[403]
[404]
[405]
References
(1960). An application to Jensens formula to polynomials. Mathematika, 7,

(1961). Lectures on Diophantine approximation, Part 1: g-adic numbers and Roths theorem. University of Notre Dame, Ann Arbor. (Cited in
Chapter 2.)
(1962). On some inequalities for polynomials in several variables. J.
London Math. Soc., 37, 341344. (Cited in Appendix A.)
(1963). On the approximation of algebraic numbers by algebraic integers.
J. Austral. Math. Soc., 3, 408434. (Cited in Chapter 2.)
(1964). An inequality for the discriminant of a polynomial. Michigan Math.
J., 11, 257262. (Cited in Appendix A.)
(1965). Arithmetic properties of lacunary power series with integral coefcients. J. Austral. Math. Soc., 5, 5664. (Cited in Chapter 7.)
(1971). On the order function of a transcendental number. Acta Arith., 18,
(1971). Lectures on transcendental numbers. 1969 Number Theory Institute
(Proc. Sympos. Pure Math., Vol. XX, State Univ. New York, Stony Brook, NY,
1969), pp. 248274. Amer. Math. Soc., Providence, RI. (Cited in Chapters 8 &
10.)
(1973). The classication of transcendental numbers. Analytic Number Theory (Proc. Sympos. Pure Math., Vol. XXIV, St. Louis Univ., St. Louis, 1972),
pp. 175179. Amer. Math. Soc., Providence, RI. (Cited in Chapter 8.)
(1975). A necessary and sufcient condition for transcendency. Math.
Comp., 29, 145153. (Cited in Chapter 8.)
(1976). On a class of transcendental decimal fractions. Comm. Pure Appl.
(1981). p-adic Numbers and their Functions, second edition. Cambridge
Tracts in Mathematics, 76, Cambridge University Press. (Cited in Chapter 9.)
(1982). Fifty years as a mathematician. J. Number Theory, 14, 121155.
(1984). Some suggestions for further research. Bull. Austral. Math. Soc.,
(1986). The successive minima in the geometry of numbers and the distinction between algebraic and transcendental numbers. J. Number Theory, 22,
Mahler, K. & Szekeres, G. (1966/67). On the approximation of real numbers
by roots of integers. Acta Arith., 12, 315320. (Cited in Chapter 3.)
Maillet, E. (1906). Introduction a` la Theorie des Nombres Transcendants et
des Proprietes Arithmetiques des Fonctions. Gauthier-Villars, Paris. (Cited in
Chapters 2, 3, 7 & 10.)
(1906). Sur la classication des irrationnelles. C. R. Acad. Sci. Paris, 143,
Margulis, G. (2002). Diophantine approximation, lattices and ows on homogeneous spaces. In: A Panorama of Number Theory or The View from Bakers
Garden, G. Wustholz (ed.), Cambridge University Press, pp. 280310. (Cited in
Chapter 4.)
References
261
[406] Markoff, A. A. (1879). Sur les formes quadratiques binaires indenies. Math.
Ann., 15, 381407. (Cited in Chapter 1.)
[407] Markovich, N. I. (1986). Approximations of zero by values of quadratic polynomials. Vests Akad. Navuk BSSR Ser. Fz. Mat. Navuk, 4, 1822, 123 (in
[408] Martin, G. (2001). Absolutely abnormal numbers. Amer. Math. Monthly, 108,
[409] Mashanov, V. I. (1987). On a problem of Baker in the metric theory of Diophantine approximations. Vests Akad. Navuk BSSR Ser. Fz. Mat. Navuk, 1,
[410] de Mathan, B. (1970). Approximations diophantiennes dans un corps local.
Bull. Soc. Math. France Suppl. Mem., 21. (Cited in Chapter 9.)
[411]
(1991). Simultaneous Diophantine approximation for algebraic functions in positive characteristic. Monatsh. Math., 111, 187193. (Cited in
Chapter 10.)
[412]
(2001). A remark about Pecks method in positive characteristic. Monatsh.
[413]
(2003). Conjecture de Littlewood et recurrences lineaires. J. Theor. Nombres Bordeaux, 15, 249266. (Cited in Chapter 10.)
[414] de Mathan, B. & Teulie, O. (To appear). Problèmes diophantiens simultanes.
Monatsh. Math.. (Cited in Chapter 10.)
[415] Mattila, P. (1995). Geometry of Sets and Measures in Euclidean Spaces. Cambridge University Press. (Cited in Chapters 3, 5, 6 & 7.)
[416] Melian, M. V. & Pestana, D. (1993). Geodesic excursions into cusps in nite
volume hyperbolic manifolds. Mich. Math. J., 40, 7793. (Cited in Chapter 9.)
[417] Melnicuk, Ju. V. (1980). Hausdorff dimension in Diophantine approximation
of p-adic numbers. Ukrain. Mat. Zh., 32, 118124, 144 (in Russian). (Cited in
Chapter 9.)
[418] Menken, H. (2000). An investigation on p-adic U-numbers. Istanbul Univ.

Fen
Fak. Mat. Derg., 59, 111143. (Cited in Chapter 9.)
[419] Mignotte, M. (1976). Sur la complexite de certains algorithmes où intervient
la separation des racines dun polynome. RAIRO Informatique Theorique, 10,
[420]
(1979). Approximation des nombres algebriques par des nombres
algebriques de grand degre. Ann. Fac. Sci. Toulouse Math., (5) 1, 165170.
(Cited in Appendix A.)
[421]
(1982). Some useful bounds. In: B. Buchberger, G. E. Collins & R. Loos
(eds.), Computer Algebra. Springer-Verlag, pp. 259263. (Cited in Chapter 7.)
[422]
(1995). On the distance between the roots of a polynomial. Applic. Alg.
Eng. Comput. Comm, 6, 327332. (Cited in Appendix A.)
[423] Mignotte, M. & Payafar, M. (1979). Distance entre les racines dun polynome.
RAIRO Anal. Numer., 13, 181192. (Cited in Appendix A.)
[424] Miller, W. (1982). Transcendence measures by a method of Mahler. J. Austral.
Math. Soc., 32, 6878. (Cited in Chapter 8.)
[425] Mkaouar, M. (2001). Continued fractions of transcendental numbers. Bull.
Greek Math. Soc., 45, 7985. (Cited in Chapter 3.)
262
References
[426] Molchanov, S. M. (1983). On the p-adic transcendence measure for the values
of functions which satisfy some functional equations. Vestnik Moskov. Univ.
Ser. I Mat. Mekh., 2, 3137 (in Russian). English transl. in Moscow Univ. Math.
Bull., 38 (1983), No.2, 3643. (Cited in Chapter 9.)
[427] Montgomery, H. M. (1994). Ten Lectures on the Interface between Analytic
Number Theory and Harmonic Analysis. American Mathematical Society, Providence, RI. (Cited in Chapters 1 & 10.)
[428] Moran, W., Pearce, C. E. M. & Pollington, A. D. (1992). T-numbers form an
M0 set. Mathematika, 39, 1824. (Cited in Chapter 7.)
[429] Mordell, L. J. (1934). On some arithmetical results in the geometry of numbers. Compositio Math., 1, 248253. (Cited in Appendix B.)
[430] Morduchai-Boltovskoj, D. (1934). On transcendental numbers with successive
approximations dened by algebraic equations. Rec. Math. Moscou, 41, 221
[431] Morotskaya, I. L. (1987). Hausdorff dimension and Diophantine approximations in Q p . Dokl. Akad. Nauk BSSR, 31, No.7, 597600, 668 (in Russian).
[432] Morrison, J. F. (1978). Approximation of p-adic numbers by algebraic
numbers of bounded degree. J. Number Theory, 10, 334350. (Cited in
Chapter 9.)
[433] Muller, P. (2002). Finiteness results for Hilberts irreducibility theorem. Ann.
Inst. Fourier (Grenoble), 52, 9831015. (Cited in Chapter 7.)
[434] Nakada, H. (1988). On metrical theory of Diophantine approximation over
imaginary quadratic eld. Acta Arith., 51, 393403. (Cited in Chapter 9.)
[435] Narkiewicz, W. (1990). Elementary and Analytic Theory of Algebraic Numbers.
Springer-Verlag, Berlin. (Cited in Chapter 2.)
[436] Nesterenko, Yu. V. (1974). An order function for almost all numbers. Mat.
Zametki, 15, 405414 (in Russian). English transl. in Math. Notes, 15 (1974),
[437] Nesterenko, Yu. V. & Waldschmidt, M. (1996). On the approximation of the
values of exponential function and logarithm by algebraic numbers. Mat. Zapiski 2, Diophantine approximations. Proceedings of papers dedicated to the
memory of Prof N.I. Feldman. Yu. V. Nesterenko (ed.), Centre for applied
research under Mech Math. Faculty of MSU, Moscow, 2342 (in Russian).
English transl. available at www.math.jussieu.fr/miw/articles/ps/Nesterenko.ps
[438] Nishioka, Ku. (1991). Algebraic independence measures of the values of
Mahler functions. J. reine angew. Math., 420, 203214. (Cited in Chapters 3,
7 & 8.)
[439]
(1996). Mahler functions and transcendence. Lecture Notes in Math., 1631,
Springer-Verlag, Berlin. (Cited in Chapters. 3, 7 & 8.)
[440] Nishioka, Ku. & Topfer, T. (1991). Transcendence measures and non-linear
functional equations of Mahler type. Arch. Math., 57, 370378. (Cited in Chapters 3 & 8.)
[441] Obreskov, N. (1950). On Diophantine approximations of linear forms for positive values of the variables. Dokl. Akad. Nauk SSSR, 73, 2124 (in Russian).
References
263
[442] Oryan, M. H. (1980). Uber

gewisse Potenzreihen, die fur algebraische Argu
mente Werte aus den Mahlerschen Unterklassen Um nehmen. Istanbul Univ
Fen Fak. Mec. Seri, 45, 142. (Cited in Chapter 7.)
[443]
(1980). Uber
die Unterklassen Um der Mahlerschen Klasseneinteilung der
transzendenten formalen Laurentreihen. Istanbul Univ

Fen Fak. Mec. Seri, 45,
[444]
(198386). Uber
gewisse Potenzreihen, deren Funktionswerte fur Argumente aus der Menge der Liouvilleschen Zahlen U-Zahlen vom Grad m sind.
Istanbul Univ
Fen Fak. Mec. Seri, 47, 1534. (Cited in Chapter 7.)
[445]
(198386). Uber
gewisse Potenzreihen, deren Funktionswerte fur Argumente aus der Menge der p-adischen Liouvilleschen Zahlen p-adische U-Zahlen
vom Grad m sind. Istanbul Univ

Fen Fak. Mec. Seri, 47, 5367. (Cited in
Chapter 9.)
[446]
(1989). On power series and Mahlers U-numbers. Math. Scand., 65,
[447]
(1990). On the power series and Liouville numbers. Doga Math., 14, 7990.
[448] Ozdemir,
A. S. (1996). On the partial generalization of the measure of transcendence of some formal Laurent series. Math. Comput. Appl., 1, 100110.
[449]
(2000). On the measure of transcendence of certain formal Laurent series.
Bull. Pure Appl. Sci. Sect. E Math. Stat., 19, 541550. (Cited in Chapter 9.)
[450] Peck, L. G. (1961). Simultaneous rational approximations to algebraic numbers. Bull. Amer. Math. Soc., 67, 197201. (Cited in Chapters 9 & 10.)
[451] Pereverzeva, N. A. (1987). A lower bound for the Hausdorff dimension of some
sets of the Euclidean plane. Vests Akad. Navuk BSSR Ser. Fz Mat. Navuk, 6,
[452]
(1989). A set-theoretic characteristic of sets in R2 that admit a given order of
approximation by algebraic numbers. Vests Akad. Navuk BSSR Ser. Fz Mat.
Navuk, 4, 3438, 124 (in Russian). (Cited in Chapter 5.)
[453] Perna, A. (1914). Sui numeri transcendenti in generale e sulla loro costruzione
in base al criterio di Liouville. Giorn. Mat. Battaglini, 52, 305365. (Cited in
Chapter 3.)
[454] Perron, O. (1929). Die Lehre von den Ketterbruchen. Teubner, Leipzig. (Cited
in Chapters 1, 2 & 7.)
[455] Petruska, G. (1992). On strong Liouville numbers. Indag. Math. (N. S.), 3,
[456] Philippon, P. (1994). Classication de Mahler et distances locales. Bull. Austral. Math. Soc., 49, 219238. (Cited in Chapter 8.)
[457]
(2000). Approximation algebrique des points dans les espaces projectifs, I.
[458] Pollington, A. D. (1979). The Hausdorff dimension of a set of non-normal well
approximable numbers. Number Theory, Carbondale (1979), Lectures Notes in
Math., 751, Springer-Verlag, Berlin, 256264. (Cited in Chapter 5.)
[459]
(1993). Sum sets and U-numbers. Number theory with an emphasis on the
Markoff spectrum. (Provo, UT, 1991), 207214, Dekker, New York. (Cited in
Chapter 7.)
264
References
[460] Pollington, A. D. & Velani, S. L. (2000). On a problem in simultaneous Diophantine approximation: Littlewoods conjecture. Acta Math., 185, 287306.
[461]
(2002). On simultaneously badly approximable numbers. J. London Math.
[462] van der Poorten, A. J. (1986). An introduction to continued fractions. Diophantine analysis. (Kensington, 1985), 99138, London Math. Soc. Lecture Note
Ser., 109, Cambridge University Press (Cited in Chapter 1.)
[463] Popken, J. (1929). Zur Transzendenz von e. Math. Z., 29, 525541. (Cited in
Chapter 3.)
[464] Queffelec, M. (1998). Transcendance des fractions continues de ThueMorse.
[465] Queffelec, M. & Ramare, O. (2003). Analyse de Fourier des fractions continues
a` quotients restreints. Enseign. Math., 49, 335356. (Cited in Chapter 1.)
[466] Raisbeck, G. (1950). Simultaneous Diophantine approximation. Canadian J .
[467] Ramharter, G. (1983). Some metrical properties of continued fractions. Mathematika, 30, 117132. (Cited in Chapter 5.)
[468]
(1994). On the fractional dimension theory of a class of expansions. Quart.
J. Math. Oxford, 45, 91102. (Cited in Chapter 5.)
[469] Ratliff, M. (1978). The ThueSiegelRothSchmidt theorem for algebraic
functions. J. Number Theory, 10, 99126. (Cited in Chapter 9.)
[470] Ridout, D. (1957). Rational approximations to algebraic numbers. Mathematika, 4, 125131. (Cited in Chapter 2.).
[471] Rieger, G.-J. (1975). Uber

die Losbarkeit von Gleichungssystemen durch Liouville Zahlen. Arch. Math., 26, 4043. (Cited in Chapter 1.)
[472] Riesz, F. & Sz.-Nagy, B. (1990). Functional Analysis. Translated from the second French edition by Leo F. Boron. Reprint of the 1955 original. Dover Publications Inc., New York. (Cited in Chapters 3 & 6.)
[473] Robert, A. M. (2000). A Course in p-adic Analysis. Graduate Texts in Mathematics, 198, Springer-Verlag, New York. (Cited in Chapter 9.)
[474] Rockett, A. M. & Szusz, P. (1992). Continued Fractions. World Scientic, Singapore. (Cited in Chapter 1.)
[475] Rogers, C. A. (1951). The signatures of the errors of simultaneous Diophantine
approximations. Proc. London Math. Soc., 52, 186190. (Cited in Chapter 1.)
[476]
(1970). Hausdorff Measures. Cambridge University Press. (Cited in
Chapter 5.)
[477] Roth, K. F. (1955). Rational approximations to algebraic numbers. Mathematika, 2, 120; corrigendum, 168. (Cited in Chapter 2.)
[478] Roy, D. (2001). Approximation algebrique simultanee de nombres de Liouville. Canad. Math. Bull., 44, 115120. (Cited in Chapter 8.)
[479]
(2003). Approximation simultanee dun nombre et son carre. C. R. Acad.
Sci. Paris, 336, 16. (Cited in Chapter 3.)
[480]
(2004). Approximation to real numbers by cubic algebraic numbers, I. Proc.
London Math. Soc., 88, 4262. (Cited in Chapters 3 & 10.)
[481]
(2003). Approximation to real numbers by cubic algebraic numbers, II. Ann.
of Math., 158, 10811087. (Cited in Chapters 3 & 10.)
References
[482]
[483]
[484]
[485]
[486]
[487]
[488]
[489]
[490]
[491]
[492]
[493]
[494]
[495]
[496]
[497]
[498]
[499]
[500]
265
(To appear). Diophantine approximation in small degree. In: Proceedings

of the seventh conference of the Canadian number theory association. (Cited in
Chapter 3.)
Roy, D. & Waldschmidt, M. (1997). Approximation diophantienne et
independance algebrique de logarithmes. Ann. Sci. Ecole

Norm. Sup., 30, 753
(To appear). Diophantine approximation by conjugate algebraic integers.
Compositio Math. (Cited in Chapters 3, 8 & 9.)
Rump, S. M. (1979). Polynomial real root separation. Math. Comp., 33, 327
Rynne, B. P. (1992). The Hausdorff dimension of certain sets arising from Diophantine approximation by restricted sequences of integer vectors. Acta Arith.,
(1992). Regular and ubiquitous systems and Ms -dense sequences. Mathematika, 39, 234243. (Cited in Chapters 5 & 6.)
(1998). The Hausdorff dimension of sets arising from Diophantine approximation with a general error function. J. Number Theory, 71, 166171. (Cited
in Chapters 5 & 6.)
(1998). Hausdorff dimension and generalised simultaneous Diophantine approximation. Bull. London Math. Soc., 30, 365376. (Cited in Chapter 5.)
(2003). Simultaneous Diophantine approximation on manifolds and
Hausdorff dimension. J. Number Theory, 98, 19. (Cited in Chapter 5.)
Rynne, B. P. & Dickinson, H. (2000). Hausdorff dimension and a generalized form of simultaneous Diophantine approximation. Acta Arith., 93, 2136.
Sakovich, N. V. (1995). The Hausdorff dimension and the distribution of values of integral polynomials over C. Vests Akad. Navuk Belarus Ser. Fz Mat.
Navuk, 2, 3946, 124 (in Russian). (Cited in Chapter 9.)
Saradha, N. & Tijdeman, R. (2003). On the transcendence of innite sums
of values of rational functions. J. London Math. Soc., 67, 580592. (Cited in
Chapter 3.)
Schikhof, W. H. (1984). Ultrametric Calculus: An Introduction to p-adic Analysis. Cambridge Studies in Advanced Mathematics 4, Cambridge University
Press. (Cited in Chapter 9.)
Schlickewei, H.-P. (1976). On products of special linear forms with algebraic
coefcients. Acta Arith., 31, 389398. (Cited in Chapter 9.)
(1981). p-adic T-numbers do exist. Acta Arith., 39, 181191. (Cited in
Chapter 9.)
Schmeling, J. & Troubetzkoy, S. (2003). Inhomogeneous Diophantine approximation and angular recurrence for polygonal billiards. Mat. Sbornik, 194, 295
Schmidt, A. L. (1974). On the approximation of quaternions. Math. Scand.,
(1975). Diophantine approximation of complex numbers. Acta Math., 134,
Schmidt, W. M. (1960). A metrical theorem in Diophantine approximation.
Canad. J. Math., 12, 619631. (Cited in Chapter 1.)
266
[501]
[502]
[503]
[504]
[505]
[506]
[507]
[508]
[509]
[510]
[511]
[512]
[513]
[514]
[515]
[516]
[517]
[518]
[519]
[520]
References
(1961). Bounds for certain sums; a remark on a conjecture of Mahler. Trans.

(1964). Metrical theorems on fractional parts of sequences. Trans. Amer.
Math. Soc., 110, 493518. (Cited in Chapter 1.)
(1966). On badly approximable numbers and certain games. Trans. Amer.
Math. Soc., 123, 178199. (Cited in Chapters 5 & 9.)
(1967). On simultaneous approximations of two algebraic numbers by rationals. Acta Math., 119, 2750. (Cited in Chapter 2.)
(1969). Badly approximable systems of linear forms. J. Number Theory, 1,
(1970). Simultaneous approximations to algebraic numbers by rationals.
Acta Math., 125, 189201. (Cited in Chapter 2.)
(1970). T-numbers do exist. Symposia Math. IV, Inst. Naz. di Alta Math.,
Rome 1968, Academic Press, pp. 326. (Cited in Chapters 1, 3, 7 & 9.)
(1971). Mahlers T-numbers. 1969 Number Theory Institute, (Proc. of
Symposia in Pure Math., Vol. XX, State Univ. New York, Stony Brook, NY,
1969), pp. 275286, Amer. Math. Soc., Providence, RI. (Cited in Chapters 3,
7 & 9.)
(1971). Diophantine approximation and certain sequences of lattices. Acta
(1971). Approximation to algebraic numbers. Monographie de
lEnseignement Mathematique, 19, Genève. (Cited in Introduction and
Chapter 2.)
(1976). Two questions in Diophantine approximation. Monatsh. Math., 82,
(1980). Diophantine approximation. Lecture Notes in Math., 785, Springer,
Berlin. (Cited in Introduction, in Chapters 1, 2, 3, 5, & 10, and in Appendices
A & B.)
(1983). Open problems in Diophantine approximation. In: Approximations
Diophantiennes et Nombres Transcendants. Luminy, 1982. Progress in Mathematics, Birkhauser. (Cited in Chapter 10.)
(1983). Diophantine approximation properties of certain innite sets. Trans.
(2000). On continued fractions and Diophantine approximation in
power series elds. Acta Arith., 95, 139166. (Cited in Chapters 9
& 10.)
Schmidt, W. M. & Wang, Y. (1979). A note on a transference theorem of linear
forms. Sci. Sinica, 22, 276280. (Cited in Chapter 4 and Appendix B.)
Schneider, T. (1957). Einfuhrung in die Transzendenten Zahlen. Springer-Verlag,
BerlinGottingenHeidelberg. (Cited in Introduction, and Chapters 3 & 9.)
Schwarz, W. (1977). Liouville Zahlen und der Satz von Baire. Math. Phys.
Semesterber., 24, 8487. (Cited in Chapter 1.)
Schweiger, F. (1995). Ergodic Theory of Fibred Systems and Metric Number
Theory. Oxford Science Publications, The Clarendon Press, Oxford University
Press, New York. (Cited in Chapter 1.)
Serret, J.-A. (1877). Cours dAlgèbre superieure, quatrième edition. GauthiersVillars, Paris. (Cited in Chapter 1.)
References
267
[521] Shallit, J. (1992). Real numbers with bounded partial quotients: a survey. Enseign. Math., 38, 151187. (Cited in Chapter 10.)
[522] Shidlovskii, A. B. (1989). Transcendental Numbers. Translated from the Russian by Neal Koblitz, de Gruyter Studies in Mathematics 12, Walter de Gruyter
& Co., Berlin. (Cited in Chapters 3 & 8.)
[523] Shorey, T. N. (1983). Divisors of convergents of a continued fraction. J. Number Theory, 17, 127133. (Cited in Chapter 2.)
[524] Siegel, C. L. (1921). Approximation algebraischer Zahlen. Math. Z., 10, 173
[525] Slater, N. B. (1967). Gaps and steps for the sequence n mod 1. Proc. Cambridge Philos. Soc., 63, 11151123. (Cited in Chapter 5.)
[526] Slesoratene, R. (1973). An analogue of a theorem of Mahler and Sprindzuk
for polynomials of degree two in two p-adic variables. Litovsk. Mat. Sb., 13,
177188 (in Russian). English transl. in Lithuanian Math. J., 13 (1973), 291
[527]
(1987). An analogue of the MahlerSprindzuk theorem for polynomials of fourth degree in two variables. Litovsk. Mat. Sb., 27, 165171 (in
Russian). English transl. in Lithuanian Math. J., 27 (1987), 9499. (Cited in
Chapter 4.)
[528] Sorokin, V. N. (1991). HermitePade approximations of sequential powers of
a logarithm and their arithmetic applications. Izv. Vyssh. Uchebn. Zaved. Mat.,
No. 11, 6674 (in Russian). English transl. in Soviet Math. (Iz. VUZ), 35 (1991),
No. 11, 6774. (Cited in Chapter 3.)
[529] Sos, V. T. (1957). On the theory of Diophantine approximations, I. Acta Math.
Hung., 8, 461472. (Cited in Chapter 5.)
[530] Sprindzuk, V. G. (1962). On some general problems of approximating numbers
by algebraic numbers. Litovsk. Mat. Sb., 2, 129145 (in Russian). (Cited in
Chapter 3.)
[531]
(1962). On theorems of Khintchine and Kubilius. Litovsk. Mat. Sb., 2, 147
[532]
(1962). On a classication of transcendental numbers. Litovsk. Mat. Sb., 2,
[533]
(1962). On a conjecture of K. Mahler concerning the measure of the
set of S-numbers. Litovsk. Mat. Sb., 2, 221226 (in Russian). (Cited in
Chapter 4.)
[534]
(1963). On algebraic approximations in the eld of power series. Vestnik
Leningrad Univ. Ser. Mat. Meh. Astronom., 18, 130134 (in Russian). (Cited in
Chapter 9.)
[535]
(1963). On the measure of the set of S-numbers in the p-adic eld. Dokl.
Akad. Nauk SSSR., 151, 1292 (in Russian). English transl. in Soviet Math. Dokl.,
4 (1963), 12011202. (Cited in Chapter 9.)
[536]
(1964). On Mahlers conjecture. Dokl. Akad. Nauk SSSR., 154, 783786 (in
Russian). English transl. in Soviet Math. Dokl., 5 (1964), 183187. (Cited in
Chapter 4.)
[537]
(1964). More on Mahlers conjecture. Dokl. Akad. Nauk SSSR., 155, 5456.
(in Russian). English transl. in Soviet Math. Dokl., 5 (1964), 361363. (Cited in
Chapter 4.)
268
[538]
[539]
[540]
[541]
[542]
[543]
[544]
[545]
[546]
[547]
[548]
[549]
[550]
[551]
[552]
[553]
[554]
[555]
References
(1965). A proof of Mahlers conjecture on the measure of the set of

S-numbers. Izv. Akad. Nauk SSSR. Ser. Mat., 29, 379436 (in Russian). English
transl. in Amer. Math. Soc. Transl., 51 (1966), 215272. (Cited in Chapters 3, 4
& 9.)
(1967). Mahlers Problem in Metric Number Theory. Izdat. Nauka i
Tehnika, Minsk, (in Russian). English translation by B. Volkmann, Translations of Mathematical Monographs, 25, American Mathematical Society, Providence, RI, 1969. (Cited in Introduction, and Chapters 3, 4, 6 & 9.)
(1977). Metric Theory of Diophantine Approximation. Izdat. Nauka,
Moscow, (in Russian). English translation by R. A. Silverman, Scripta Series
in Mathematics, John Wiley & Sons, 1979. (Cited in Introduction, and Chapters
1 & 4.)
(1980). Achievements and problems of the theory of Diophantine approximations. Uspekhi Mat. Nauk, 35, 368, 248 (in Russian). English transl. in
Russ. Math. Survey, 35 (1980), 180. (Cited in Chapter 9.)
Starkov, A. N. (2000). Dynamical systems on homogeneous spaces. Translations of Mathematical Monographs, 190, American Mathematical Society,
Providence, RI. (Cited in Chapter 4.)
Steinig, J. (1992). A proof of Lagranges theorem on periodic continued fractions. Arch. Math. (Basel), 59, 2123. (Cited in Chapter 1.)
Stewart, C. L. (1984). A note on the product of consecutive integers. Topics in
classical number theory, Vol. I, II (Budapest, 1981), 15231537, Colloq. Math.
Soc., Janos Bolyai 34, North-Holland, Amsterdam. (Cited in Chapter 2.)
Stokolos, A. M. (2001). Some applications of Gallaghers theorem in harmonic
analysis. Bull. London Math. Soc., 33, 210212. (Cited in Chapter 4.)
Stolarsky, K. B. (1974). Algebraic numbers and Diophantine approximation.
Pure and Applied Mathematics, No. 26. Marcel Dekker Inc., New York. (Cited
in Chapter 3.)
Sullivan, D. (1982). Disjoint spheres, approximation by imaginary numbers,
and the logarithm law for geodesics. Acta Math., 149, 215237. (Cited in Chapters 1 & 9.)
Teulie, O. (2001). Approximation de nombres reels et p-adiques par des nombres algebriques. Thèse, Universite de Bordeaux. (Cited in Chapter 9.)
(2001). Approximation dun nombre reel par des unites algebriques.
Monatsh. Math., 133, 169176. (Cited in Chapters 3 & 10.)
(2002). Approximation dun nombre p-adique par des nombres algebriques.
Acta Arith., 102, 137155. (Cited in Chapter 9.)
(Preprint). Approximation dun nombre p-adique par des nombres quadratiques. (Cited in Chapter 9.)
(2002). Approximations simultanees de nombres algebriques de Q p par des
rationnels. Monatsh. Math., 137, 313324. (Cited in Chapter 9.)
Thue, A. (1909). Uber

Annaherungswerte algebraischer Zahlen. J. reine
angew. Math., 135, 284305. (Cited in Chapter 2.)
Thurnheer, P. (1981). Un rafnement du theorème de Dirichlet sur
lapproximation diophantienne. C. R. Acad. Sci. Paris Ser. I Math., 293, 623
(1982). Eine Verscharfung des Satzes von Dirichlet u ber diophantische Approximation. Comment. Math. Helv., 57, 6078. (Cited in Chapters 1 & 10.)
References
[556]
[557]
[558]
[559]
[560]
[561]
[562]
[563]
[564]
[565]
[566]
[567]
[568]
[569]
[570]
[571]
[572]
[573]
269
(1984). Zur diophantischen Approximation von zwei reellen Zahlen. Acta

(1985). Approximation diophantienne par certains couples dentiers.
C. R. Math. Rep. Acad. Sci. Canada, 7, 5153. (Cited in Chapters
1 & 10.)
(1990). On Dirichlets theorem concerning Diophantine approximation.
Acta Arith., 54, 241250. (Cited in Chapters 1 & 10.)
Tichy, R. F. (1979). Zum Approximationssatz von Dirichlet. Monatsh. Math.,
Tishchenko, K. I. (1996). Systems of linearly independent polynomials with
height gaps and the Wirsing problem. Vests Nats. Akad. Navuk Belarus Ser.
Fz Mat. Navuk, 4, 1622, 140 (in Russian). (Cited in Chapter 3.)
(2000). On approximation to real numbers by algebraic numbers. Acta
(2000). On the approximation of complex numbers by algebraic numbers of
bounded degree. Dokl. Nats. Akad. Nauk Belarus, 44, No.3, 2528 (in Russian). (Cited in Chapter 9.)
(2000). On the application of linearly independent polynomials in the Wirsing problem. Dokl. Nats. Akad. Nauk Belarusi, 44, No.5, 3436 (in Russian).
(2000). On some special cases of the Wirsing conjecture. Vests Nats. Akad.
Chapter 3.)
(2000). On new methods in the problem of the approximation of real
numbers by algebraic numbers of degree at most three. Vests Nats. Akad.
Chapter 3.)
(2001). On the problem of the approximation of p-adic numbers by algebraic
numbers. Vests Nats. Akad. Navuk Belarus Ser. Fz. Mat. Navuk, 4, 129130,
(2001). On the approximation of real numbers by algebraic integers of the
third degree. Dokl. Nats. Akad. Nauk Belarus, 45, No.1, 1719 (in Russian).
(2002). On simultaneous approximation of two real numbers by real numbers. Dokl. Nats. Akad. Nauk Belarus, 46, No.2, 2933 (in Russian). (Cited in
Chapter 3.)
Topfer, T. (1995). Algebraic independence of the values of generalized Mahler
functions. Acta Arith., 70, 161181. (Cited in Chapters 3 & 8.)
(1995). An axiomatization of Nesterenkos method and applications on
Mahler functions, II. Compositio Math., 95, 323342. (Cited in Chapters 3 &
8.)
Tricot, C. (1980). Rarefaction indices. Mathematika, 27, 4657. (Cited in
Chapter 5.)
(1982). Two denitions of fractional dimension. Math. Proc. Cambridge
Philos. Soc., 91, 5474. (Cited in Chapter 5.)
Turkstra, H. (1936). Metrische bijdragen tot de theorie der Diophantische approximaties in het lichaam der p-adische getallen. Dissertation, Vrije Univ.,
Amsterdam. (Cited in Chapter 9.)
270
References
[574] Vaa nanen, K. (1989). Application of HermiteMahler polynomials to the approximation of the values of p-adic exponential function. Monatsh. Math., 108,
[575] Vahlen, K. Th. (1895). Uber

Naherungswerte und Kettenbruche. J. reine
angew. Math., 115, 221233. (Cited in Chapter 1.)
[576] Vallee, B. (1998). Dynamique des fractions continues a` contraintes
periodiques. J. Number Theory, 72, 183235. (Cited in Chapter 5.)
[577] Vilchinski, V. T. (1990). On the metric theory of non-linear Diophantine approximation. Dokl. Akad. Nauk BSSR, 34, No.8, 677680 (in Russian). (Cited
in Chapter 5.)
[578] Vilchinskii, V. T. & Dombrovski, I. R. (1990). The Hausdorff dimension and
estimates for trigonometric sums over prime numbers. Vests Akad. Navuk
BSSR Ser. Fz. Mat. Navuk, 1, 36, 123 (in Russian). (Cited in Chapter 5.)
[579] Vinogradov, A. I. & Chudnovsky, G. V. (1984). The proof of extremality of
certain manifolds. Appendix in: G.V. Chudnovsky Contributions to the Theory of Transcendental Numbers. Translated from the Russian by J. A. Kandall,
Math. Surveys Monographs, 19, Amer. Math. Soc., Providence, RI. (Cited in
Chapter 4.)
[580] Viola, C. (1979). Diophantine approximation in short intervals. Ann. Scuola
Norm. Sup. Pisa Cl. Sci., (4) 6, 703717. (Cited in Chapter 1.)
[581] Volkmann, B. (1960). Zur Mahlerschen Vermutung im Komplexen. Math.
Ann., 140, 351359. (Cited in Chapter 9.)
[582]
(1961). The real cubic case of Mahlers conjecture. Mathematika, 8, 5557.
[583]
(1962). Zur metrischen Theorie der S-Zahlen. J. reine angew. Math., 209,
[584]
(1963/64) Zur metrischen Theorie der S-Zahlen, II. J. reine angew. Math.,
[585] van der Waerden, B. L. (1953). Modern Algebra, Vol. I, Revised English edition.
Ungar. (Cited in Chapter 7.)
[586]
(1991). Algebra, Vol. I. Springer-Verlag, New York (Cited in Chapter 3 and
Appendix A.)
[587] Waldschmidt, M. (1971). Independance algebrique des valeurs de la fonction
exponentielle. Bull. Soc. Math. France, 99, 285304. (Cited in Chapter 8.)
[588]
(1974). Nombres transcendants. Lecture Notes in Mathematics, 402,
Springer-Verlag, BerlinNew York. (Cited in Chapter 8.)
[589]
(1978). Transcendence measures for exponentials and logarithms. J .
Austral. Math. Soc. (Series A), 25, 445465. (Cited in Chapter 8.)
[590]
(2000). Conjectures for large transcendence degree. Algebraic Number Theory and Diophantine Analysis (Graz, 1998), 497520, de Gruyter, Berlin. (Cited
in Chapter 8.)
[591]
(2000). Diophantine Approximation on Linear Algebraic Groups. Transcendence properties of the exponential function in several variables. Grundlehren
der Mathematischen Wissenschaften, 326, Springer-Verlag, Berlin. (Cited in
Chapters 2 & 8, and Appendix A.)
[592]
(2004). Open Diophantine Problems. Moscow Math. J., 4. (Cited in Chapter
10.)
References
271
[593] Wang, Y. & Yu, K. (1981). A note on some metrical theorems in Diophantine
approximation. Chinese Ann. Math., 2, 112. (Cited in Chapter 4.)
[594] Wang, Y., Yu, K. & Zhu, Y. C. (1979). A transfer theorem for linear forms. Acta
Math. Sinica, 22, 237240 (in Chinese). (Cited in Chapter 4 and Appendix B.)
[595] Wass, N. Ch. (1990). Algebraic independence of the values at algebraic points
of a class of functions considered by Mahler. Dissertationes Math. (Rozprawy
Mat.), 303, 61 pp. (Cited in Chapters 3 & 8.)
[596] Weiss, B. (2001). Almost no points on a Cantor set are very well approximable.
R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci., 457, 949952. (Cited in Chapter 5.)
[597]
(2002). Dynamics on parameter spaces: submanifold and fractal subset
questions. Rigidity in Dynamics and Geometry (Cambridge, 2000), 425440,
Springer, Berlin. (Cited in Chapter 5.)
[598] Wirsing, E. (1961). Approximation mit algebraischen Zahlen beschrankten
Grades. J. reine angew. Math., 206, 6777. (Cited in Chapters 2, 3, 8, 9 &
10, and Appendix A.)
[599]
(1971). On approximations of algebraic numbers by algebraic numbers of
bounded degree. 1969 Number Theory Institute (Proc. of Symposia in Pure
Math., Vol. XX, State Univ. New York, Stony Brook, NY, 1969), pp. 213247,
Amer. Math. Soc., Providence, RI. (Cited in Chapter 2.)
[600] Wolfskill, J. (1983). A growth bound on the partial quotients of cubic numbers.
J. reine angew. Math., 346, 129140. (Cited in Chapter 2.)
[601] Worley, R. T. (1981). Estimating | p/q|. J. Austral. Math. Soc. (Series A),
[602] Xin, X. L. (1989). Mahlers classication of p-adic numbers. Pure Appl.
Math., 5, 7380 (in Chinese). (Cited in Chapter 9.)
[603] Xu, G. S. (1986). On the arithmetic properties of the values of p-adic Efunctions. Acta Math. Sinica, 29, 444453 (in Chinese). (Cited in Chapter 9.)
[604]
(1988). Diophantine approximation and transcendental number theory.
Number Theory and its Applications in China, 127142, Contemp. Math., 77,
Amer. Math. Soc., Providence, RI. (Cited in Chapter 4 and Appendix B.)
[605] Yanchenko, A. Ya. (1982). The measure of mutual transcendence of almost all
pairs of p-adic numbers. Vestnik Moskov. Univ. Ser. I Mat. Mekh., 3, 6265,
111 (in Russian). English transl. in Moscow Univ. Math. Bull., 37, No.3, 7174.
[606] Yilmaz, G. (1996/97). Arithmetical properties of the values of some power
series with algebraic coefcients taken for Um -numbers arguments. Istanbul
Univ.
Fen Fak. Mat. Derg., 55/56, 111144. (Cited in Chapters 3, 7 & 9.)
[607] Yu, K. (1981). Hausdorff dimension and simultaneous rational approximation.
J. London Math. Soc., 24, 7984. (Cited in Chapter 5.)
[608]
(1981). A note on a problem of Baker in metrical number theory. Math.
Proc. Cambridge Philos. Soc., 90, 215227. (Cited in Chapters 4, 5 & 9.)
[609]
(1987). A generalization of Mahlers classication to several variables. J.
reine angew. Math., 377, 113126. (Cited in Chapter 3.)
[610] Zeren, B. M. (1980). Uber

einige komplexe und p-adische Luckenreihen mit
Werten aus den Mahlerschen Unterklassen Um . Istanbul Univ.

Fen Fak. Mec.
Seri., 45, 89130. (Cited in Chapters 7 & 9.)
272
[611]
[612]
[613]
[614]
[615]
[616]
[617]
References
(1991). Uber
eine Klasse von verallgemeinerten Luckenreihen, deren Werte
fur algebraische Argumente transzendent, aber keine U-Zahlen sind, I, II, III.
Istanbul Univ.
Fen Fak. Mat. Der., 50, 7999; 101114; 147158. (Cited in
Chapters 3 & 7.)
Zheludevich, F. F. (1983). Simultaneous approximations of numbers by roots of
integral polynomials. Vests Akad. Navuk BSSR Ser. Fz. Mat. Navuk, 4, 1418
(1984). The simultaneous approximation of numbers by the roots of integer polynomials. Vests Akad. Navuk BSSR Ser. Fz. Mat. Navuk, 2, 1420 (in
(1986). Simultane diophantische Approximationen abhangiger Grossen in
mehreren Metriken. Acta Arith., 45, 8798. (Cited in Chapter 9.)
(1987). Approximations of S-adèles by algebraic numbers of bounded
degree. Vests Akad. Navuk BSSR Ser. Fz. Mat. Navuk, 6, 1922, 124
Zhu, Y. C. (1981). On a problem of Mahler concerning the transcendence of
a decimal fraction. Acta Math. Sinica, 24, 246253 (in Chinese). (Cited in
Chapter 2.)
Zong, C. M. (1994). Analogues of K. Mahlers theory. Acta Math. Sinica
(N. S.), 10, Special Issue, 141154. (Cited in Chapter 9.)
Index
badly approximable number, 11, 11, 16, 22,

25, 26, 71, 104, 114, 119, 120, 124, 137,
165, 193, 201, 211213, 216
badly approximable system of linear forms,
120
BorelCantelli Lemma, 13, 125
A-number, 42
A -number, 48
A-number,
167
M0 set, 165
S-number, 42, 71, 73, 75, 111, 112, 152,
161, 198, 202, 203, 207, 208
S -number, 48, 105, 109, 110, 136, 206
s-set, 93, 98, 119, 135
T -number, 43, 71, 72, 94, 140, 151, 152,
163165, 193, 199, 207, 213
T -number, 48
U -number, 43, 71, 72, 94, 141, 153, 165,
184, 202, 203
U -number, 48
Um -number, 153, 153, 154, 155, 158, 164,
165, 199, 200
H f -measure, 91, 152, 154
-covering, 91
S-number,
167, 188
T -number, 167
U -number, 167
Champernowne number, 39, 71, 165

classication of numbers, 41, 171, 188, 191,
194, 200
continued fraction, 5, 72, 158, 161, 163, 164,
193, 201
convergent, 7, 40, 113
criterion of irrationality, 3, 28
criterion of transcendence, 29, 189
dimension function, 91
Dirichlets Theorem, 1, 22, 23, 67, 121, 201
discrepancy, 103
discriminant, 223
distance function, 118
DufnSchaeffer conjecture, 18, 76, 119, 214
absolute height, 181

algebraic integer, 35, 40, 62, 64, 65, 67,
73, 124, 201, 209, 233
algebraic number, 3
algebraic unit, 64, 124, 209, 209
approximable at order , 12
approximation function, 12, 101, 138
approximation of a linear form, 23, 119,
203
approximation of a system of linear forms, 23,
24, 118, 119, 138, 201, 202, 239
asymptotic formula, 24, 25, 88, 89
Fibonacci word, 66
Fourier-Stieltjes transform, 118, 165
Frostman Lemma, 96
full, 12
badly approximable n-tuple, 24, 72

badly approximable linear form, 203
Hausdorff dimension, 92, 117, 170, 174, 198,

218
exact logarithmic order, 134

extremal curve, 88
Gelfonds criterion, 188

Gelfonds Lemma, 221
Golden Section, 20, 66
273
274
Index
HausdorffCantelli Lemma, 95
height, 31, 219
order function, 172

order of approximation, 134, 138, 212
inhomogeneous approximation, 25, 62, 68, 89,

103, 118, 120, 136, 138, 202
intersective sets, 130
irrationality measure, 27
partial quotient, 7, 40, 164, 214
Khintchines Theorem, 12, 24, 25, 132, 193,

202, 203
Khintchines Transference Theorem, 64, 65,
201, 208, 237
Koksmas classication, 47
Lagrange spectrum, 20
length, 171
Liouville number, 4, 22, 25, 26, 44, 51, 73,
120, 141, 154, 164, 175177, 190, 202,
203, 215
Liouvilles inequality, 32, 39, 152, 154, 155,
159, 222, 234
Littlewoods Conjecture, 165, 204, 217,
218
lower order at innity, 129
Mahler measure, 219
Mahlers classication, 42, 73
Main Problem, 206
mass distribution principle, 96
measure of algebraic approximation, 180
minimal polynomial, 220
multidimensional approximation, 188
multidimensional problem, 73, 118
multiplicative approximation, 77, 88, 89, 112,
118, 201, 202, 213
regular system, 98, 103, 119, 193

restricted approximation, 2426, 89, 113, 119,
120, 216
resultant, 223
Ridouts Theorem, 29, 40, 120, 215
Salem set, 118
Schmidts Subspace Theorem, 32, 39, 141,
162, 192, 198
Schubfachprinzip, 2
semi-discriminant, 229
separable polynomial, 223
simultaneous approximation, 2325, 62,
64, 72, 119121, 138, 190, 201,
202, 204, 216
size, 171
Sprindzuks classication, 167
strongly extremal curve, 88
Sturmian word, 39
successive minima, 236
Three Distance Theorem, 117, 137
ThueMorse word, 39
transcendence measure, 180
transcendence type, 174
transcendental number, 3
transference theorem, 77, 202
triadic Cantor set, 98, 119, 165, 213
type, 43, 50, 167
ubiquitous system, 118, 138
nave height, 219

normal number, 25, 119, 213
null, 12
optimal regular system, 123, 124, 137, 198
order, 167
very well approximable, 118

very well approximable number, 13,
104, 213
Wirsings Conjecture, 206

Approximation by Algebraic Numbers PDF

Uploaded by

Copyright:

Available Formats

Approximation by Algebraic Numbers PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Approximation by Algebraic Numbers PDF

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

CAMBRIDGE TRACTS IN MATHEMATICS

W. FULTON, A. KATOK, F. KIRWAN, P. SARNAK, B. SIMON

Approximation by Algebraic Numbers

Approximation by Algebraic Numbers

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, So Paulo

978-0-511-26591-4 eBook (NetLibrary)

Approximation by rational numbers

Approximation to algebraic numbers

The classications of Mahler and Koksma

Mahlers Conjecture on S-numbers

Hausdorff dimension of exceptional sets

Deeper results on the measure of exceptional sets

On T-numbers and U -numbers

7.8 Brief summary of the results towards the Main Problem

Other classications of real and complex numbers

Approximation in other elds

Conjectures and open questions

Appendix A Lemmas on polynomials

Maurice Mignotte and Michel Waldschmidt encouraged me constantly.

The following statements are not proved in the present book:

Frequently used notation

, means that there is an implied constant.

Frequently used notation

x the n-tuple (x1 , . . . , xn ).

A-, S-, T -, U -numbers Ch. 8.

1.1 Dirichlet and Liouville

Approximation by rational numbers

We conclude by setting p := m  m k and q :=  k, and by noticing that q

1.1 Dirichlet and Liouville

Approximation by rational numbers

Hence, we have |t | 1 and

P ROOF. Since its decimal expansion is not ultimately periodic, is irrational.

and is not algebraic of degree greater than or equal to 2, by Theorem 1.2.

for innitely many rationals p/q.

1.2 Continued fractions

Approximation by rational numbers

Since a1 1 and r is an integer, we deduce that n = 1 and a1 = 1, thus

We then have = a0 + 1/1 . For any positive integer n, we dene inductively

and n+1 = 1/{n },

1.2 Continued fractions

we have, for any positive integer n,

P ROOF. We proceed by induction. Since p1 /q1 = a0 + 1/a1 = (a0 a1 +

It follows from (1.4) with j = n + 1 and the inductive hypothesis applied to

Approximation by rational numbers

whence, by (1.4) with j = n and j = n 1, we get qn+1 = an+1 qn + qn1

qn pn2 pn qn2 = (1)n1 an .

and, for all n 1,

1.2 Continued fractions

Approximation by rational numbers

Furthermore, the set of real numbers having a continued fraction expansion

1.2 Continued fractions

Approximation by rational numbers

Conversely, if the partial quotients of are not greater than a constant M,

1.3 The theorem of Khintchine

1.3 The theorem of Khintchine

An alternative proof of Theorem 1.10, valid under slightly less restrictive

Approximation by rational numbers

We conclude by setting p := m m k and q := k, and by noticing that q

Lebesgue measure zero. In particular, taking for the functions x x 21/k ,

large enough in terms of . Since is non-increasing, we get that, for almost

where denotes the function distance to the nearest integer. Clearly, ( ) =

for any algebraic number = of