Professional Documents
Culture Documents
Textbook Auxiliary Polynomials in Number Theory 1St Edition Masser Ebook All Chapter PDF
Textbook Auxiliary Polynomials in Number Theory 1St Edition Masser Ebook All Chapter PDF
https://textbookfull.com/product/number-theory-1st-edition-
robert-freud/
https://textbookfull.com/product/invitation-to-number-theory-1st-
edition-ore/
https://textbookfull.com/product/statistical-independence-in-
probability-analysis-number-theory-kac/
https://textbookfull.com/product/quaternion-algebras-in-number-
theory-spring-2016-kimball-martin/
Algebraic Number Theory - A Brief Introduction 1st
Edition J.S. Chahal
https://textbookfull.com/product/algebraic-number-theory-a-brief-
introduction-1st-edition-j-s-chahal/
https://textbookfull.com/product/invitation-to-number-theory-2nd-
edition-oystein-ore/
https://textbookfull.com/product/number-theory-revealed-an-
introduction-andrew-granville/
https://textbookfull.com/product/research-directions-in-number-
theory-women-in-numbers-iv-jennifer-s-balakrishnan/
https://textbookfull.com/product/anomalies-in-net-present-value-
returns-and-polynomials-and-regret-theory-in-decision-making-1st-
edition-michael-c-i-nwogugu-auth/
C A M B R I D G E T R AC T S I N M AT H E M AT I C S
General Editors
GENERAL EDITORS
B. BOLLOBÁS, W. FULTON, A. KATOK, F. KIRWAN,
P. SARNAK, B. SIMON, B. TOTARO
DAV I D M A S S E R
University of Basle, Switzerland
University Printing House, Cambridge CB2 8BS, United Kingdom
www.cambridge.org
Information on this title: www.cambridge.org/9781107061576
© David Masser 2016
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2016
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloguing in Publication Data
Names: Masser, David William, 1948–
Title: Auxiliary polynomials in number theory / David Masser, Universitat Basel, Switzerland.
Description: Cambridge : Cambridge University Press, 2016. | Series: Cambridge tracts
in mathematics ; 207 | Includes bibliographical references and index.
Identifiers: LCCN 2015050947 | ISBN 9781107061576 (Hardback : alk. paper)
Subjects: LCSH: Number theory. | Polynomials.
Classification: LCC QA241 .M395 2016 | DDC 512.7/4–dc23 LC record
available at http://lccn.loc.gov/2015050947
ISBN 978-1-107-06157-6 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-party Internet Web sites referred to in this publication
and does not guarantee that any content on such Web sites is, or will remain,
accurate or appropriate.
Contents
1 Prologue 1
2 Irrationality I 7
3 Irrationality II – Mahler’s Method 20
4 Diophantine equations – Runge’s Method 30
5 Irreducibility 50
6 Elliptic curves – Stepanov’s Method 64
7 Exponential sums 76
8 Irrationality measures I – Mahler 88
9 Integer-valued entire functions I – Pólya 101
10 Integer-valued entire functions II – Gramain 111
11 Transcendence I – Mahler 123
12 Irrationality measures II – Thue 133
13 Transcendence II – Hermite–Lindemann 158
14 Heights 166
15 Equidistribution – Bilu 193
16 Height lower bounds – Dobrowolski 200
17 Height upper bounds 212
18 Counting – Bombieri–Pila 218
19 Transcendence III – Gelfond–Schneider–Lang 228
v
vi Contents
References 334
Index 342
Introduction
Ever since it was invented, arguably by Runge, the method of auxiliary poly-
nomials has been vital to (and of unreasonable effectiveness in) the modern
development of key aspects of number theory. The aim of this book is to give
an account of the method in many of its forms, focusing almost exclusively on
those polynomials which cannot be written down explicitly.
I well remember (standing in Heffers bookshop Cambridge around 1970)
reading about this method in the foreword to Lang’s book on transcendental
numbers, and experiencing disbelief that anything so far-fetched could work at
all. So I will not attempt any explanation at this point.
Instead, I (or from now on, the authorial we) treat the method as the union
of its examples, and there is no shortage of these.
Here is the plan of this book (Mike Tyson said that everyone has a plan until
you punch them in the face – then they don’t have a plan). The general strategy
is to present in each chapter an application of the method to a different sort of
problem, often the simplest in its area. Then at the end of each chapter we give
a brief account of subsequent developments in the area.
We start with a short Prologue (Chapter 1) where we show that the basic idea
can be used in rather simple situations which have nothing to do with number
theory.
Then in Chapter 2 we commence our diophantine considerations with a
discussion of irrationality. We quickly dispose of the number e by the standard
truncation argument and we show also that e is not a quadratic irrational.
Here we meet a small problem, which can be rather quickly solved; how-
ever, it is typical of the problems that arise in later applications and in some
examples its solution can be distinctly non-trivial. Thus Roth, in showing that
irrational algebraic numbers cannot be approximated to within an order of
q−2− by rationals p/q, had to solve such a problem. The solution is called
vii
viii Introduction
Roth’s Lemma, and it was certainly one of the achievements that gained him
a Fields Medal. We do not prove Roth’s Theorem here but we do treat Thue’s
Method in Chapter 12.
We postpone to Chapter 13 a proof that
∞
αk 1 1 1
eα = = 1 + α + α2 + α3 + α 4 + · · ·
k! 2 6 24
k=0
is irrational and even transcendental for all rational and even algebraic α = 0,
as this requires elements of algebraic number theory. Truncation gives only
irrationality
√ and that essentially only for α = 1, 2, 4 (and slightly surprisingly
α = 2).
In the same Chapter 2 we construct our first auxiliary polynomial with a
diophantine purpose: namely, to show that for any rational α = 0 the classical
series
∞
αk 1 1 1
= 1 + α + α2 + α3 + α4 + · · ·
2k(k−1)/2 2 8 64
k=0
for every rational α with 0 < |α| < 1. The proof is quite similar to that
in Chapter 2 but a little more analysis is needed, and further the auxiliary
polynomial is more complicated, in fact already of a fairly typical sort; still we
calculate some more examples. Mahler’s Method has been greatly developed
and some recent applications refer to the famous Mandelbrot set. It also
√ played
a transient role in proving that the decimal digits of numbers like 2 cannot
be generated by a finite automaton.
Introduction ix
proof with the help of some easier intermediate results. Not even the definition
of elliptic curve is needed here, let alone any properties. Thus all we do is count
the integer solutions (x, y) modulo p of an equation y2 = x3 + ax2 + bx + c
modulo p. There are many generalizations (and Schmidt wrote an entire book
about them) but none has quite the same appeal. With rather little extra effort
one can treat y2 = x5 + · · · and worse; in the geometric context this is far from
simple because it would involve curves of genus 2 and worse.
In Chapter 7 we make another jump which seems even bigger, to the topic
of exponential sums. The best known is Gauss’s
p
2π ik2
exp ,
p
k=1
also for prime p, whose absolute value p1/2 is much smaller than the number of
its terms. One of these sums, due to Heilbronn, resisted for some time all efforts
to prove its smallness until Heath-Brown in 1996 achieved this. His beautiful
proof imitated Stepanov’s auxiliary polynomial in a kind of analytic context
involving a logarithm-like function. Some arguments had been anticipated by
Mitkin in 1992. We give the details. Specifically
p
2π ikp
exp ≤ 4p11/12 .
p2
k=1
As far as I know, these arguments have not been developed very far since then,
despite some interesting features involving differential equations.
In Chapter 3 we proved the irrationality of the values μ of Mahler’s series
at non-zero rationals. Thus the quantity |μ − p/q| is positive for all integers p
and q ≥ 1. A natural question is: “How small can this quantity get?” Indeed
with an algebraic irrational in place of μ this question is fundamental in the
theory of diophantine equations, as we will see in Chapter 12. Our answer
in Chapter 8 requires refining the arguments of Chapter 3. There are two key
steps. One is a “zero estimate” asserting that not too many things can vanish;
such estimates play a major role in more recent developments. The other, more
classical, is an estimate for the coefficients of the auxiliary polynomial; this
involves the famous Siegel Lemma, which will be used over and over again
in the sequel. We also make a simple application of the maximum modulus
principle for analytic functions. This too will be used frequently later, under
the popular name of the Schwarz Lemma. In this way we will prove that there
exist c = c(μ) > 0 and κ = κ(μ) such that
μ − p ≥ c .
q qκ
Introduction xi
For example with μ = ∞ 2k
k=0 (2/3) we can take κ = 77.
There is a famous result of Pólya on entire functions mapping the natural
numbers to the rational integers; this may have influenced Gelfond in his pio-
neering work on the transcendence of αβ (see Chapter 19). Pólya’s original
proof used interpolation formulae and gave the best possible constant. Much
later Waldschmidt gave a version by auxiliary polynomials, which sadly gives
a worse constant. The proof is nevertheless illuminating; it needs binomial
coefficients to avoid factorials, one of the key ideas in Thue’s famous proof
(see Chapter 12). More precisely, we show in Chapter 9 that an entire function
f with
f (0), f (1), f (2), . . . , f (n), . . .
all in Z must be a polynomial if |f (z)| grows of order at most C |z| for a certain
C > 1. Pólya could take any C < 2; and the standard example 2z shows that
nothing better is possible. Or reformulated: if a non-polynomial entire function
f has this growth, then at least one of f (0), f (1), f (2), . . . must be non-integral.
Gelfond’s step from non-integrality to transcendence needed many more ideas,
all of which will be developed in this book.
The rather natural generalization to the Gaussian integers G = Z + Zi with
f mapping G into itself also played a similar historical role; for example it
probably directly inspired Gelfond’s proof of the transcendence of eπ . But the
best possible constant did not appear until a relatively recent paper of Gramain;
paradoxically enough, his proof involves an auxiliary polynomial (or better an
auxiliary function). More precisely, f itself must be a polynomial if |f (z)| now
grows of order at most C|z| for a certain C > 1. Gelfond considered this
2
into the more sophisticated “height”, which will then be studied for its own
sake, for example with reference to Lehmer’s Question of 1933 in connexion
with factorization problems.
At last in Chapter 12 we prove the famous Thue improvement of Liouville’s
classical result. The proofs here start getting more elaborate, and another key
element is dealing with the dangerously heavy factorials that threaten to sink
the method; however this problem has been solved in Chapter 9. Yet another
feature is a simple form of zero estimate. These have proved crucial in later
developments involving Roth, Schmidt, Schlickewei, and others. More pre-
cisely, given any algebraic number α of degree d ≥ 3 and any κ > d2 + 1, we
show that there is a positive constant c = c(α, κ) such that
α − p ≥ c
q qκ
for all integers p and q ≥ 1. The Liouville result was for κ = d, and the
later Roth estimate was for any κ > 2. Here we try to break the proof into
molecules, and we also speculate on how Thue may have arrived at his proof;
there are interesting connexions with Newton’s Method in numerical analysis
and later improvements by Halley and others. We also give the applications to
diophantine equations. Here we encounter the uncomfortable phenomenon of
ineffectivity for the first time.
Then in Chapter 13, using the machinery of the previous chapter, we
prove the Hermite–Lindemann result on the transcendence of the values of
the exponential function at algebraic numbers; thus eα is transcendental for
every algebraic α = 0. Our proof is a kind of ad hoc development of the
auxiliary polynomial techniques introduced so far; we have by now illustrated
so many of these techniques that several proofs are available. We choose the
one most suited for generalization to the Schneider–Lang Theorem later on in
Chapter 19.
Chapter 14 is where we develop the size in Chapter 11 to the absolute height
H(α) ≥ 1 or the logarithmic version h(α) = log H(α) ≥ 0. This is rather easy
to define, but to establish properties like H(α 2 ) = H(α)2 , we need quite a
bit of algebraic number theory, and we will sketch the details. The motivation
is two-fold: first, the results of the next two chapters are about heights per
se, and second, the proof of the later Schneider–Lang result then becomes
fairly streamlined. We also give a version of the Siegel Lemma in the heights
language. This requires essentially defining the height of a vector (α1 , . . . , αn )
of algebraic numbers. To break the monotony, we prove on the way some easy
results on lower and upper bounds for heights that have led to some lively
modern developments.
Introduction xiii
The results involve a Weierstrass function ℘ (z) with invariants g2 , g3 that are
themselves algebraic; the analogue of Hermite–Lindemann then asserts the
transcendence of ℘ (α) for any algebraic α = 0. Already the elliptic ana-
logue of Gelfond–Schneider has consequences for the modular function j(τ )
defined on the upper half-plane: namely that j(α) is transcendental whenever
α is algebraic but not quadratic; this we postpone to the next chapter. But as
Schneider discovered, there are several other interesting consequences; and
even he overlooked one of them. This chapter is the longest in the book, due to
our supplying the main details for the proofs of most of these consequences. It
might get shorter if we could use facts about commutative group varieties, but
Introduction xv
that would introduce too much algebraic geometry not in the elementary spirit
of the book.
In 1969 Mahler conjectured that the alternative modular function
(the pattern of coefficients is quite self-evident), defined for all q in the unit
disc, takes transcendental values at non-zero algebraic q. In 1996 Barré, Diaz,
Gramain and Philippon proved this using an auxiliary polynomial directly
on J itself. This was not only the first auxiliary polynomial of its kind (as
Schneider had wanted many years ago in his Second Problem), but it led soon
afterwards to Nesterenko’s unexpected breakthrough implying the algebraic
independence of π and eπ . This is a most attractive area where aspects of
elliptic, modular and exponential functions blend into each other. We give
a proof of Mahler’s Conjecture in Chapter 21, after deducing the analogous
result for j(τ ) = J(e2π iτ ) from the results of the previous chapter.
Up to now we never discussed problems of algebraic independence. Maybe
the reader’s curiosity for this topic has been awakened at the end of the pre-
vious chapter, and now she gets a classical example. After the famous Linde-
mann–Weierstrass result (not covered in this book), which was generalized to
E-functions by Siegel and Shidlovsky (likewise not here), the most spectacular
2
was the algebraic independence of α β and α β , for algebraic α = 0 and cubic
β, due to Gelfond in 1949. But as a lot of the machinery is already available,
our proof in Chapter 22 will not be too long. Here too one needs “large radius”.
Finally in an Appendix we prove exotic height results like
3ξ − 4 ξ 3 + 3ξ + 4 + 8
h ≤ h(ξ ) + 10000( h(ξ ) + 1)
ξ2
where a crude estimate would give at least 2h(ξ ) on the right-hand side. Indeed
if we replace 8 in the numerator by 7 this is unavoidable. The square root here,
traditionally associated with the quadratic nature of Néron–Tate heights on
abelian varieties, is actually needed.
Let us mention here yet another use for auxiliary polynomials: to show
that certain algebraic numbers arising from commutative group varieties have
“large degree”. It is well-known that the root of unity e2π i/n has degree φ(n)
the Euler φ-function, and also that for any θ < 1 there is a positive con-
stant c, of course effectively computable, with φ(n) ≥ cnθ . By a famous
result of Serre the elliptic analogue has any θ < 2, but only recently has
this been made effective, in an elaborate proof involving, among other things,
xvi Introduction
does not contain 0 as it might in some other cultures. But Z, Q, R, C and the
fields Fp = {0, 1, . . . , p − 1} are unambiguous.
And what about the exercises? These are at the end of each chapter, where
they are divided into two parts by a starred line. Those above the line need only
the prerequisites above and what’s in the book so far, and they are essentially
what were fed to students as homework reinforcing the lectures. Those below
the line go further, and sometimes need extra knowledge; they are of varying
levels of difficulty, sometimes hinted at. Concerning the lectures themselves,
there are many possibilities; for example I covered Chapters 1, 2, 3, 4, 5
together with some algebraic number theory in a single semester, then followed
up with Chapters 6, 8, 9, 11, 12, 13 and more algebraic number theory, and so
was able to start a third semester with Chapter 14 in detail, then Chapters 15,
16, 18, 19 and bits of Chapter 20.
xviii Introduction
L
P= pi X i (1.1)
i=0
1
2 Auxiliary Polynomials in Number Theory
(in Basle, after earlier attempts in Ann Arbor, Constance, Hong Kong,
Heraklion and Vienna).
Maple doesn’t respond for 47 seconds; and then gives an incomprehensible
error message (the resultant that you are seeking does not exist). How can we
find this relation P(x, y) = 0?
Let us write
L L
P(X, Y) = pij X i Y j (1.6)
i=0 j=0
for unknown L and undetermined coefficients pij which are presumably inte-
gers as in (1.3) and (1.4). If we substitute (1.5) into P(x, y), then we obtain a
polynomial in t of degree at most 3239L. Its coefficients are linear forms in the
pij . If we equate these coefficients to zero, then we get a system of 3239L + 1
homogeneous linear equations in the (L + 1)2 unknowns pij . By linear algebra
this system is solvable non-trivially, provided
(L + 1)2 > 3239L + 1
for example if L = 3238.
This proves something: namely that there is a non-trivial relation of degree
at most 3238 in each variable. There may well be (L + 1)2 = 10491121 terms
in the relation, which accounts for Maple’s chickening out. And the trouble
for anyone, of course, is that we now have to solve 10487883 equations in
10491121 unknowns.
The reader may see first that this works for any two polynomials F, G in t
instead of (1.5), and second that it generalizes to more variables; for example
between any three polynomials in two variables there is a non-trivial alge-
braic relation (as would follow more simply by consideration of transcendence
degree).
This example is perhaps more typical of those to follow in these pages. After
the substitution (1.5) we may regard the function P(x, y) as having a large order
of vanishing at t = 0; so large, indeed, that it must vanish identically.
In both examples the goal is practically the auxiliary polynomial itself; so it
is hardly “auxiliary”.
We will see many more and subtler applications in this book (sadly not
Siegel’s Theorem (Siegel, 1955) about functions on compact manifolds, which
is a very sophisticated generalization of the second example – see however
Lemma 20.4 and Exercise 20.67). But before we start, let us ask: since there
appear to be so many terms in the relation connecting (1.5), what are the coef-
ficients like? If we normalize them to be integers, how big are they? We note
that the Cramér formulae for solving linear equations involve determinants
4 Auxiliary Polynomials in Number Theory
Exercises
1.1 Show that there is P = 0 in C[X] such that P(X)(X 1000 + X 100 + X 3 ) has
n2
the form Nn=0 an X .
1.2 Let F be in C[X] with degree at most D. Show that there is P = 0 in C[X]
n2
with degree at most D2 − D such that PF has the form N n=0 an X .
1.3 Let F be in C[X]. Find P = 0 in C[X] such that PF has the form
N 2n
n=0 an X .
1.4 Let F be in C[t] with degree at most D ≥ 1 and let G be in C[t] with
degree at most E ≥ 1 . Show that there is P = 0 in C[X, Y] with degree at most
D + E − 1 in each variable such that P(F, G) = 0.
1.5 Let F be in C[t] with degree at most D ≥ 1 and let G be in C[t] with
degree at most E ≥ 1. Show that there is P = 0 in C[X, Y] with degree at most
E in X and degree at most DE − E + 1 in Y such that P(F, G) = 0.
1.6 Let t, u be independent variables, and let F, G, H be in C[t, u]. Show that
there is P = 0 in C[X, Y, Z] such that P(F, G, H) = 0.
∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗
1.7 Show that there is an absolute constant c (that is, not depending on any
parameters) with the following property. Let F be in C[X] with degree at most
D ≥ 2. Then there is P = 0 in C[X] with degree at most cD log D such that PF
has the form p ap X p , where p runs over the set of primes.
1.8 Let F be in C[X]. Find P = 0 in C[X] such that PF has the form
N 3n
n=0 an X .
1.10 Let F be in C[t] with degree at most D ≥ 1 and let G be in C[t] with
degree at most E ≥ 1. Show that there is P = 0 in C[X, Y] with degree at most
E in X and degree at most D in Y such that P(F, G) = 0 [Hint: resultants].
1.11 Let P = 0 in C[X, Y] be such that P(t1948 + t666 + 1, t1291 + t163 + t) =
0. Show that P has degree at least 1291 in X and degree at least 1948 in Y
(compare Exercise 5.7).
1.12 Can one essentially improve the D2 − D in Exercise 1.2? I don’t know.
1.13 Let F, G be in C(t) (rational functions). Show that there is P = 0 in
C[X, Y] such that P(F, G) = 0.
1.14 Let
(t2 − t + 1)3 (t2 + t + 1)3
F = 256 , G = 256 .
t2 (t − 1)2 t2 (t + 1)2
Show that P(F, G) = 0 for
P = X 3 Y − 2X 2 Y 2 + XY 3 − 1728(X 3 + Y 3 ) + 1216(X 2 Y + XY 2 )
(This is related to the differential equation for the modular function – see
Exercise 21.15.)
1.17 With L = 3238 show that (t1948 + t666 + 1)L (t1291 + t163 + t)L has a
coefficient at least
32L
> 103082 .
L +L+1
2
7
8 Auxiliary Polynomials in Number Theory
n
n ∞
∞
1 (−1)k 1 (−1)k
fn = r + s +t = −s −t ,
k! k! k! k!
k=0 k=0 k=n+1 k=n+1
we get
2 2
|n!fn | ≤ |s| + |t| .
n+1 n+1
Hence the n!fn are integers tending to zero as n tends to infinity. But we no
longer know that these integers are non-zero as in (2.1).
In fact it is not too hard to show that
(n + 1)!(fn+1 − fn ) = s + (−1)n+1 t
and so
(n + 2)!(fn+2 − fn+1 ) = s − (−1)n+1 t.
from the wonderful book of Conway √ and Guy (1996), p. 253) that these ideas
also give the irrationality of λ = e 2 via
2l ∞
1
λ+ = 2
λ (2l)!
l=0
is irrational and even transcendental for any rational and even algebraic α = 0.
In the present chapter we will consider the series
∞
zk 1 1 1
= 1 + z + z2 + z3 + z4 + · · · , (2.3)
2k(k−1)/2 2 8 64
k=0
which similarly converges for all real and even complex z to a function f (z).
It converges faster than the series of ez , so let us see what truncation of f (α)
gives.
Let us start with α = a/b, for simplicity taking a ≥ 1, b ≥ 1 in Z. The
truncations are
n ∞
αk αk
fn = f (α) − = .
2k(k−1)/2 2k(k−1)/2
k=0 k=n+1
Again there is domination on the extreme right, and if n is large enough we get
α n+1 2an+1
|fn | ≤ 2 = . (2.4)
2n(n+1)/2 2n(n+1)/2 bn+1
Taking into account a common denominator 2n(n−1)/2 bn , we deduce
2an+1 2a a n
|2n(n−1)/2 bn fn | ≤ = .
2n b b 2
The proof works if the estimate tends to zero as n tends to infinity (assuming
we can rule out the snag fn = 0 as in (2.2) above). Unfortunately this is the
case only for a = 1.
Another random document with
no related content on Scribd:
CHAPTER VII.
A STRANGE FANCY.