S.D.

Adhikari

Aspects of
Combinatorics
and
Combinatorial
Number Theory
 ‘Oh! I have looked for miracles, and now that I don’t look
for them any more, I seem to see them everywhere... Oh! J’ai
cherche des miracles, et maintenant que je n’en cherche plus, il
me semble en voir partout. They say “chance”, but what does it
mean? The smallest of these chances shines like a star in the great
forest of the world; and sometimes I feel that a casual gesture, a
small second of inattention, a hop to the right instead of the left, a
bird’s feather, a mere nothing which fleets by, contains a world of
vertiginous premeditation - and perhaps ... Perhaps we do not see
. the invisible thread which. connects this dazzled second,
this sudden crossing of roads, this winged thistle seed, to another
unfinished story, an old unfulfilled promise, a forgotten hill .
Where is the begining of the story? ... Ou, est le commencement de
VhistoireV

— ‘By the body of the earth : A perpetual story’ : Satprem

‘The mystery as well as the glory of mathematics lies not so

much in the fact that abstract theories turn out to be useful in
solving problems but in the fact that - wonder of wonders - a theory
meant for one type of problem is often the only way of solving
problems of entirely different kinds, problems for which the theory
was not intended.’

‘Indiscrete thoughts ’ : Gian-Carlo Rota

 Digitized by the Internet Archive
in 2019 with funding from
Kahle/Austin Foundation

https://archive.org/details/aspectsofcombinaOOOOadhi
 Sukumar Das Adhikari

Aspects of
Combinatorics and
Combinatorial
Number Theory

Thomas i. Bata Library

TRENT UNIVERSITY
PETERBOROUGH, ONTARIO

CRC Press
Boca Raton London New York Washington, D.C.

Narosa Publishing House

New Delhi Chennai Mumbai Kolkata
(5.A.

Sukumar Das Adhikari

Harish-Chandra Research Institute
Chhatnag Road, Jhusi
Allahabad-221 019, India

Library of Congress Cataloging-in-Publication Data:

A catalog record for this book is available from the Library of Congress.

All rights reserved. No part of this publication may be reproduced, stored in retrieval system
or transmitted in any form or by any means, electronic, mechanical, photocopying or
otherwise, without the prior permission of the copyright owner.

This book contains information obtained from authentic and highly regarded sources.
Reprinted material is quoted with permission, and sources are indicated. Reasonable efforts
have been made to publish reliable data and information, but the author and the publisher
cannot assume responsibility for the validity of all materials or for the consequences of
their use.

Exclusive distribution in North America only by CRC Press LLC

Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton,
Florida 33431. Email: orders@crcpress.com

Copyright © 2002 Narosa Publishing House, New Delhi-110 002, India

No claim to original U.S. Government works

International Standard Book Number 0-8493-0974-3
Printed in India
Gratefully dedicated to

Prof. K. Ramachandra
 Preface

The present volume has its origins in the notes which were
prepared initially for the course of lectures delivered by the au¬
thor in a workshop at Pune, organized jointly by Bhaskaracharya
Pratishthana and the Department of Mathematics, University of
Pune. Several other chapters grew from the notes prepared for
the lectures delivered at Witwatersrand University (Johannesburg,
South Africa), Vivekananda College ( Chennai), Institute of Math¬
ematical Sciences ( Chennai), Indian Statistical Institute (Banga¬
lore), Tata Institute of Fundamental Research (Mumbai), University
of Mysore and some universities in Japan and Europe.

A large part of this book concerns Ramsey-type results in com¬

binatorial number theory. These results talk about ‘unavoidable
regularities’. “Complete disorder is impossible”: to describe the
philosophy of Ramsey Theory in a nutshell, perhaps nothing can
be better than this statement of Motzkin. To be more precise, in
many cases such a result says that if a large structure is divided
into finitely many parts, at least one of the parts will retain cer¬
tain regularity properties of the original structure. In some cases,
‘large’ substructures will be seen to have certain regularities. That
is, if the ‘size’ of a structure is big enough, certain regularities are
unavoidable. Attempt has been made to touch upon all the classi¬
cal results in this area and, at the same time, to give glimpses of
various techniques used to tackle these problems. However, most of
our proofs rely on combinatorial arguments. As for methods based
on Ergodic Theory and Topological Dynamics, we have limited our¬
selves to giving some simple examples and a list of references.

While writing this volume, effort has been made to minimize

overlap with material in available texts. Therefore, along with sev¬
eral recent results, many new proofs as well as simplified versions
of known proofs of well known results have been made available.
viii PREFACE

Again, at many places readers are directed to the available texts for
certain results.
Starting from a discussion on the pigeonhole principle, of which
the classical theorem of Ramsey can be thought of a generaliza¬
tion, in the first chapter we first discuss the theorems of Ramsey
and Erdos-Szekeres and proceed to present the classical results of
Schur, van der Waerden and Hilbert, the early results in the area
of ‘Ramsey-type theorems in combinatorial number theory’. The
theorem of Hales and Jewett and some variations of the van der
Waerden’s theorem are given in Chapter 2. In Chapter 3, we talk
on some generalizations of Schur’s theorem. Topological methods
are used in Chapter 4 and also in the proof of Hilbert’s theorem
in Chapter 1. In Chapter 5, we give an introduction to Euclidean
Ramsey theory which had its inception in the pioneering papers of
Erdos, Graham, Montgomery, Rothschild, Spencer and Straus. In
the first part of Chapters 6, we shall see some Ramsey-type theo¬
rems in additive number theory which include the Erdos-Ginzburg-
Ziv Theorem and related theorems. In the second part of Chapter
6, we see instances of application of Ramsey’s theorem to certain
number theoretic problems. In Chapter 7, which deals with con¬
gruence of the partition function p(n), we give some recent results
on the parity of the partition function. In Chapter 8, which deals
with Ramsey-type results in partially ordered sets, we shall prove a
theorem of Harzheim which says that for a choice function / on the
set of non-empty subsets of a large set will contain large chains on
which / is constant; we shall also consider some generalizations of
this result of Harzheim. In the last section of each of these chapters
we provide some notes on the background for the problems discussed
in the particular chapter as well as give some indications regarding
further developments.

Solutions to selected exercises are given in the last chapter.

As for prerequisites for this book, one needs some knowledge of
elementary number theory, linear algebra, the rudiments of point-
set topology and some basic results in the theory of finite fields,
at the level of the topics covered in the books of Hardy & Wright
[91], Herstein [98] and Munkres [119] cited in the bibliography.
However, whenever we use a theorem, belonging to one of these
subjects, which could be beyond an undergraduate course, we try
to supply a proof of that theorem in this book.
 PREFACE IX

While numbering a theorem (or a definition or a remark), the

chapter number is not indicated. Therefore, while reading through
a chapter, if one comes across a reference to a theorem where only
the theorem number is given, it would mean that it is a theorem in
that particular chapter; otherwise while referring to a theorem, we
also mention the Chapter in which it appears.
I would like to thank: Prof. Mahesh Nerurkar, Prof. Vitaly
Bergelson, Prof. R. Balasubramanian, Prof. S. A. Katre and several
other mathematicians in the Pune workshop, for many stimulating
lectures and discussions; Prof. J. N. Ridley, Prof. Arnold Knopf-
macher, Prof. A. Schinzel, Prof. T. Nakahara, Prof. S. Kanemitsu,
Prof. K. Gyory, Prof. T. N. Shorey and many other mathematicians
who were present in the audience during my lectures in Johannes¬
burg and other places, for their valuable comments; Prof. Bhaskar
Bagchi, Prof. Sridhar Inamdar and Dr. R. Thangadurai, for their
help and suggestions during the preparation of the manuscript; Dr.
D. Surya Ramana, Mr. Anirban Mukhopdhyay, Mr. Shripad M.
Garge, Mr. Purushottam Rath, Mr. Anupam Kumar Singh and
Ms. Sanoli Gun, for helping me with the proof-reading and making
some suggestions; Dr. B. Ramakrishnan, for constant encourage¬
ment.
Further, I would like to thank Prof. H. S. Mani, Prof. Ravi S.
Kulkarni and Prof. D. Prasad for their encouragement.

October, 2001 Sukumar Das Adhikari

Allahabad.
 Notations and terminologies

Throughout the text, the symbols Z+,N, Z, Q and R will de¬

note respectively the set of positive integers, the set of non-negative
integers, the set of integers, the set of rationals and the set of real
numbers.
As usual, $ will denote the empty set.
For a subset A of a set B, the notation A C B will not mean
strict inclusion. A strict inclusion will be indicated by the notation
A^B.
Usually, for a positive integer n, the symbol [n] will denote the
set {1, 2, • • • n}. In some places, when for a real number x, we write
[x] to denote the largest integer < x, we shall explicitly mention it.
If r G Z+, then an r-colouring of a set S is a map x ’■ S —>
{1, • • • , r}. If s is an element of 5, then x(s) is called the colour of
5. A set T C 5 is called monochromatic with respect to a colouring
X if X is constant on T.
One observes that writing S — x-1(l) Ux_1(2) U • • •Ux'H?'), an
r-colouring of a set 5 is nothing but a partition of S into r parts.
A given Diophantine equation is said to be r-regular if for any r-
colouring of Z, there is a monochromatic solution of that equation.
The Diophantine equation is said to be regular if it is r-regular for
each r G Z+.
For a finite set 5, |5| will denote the number of elements of S.
If for a positive integer k, X is a A>element subset (that is,
\X\ = k) of a set Y, we shall simply say that X is a A;-subset of Y.
The collection of all A;-subsets of a set Y will be denoted by Q.).
For any prime power q, Fq will denote the finite field with q
elements and the symbol F* will denote the multiplicative group of
non-zero elements of Fq. In general, for any field K, K* will denote
the multiplicative group of its non-zero elements.
 Contents

Preface vii

Notations and terminologies xi

Chapter 1. Classical Ramsey-type theorems 1

1. Introduction 1
2. The Pigeonhole Principle 1
3. Theorems of Ramsey and Erdos-Szekeres 4
4. van der Waerden’s Theorem 9
5. Schur’s Theorem 14
6. A theorem of Hilbert 14
7. Notes 19

Chapter 2. van der Waerden revisited 23

1. Introduction 23
2. Hales-Jewett theorem 24
3. Some variations of van der Waerden’s theorem 28
4. Notes 32

Chapter 3. Generalizations of Schur’s theorem 37

1. Introduction 37
2. Rado’s theorem 37
3. Folkman’s theorem 41
4. Notes 46

Chapter 4. Topological methods 49

1. Introduction 49
2. Compact semigroups and idempotents 49
3. Ramsey-type theorems 53
4. Notes 57

Chapter 5. Euclidean Ramsey theory 61

1. Introduction 61
2. Preliminaries 62
3. Fundamental Theorems 64
XIV CONTENTS

4. Further Ramsey configurations 67

5. A theorem of Graham 70
6. Notes 72

Chapter 6. Additive number Theory and related questions 75

1. Introduction 75
2. Zero-sum problems 75
3. A question of Sidon and some related problems 90
4. Notes 93

Chapter 7. Partitions of integers 95

1. Introduction 95
2. Preliminaries 96
3. Generating functions 99
4. An identity of Euler 102
5. Parity of partition function 107
6. Notes 114

Chapter 8. Ramsey-type results in posets 117

1. Introduction 117
2. Preliminaries 118
3. A theorem of Harzheim 118
4. Generalizations 121
5. Notes 126

Chapter 9. Solutions to selected exercises 129

Bibliography 143

Index 153
 CHAPTER 1

Classical Ramsey-type theorems

1. Introduction

In this chapter, we shall be dealing with some of the earliest

Ramsey-type theorems. These include the theorem of Ramsey; the
other results being those of Schur, van der Waerden and Hilbert,
which share the credit of preceding the result of Ramsey in the class
of Ramsey-type theorems.

In Section 2 of this chapter, we shall briefly discuss the pigeon¬

hole principle, of which the classical Ramsey theorem is a general¬
ization. Before embarking on our Ramsey-type results, in the last
remark of Section 2, we try to explain the general philosophy and
common feature of these results.

Section 3 will be devoted to the theorems of Ramsey and Erdos-

Szekeres.

In Section 4 we shall be dwelling upon van der Waerden’s theo¬

rem. In fact, we shall be proving a generalized version of it.

Finally, in Sections 5 and 6, we shall take up the theorem of

Schur and that of Hilbert respectively.

2. The Pigeonhole Principle

In its simplest version, the ‘pigeonhole principle’ says that if n+1

objects are put in n pigeonholes, then there will be a pigeonhole
containing more than one object. Interesting applications of this
simple looking statement dates back to the days of Dirichlet.
2 1. CLASSICAL RAMSEY-TYPE THEOREMS

There are many versions of the pigeonhole principle which gen¬

eralizes the statement given above in various ways ; not straying far
from it at the same time. Here we state one of them.
Theorem 2.1. (Generalized pigeonhole principle) Let A
and B be finite nonempty sets where B — {b\,b2,--- 5^r}- Let
f be a function from A to B. Then, for non-negative integers
ai, a2, • • • , ar, the equality

\A\ — a\ + 0.2 T ''' T ar — p T 1

would imply that 3 i G {1, • • • , r} such that \f~1{bl)\ > a^.

Proof. If possible, let there be no such i. Then

\A\ = \f~l{bi)\ < - 1 ) = J2ai-r,

i—1 i= 1 i=l
which contradicts our assumption. □

Remark 2.1. It is clear that the theorem above is interesting

only when |H| > \B\.
Remark 2.2. A, B and / being as in the statement of Theo¬
rem 2.1 above, if |H| = |B| + 1, then from the theorem it follows that
there is bx £ B with |/_1(6j)| > 2 which is the commonly known
version of the pigeonhole principle stated at the beginning of this
section.
Remark 2.3. Two of the main tools of combinatorics are count¬
ing argument and induction. While (as can be seen from the proof
of Theorem 2.1) the pigeonhole principle is of the first type, we shall
soon have many instances of very clever applications of the second.
Exercise 2.1. (Dirichlet) Show that for all real numbers a and
n G Z+, there are p,q G Z, with 1 < q < n satisfying

p| 1 1
a — - < — < — .
q qn ql
Exercise 2.2. Show that given any five integer lattice points
on the plane the midpoint of the line segment determined by some
two distinct points among them will also be an integer lattice point.
Exercise 2.3. Writing n(4) to be the minimum number of in¬
teger lattice points in the plane so that some four of them must
determine an integer lattice point centroid, by applying Exercise
2.2 above, show that n(4) = 13.
 2. THE PIGEONHOLE PRINCIPLE 3

Definition 2.1. In what follows, a graph G will consist of a

non-empty set V(G) of elements, called vertices and a set (possibly
empty) E(G) of elements called edges, together with a relation of
incidence that associates with each edge two distinct (our definition
avoids loops) vertices, called its ends. If two vertices are incident
to an edge, they are called adjacent.

A graph G is called finite if both V(G) and E(G) are finite sets.
Henceforth, by a graph we shall always mean a finite graph. The
integers |V(G)| and |£(G)| are called respectively the order and the
size of the graph G.

The degree d(v) of a vertex v in a given graph G is the number

of edges of G to which v is incident.

A graph is called simple if no two of its edges have the same pair
of ends.

The complete graph Kn is a simple graph which has order n and

size (2); ^at is, each pair of distinct vertices are adjacent.

In a graph, a subset of the vertices is called a clique, if every

vertex in the subset is adjacent to every other vertex in the subset.
A subset of the vertices is called an independent set, if every vertex
in the subset is nonadjacent to every other vertex in the subset.

A graph H is called a subgraph of a graph G if V{H) C V(G),

E(H) C E(G) and an edge in H has the same pair of vertices as its
ends as in G.
Exercise 2.4. (Mantel) Let n > 2 be a positive integer. Show
that if a simple graph G of order 2n contains n2 + 1 edges, then G
contains a triangle (K3).
EXERCISE 2.5. (Erdos and Szekeres) Given a sequence of mn+1
distinct real numbers, show that if it does not contain a monotone
increasing subsequence of length m + 1, then it must contain a
monotone decreasing subsequence of length n-1-1.

Remark 2.4. Apart from the fact that all of them can serve
as beautiful examples of application of the pigeonhole principle, the
last few exercises above, have one more thing in common and this is
the fact that all of them are instances of certain unavoidable regular¬
ities. As we shall see as we proceed, unavoidable regularities, or in
other words, existence of regular substructures in general combina¬
torial structures is the phenomena which can be said to characterize
4 1. CLASSICAL RAMSEY-TYPE THEOREMS

the subject of Ramsey theory. Most often, we shall come across re¬
sults saying that if a large structure is divided into finitely many
parts, at least one of the parts will retain certain regularity prop¬
erties of the original structure. In some results in Ramsey theory,
‘large’ substructures will be seen to have certain regularities.

3. Theorems of Ramsey and Erdos-Szekeres

We are now going to state Ramsey’s theorem which originally

appeared as a lemma in a paper [148] of F. P. Ramsey on Mathe¬
matical logic.
Theorem 3.1. (Ramsey’s Theorem ) Let k,r,l(> k) be posi¬
tive integers. Then there exists a positive integer n — n(k, r, l) such
that for any r-colouring of the k-subsets of the set [n], there is an
l-subset of [n] all of whose k- subsets are of the same colour.
Remark 3.1. The smallest n for which Theorem 3.1 holds will
be denoted by R(k,r,l).
We shall prove the following general version.
Theorem 3.2. ( Generalized Ramsey’s Theorem) Given
positive integers k,r,li,--* ,lr, there exists a positive integer n =
n(k, r; /1; • • • ,/r) such that for any r-colouring of the k-subsets of
the set [n] with colours Ci, • • • , cr , for some i E {1, • • • , r}, there is
an L-subset of [n] all of whose k- subsets are of the colour c*.

Proof. If Theorem 3.2 holds for positive integers k, r, G, •• • , t7r,

we shall denote the smallest positive integer n for which it holds
by R*{k, r; lu ■ ■ ■ lr). For the proof, we proceed by induction. For
k = 1, it is Theorem 2.1. Therefore, let k > 1 and we assume that
the numbers R*{k - 1, r; lx, • • • ,lr) are defined for all If s. For any
given k, if lz = k for all i E (1, • • • , r}, then again the result follows
trivially.
Now, assuming that the numbers

R* {k, r, l\, , li—1, li 1, h+l, ' ' ' i L)

are defined for i = 1, • • • , r, we have to show that R*(k, r; llt l2, • • • , lr)
exists.
Let n — 1 + R*(k — 1, r; ai, • • • , ar).

Let an r-colouring c of the k-subsets of the set [n] with colours

c\, ■ ■ • Tr be given. Let m E [n] and Y = [n] \ {m}. An induced
 3. THEOREMS OF RAMSEY AND ERDOS-SZEKERES 5

r-colouring c* of the (k - l)-subsets of Y is now defined in the

following manner. Given a (k - l)-subset S of Y, the colour of S is
c* if and only if that of S U {m} is cl in the original colouring.

By the definition of n, for some i there is a set At (c [n]\{m})

of size at all of whose (k — l)-subsets have the same colour c* in the
c*- colouring.

By the definition of at, the set At will contain either a set of size
lj all of whose fc-subsets have colour cj for some j ^ i or a set X
say of size lt — 1 all of whose k-subsets have colour Cj.

In the first case we are through. In the second case, X U {m}

is a set of size lt. Given a k-subset K of X U {m}, if m does not
belong to K then by the assumption in the above paragraph, K is
of colour Ci. On the other hand, if m G if, then since all (k - 1)
-subsets of At are of colour c*, so is the set K \ {m} and therefore
by the definition of the induced colouring c*, K is of colour ct. This
completes the proof. □

Remark 3.2. The smallest n for which Theorem 3.2 ( General¬

ized Ramsey’s Theorem) holds will be denoted by R(k, r; /1; • • • , lr).
Taking l\ = • • • = lr = l, we observe that

R(k, r; /,•••,/) = R(k, r, /).

'-v-'
r times

When k = 2 and r = 2, often the symbol R(p,q) is used in

place of R(2,2;p,q). In the case p = q = m, one writes R(m) for
R(2, 2;m, m) — R(2, 2, m) = R(m,m). The situation where k = 2
and r — 2, has a graph theoretic interpretation. For a 2-subset of
the set of vertices, the first colour can be interpreted as the case
where the two vertices are joined by an edge and the second colour
as the case where they do not have an edge joining them. Then
any graph with R(p, q) vertices contains either a Kp (a clique on p
vertices) or an independent set of q vertices.
Exercise 3.1. Using Theorem 3.1, show that every finite semi¬
group S contains an idempotent, that is, an element a satisfying
a2 = a. (Recall that a semigroup 5 is a set with an associative law
of composition.)
We now give an application of Ramsey’s theorem which played
a major role in popularizing Ramsey’s theorem and the subsequent
emergence of a new branch of Combinatorics.
6 1. CLASSICAL RAMSEY-TYPE THEOREMS

Theorem 3.3. (Erdos-Szekeres Theorem) For a given pos¬

itive integer n(> 3), there is an integer N — N(n) such that any
collection of N or more points in the plane, no three on a line, has
a subset of n points forming a convex n-gon.

Proof. Let n > 3 be given. Given a finite set S of points in the

plane (no three points of which are on a line), we colour the three
point subsets with two colours red and green such that the colour
of the set {A, B, C} is red if the number of points in the interior
of triangle ABC is even and is green otherwise. Now, by Ramsey’s
theorem, there is a positive integer N — N(n) such that for any set
S of N points in the plane, no three of which are collinear, there is
an n-subset J of S such that the set of all 3-subsets of J are of the
same colour.

We claim that the points of J are the vertices of a convex n-gon.

To establish our claim, it will be enough to prove that given any

three points A, B, C of J there can not be any point of J in the
interior of the triangle ABC. For, in the plane, if a point P lies in
the convex hull of some set T, then P will lie in the convex hull of
some subset of T that contains at most three points. For a general
version of this result one may look into [152]' Chapter 3.
Now, given four distinct points A, B, C and D in the plane such
that no three are on a line and D lies in the interior of the triangle
ABC, the number of points in the interior of the triangle ABC is
one more than the total number of points in the interiors of the
triangles ABD, ACD and BCD (keeping in mind that no three
points of 5 are on a line). In other words, if we denote the number
of points in the interior of a triangle XYZ as f(XYZJ, then

f(ABC) = f(ABD) + f(ACD) + f(BCD) + 1.

Since the above equation shows that the numbers f(ABC),

f(ABD), f(ACD) and f(BCD) can not be all even or all odd,
it follows that given any three points A, B,C of J there can not be
any point of J in the interior of the triangle ABC. This proves our
claim. □

The above proof is due to S. Johnson [104], We now sketch the

original proof of Erdos and Szekeres [58],
 3. THEOREMS OF RAMSEY AND ERDOS-SZEKERES 7

Proof. First, we observe that given any five points in the plane,
no three on a line, some four of the points form a convex quadrilat¬
eral. To see this, we take any three of the points, say A, B, C. If one
of the remaining points is outside the triangle ABC we are through.
If both of the remaining points D and E is within the triangle, then
the line DE will meet two sides of the triangle (since no three points
are collinear), say AB and AC. Clearly, the quadrilateral BCDE
is convex.
Our second observation is that if we have n points in the plane,
no three on a line, such that every quadrilateral formed by taking
any four of the points is convex, then the given set of points form a
convex n-gon.
We claim that N = i?*(4, 2; 5, n). Given N points in the plane
we colour the four subsets of it in the following way. We colour a
4-subset red if it forms a convex quadrilateral and green otherwise.
By our first observation, there is no 5-set all of whose 4-subsets are
green. Hence there is an n-set all of whose 4-subsets are red and by
our second observation, it forms a convex n-gon. □

We now state the infinite version of Ramsey’s theorem.

THEOREM 3.4. (Ramsey) [148] If r and k are two given posi¬
tive integers, then for any r-colounng of the k-subsets of Z+, there
is an infinite subset S of Z+ with all its k-subsets having the same
colour.
We observe that the case k — 1 is trivial; it is nothing but an
infinite version of the pigeonhole principle.
Here we shall prove a weaker version of the above theorem. More
precisely, we shall prove it in the particular case k = 2. The general
proof can follow ( see [78], for instance ) by repeated application of
the same idea.
THEOREM 3.5. For any finite colouring of the 2-element subsets
of Z+, there is an infinite subset S o/Z+ with all its 2-subsets having
the same colour.

Proof. Let r e Z+. Suppose an r-colouring

of Z+.
8 1. CLASSICAL RAMSEY-TYPE THEOREMS

Consider the subset T C (z2+) defined by:

T = {{1, a} : a € Z+,a > 1}.

Since T is infinite, some colour, say cn, will occur infinitely often.
Let {l,ai}{l,a2}, {l,a3}, • • • be the elements of T which receive
the colour cn. Here it is assumed that a\ < a2 < a3 < • • • .

Now, consider the subset 7\ C (Z2 ) defined by:

7\ = {{ai, di} : i > 1}.

Once again, since Tx is infinite, some colour, say cJ2, will occur
infinitely often. Let {ai, ^>i}, {ai, b2}, {ai, 63}, • • • be the elements
of T\ which receive the colour c;2. As before, it is assumed that
61 < 62 < &3 < • •• •
Next, we consider the subset T2 C (z2+) defined by:

T2 = {{b1,bl}:i> I}-

Continuing in this way, we obtain an infinite set

V = {1,0!, 61, • • • }.

We observe that the colour of any pair {a, /3} G will depend
on the smaller one among a and (3. Therefore, for each element
d £ V we can define an r-colouring x* as follows:

X*(d) = X({d,d'}), for any d' > d.

By the pigeonhole principle, elements of some infinite subset Vi
of V will receive the same colour in the colouring x*- From the
definition of x* it follows that x({d, d'}) will be the same for any
{d,d'}€ (P2'). □

Though in these notes we will be mainly concerned with the

structural part of the theory, we make few exceptions to give a
flavour of different techniques which find their way into Ramsey
Theory.
Here is a result which gives the flavour of the Probabilistic Method
developed by Erdos.
Theorem 3.6. (Erdos) [46]

R(l) = R(IJ) = R(2, 2, /) > 2(z“2>/2,

where R(k,r,l) is as defined in Remark 3.1, following Theorem 3.1

and the other notations are as in Remark 3.2.
 4. VAN DER WAERDEN’S THEOREM 9

Proof. The number of possible 2-colourings of the 2-subsets of

a given set S of n points is 27^7l-1)/2_

Given an /-subset L, the number of ways in which the 2-subsets

of S can be 2-coloured so that the 2-subsets of L are monochromatic,

2.2(a)~(*).
Since there are ways of choosing an /-subset, if

2(2) —< 2n(n~1)/2^

that is,

then there exists a 2-colouring of S for which there is no / subset

all the 2-subsets of which are monochromatic.
Now,
2(Z-2)/2
n <
nl < 2«l-2)/2 < 2(i)-l>

the result follows. □

Remark 3.3. There is a big gap between the known constructive
lower bound (see [78], [61], [39]) for R(2,2,l) and the existential
one proved above. Also, despite of efforts very little is known about
the numbers R(k, r, /) in general. In the notes in Section 7, we shall
give some references for bounds for classical Ramsey numbers.

4. van der Waerden’s Theorem

Rightly finding its place among the ‘pearls’ that Kinchin [106]
presents in his ‘Three pearls of Number theory’, the theorem of van
der Waerden [176] we are going to state now, led to many interesting
developments in combinatorics and number theory.
Theorem 4.1. (van der Waerden’s Theorem) Given k,r £
Z+, There exists W(k,r) £ Z+ such that for any r-colouring of
{1, 2, • • • \V(k, r)}, there is a monochromatic arithmetic progression
( A.P.) of k terms.
10 1. CLASSICAL RAMSEY-TYPE THEOREMS

We prove the following generalization of Theorem 4.1 to higher

dimension as it is not much, harder to do so. Following Anderson
[13] (given also in [111]), we present the proof of this generalization
which is due to G. Griinwald (Gallai).
Theorem 4.2. (Griinwald’s Theorem) Let d, r G Z+. Then
given any finite set S C (Z+)d, and an r-colouring of (Z+)d, there
exists a positive integer ‘a’ and a point ‘v’ in (Z+ )d such that the
set aS + v is monochromatic.

Proof. We have to prove the following statement for all finite

sets S C (Z+)d.
A(S): For each k G Z+, 3 n = n(k) such that for every k-
colouring of Bn =f {(ai, • • • , a^) : az G Z+, 1 < at < n}, Bn contains
a monochromatic subset of the form aS + v for some a G Z+ and
v e Bn.
We proceed to prove the statement A(S) by induction on |S|.
Clearly, A(S) is true if |S| = 1. We show that A(5) =>• A(5 U {5})
for any s G (Z+)d.
For that purpose, we introduce the following intermediate state¬
ment corresponding to a positive integer p.
C(p): Let S C (Z+)d be fixed for which A'(S) is true. Then for
given k G Z+ and s G (Z+)d, 3n — n(p,k,s) G Z+ such that for
each /c-colouring of Bn, there are positive integers a0, al5 • • • , ap and
a point u G (Z+)d such that the each of the (p + 1) sets

Tq d= u + ( at j S + [ ai) 0 < Q <

\0<i<q ) \9GlTP /
are monochromatic subsets of Bn.
The statement C(p) is trivial for p = 0.
We now prove that C(p) => C(p + 1).
Let n — n(p, k, s) be the integer specified for C(p). Let k' d= knd.
If k colours are available, then there are k' ways to colour the cube
Bn. Now, given a /c-colouring of (Z+)d, we define a new colouring
of (Z+)d such that two points u and v will have the same colour in
the new colouring iff the cubes u + Bn and v + Bn are identically
coloured in the original fe--colouring of (Z+)d. We call this the
associated colouring of the given k- colouring.
Clearly, this associated colouring of (Z+)d is a k'~ colouring.
 4. VAN DER WAERDEN’S THEOREM 11

Now, from A(S), it follows that 3 an integer n' = n'(k') such

that for every ^'-colouring of Bn>, Bn> contains a monochromatic
subset of the form a'S + v' for some a'(^ 0) G Z+ and v' € Bn>.

Let N — n + n' + 1. Let a /c-colouring of Bn be given. In

an arbitrary way we extend this to a /c-colouring of (Z+)d. Now,
corresponding to the associated k'- colouring of (Z+)d, Bn' contains
a monochromatic subset of the form a'S + v'. This means that the
|S|-cubes Bn 4- a't + v',t G S are coloured identically in the original
k- colouring. We observe that all these cubes lie in BN■ By C(p),
for any t G 5, the cube Bn + a't + v' contains monochromatic sets

Tg(t) — a't + v' + u + j ^2 ai j $ + [ X! 0,1 ) s’ 0 < q < p-

i0<i<q ^q<i<p

Setting bo = a' and 6, = aj_ i, l<z<p-(-l, we claim that the

sets

Tg = W + u) + X bi 5+ X bi 5’ 0 - ^ - P + 1
\0<i<q J \9<*<P+1 /

are monochromatic.

For q = 0, Tq=(v' + u) + (Eo<z<p+i s is a singleton and the

claim is established.
For q > 1, Tg = UtesTq-i(t). Since Bn + a't + v' are identi¬
cally coloured for different f s belonging to S, it follows that the
monochromatic sets Tq-\{t) are of the same colour and hence T'q =
UtesTq-\(t) is a monochromatic subset of BN-
Thus C(p + 1) holds with n(p + 1, k, s) = N.

Now that C(p) is established for all integers p > 0, the particular
case p = k gives us an integer n = n(k, k, s) such that given any
/c-colouring of Bn, 3(k + 1) monochromatic sets T0, • • • ,Tk in Bn.
By pigeonhole principal, two of these sets, say Tr,Tq with r < q are
of the same colour.

Writing

Tr — u + X a‘) S+ ( X
0<i<r J \r<i<q )
s+ ( X a*)s
\q<i<k+l J

and
12 1. CLASSICAL RAMSEY-TYPE THEOREMS

T, = u+( £ a.) S + f£ 5+ ( E
\0<i<r / \r<i<q ) \q<i<k+1 /

and choosing Sq £ S (S being nonempty), it follows that the set

T = u + l X ai ] so + ( X ) (5* hJ {s}) +1 X ai | s
\0<i<r / \r<i<q ) \q<i<k+1 )

is contained in BN and is monochromatic.

Setting a = Ylr<i<q ai anc^ v — W+(Eo<kr ai) s0+(llq<i<k+l ai) Si

T = a(S U {s}) + v and this establishes A(S U {s}). □

We note that when d = 1, putting 5 = {1, • • • , A;}, Theorem 4.2

implies Theorem 4.1.
Once again, we shall now stray out of the territory of the struc¬
tural Ramsey-theory, to prove a lower bound of van der Waerden
numbers in some particular cases, to give some flavour of algebraic
techniques.

Theorem 4.3. If p is a prime, W(p+ 1,2) > p(2P - 1).

Proof. As usual let F2v denote the finite field with 2P elements.
Let a be a generator of the group Ffp.
We fix a basis iq, • • • , vp for F2p over F2.
Now, for j G {1,2, •• • ,p(2P - 1)}, let

= QijVi + • • • + cipjVp, ciij € F2.

We now define a 2-colouring [p(2p — 1)] —> {red, green} as follows.

Let the green patch consist of elements j 6 [p(2p — 1)] with a\3 = 0
and the red portion all the remaining elements j with a\j — 1.
We claim that with the above 2-colouring, the set [p(2P — 1)] =
{1,2, • • • , p(2P — 1)} does not contain any monochromatic A.P. of
length p + 1.
If possible, suppose that there is a monochromatic A.P.

{a, a + 6, • • • , a + pb} C {1, 2, • • • ,p(2p - 1)}.

Let (3 ■= <Ta,7 = ab. Since, l<a<a+p6< p( 2P - 1), we

have b < 2P — 1 and hence 7 / 1. It follows that the elements
(3, j3j, ■ ■ • , (3y have the same first coordinate as vectors.
 4. VAN DER WAERDEN’S THEOREM 13

Case I (The A.P. is green)

The first coordinate of the p vectors /?,/Fy, • • • ,is zero.
Since they belong to a vector space of dimension p — 1, they are
linearly dependent. In other words, 3 a0, ai, • • • , ap_j in F2 not all
zero such that

= 0
i=0
p-1

=> - 0
i=0

Now 7 6 F2p, 7 ^ F2 and 7 satisfies a nonzero polynomial of

degree < p over F2 — contradiction.

Case II (The A.P. is red)

In this case, wre consider the vectors (7 — 1), - - - , (3(yp - 1) and
apply the same argument as in Case I.

Thus, there are ai, • • • , ap € F2 not all zero such that

^a,(/3(7l - 1)) = 0.
i=i

Dividing by 7 — 1, we see that 7 satisfies a nonzero polynomial of

degree less than p - contradiction. □

Remark 4.1. Several special colourings in finite fields and the

field of rationals have played important roles in Ramsey Theory.
Some such instances are recorded in [2],

Exercise 4.1. Show that whichever way we 3-colour the points

of the Euclidean plane, there will be two points at a distance 1
apart, receiving the same colour.

EXERCISE 4.2. Show that Theorem 4.1, is equivalent to the fol¬

lowing statement.

V(k,r): For any two positive integers k and r, if the set Z+ of

positive integers is partitioned into r subsets, then at least one of
the subsets will contain an arithmetic progression of k terms.
14 1. CLASSICAL RAMSEY-TYPE THEOREMS

5. Schur’s Theorem

Theorem 5.1. (Schur’s Theorem ) For any r-colouring of

Z+, 3 a monochromatic subset {x, y, zj of Z+ such that x + y — z.
(The situation is described by saying that the equation x + y = z
has a monochromatic solution. )

Proof. We shall deduce Theorem 5.1 from Theorem 3.1. Let

N = R(k, r, /), where R(k, r, l) is as defined in Remark 3.1 following
Theorem 3.1.
An r-colouring x
'■ [-W] —> [r] induces an r-colouring x* °f the
collection of 2-element subsets of [N]:

By definition, 3 a 3-element subset {a, 6, c} of [N] with a < b < c

such that x*({a,6}) = X*({^c}) = X*({c>a})> that is, x(& — a) =
X{c-b) = x{c~a).
Since (b — a) + (c - b) = (c - a), we get a monochromatic solution
of x + y = z. ID

Exercise 5.1. (Schur) Using Theorem 5.1, show that for every
positive integer m, the congruence

xm + ym = zm (mod p) '

has nontrivial solutions for all sufficiently large primes p.

Exercise 5.2. Find a finite colouring of Z+ such that the equa¬
tion x + y = 5z has no monochromatic solution.

6. A theorem of Hilbert

In this section, at first we give a proof of a theorem of Hilbert by

topological methods. We shall not have much scope to dwell on the
application of dynamical systems to Ramsey Theory, an interesting
area which is very active today (see [66] and [22] for instance, fur¬
ther references will be given in Chapter 4). However, in this section,
one can get a glimpse of that.
We shall also sketch the original proof of Hilbert at the end of
this section.

We go through a sequence of definitions and observations before

we come to the theorem of Hilbert.
 6. A THEOREM OF HILBERT 15

Definition 6.1. Let T be a continuous map of a topological

space into itself. Then a point x £ X is called a recurrent point
for T if for any neighbourhood V of x, 3 n > 1 with Tn(x) £ V.
Now, let X be a compact topological space and T : X -» X
a continuous map. Let T denote the family of nonempty closed
subsets of X invariant under T.
Ordering by inclusion, we observe that because of compactness
of A, by the finite intersection property, the intersection of a totally
ordered chain in X belongs to T. Hence, by Zorn’s lemma, X has
a minimal element T0.
We claim that each point of T0 is a recurrent point for T.
Let y E Y0 and Y={Tn(y) : n > 1}, the forward orbit closure of
y. Now, Yq being T invariant, {Tn(y) : n > 1} C To and therefore,
T0 being closed, {Tn(y) : n > 1} C T0. But by definition, T is
nonempty and closed. Further, since T is continuous, T is invariant
under T. Therefore, T £ T and therefore by minimality Y — Yq.
Hence y £ T which means that each neighbourhood of y contains
Tn(y) for some n > 1.
Thus in particular, we have
Proposition 6.1. For a continuous map T from a compact
topological space X into itself, the set of recurrent points for T is
nonempty.
Let 0. be the collection of sequences made of finitely many sym¬
bols. In other words, if we denote the finite set of symbols by A,
then D consists of all maps / : Z+ —> A.
For a, b £ A the metric d defined below gives the discrete topol¬
ogy on A.

The space Q — Az+ with the product topology is metrizable. If

for ago/ £ fi one defines

OO
d (cj(n),ca'(n))
D{u,u') -
n— 1 2n
then it is easy to verify that the metric D corresponds to the
product topology on D. By Tychonoff’s theorem, (Q, D) is therefore
compact.
16 1. CLASSICAL RAMSEY-TYPE THEOREMS

Also the semigroup Z+ acts on the elements of A in the obvious

way by shifting. That is by the action of n 6 Z+, an element
cd G fi goes to u' G n where uj'(m) = uj(m + n). The map on Q
corresponding to n = 1 (which in fact determines the action of the
semigroup Z+) will be called the shift map and we shall denote it
by a. The map a : Q —» D is continuous.
The space D endowed with the metric D and the Z+ action is
a symbolic flow in the terminology of Dynamical Systems. In a
symbolic flow, by saying that a point is recurrent one means that it
is recurrent for the shift map.
A is sometimes called the alphabet and a finite sequence of ele¬
ments in A is called a word. Since it is clear that two points u,u' G D
are close if they agree on a large block of numbers (1, 2, • • • , N), we
have

Proposition 6.2. In a symbolic flow, a sequence la £ Q. is

recurrent if and only if every word occurring in u occurs a second
time.

Remark 6.1. We further note that a most general recurrent

point ui will look like

uj — [(aw(1)a)a;(2)(acc(1)a)]cc(3)[(aci;(1)a)a;(2)(aw(1)a)] • • • (1)

where a = cu(l) G A and cA1) ’§ are arbitrary words composed of

elements of A.

Theorem 6.3. (Hilbert) Given a finite colouring on Z+ and

a positive integer l, one can find l elements mx < m2 <■■■< mi in
Z+ such that if P(mx, ■ • • ,m{) denotes the set of sums ]T-=1 c(i)mi}
c(i) = 0 or 1, then infinitely many translates of P{mx, • • • , mfl are
of the same colour.

Proof. Let X '■ Z+ —> {ci, c2, • • • , cq} be a (/-colouring on Z+.

Let A = {1, 2, • • • , q} and D be defined as above. We consider the
element (G 2 where

f(n) = i <*=» n G x_1 (cj), i = 1,2 ■ • ■ q.

 n
6. A THEOREM OF HILBERT 17

Case I. (£ is a recurrent point)

Let £ has the form given in (1) and

Wo = a
Wi = Woo/^Wo
W2 = <jJ\

^n ' Wn_jW^

Now, denoting the length of the word wncc/n+1) by ran+1, we see

that if some symbol occurs at position p in wn, then it occurs at
positions p and p + mn+1 in wn+1 = wnw(n+1)wn. Thus the symbol
a occurs at positions 1,1 + ml51 -f m2,1 + m\ + m2 and in general
at positions belonging to 1 + P{m\, ■ ■ - mi) for any l. Since every
finite configuration occurs infinitely often in £, it is now clear that
X_1(a) contains infinitely many translates of P(mi, • • -m<).

Case II. (£ is not a recurrent point)

In this case we consider the forward orbit closure X of £ in
Q. The shift operator takes points of Ar to X and therefore by
Proposition 6.1 there is a recurrent point say w for the shift operator
a in X. Let w be of the form given in (1). Therefore, there exists a
sequence of positive integers {n^} such that crnfc(£) w. If a is the
leading symbol in w, then arguing as before, a occurs at positions
belonging to 1 + P(ml5 • • - mi). Choose k such that crn'c(£) agrees
with w for (l+mi+- • --Tra/) terms. Then £(n,fc+p) = a whenever p €
(1 + P(mi, • • • mi)). One can assume that nk —>• oo. For, otherwise, a
finite translate of £ would be recurrent and one could then invoke the
first case. Therefore we obtain that 1 +nk + P(mi, • • • mi) G X-1 (a)
for a sequence {n^} where rik —> oo and this proves the theorem. □

We now give a sketch of Hilbert’s original proof [99] of The¬

orem 6.3 which consists of repeated application of the pigeonhole
principle.

Proof. Let x '■ z+ {ci, c2, • • • , cq} be a ^-colouring of Z+.

If we consider the q + 1 integers 1, 2, • • • , q + 1, then by pigeonhole
principle, one among the symbols ci,c2,-- - ,c9 (say h) will be re¬
peated twice in the word y(l)x(2) • • • x Q { + !)• Therefore, this word
will have one of the following types of words as a subword:
18 1. CLASSICAL RAMSEY-TYPE THEOREMS

H\ = hh
H\ = h.h

H. h---h

where H\ is an word of length s with h as its first and last elements.

We observe from above that there are q types for each of the
ci1 s. So as h varies over the ct’s there are q2 types. So if we consider
the word x(l)*(2) • • - x((<7 + l)3) of length {q + l)3, then for some
u( 1) e {2, • • • ,q + 1}, one of the following types of subwords will
always appear.

#0) • • ‘ #(i)

where H2 is an word of length s with Hu^ at the beginning and

the end.

We proceed as before and repeat the process l times to conclude

that for a sufficiently large integer n (depending only on l and q),
the word x(l)x(2) • • • xin) will contain a subword of the type H[^V) —

where ‘ and so on- Here

H[^ is a word of length i(l), a word of length i(l — 1) etc.

Since there are finite number of such types, and one occurs in
every length n, one of them will be repeated infinitely many times.
If H[^ as described above is the one which is repeated infinitely
many times, then writing rri\ — i(l) — 1, m2 = i(2) — z(l), • • ■ , m; =
i(l) — i(l — 1), it is easy to observe that infinitely many translates
of P(mi, ■ • • ,ra/) are of the same colour. □
 7. NOTES 19

7. Notes

M. J. Erickson [60] gives several versions of the pigeonhole prin¬

ciple including Theorem 2.1 and derives some beautiful consequences
as well.

The theorem of Erdos and Szekeres was suggested by the curious

observation made by Esther Klein (later to become Mrs. Szekeres)
that from five points of the plane of which no three lie on the same
straight line, it is always possible to select four points determin¬
ing a convex quadrilateral. The classical paper [58] of Erdos and
Szekeres had a second proof of their theorem using the monotone
subsequence argument. For the one sketched in Section 3, they had
to independently rediscover a finite version of Ramsey’s theorem as
the paper of Ramsey which had been published some years before,
was still little known.
Actually, it is only after the result of Ramsey was rediscovered
by Erdos and Szekeres, the branch of combinatorics called Ramsey
Theory grew up and with hindsight we can now see the unifying
feature of the early Ramsey-type theorems which are seemingly un¬
related. It is therefore quite justified that results which had been
proved much before the paper of Ramsey saw the light of the day,
belong to a subject named after Ramsey. Commenting on Ramsey’s
being called as ‘eponymous’ by Mellor, Harary in his tribute [87]
says:
“ Of course! The study of ramsey theory has become so im¬
portant that his name has become an adjective, along with other
immortal mathematical lower case adjectives, including abelian,
boolean, cartesian, ....”
Frank P. Ramsey was mainly interested in philosophy and made
remarkable contributions to mathematical economics. In this con¬
nection, Mellor [116] says:
“...his (Ramsey’s) other achievements- several also eponymous
- were no less remarkable, ....”
One may look into the article [116] of Mellor, referred above for
an account of Ramsey’s contribution to subjects other than math¬
ematics. An article [163] of Joel Spencer appearing in the same
volume of Journal of Graph Theory gives ‘an anecdotal account
of some of the events and people that have helped shape Ramsey
Theory’.
20 1. CLASSICAL RAMSEY-TYPE THEOREMS

Regarding bounds for classical Ramsey numbers, from a recur¬

sion formula (see [83]) one can derive

*<*■')* (**-T2)- (2)

where R(k, l) is as defined in Remark 3.2.
By a constructive method, Greenwood and Gleason [85] deter¬
mined exact values of R(k, l) for certain small values of k and l. Few
other known exact values and good bounds are recorded in [41] and
[78]-
For the number R(3,k), Erdos [47] proved by probabilistic ar¬
guments that

R(3'
for some absolute constant c > 0.
In the other direction, the upper bound of Graver and Yackel
[84] has been improved by Ajtai, Kolmos and Szemeredi in a very
important paper [8]. The latters obtained

Apart from the specific cases mentioned above, for general R(k, /),
the estimate given by (2) has been improved appreciably by Rodl
and Thomason independently. Thomason’s result [174], which is
stronger than that of Rodl, is as follows:

R{k, l) < fc(-? Wi°s* > 1 ~ 2),

for some absolute constant A > 0 and all k > l > 2.

For further information regarding bounds of Ramsey numbers,

one may look into [83], [41] and [81]. For some results and open
questions related to certain generalized Ramsey numbers, one may
look into [36].

Apart from the two proofs supplied by us, at least one more way
of using Ramsey’s theorem to derive Theorem 3.3 is available in the
literature. That is due to M. Tarsy (see [83]).
 7. NOTES 21

There are several natural ways in which Ramsey’s theorem gen¬

eralizes to the infinite version. For a nice short account, one may
look into [37], For a detailed account of Canonical Ramsey-theorems,
one may consult [83] or look into the original paper of Erdos and
Rado [57]. The project which originated in this paper of Erdos
and Rado, led to the development of Partition Calculus of cardinal
numbers, an account of which can be found in the book written by
Erdos, Hajnal, Mate and Rado [56].
Regarding van der Waerden’s theorem (Theorem 4.1), we rec¬
ommend van der Waerden’s personal account [177] of finding its
proof. It contains the formulation of the problem with the valu¬
able suggestions due to Emil Artin and Otto Schreier and depicts
how the sequence of basic ideas occurred as an elaboration of the
psychology of invention. The result was originally (see [83] for in¬
stance) conjectured by Schur; since van der Waerden came to know
it through Baudet, he calls it Baudet’s conjecture.
For a proof of Theorem 4.1, utilizing the methods of topologi¬
cal dynamics (originally given by Furstenberg and Weiss [71]) one
may look into [83] or [111]. A topological dynamics version of
Griinwald’s Theorem (Theorem 4.2) can be found in [115].
The original proof of Schur ([158]) of Theorem 5.1 was effective.
More precisely, if we define the Schur Number S(r) to be the least
integer such that for any r-colouring x of the set {1,2, • • • , S(r)},
there exist integers x,y such that x(t) = x(y) — x(x + y)> then
Schur’s proof yielded S(r) < [er!], where [er!] denotes the largest
integer < er\. By the proof given here in Section 5, bounds for
N = R(2, 3, r) provide bounds for S{r) and one gets slight improve¬
ment on the bound [er!]. For even r, a recursive bound yields the
following best known result due to Honghui Wan [179]:

For even r > 6,

In the other direction, the lower bound for S(r) obtained by

Schur was later improved:
(I) (Abbott and Hanson [1])

S(r) > c(89)r/4,

22 1. CLASSICAL RAMSEY-TYPE THEOREMS

for some absolute constant c > 0.

(II) (Fredericksen [64]) For r > 5,

S(r) > c(315)r/5,

for some absolute constant c > 0.

Proofs for both of these results depend upon a recurrence rela¬
tion due to Abbott and Hanson [1]. Some information related to the
above bounds, as well as a proof in the line of the original one due
to Schur (with the effective bound mentioned above), is available in
[161].

In the dynamical proof of Theorem 6.3 we have closely followed

the treatment in [66], Quantitative forms of this theorem are ob¬
tained in [34].
 CHAPTER 2

van der Waerden revisited

1. Introduction

The theorem of van der Waerden as stated in Theorem 4.1 of

Chapter 1, was a statement dealing with integers. In Section 2 of
this chapter, we are going to discuss about Hales-Jewett theorem
[89]. This theorem of Hales and Jewett, revealing the combinatorial
nature of van der Waerden’s theorem would claim that this ‘pearl
of number theory’ belonged to the ancient shore of combinatorics.
Perhaps nothing can better describe the role of this theorem as has
been done in the following statement in [83]:

“the Hales-Jewett theorem strips van der Waerden’s theorem of

its unessential elements and reveals the heart of Ramsey theory”.

We present S. Shelah’s proof [160] of Hales-Jewett theorem,

which, apart from having some fundamentally new ideas, improved
upon the previously known upper bounds for the Hales-Jewett and
van der Waerden functions in a significant way. For the proof, we
closely follow the exposition in [83].

However, on the other hand, van der Waerden’s theorem was a

prelude to a theme which essentially belongs to the realm of Number
Theory. The development of this theme, culminating in the theorem
of Szemeredi, followed by the ergodic proof of Szemeredi’s theorem
due to Furstenberg and holding some yet unanswered questions to
its bosom will be briefly surveyed in Section 4.

23
24 2. VAN DER WAERDEN REVISITED

2. Hales-Jewett theorem

Let
Ct = {xxx2---xn : Xi E {1,2, •••<}}

be the collection of words of length n over the alphabet of t- symbols

Definition 2.1. A subset L c C” will be called a Shelah line

if \L\ — t and we can write L — {ii,i2,” • >M with 4 = Xki • • • Tjtn
where for some fixed z and j, 0 < i < j < n,

' t -1 if s < i
%ks k if i < s < j
> t if j < s.

Among themselves, the different words 4 in L differ only in the

positions {s : i < s < j}. These positions will be referred to as the
moving co-ordinates of the Shelah line.

Definition 2.2. A word x\ •• -xn E C(n will be called a Shelah

point if it belongs to some Shelah line.

Definition 2.3. Let ni,n2, , ns be given positive integers

and n = Yli=ini- Let Lj be a Shelah line of C™J for 1 < j < s.
Then we call L\ x L2 x • • • x Ls a Shelah s-space of C(n.

Definition 2.4. If a3 is the value of the moving co-ordinates

in the j-th block of f E Lx x L x
2 • • • x Ls, then

(f) : L\ x L2 x • • • x Ls —> Cl

defined by </>(£) = ai • ■ • is the canonical isomorphism between

L\ x L2 x • • • x Ls and C(s.

Definition 2.5. A colouring x of Cf will be called fliptop if it

has the property that given any two points P, Q of Cst such that
they differ exactly at one place where one of them takes the value
t — 1 and the other one the value t, P and Q are coloured identically,
that is, x{P) = X(Q)-
Definition 2.6. Let <j> : Lx x L2 x • • • x Ls —> Q be the
canonical isomorphism (see Definition 2.4). Then a colouring x of
the Shelah s-space L\ x L2 x • • • x Ls will be called fliptop if the
colouring x'(P) d= x(</>_1(P)) of Cst is fliptop according to Defini¬
tion 2.5 above.
 2. HALES-JEWETT THEOREM 25

Remark 2.1. We observe that a colouring of a Shelah line will

be fliptop according to Definition 2.6 above if and only if the last
two points have the same colouring.
Remark 2.2. If a colouring x °f L\ x L2 x • • • x Ls is fliptop,
sometimes, by abuse of language we say that L\ x L2 x • • • x Ls
is fliptop under x or just that L\ x Z,2 x • • • x Ls is fliptop if the
colouring is well understood.
We need some lemmas.
Lemma 2.1. Let C(n be c-coloured arbitrarily for some c < n.
Then there exists a fliptop Shelah line.

Proof. For 0 < i < n, we define Pt to be the word xu ■ ■ ■ Xin

where
t — 1 if j < i
Xij —
t if j > i.

Since n + 1 > c, by Pigeonhole principle, there are /, g €

{0,1, • • • , n}, / < g, such that Pf and Pg have the same colour.
Now, we consider the Shelah line {^1, ^2, • • • , G} with

@k — %k\ ' ' ' ^kn

where
ft — 1 if s < f
xks = < k if f < s < g
{ t if g < s.

Since the last two points on this Shelah line are Pf and Pg which
have the same colour in the given c-colouring, by Remark 2.1 we
are through. ^

Lemma 2.2. Given three positive integers r, s and t, we define

integers n1; n2, • ■ • , ns inductively as follows. We take
«s_1
ni = r ,
n2 = rv 2 '

and in general, for i > l, after having defined nu n2, ■ ■ • ,nit

S— 1
nl+i — rA% where Al = x t

Let n = nx + n2 H-hns and suppose an arbitrary r-colouring x

of C? has been given. Then there is a fliptop Shelah s-space of Q1.
26 2. VAN DER WAERDEN REVISITED

Proof. We write a point y G C" as y — yx • • • ys where y3 G

Ctj.
On Ct% we define an equivalence relation =a as follows:
For ys and y's in Ctn% we say

ys =s y's iff x{Vi ■ ■ ■ Vs-iVs) - x(yi *' * Va-M

for all Shelah points yi, ■ • • , ys-\ where y3 G C?J.

For each j, 1 < j < s — 1, there are at most choices

for 2/j.
Hence there are at most f]j=i aj = -^s-i choices for yi • • -ys_i
and therefore there are at most ns — rAs~x equivalence classes. Thus
the equivalence relation gives rise to an ns-colouring of C?s. By
Lemma 2.1, there exists a Shelah line Ls C Ctns, fliptop under Xs-
We proceed by induction to find Ls_i, Ls_2, • • • as follows.
Once we have found Ls, Ls_i, • • • , L/+i, we define an equivalence
relation =/ on C?‘, by setting

yi =/ y[ iff xiVi • • • - • ■ zs) = x(yi • • • * • • ,*«),

for all Shelah points yi, • • • ,yi-i where y3 G Ctnj and all choices of
2:^ G Tj, / ~t~ 1 < 7 ^ S.

Each 2/j, 1 < j < l — 1, can be chosen in at most a3 = (nj2+1)f

ways as before and since the lines L/+1,--- , Ls are already fixed,
each zj can be chosen in t ways. Hence there are at most (rijTi •
ts~l — Ai—i choices for yx ■ ■ -yi-\, Zi+1, • • • ,zs and therefore there
are at most n/ = r4'-1 equivalence classes. Thus the equivalence
relation =/ gives rise to an rq- colouring xi of Ctn'. Therefore, again
by Lemma 2.1, there exists a Shelah line Li C Cfn', fliptop under xi-
If for a fixed z, 1 < i < s, yt and y\ are the last two points of
then since by construction, Lt is fliptop under Xi we have

Vi =i y[

=> x{y\ ■ ■ ■ yi-iyiZi+i, • • • zs) = x(yi • • • yi-iy'^+i, ■ ■ ■, zs)

for all Shelah points yi, • • • , yt-\ and all z3 G L3, i + 1 < j < s.
Therefore, in particular considering Shelah points z3 G for
1 < j < i, we have

X(~l ’ ' ' —1 yi^i+l i ' ’ ' zs) — X(^l ■ • ■ ^-iy^i+1, • ■ • , -Zs)

for G Lj, for 1 < j < s, j ^ i.

 2. HALES-JEWETT THEOREM 27

This shows that the Shelah s-space L\ x • • • x Ls is fliptop under

X- □

Definition 2.7. By a combinatorial line in C” we mean a set of

t points in Ctn ordered as X\, X2, • • • , Xt where Xi = xnxl2 ■ ■ -xin
such that for j belonging to a nonempty subset / of {1, • • - n} we
have xSj = s for 1 < s < t and X\j — ■ • • = xtJ — Cj for some
Cj E {1, • • -t} for j belonging to the complement (possibly empty)
of I in {1, • • • n}.

Remark 2.3. For t = 3 and n = 5, the following is an example

of a combinatorial line in Cf:

12123
22223
32323

We see that the collection of words in Cf are in one-to-one

correspondence in the obvious way with a subset of the integers
1,2,*-- ,33333 where the integers have their usual expression in
decimal system, that is, with base 10. Thus, the three words in the
above combinatorial line correspond to an arithmetic progression
with common difference 10100.
Lemma 2.3. If there exists a positive integer s = HJ{r, t — 1)
such that for any r-colouring ofCfll, there exists a monochromatic
combinatorial line, then under any fliptop r-colouring of C*, there
exists a monochromatic combinatorial line.

Proof. Let the condition in the lemma hold. Let x be any

fliptop r-colouring of Cfl We consider the restriction of X on Cf_x
contained in C* consisting of the words which involves the symbols
from the set {1,2, •• • , (t — 1)} only. Then there is a monochromatic
combinatorial line {I\, • • • , It-1}- Now, ■ • • , It-iflt} is a
combinatorial line in Cf where lt is obtained from lt_i by putting
t in place of (t — 1) in the moving co-ordinates of the combinato¬
rial line. Now, £t can be obtained from £t_i by a finite sequence
of changes, each change involving a replacement of t - 1 by t in
exactly one place. Now, the colouring being fliptop, each such
change preserves the colour and thus it and lt_\ have the same
colour. Since {^,4, • • • flt-1} have the same colour to start with,
{^i, i2, • • • , it-\, It} f°rm a monochromatic combinatorial line. □
28 2. VAN DER WAERDEN REVISITED

Theorem 2.4. ( Hales-Jewett Theorem ) Given any two

positive integers r and t, there exists n = HJ(r,t) such that if Cf
is r-coloured then there exists a monochromatic combinatorial line.

Proof. The proof is by induction on t. Since Cf is a singleton,

HJ(r, 1) can be taken as 1.
Now, suppose s = HJ(r, t — 1) exists. Taking n as in Lemma 2.2
corresponding to the integers r, s and t here, given any r- colouring
of Cf there exists a fliptop Shelah s-space Li x • • • x Ls. Let x' be
the colouring on C/ defined by x'iv) = x{4,~1(y)) where <j> : Lx x
• • • x Ls —* Cf is the canonical isomorphism. Then x' is fliptop and
because of the induction hypothesis we can now apply Lemma 2.3
to conclude that there exists a monochromatic combinatorial line
L C Cl. Now, ) C Li x- • -xLs is the required monochromatic
combinatorial line in Cf. □

Remark 2.4. Keeping Remark 2.3 in mind, it is easy to see

that if we consider an r-colouring of Cf (with suitably large t),
induced from a given r-colouring of the integers which have base
d representation with d > t, Theorem 2.4 would imply the exis¬
tence of a monochromatic arithmetic progression of t terms. Thus,
Theorem 2.4 implies van der Waerden’s theorem.
Exercise 2.1. Show that Theorem 2.4 implies Griinwald’s the¬
orem (Theorem 4.2 of Chapter 1).

Exercise 2.2. Show that for any finite semigroup 5, there exists
a positive integer n and an element x € 5 such that the set {ns + x :
s E 5} is a singleton.

3. Some variations of van der Waerden’s theorem

In this section, we shall see some variations of van der Waer¬

den’s theorem. We shall also discuss about some implications of
the theorem.

First we state the following strengthening of van der Waerden’s

theorem (Theorem 4.1 of Chapter 1); this will be useful in the next
chapter.

Theorem 3.1. Gwen k, r, s e Z+, there exists N = N{k, r, s) e

Z+ such that for any r-colouring of [AT], there are a,d G Z+ such
 3.1S0ME VARIATIONS OF VAN DER WAERDEN’S THEOREM 29

that the set

{a, a + d, • • • , a + kd} U {sd} C [N]

is monochromatic.

Proof. The proof is by induction on r. When r = 1, N can be

just chosen to be equal to the larger one of the two numbers k + 1
and s. We assume the result to be true for r — 1 colourings (with
any k and s).

Now, for given k and s we have to prove the result for the triple
k, r, s.
Let m - N(k, r — 1, s).
By van der Waerden’s theorem, there is a positive integer W(km+
l, r) such that for any r-colouring of [W(km + l,r)j, there is a
monochromatic arithmetic progression of length km + 1. We claim
that the number sW(km + 1, r) can be taken as N(h, r, s).
Given any r-colouring of [sW(km + l,r)], there is a monochro¬
matic arithmetic progression

{a, a + b', • • ’ ,a + kmb'} C [W (km -I- 1, r)].

If for some j, 1 < j < m, sb'j happens to be of the same colour

as the elements of the above monochromatic progression, then we
are through with d = jb'. Otherwise, the set {sb'j|1 < j < m} is
(r — l)-coloured and by the correspondence between the sets [m]
and s[m] through multiplication by s and the induction hypothesis,
we are through. □

Remark 3.1. We should observe that Theorem 3.1, of which

the statement is stronger than that of van der Waerden’s theorem,
is a strengthening of Schur’s theorem (Theorem 5.1 of Chapter 1) as
well. More precisely, taking k — 1 and s — 1 in the above theorem,
we get a monochromatic set {a, a + d, d) for any r-colouring of the
set [N(l,r, 1)].
In the next theorem, we give a proof of yet another variation of
van der Warden’s theorem, following T. C. Brown [33].
THEOREM 3.2. The following statement is equivalent to van der
Waerden’s theorem as stated in Theorem 4-1 of Chapter 1.
S(k,m): Given any two positive integers k and m, there is a
positive integer p = p(k,m) such that for any p positive integers
30 2. VAN DER WAERDEN REVISITED

ax < a2 < ■ ■ ■ < ap with the property that the difference between any
two consecutive elements at and al+1 satisfies al+x — a* < m, the set
ax, a2, • • • , ap contains an arithmetic progression of length k.

Proof. First, we prove that Theorem 4.1 of Chapter 1 implies

the statement S(k,m).
Let two positive integers m and k be given. We take

p — W(k,m) — (m — 1),

where W(k,m) is the van der Waerden number.

Now, consider a set A0 = {ax,a2,--- ,ap} of positive integers,
where ax < a2 < ■ ■ • < ap and al+x — a, < m, for 1 < i < p — 1.
We shall show that A0 contains an arithmetic progression of
length k.
We define the following sets:

A\ — {ai + 1, a2 + 1, • • • , ap + 1} \ Ao
A2 — {ai + 2, a2 + 2, • • • , ap + 2} \ (Aq U Aj)

Am-1 - {ai + m — 1, a2 + m — 1, • • • ,ap + m— 1}

\(J40 U dj U • • ■ v4m_2).

Now, we have got a partition

(ai, ai + 1) + 2, • • • ,ap + m— 1} — Aq U A\ U • • • Am_i

of the set {ai, ax + 1, ax + 2, • • • , ap + m — 1} into m parts.

Since

ap + m — 1 — (ax — 1) = ap + m — ax

> (p + ax — 1) + m — ax
= W(k,m),

at least one of the sets among • • • ,Am_\, say Aj, contains

an arithmetic progression of length k.
 3.''SOME VARIATIONS OF VAN DER WAERDEN’S theorem 31

But,

Ai C {a! + j, a2 -f j, • • • , ap + j},

where {a! + j, a2 + j, ■ ■ ■ , ap + j} is a translation of the set A0 =

{al5a2, ••• , ap}. This shows that A0 contains an arithmetic pro¬
gression of length k.

Now we suppose that the statement S(k,m) holds for any two
given positive integers k and m. We have to show that W(k,r)
exists for positive integers k and r. We fix k and demonstrate the
existence of W(k,r) by induction on r. The case r = 1 is trivial.

We assume that W(k,r) exists and establish that W(k,r + 1)

exists. At this point, we take the help of Exercise 4.2 of Chapter
1. Namely, we establish the statement V(k,r + 1) of that Exercise,
which would then imply the existence of W(k,r + 1).
Suppose the set Z+ of positive integers is partitioned into r + 1
subsets:
Z+ = Xi U Ar2 U • • -\JXr+i.

At least one of the disjoint sets A^l < i < r + 1, has to be

infinite. Without loss of generality, assume that the set Xr+i is
infinite. Let
AC+i = {ai, a2, 03, • • • }.

If there is a positive integer m such that al+1 — at < m, for all

i > 1, then by S(k,m), Xr+\ contains an arithmetic progression of
k terms.
On the other hand, if there is no such integer m, then there are
arbitrarily large sets of consecutive positive integers which will be
contained in X\ U X2 U • • • U Xr. Therefore, by existence of W(k,r),
one of the sets X{, 1 < i < r will contain an arithmetic progression
of k terms, thus establishing the statement V(k,r + 1). □

Exercise 3.1. Using van der Waerden’s theorem, prove that

given any two positive integers r and k, there is a positive integer
n = n(r, k) such that if [n] is r coloured, there are positive integers
ax < a2 < ■ ■ • < ak satisfying Ei<i<k O < n such that for any subset
I of {1,2, ••• ,ifc}, the colour of the element Eiei ai depends only
on max(7).
32 2. VAN DER WAERDEN REVISITED

4. Notes

For information regarding the remarkable way Shelah’s proof

[160] improves the upper bounds for the van der Waerden number
W(k,r), the Hales-Jewett number HJ(r,t), etc., and brings down
the estimate of their growth rates in the Ackermann hierarchy, one
may consult [83] (second edition), apart from Shelah’s original pa¬
per [160].

A remarkable extension of Hales-Jewett theorem was obtained

by Graham and Rothschild [82]. Their result is a partition theorem
for k-parameter sets, where, roughly speaking, parameter sets are
higher dimensional analogues of combinatorial lines.

Thus, O-parameter sets are elements of Cf (that is, n-tuples)

and 1 -parameter sets are the combinatorial lines occurring in Hales-
Jewett theorem.

Graham-Rothschild theorem for parameter sets generalizes the

Hales-Jewett theorem in the same way Ramsey’s theorem general¬
izes the pigeonhole principle. This result of Graham-Rothschild,
which Nesetril [122] calls the ‘first theorem of the new Ramsey the¬
ory age’, provides a natural framework for working with seemingly
different structures.

Besides many other applications, Graham-Rothschild theorem

for parameter sets led to the solution of Rota’s conjecture about es¬
tablishing the analogue of Ramsey’s theorem for finite vector spaces.
More precisely, Graham, Leeb and Rothschild [80] proved the fol¬
lowing theorem:

For a finite field F, given positive integers r, n, t, there exists a

positive integer N = GLR(r, fc, t) such that for every r-colouring of
the set of all t-dimensional vector subspaces of an N-dimensional
vector space V over F, there exists a k-dimensional vector subspace
all of whose t-dimensional vector subspaces have the same colour.

A simplified proof of the above result was later supplied by

Spencer [162]. This simplified version is also given in [83].

For further information, apart from the very good article of

Nesetril [122] already mentioned, one may look into an article of
Promel and Voigt [135] which gives a nice survey of the development
related to the structure of Graham-Rothschild parameter sets.
 4. NOTES 33

The theorem of van der Waerden gave rise to several questions

and conjectures and a great amount of research work followed. In
the thirties, Erdos and Turan [59] conjectured that for a subset of
the set of positive integers, the property of possessing arithmetic
progressions of arbitrary length actually depends on the ‘size’ of
the set.
More precisely, they conjectured that any subset of Z+ with
positive upper natural density will have the property. For A € Z+,
the upper natural density d(A) of A is defined by

l^n[jV]|
d(A) — lim sup
N—too N

One can observe that in this regard, the situation is quite differ¬
ent in the case of Schur’s theorem (Theorem 5.1 of Chapter 1). For,
even though the set of odd integers and the set of even integers have
the same upper natural density (namely, 1), the set of odd integers
do not possess any solution to the equation x + y = z.

The first progress towards the Erdos-Turan conjecture was made

by K. F. Roth [151], who proved that any subset A of the set Z+
of positive integers with positive upper natural density will always
contain a three-term arithmetic progression.

After that Szemeredi first improved [166] Roth’s result to that

of A possessing a four-term arithmetic progression. Later in 1974,
in a famous paper [167] Szemeredi proved the general Erdos-Turan
conjecture by combinatorial method.

Though the above paper [167] of Szemeredi is very difficult,

one can get some flavour of it from the proof of Roth’s result by
Szemeredi’s method as given in [78].

Later, Furstenberg [67] gave an ergodic theoretic proof of Sze¬

meredi’s theorem which opened up the subject of Ergodic Ramsey
Theory. In the notes in Chapter 4, we shall mention some recent
developments in this area and supply further references.

The following conjecture of Erdos is still open.

Conjecture 4.1 (Erdos) If A c Z+ satisfies

OO,
£-
Tj a
a£A

then A contains arithmetic progressions of arbitrary length.

34 2. VAN DER WAERDEN REVISITED

Erdos offered (see [51] for instance) $5000 for a proof (or dis¬
proof) of the above conjecture.

If the above conjecture is proved, we shall have in particular

that the set of prime numbers contains arithmetic progressions of
arbitrary lengths. Even though experts in the field differ in their
opinions regarding the veracity of the above conjecture, all of them
believe that the result is true for the case of the set of primes.

It should be noted that regarding the above conjecture, for A C

Z+ satisfying J2a&A ~ ~ 00> ^ is n°t even known whether such a set
A will necessarily contain an arithmetic progression of length three.

However, in the case of primes it is known that the set of primes

contains infinitely many triples in arithmetic progression. Specific
references for this result are available in [96]. The best known
result in this direction is due to Heath-Brown [96], who proves
the existence of infinitely many quadruples a, b, c, d in arithmetic
progression, where a, 6, c are primes and d is an almost-prime (that
is, d is either a prime or a product of two primes).

In 1993, the following sequence of primes in arithmetic progres¬

sion consisting of 22 terms was found [134] (see also Ribenboim
[149] ):

11410337850553 + 4609098694200/c, 0 < k < 21.

In [77], along with Conjecture 4.1 above, another related conjec¬

ture is mentioned. This conjecture, which is expected to be easier
than Conjecture 4.1, is as follows:

Conjecture 4.2 If A c (Z+)2 satisfies

1
£
0d)eA i2 + j2
= oo,

then A contains 4 vertices of a square.

Defining rk{n) to be the smallest integer such that whenever

A C [n] satisfies |.4| > rk(n), .4 contains an arithmetic progression
of k terms, Szemeredi’s result [167] implies that

Dt(n) = o(n).
 4. NOTES 35

For the case k — 3, Roth’s proof [151] gave

This was later improved by Heath-Brown [97]:

t3(n) = O -— , for some absolute constant c > 0.

\ I(log
ncr n)
n lc /

Szemeredi [168] showed that c = | is admissible in the above

estimate.
Recently, Bourgain [31] has proved:

In the other direction, it was proved by Salem and Spencer [155]

(see also Behrend [20]) that

r3(n) > ne cVlozn for some absolute constant C > 0.

At this point, we should mention a related paper of Chung and

Goldwasser [40] completely describing the maximal sets Sn and Tn
of [n] containing no solution of x+y = 2 and x+y = 3z, respectively.
Regarding estimates of rk(n) for k > 3, Gowers [74] has made a
remarkable breakthrough while establishing

for some absolute constant d > 0

where the method seems to go through for rk(n) for k > 4.

Apart from the original paper [74], we recommend two beautiful
articles [75] and [31] for getting an idea as well as the background
of the proof. Both these articles contain some open problems as
well.
'
 CHAPTER 3

Generalizations of Schur’s theorem

1. Introduction

In this chapter, we shall talk about some generalizations of

Schur’s theorem (Theorem 5.1 of Chapter 1). These include Rado’s
theorem and Folkman’s theorem.
We shall take up Rado’s theorem in Section 2.
Successful investigations of Rado ([138], [139], [140]) provided
necessary and sufficient conditions for a system of homogeneous
linear equations over Z to possess monochromatic solutions in any
finite colouring of Z+.
We shall give the proof only for the special case of a single equa¬
tion which already is a generalization of Schur’s theorem. However,
we shall give the statement of the general case, from which, as one
will see, van der Waerden’s theorem too follows as an easy corollary.
In Section 3, we shall take up Folkman’s theorem, which is yet
another generalization of Schur’s theorem. We shall also discuss
about results related to Folkman’s theorem.

2. Rado’s theorem

Schur’s Theorem (Theorem 5.1 of Chapter 1) says that for any

finite colouring of Z+, 3 a monochromatic subset {x,y,z} of Z+
satisfying the equation
x + y = z.
Again, for any finite colouring of Z+, monochromatic arithmetic
progressions of arbitrary length are guaranteed to exist by van der
Waerden’s Theorem (Theorem 4.1 of Chapter 1). In particular,
37
38 3. GENERALIZATIONS OF SCHUR’S THEOREM

for any finite colouring of Z+, the existence of a monochromatic

arithmetic progression of three terms would imply the existence
of a monochromatic subset {x, y, z} (with x, y, z distinct) of Z+
satisfying the equation
x + z — 2y.
These results gives rise to the natural question about the exis¬
tence of monochromatic solutions of the general linear homogeneous
equation

C\X\ -4-f cnxn = 0, Ci(± 0) G Z, (3)

corresponding to finite colourings of Z+.

The following result gives a necessary condition for this.
Theorem 2.1. If for no non-empty subset I C {1, • • • , n}, the
number ci is zero, then 3 a finite colouring on Z+ such that
equation (3) does not have a monochromatic solution in 7r.

Proof. Let p > 0 be a prime.

Now, any q G Q* can be uniquely expressed as
ry) a
q = ——, j G Z, a G Z, b G Z+ ,p J(a,p / b and gcd (a, b) — 1.
b
We define the so-called super modulo colour Sp on Q* by:

SP(q) = ^ (mod p) G F*.

Then Sp is a (p — l)-colouring of Q* with the property that

Sp(x) = Sp{y) => Sp(ax) = Sp(ay) V a G Q*.

We claim that if for no non-empty subset / of {1, • • • , n}, ct

= 0 (mod p), then equation (3) does not have a monochromatic
solution in Z with the colour Sp.
Once this claim is established, Theorem 2.1 will be proved. To
observe this we have just to see that if the condition in the theorem
is satisfied, then {ICze/ c* : T I C [n]} is a set consisting of finitely
many non-zero integers and therefore some prime p will not divide
EieiCi for any (<f> fi)I C [n].
Assume that for no non-empty subset / of {1, - - - ,n}, Yha ci
= 0 (mod p) and if possible, let aq, • • • , xn G Z+ be a monochro¬
matic solution (w.r.t. Sp) of equation (3). Now, for any p G Q*,
px\,- ■ ■ , pxn is also a monochromatic solution and hence without
loss of generality we can assume that gcd (x1; • • • , xn) = 1.
 2. RADO’S THEOREM 39

Now, rearranging the indices if necessary, there is k, with n >

k > 1 such that

p J(xi for i — 1, • • • , k

and p\xx for k < i < n.

Reducing equation (3) modulo p,

0 = J^CiXi = p!.
i— 1 \i—l )

Since p J(xx, the above implies J2i=\ ct = 0 — contradicting our

assumption. □

In fact, the converse of Theorem 2.1 is also true and together

they constitute the following abridged version of a theorem of Rado.
Theorem 2.2. (Rado) Equation (3) has a monochromatic
solution (sq, • • • ,xn) ( where x%’s may not be distinct ) in Z+ with
respect to any finite colouring if and only if some non-empty subset
of {ci, • • • , cn} sums to zero.

Proof. Only the ‘if’ part remains to be proved.

Reordering if needed, we assume that for some k > 2,

Ci + • • • + c*; = 0. (4)

Now suppose that a finite colouring of Z+ has been given. If k = n,

then X\ — • • ■ — xn — 1 would have served as a monochromatic
solution to equation (3) and therefore we assume A: < n.
Let

A = gcd (cl5 • • • , cjt) and B = ck+l H-h cn, (5)

where B can be assumed to be nonzero, for otherwise once again

Xi = ■ • • — xn — 1 would give a monochromatic solution. Here, for
twro non-zero integers a, 6, we shall consider gcd (a, b) to be positive.
Therefore, A > 0. Also, multiplying by ( — 1), if necessary, we can
assume that B > 0.

B
Now, writing — t —
gcd (A, B) ’
we have
AB
■At = = BD, say. (6)
gcd (A, B)
40 3. GENERALIZATIONS OF SCHUR’S THEOREM

Also, 3A1} • • • , A* € Z such that

cjAi + • • • + CfcAfc = At. (7)

We choose l G Z+ such that 7* = l + Xt > 1 V i = 1, • • • k. Let r G

Z+ be such that r > max jt. Now, Theorem 3.1 of Chapter 2 gives
us a and d such that {a + d, • • • , a + rd} U {dD} is monochromatic.

Let

a + 7 jd 1 < i < k
Xi = (8)
dD k < i.

n k n

Now, Y^CiXi (a + ld)(ci-\-\-ck) + d'^2cl\l + dD ^ c,

1=1 i=l i=A:+l

0 + dAt + dDB (by equations (4), (5) and (7))

0. (by equation (6))

In other words, equation (8) provides us with a monochromatic

solution to equation (3). □

Remark 2.1. For the equations x + y = z and x + z = 2y,

writing them a.s x + y — z — 0 and x — 2y z — 0 as in equation
(3), there are non-empty subsets of the set of coefficients which sum
to zero. Therefore, in particular, the above theorem includes the
content of Schur’s theorem.
In the other direction, given equations like x + y — 5z, following
the proof of Theorem 2.1, we can construct a finite colouring (a
6-colouring for this particular equation) of Z+, so that the equation
does not have any monochromatic solution.
We now proceed to give the complete statement of Rado’s the¬
orem.

Definition 2.1. Let C = (cl3) be a matrix with entries in Z.

The system of linear equations Cx = 0 is said to be regular on Z+
if for any finite colouring of Z+, 3 a monochromatic solution for the
system. In a similar way one could have defined regularity on any
set on which a system of equations is defined.

Theorem 2.3. (Rado) For C — (ctJ) with entries in Z, the

system of linear equations Cx — 0 is regular on Z+ if and only if
the column vectors of C = (ctj) can be ordered as C i,--- , Cn and
 3. FOLKMAN’S THEOREM 41

there are k{’s with 1 < ki < • • • < kt = n such that writing

ki
A,= £ c„

one has (I) A\ is the zero vector

and (II) Ai is a Q-linear combination of Ci, • • • , CV,_j Vi, 1 < i < t.

The condition on the matrix C = (Cij) in the above theorem

is known as the columns condition. In the case of a single linear
homogeneous equation, the statement of Theorem 2.3 reduces to
that of Theorem 2.2. We do not prove Theorem 2.3 here. For
a proof, [83] is an easily accessible reference. It is worthwhile to
mention that the colouring Sp on Q* has a special significance in
the present context. To be more precise, it can be proved that if
Cx = 0 has monochromatic solutions with respect to the colourings
Sp for all primes p then C satisfies the conditions of Theorem 2.3
and therefore the system is regular by the theorem. In other words,
if some finite colouring can prevent Cx = 0 from being regular,
then some Sp can do that.

3. Folkman’s theorem

The result in the following exercise is known as Folkman’s The¬

orem in the literature.

Exercise 3.1. Derive the following statement from Exercise 3.1

of Chapter 2:

F(r,k): Given any two positive integers r and k, there is a

positive integer n = n(r, k) such that if [n] is r coloured, there
are positive integers a\ < a2 < • • • < a*, satisfying T.i<i<kai S n
such the elements a, are identically coloured as I varies over
different non-empty subsets of {1, 2, • • • , k}.

REMARK 3.1. Schur’s theorem can be restated by saying that

for any positive integer r, there is a positive integer n such that for
any r-colouring of [n], 3 x,y such that the three elements x, y and
x + y are in [n] and of the same colour. Thus Schur’s theorem is
nothing but the statement F(r, k) in the above exercise, with k - 2.

The following infinite version is due to Hindman [100].

42 3. GENERALIZATIONS OF SCHUR’S THEOREM

Theorem 3.1. (Hindman ) Given any finite colouring of Z+,

there always exist an infinite subset of Z+ all the finite sums of
which are of the same colour.
We do not prove Theorem 3.1 here. The original proof of Hind¬
man was very complicated. A shorter proof due to Baumgartner
[19] can be also be found in [101] and [83]. In fact, [101] and [83]
contain other proofs of Hindman’s theorem by topological methods;
we shall refer to these proofs in Chapter 4.
Remark 3.2. It is easy to see that Hindman’s theorem implies
Hilbert’s theorem (Theorem 6.3 of Chapter 1).

We now state the Finite Union Theorem (often called Disjoint

Union Theorem in literature), which is the set theoretic analogue of
Folkman’s Theorem.
Theorem 3.2. (Finite Union Theorem) Given any two posi¬
tive integers r and k, there is a positive integer u — u(r, k) such that
ifV([u}) is r-coloured, where V([u}) denotes the family of non-empty
subsets of [u], then there exists a collection V = {Dt W c P(M)
of disjoint sets with |/| = k such that

FU(V) = {Ui€J Di : $ * T C /}
is monochromatic.
Exercise 3.2. Deduce Folkman’s theorem from the finite union
theorem.
The converse of the above exercise is also true. Namely, the
finite union theorem can be deduced from Folkman’s theorem as
well. But, the proof for this is more difficult. Interested reader may
look into [83] for a proof of the same.
Here we present a proof of Theorem 3.2 due to Taylor [171].-
This proof does not require the use of van der Waerden’s theorem.
After defining some notations we prove some lemmas before we take
up the proof of Theorem 3.2.
If T is a collection of pairwise disjoint non-empty sets, then
NU(T) will denote the set of non-empty unions of elements of T.
It is easy to see that if T and V are collections of pairwise disjoint
non-empty sets, such that V C NU(T), then NU(T>) C NU(T).
Lemma 3.3. (Taylor) For each pair of positive integers r and k,
there is a positive integer b — b(r, k) such that ifV = {\'\, • • • Vj,} is a
collection of pairwise disjoint non-empty sets, then for any partition
 3. FOLKMAN’S THEOREM 43

of NU(V) into r parts, there exists a collection T> = {D\, • • • Dk] C

NU(V) of pairwise disjoint sets, such that for any S G NU{V), the
sets D\ and Di U S belong to the same part.

PROOF. Giving a partition of a set, is equivalent to giving an

equivalence relation ‘=’ on the set such that two elements x and y
belong to the same part if and only if x = y.
Sometimes, for partitions into r parts, it is convenient to speak
in terms of equivalence relations with r equivalence classes.

We proceed by induction on k.

For k = 1, we can take 6(r, 1) = 1.

Again for r = 1, the result holds trivially for all k and there
is nothing to prove. Therefore, assuming that b(r,k)'s exist for
all r > 1, we have to establish that for any given integer r > 2,
b(r, k + 1) exists.

We claim that for k > 1,

b(r, k + 1) < (r + 1) + b Q(r2 + r3), h'j . (9)

For the proof of (9), writing

b = (r + 1) + b Q(r2 + r3), kj ,

let
V={Vu---Vb}

be a collection of pairwise disjoint non-empty sets and

NU{V) = A U • • • U A

a partition of NU(V). Let ‘=’ denote the equivalence relation on

NU(V) corresponding to this partition.

For each T G NU({Vr+2,-- ■Vb}), by pigeonhole principle, we

can choose a triple of positive integers f(T) = {p, q, i) with 1 < p <
q < r + 1 and 1 < i < r such that both the sets (Upj=lVj) U T and
(U9j=\Vj) U T belong to A-
Now we define a new equivalence relation on NU({Vr+2, ■ ■ • Vj,})
by:
7j ~ T2 if and only if /(Tj) = f{T2).
44 3. GENERALIZATIONS OF SCHUR’S THEOREM

It is easy to see that the equivalence relation defines a par¬

tition of NU({K-i-2 5 • • • H}) info

(T)-r = 5<r2 + r3)

parts and hence according to our choice of b there exists a collection
T={TU - Tr) C NU({Vr+2, ■ ■ ■ H}) of pairwise disjoint sets, such
that for any S' G NU(T),

7\ ~ 7\ U S'.

This means, there is a pair p, q with l<p<q<r+l such that

for every S' G NU(T),

(Upl=lVi)UTiUS' = (U?=1Vi)UTi = (U^iVSjUT! =

(10)

Now, let V — {Di, • • • Dk+i}, where

(U?=1^-)U if i = 1
Ti if 2 < i < k
(ULP+1L) if i = k + 1.

Once we show that for any S G NU(V), D\ = D\ U 5, we shall

be through.

Now, for given 5 G NU(V), if we write S' = S \ (U^K),

then S' G NU(T). We observe that S' will differ from S only if 5
contains D\ or Dk+1-
If S' does not contain Dk+1, then from our definition of Dt and
(10),

DiUS — (uf=1Vj) U T, U S' = (Uf=1K) UT, = Di.

On the other hand, if S' contains then

£>l = (Uf=1Vi)uT, (uL.Our,

(u’=iC)ur,u5'
(Uf=i Vi) u T, u (uf=p+1V)) u S'
Di U S.

□
 3. FOLKMAN S THEOREM 45

Lemma 3.4. (Taylor) For each pair of positive integers r and k,

there is a positive integer c = c(r, k) such that ifV = {Vi, • • • Vc} is a
collection of pairwise disjoint non-empty sets, then for any partition
of NU(V) into r parts, there exists a collection £ = {Ei, • • • Ek)
C NU(V) of pairwise disjoint sets, such that for any S G NU(£),
and i G (1,2, •••,&}, the sets E, and Et U S belong to the same
part, that is,
Ei = Ei U 5,

where = ’ denotes the equivalence relation on NU(V) corresponding

to the given partition of NU(V).

PROOF. The proof is by induction on k, starting with the ob¬

servation that for A; = 1, we can take c(r, 1) = 1.
We claim that for A: > 1,

c{r,k) <b(r,l + c(r,k- 1)), (11)

where the notation b(r, k) is as defined in Lemma 3.3 above.

For the proof of (11), writing

c = b(r, 1 + c{r,k - 1)),

let
V = {VU---VC]
be a collection of pairwise disjoint non-empty sets and

NU(V) = Ax U • • • U Ar
a partition of NU(V). Let ‘=’ denote the equivalence relation on
NU(V) corresponding to this partition.
Bv Lemma 3.3 and our choice of c, there exists a collection
V = {Ei, Di, • • • L>c(rA-1)} C NU(V) of pairwise disjoint sets, such
that for any 5 G NU(T>), we have E\=E\ U S.
Now, applying the induction hypothesis to the collection W =
[Di, ■ ■ • Dc(r,jfc-i)}> there exists a collection S' = {E2,---Ek} C
NU(W) C NU(V) of pairwise disjoint sets, such that for any
S G NU{£'), and i G {2, • • • , k), El = El U 5.

Clearly, £ - [Ex, E2, • • • Ek) has the desired property.

□
Now, we proceed to prove Theorem 3.2.
46 3. GENERALIZATIONS OF SCHUR’S THEOREM

Proof. We claim that

u(r, k) < c(r,rk — r + 1)), (12)

where the notation c(r, k) is as defined in Lemma 3.4 above.

For the proof of the claim (12), writing

n = c (r,rk — r + 1)),

let
{.4i, • • • Ar}
be a partition of the collection of non-empty subsets of {1, • • • , n}.
Let V = {{1}, • • • , {n}} and let ‘=’ denote the equivalence relation
on NU(V) given by

S = T if and only if S G At and T E At for some i G {1, 2, • • • , r}.

Then, we have a collection S = {£j, ■ ■ ■ -EVfc-r+i} C NU(V) of

pairwise disjoint sets, as guaranteed to exist by Lemma 3.4.
Now, for our equivalence relation ‘=’, the number of equivalence
classes is r. Since E contains rk — r + 1 — r(k — 1) +1 sets, it follows
that at least one of the equivalence classes will contain more than
(A; — 1) elements. In other words, there is a subset V = {Eix, • • • Elk}
of E such that
Elp = Eiq, for every p,q < k.
Therefore, by the property of the elements of £, this would imply
that
NU(V) C Az for some i 6 {1, 2, • • • , r}.
This completes the proof. □

4. Notes

As has been mentioned in Remark 2.1, Theorem 2.2 generalizes

Schur’s theorem. It can be seen that Theorem 2.3 implies van der
Waerden’s theorem. It may not be that obvious that Folkman’s
theorem also follows from Theorem 2.3.
Rado [141] later proved strengthening of Theorem 2.3 for regu¬
larity of Cx — 0, where C is an m x n matrix over R, and R is any
subring of the complex numbers.
In [23], it was shown that a version of Rado’s theorem is valid
for vector spaces over finite fields.
 4. NOTES 47

It was shown in [24] that .the sufficiency of a suitable column

condition is valid for any commutative ring.

In [139], Rado made the following conjecture: There is a func¬

tion r : Z+ —> Z+ such that given any equation C\X\ -I-1-cnxn — 0
with integer coefficients which is not regular over Z+, there exists
a partition of Z+ into at most r(n) parts with no part containing a
solution to the equation.
Though the above conjecture is still open, it was shown in [24]
that the natural analogue of Rado’s conjecture fails for general com¬
mutative rings.

If G is an abelian group and A is an m x n matrix, then Deuber

[42] gave conditions for partition regularity of Ax = 0 in G. That
is, he gave conditions for the existence of monochromatic solutions
of the system Ax = 0 for any finite colouring of G \ {0}. One may
also look into [83] for these results.

We shall encounter regularity of certain nonhomogeneous linear

systems in Chapter 5.

Going back to Schur’s theorem once again, this theorem is equiv¬

alent to the statement that if Z+ = UTl=lCi is a partition of Z+, then
there exists i, 1 < i < r and n G Cx such that Cx D {Cx - n) ^ 4>.
That is, there exists mGCj such that m + n € Ct.
One may ask how many such m, n are there in Cx for which
m + n is also in Cx. A result of Vitaly Bergelson [21] answers to
this question (the statement of this result involves the notion of
upper natural density which was defined in the notes given in the
last section of Chapter 2):

Let Z+ = \Jri=lCi be a finite partition of Z+. Then, there exists

i,l < i < r such that d(Ci), the upper natural density of Cx, is
positive and for any e > 0, the set {n 6 Ci\d(Ci H (Cx — n)) >
(id(Ci))2 - e} has positive upper density.

In the spirit of the above result of Bergelson, an iterated density

version of the ‘strong van der Waerden theorem’ (the statement
obtained by putting s = 1 in that of Theorem 3.1 of Chapter 2) was
obtained in [62].

The result of Rado completely characterizes the regularity of sys¬

tem of linear homogeneous equations over integers. For equations
of higher degree, which have sufficiently ‘large’ number of solutions,
48 3. GENERALIZATIONS OF SCHUR’S THEOREM

one may ask similar questions. In this direction, even for the sim¬
plest of equations like x2 + y2 = z2 it is not known whether the
equation is regular or not.
However, for a certain class of nonlinear equations, results can
be derived from ‘corresponding’ linear equations.
In this direction, answering a question of Erdos [50], it was
proved by several authors ([35], [110], [3]) the equation 1/x + l/y =
\/z is partition regular.
In [3], it is shown that Ramsey’s theorem implies the partition
regularity of the above equation. The papers of Brown and Rodl
[35] and Lefmann [110] contain some general results which imply
the above result as a particular case.
More precisely, considering the Diophantine equation

^ + ^ + ... + ^ = 0, (13)
V\ V2 Vn

where a*’s are non-zero integers and yt’s are n unknown variables,
Brown and Rodl show that (13) is regular if and only if the corre¬
sponding linear equation a^xi + a2x2 + • • • + anxn = 0 is regular,
while Lefmann proves by a different method (but still depending on
Rado’s result) that (13) is regular if and only if for some non-empty
subset / of {1, 2, • • • , n}, J+€/ az = 0.
Thus the equation 2/z = 1/x+l/y is regular whereas the Erdos-
Straus’ equation 4/x4 = l/xi + l/x2 + l/x3 and Sierpinski’s equation
5/x4 = 1 /X\ + l/x2 + l/x3 are not.

In [6], lower bounds for the function f(r) is derived, where f(r)
is defined to be the minimal positive integer such that whenever the
set {1,2, ••• ,/(r) + 1} is partitioned into r parts, at least one of
the parts will contain a solution of the equation l/x + l/y = l/z.
The finite union theorem (Theorem 3.2) first appeared in a paper
[82] of Graham and Rothschild. Taylor’s proof of Theorem 3.2
actually yields upper bounds for the least integer it(r, k) satisfying
the requirements as in Theorem 3.2 (and hence, upper bounds for
the least integer n(r, k) satisfying the requirements as in Exercise
4.1, as it will satisfy n(r,k) < 2u(r,A:) ). As one can guess, the
relations (9), (11) and (12) are used to obtain these upper bounds.
For details, one may refer to the original paper [171] of Taylor.
Milliken [117] generalizes Hindman’s theorem, the way Ram¬
sey’s theorem generalizes the pigeonhole principle.
 CHAPTER 4

Topological methods

1. Introduction

In Section 6 of Chapter 1, we already had a glimpse of topolog¬

ical methods for establishing some results in combinatorial number
theory. In the present chapter, we shall see some more instances
of successful applications of topological methods to combinatorial
problems.

We shall mainly dwell on the use of the topological algebraic

structure of enveloping semigroups by Furstenberg and Katznelson.
This is done in Sections 2 and 3 of this chapter which are based on a
paper of Furstenberg and Katznelson [69]. In Section 2, we present
some results on idempotents in compact semigroups and in Section
3 we shall see applications of these results to some Ramsey-type
problems.

In Section 4, we give a small survey of ergodic and topological

methods which have been successful in dealing with Ramsey-type
problems and few other combinatorial problems.

2. Compact semigroups and idempotents

Definition 2.1. Let S be a semigroup. In other words, 5 is a

non-empty set with an associative law of composition. We denote
the law of composition by juxtaposition.

If 5 is endowed with a topology with respect to which it is a

compact Hausdorff space and for any given s G S, the map t i-> ts
is continuous, then S will be called a compact semigroup.
49
50 4. TOPOLOGICAL METHODS

We recall that an idempotent a G S is an element satisfying

2
a = a.
Remark 2.1. In Exercise 3.1 ( solution is given in the last chap¬
ter) of Chapter 1, we see that Ramsey’s theorem can be used to show
that every finite semigroup contains an idempotent. In the present
chapter we shall see how existence of idempotents in compact semi¬
groups can be used to derive combinatorial results.
Theorem 2.1. (Ellis [44]) A compact semigroup S contains an
idempotent.

Proof. Let T denote the family of compact subsemigroups of

S ordered by inclusion. We observe that according to our definition,
a subsemigroup is necessarily non-empty. Because of compactness
of 5, by the finite intersection property, the intersection of a totally
ordered chain in T belongs to T. Hence, by Zorn’s lemma, T has
a minimal element, say P.
If p G P, then by the continuity of the map t i—>■ tp, Pp is
compact. Also, it is clear that Pp is a subsemigroup. Since p G P,
we have Pp C P. Hence, by the minimality of P, we have

Pp - P.

Thus,

Q = {t e P\xp — p}
is non-empty.
Clearly, Q is a subsemigroup and once again, by the continuity of
the map t i-a tp, it follows that Q is closed. Now, 5 being Hausdorff,
we get that Q is compact and by the minimality of P,

Q — P.
This implies that p2 = p. □

Given a compact semigroup 5, by the above theorem we get that

the subset consisting of the idempotent elements of S is non-empty.
Given two idempotents a and (3 of 5, we define a relation a < (3 by
declaring

a < (3 «=> a/3 = (3a = a (14)

Clearly, (14) defines a partial order on the set of idempotents of

S. that is, the relation ‘<’ defined above is reflexive, antisymmetric
and transitive.
 2. COMPACT SEMIGROUPS AND IDEMPOTENTS 51

Definition 2.2. Let S be a compact semigroup. One says that

J (7^ <L) C S is a left ideal of S if J is closed and SJ C J.
Remark 2.2. Being a closed subset of a compact set, a left ideal
is compact. Therefore, if J C S is a left ideal of 5, then for any
s G 5, the set Js is a continuous image of a compact set and hence
compact. Since S is Hausdorff, Js is closed. The other condition
being trivially satisfied, Js is a left ideal.
Ordering by inclusion, by Zorn’s lemma, there is a minimal left
ideal. Let J be a minimal left ideal of a compact semigroup S.
Now, for x E J, we have Jx C J. Therefore, by minimality we have
Jx = J.
Definition 2.3. One defines a right ideal by making the obvious
change in the definition of a left ideal. If J C S is a left ideal as
well as a right ideal, then it is called a two-sided ideal.
Exercise 2.1. If J C 5 is a minimal left ideal of a compact
semigroup 5, then for s G 5, show that Js is also a minimal left
ideal.
Theorem 2.2. (Furstenberg and Katznelson) Let J C S
be a minimal left ideal of a compact semigroup S and a and (3 be
idempotents in S such that (3 € J and a < (3 where the partial
ordering ‘< ’ is defined in (If).
Then a = (3.

PROOF. Since a < (3, from (14) we have

a(3 = (3a — a. (15)

Now, J being a left ideal, a = a(3 e J and hence by Remark

2.2, Ja = J. Therefore, (3 = 6a for some S e J. Since a is an
idempotent,

(3a = 5a2 = Sa = (3. (16)

From (15) and (16) we have a = (3. □

Remark 2.3. Let J be a left ideal of a compact semigroup 5.

Since a left ideal is itself a compact semigroup, by Theorem 2.1
J contains an idempotent a. If J happens to be a minimal left
ideal, then by Theorem 2.2, a is a minimal element in the set 1 of
idempotents in S, endowed with the partial ordering £<’. Since a
compact semigroup S has a minimal left ideal, I contains a minimal
element for the partial ordering ‘<’.
52 4. TOPOLOGICAL METHODS

Theorem 2.3. (Furstenberg and Katznelson) Let J be a

left ideal of a compact semigroup S. If (3 is any idempotent of S,
then J(3 contains an idempotent a such that a < (3.

Proof. By Remark 2.2, J/3 a left ideal of S. Therefore, there

is an idempotent S G J/3. Let 6 = 9/3 with 6 G J■ Writing a =
(36 = (39/3, we observe that a £ J/3 and

a2 = ((39 /3) {/39/3) = (39/39/3 = P62 = (36 = a.

Since a = (39(3 and (32 — f3, we have a(3 = (3a = a, which means
a < (3. □

Theorem 2.4. (Furstenberg and Katznelson) An idempo¬

tent of a compact semigroup S is minimal if and only if it belongs
to a minimal left ideal.

Proof. In Remark 2.3, we already observed that by Theo¬

rem 2.2, idempotents belonging to minimal left ideals are minimal.
Conversely, suppose that we are given a minimal idempotent (3
in S. We take any minimal left ideal J of S. By Exercise 2.1, J/3
is also a minimal left ideal. Again, by Theorem 2.3, J/3 contains an
idempotent a such that a < (3 and hence by the minimality of (3,
we have a = (3. Thus, (3 is contained in the minimal left ideal J/3
and we are through. □

Remark 2.4. If J\ is a right ideal and J2 is a left ideal of 5,

then (<f> ^)J\ J2 C J\ (T J2.
Therefore, in particular, given any minimal left ideal /, a two-
sided ideal J will have a non-empty intersection with /. Since / fT J
is a left ideal, by the minimality of I we must have I C I fl J C J.
Therefore, by Theorem 2.4, a two-sided ideal contains all the
minimal idempotents of 5.

Theorem 2.5. (Furstenberg and Katznelson) Let J be a

two-sided ideal of a compact semigroup S. If (3 is any idempotent
of S, then J contains an idempotent o such that a < (3.

Proof. Since J is a two-sided ideal and hence a left ideal in

particular, from Theorem 2.3, we get that J/3 contains an idempo¬
tent o such that a < (3.
But J being a two-sided ideal, J/3 C J.
Hence the theorem. □
 3. RAMSEY-TYPE THEOREMS 53

Remark 2.5. Let X be a compact metric space. As usual, by

Xx we denote the set consisting of all maps / : X —> X. Now,
Xx is a semigroup under composition of maps; for /, g £ Xx we
denote this law of composition by / o g. Since X is compact, Xx
is compact with the product topology. We observe that for any
given g £ Xx, the map / i-» / o g is continuous. Thus Xx is
a compact semigroup. It is clear that an idempotent 0 £ Xx is
nothing but a map such that for y £ 0(Ar), (f)(y) = y. That is, an
idempotent equals the identity on its range. For two idempotents
0 and 9, the order relation 9 < 0 as defined in (14) would mean
9 o cj) = (f> o 9 = 9, which is equivalent to the requirement that
the range of 9 is contained in the range of 0. Thus the minimal
idempotents here are the constant maps.

3. Ramsey-type theorems

Let T be a discrete semigroup and A a finite set. Consider the

set X = Ar consisting of all maps / : T —>■ A.
In Section 6 of Chapter 1, while giving the topological proof of
Hilbert’s theorem we considered Ar with T = Z+.
Now, X is the collection of sequences made of finitely many
symbols. Often the set A consisting of the finitely many symbols is
called the space of colours. Clearly, the elements of X are nothing
but | A [-colourings (often referred to as A-colourings, specially when
one wants to mention the colours involved rather than just giving
the number of colours) of T.
In what follows, without loss of generality, we shall assume that
T has a unit element e. Because, in case T does not have one, we
can add a unit element e and extend the colouring as we please.

As had been described in the case of Y = Z+, in the general case

also T acts on A" by ‘translation’:
for a £ T, and x £ A",

(a(a)x)(b) = x(ba) V b £ T. (17)

For a fixed positive integer k, writing Y = Xk, the above action

(17) of T is extended to an action of Tk on T as follows:

a(au-• • ,ak) = (tf(ai), ■ * ■ ,cr(ajt))- (18)

54 4. TOPOLOGICAL METHODS

Definition 3.1. Given a subsemigroup G C rfc, the enveloping

semigroup of G is the closure of o(G) in Yy. We shall use the
notation E(G) to denote the enveloping semigroup of G. Thus,
E{T) will denote the closure of cr(r) in Xx. For an arbitrary subset
A of Tk, we use the notation E(A) to denote the closure of cr(A) in
YY,
Theorem 3.1. (Furstenberg and Katznelson) Suppose we
are given a subsemigroup G C Tk such that G contains the diagonal
A*(r) d= {(x,x,---x)\x e r}.
'-V-"

k times
Then, with the notations as above, if I is a two-sided ideal of G
and 6 is a minimal idempotent in E(T), we have

‘■O.-yOi 6 E(I).
k times
Proof. We claim that given two subsets A and B of

E{A)E(B) c E(AB).

Once the above claim is established, we shall in particular have,

E(I)E(G) C E{IG) C E(I) and E{G)E{I) c E{GI) C E{I),

which would imply that E(I) is a two-sided ideal of E(G).

Now we proceed to establish our claim. Let <p e E(A) and
ip G E(B). Also, let Y\ and Ti be arbitrary finite subsets of Y and
Tk respectively.
Since cp e E(A), there exists a e A such that a (a) is close
enough to 0 such that

cr(a)i/u/(7) = Wy(l) for all y G Yu 7 € Ti-

Again, since ip G E(B), there exists b G B such that

a(b)y(ja) = ipy{ja) for all y G Yx, 7 G

Therefore,

o{ab)y(~i) = cr(6)y(7a) = ipy{ja) = a(a)ipy( 7) = <f>xpy( 7),

for all ye T1,7 G Ti.

Since the above is true for arbitrary finite sets Yi C Y and
T ] C rk, we have
H G E(AB),
and hence the claim.
 3. RAMSEY-TYPE THEOREMS 55

Since a(G) acts component wise, we have E(G) C (E(r))k.

Since Afc(T) C G, E(G) contains the diagonal Ak{E(T)) of (E(T))k.
Therefore,
(^J) € E(G).
k times

Now, since E(I) is a two-sided ideal of E(G), by Theorem 2.5,

E(I) contains an idempotent

(fo, ■y<Pk) < (9^9),

k times k times

which means that fa is an idempotent with f>x < 9 for 1 < * < k.
Since 9 is a minimal idempotent in E(T), we have fa = 9. Hence
the theorem. □

Theorem 3.2. (Furstenberg and Katznelson) Let G C Tk

be a subsemigroup such that G contains the diagonal Afc(r) and let
I be a two-sided ideal of G. Then for any finite colouring of T, I
contains an element (71, • • • 7k) such that the elements 7i, 1 < i < k,
receive the same colour.

PROOF. Let C denote the range of the given finite colouring of

T. Consider the set X — Cr consisting of all C-colourings of P. Let
x G X denote the given colouring and 9 be a minimal idempotent
in £(T).
Then by Theorem 3.1,

(g,_' -:g) € E(I).

k times

We write x' = 9x.

There exists (71, • • - 7*) G / such that cr ((71, ■7fc)) and '“ 0)
k times
are close enough so that the maps
/ \
0 ((7l) * ' ' Tk)) (x, ■ ■ ■ ,x)
\ A; ti
times

and
/ \
(9^-9) (t,••• , x)
k times \ k times /

have the same value at the identity element e.

56 4. TOPOLOGICAL METHODS

This means
x(lj) = *'(e)>
for 1 < j < k. n

Remark 3.1. As Furstenberg and Katznelson remarks, in the

proof of the above theorem, the fact that

{Or^e) e E(I).
k times

where 9 is a minimal idempotent in F^T), is not used with its full

strength. More precisely, the only fact which is being utilised is
that E(I) intersects the diagonal.

Application 3.1. (Furstenberg and Katznelson) Let T be

the semigroup (N, +) of non-negative integers under addition.
We consider the subsemigroup

G =f {(a, a + 6, • • • , a + (k — 1)6)|a, 6 G N}

of the semigroup Tk.

Clearly,

I d= {(a, a + 6, • • • , a + (fc — l)6)|a € N,6 G Z+}

is an ideal of G.
By Theorem 3.2, for any finite colouring of N, there is a posi¬
tive integer b such that all the k integers a, a + b, ■ ■ • ,a+(k— 1)6
receive the same colour. In other words, we have deduced van der
Waerden’s Theorem (Theorem 4.1 of Chapter 1) from Theorem 3.2.

Application 3.2. (Furstenberg and Katznelson) Let T =

W(k) be the free semigroup generated by the k symbols {1, 2, • • • , k}.
In other words, T is the semigroup consisting of all finite words of
the form w(l)w(2) • • -w(n), n € Z+, w(j) G (1, 2, • • • , k}, the op¬
eration for the semigroup being concatenation.

We consider the subsemigroup G C Tfc, which is the span of

Afc(r) and the element (1,2, ••• ,k). A typical element of G will
look like (Di, D2, • • • , Dk) where with some finite words [Li], • • • ,
[U

Dj = [£i]j[£2]j[Z,3]j • ■ ■ j[Lt], for all j, 1 < j < k.

 4. NOTES 57

Taking / = G \ A*(T), we see that the set / consisting precisely

the combinatorial lines is an ideal of G.
Thus Hales-Jewett Theorem (Theorem 2.4 of Chapter 2) also
follows from Theorem 3.2.

Remark 3.2. The original paper of Furstenberg and Katznelson

[69] contains several other applications of this algebraic approach.
One may look into the original paper for these.

4. Notes

Originating in the investigations [67] related to the theorem of

Szemeredi on arithmetic progressions, further research of Fursten¬
berg and B. Weiss [71] initiated the application of topological dy¬
namics in the study of Ramsey-type theorems. The paper [71] of
Furstenberg and B. Weiss already contained dynamical proofs of the
classical Ramsey-type theorems and some of their generalizations.
For instance, van der Waerden’s theorem can be deduced from
the following recurrence result.
// 7\,T2, • ■ • , Tn : X X are homeomorphisms of a compact
metric space such that TfTj = TfTx for 1 < i, j < n, then there exists
x £ X and a sequence of positive integers {n\, n2, • • • }, —> oo
with d (T™3 x, Xs) —* 0 for each i = 1, 2, • • • n.
To derive van der Waerden’s theorem from the above, one works
with the space Q = Az where A = {1, 2, ■ • • , r} where the element
lj € £2 corresponding to a finite partition Z+ = C\ U C2 • • Cr can
be taken to be

. . f 7 if n > 1 and n £ C,
{ 1 if n < 0.
Then as in the dynamical proof of Hilbert’s theorem in Chapter
1, one considers the forward orbit closure of u under the shift map a
and applies the above recurrent result with T2 = a1 for i = 1, 2, • • • r.
Apart from the original paper of Furstenberg and B. Weiss [71],
details of the above proof can also be found in [83], [66] and [132],
for instance.
Essentially the same argument works for the proof of Griinwald’s
Theorem (Theorem 4.2 of Chapter 1).
58 4. TOPOLOGICAL METHODS

Taylor [172] gave a combinatorial version of the Furstenberg-

Weiss topological proof of van der Waerden’s theorem.
Furstenberg and Weiss [71] used the notion of proximality to
establish Hindman’s theorem (Theorem 3.1 of Chapter 3). As men¬
tioned in Chapter 3, the book [83] of Graham, Rothschild and
Spencer contains two topological proofs of Hindman’s theorem;
one of them is the Furstenberg-Weiss proof, the other being the
ultrafilter-proof due to Glazer. Glazer never published his proof; it
was presented by Hindman in [101]. An account of the story behind
the discovery of Glazer’s proof (or, the Galvin-Glazer proof, a name
which will be more appropriate) is available in [102].
The 1977 paper [67] of Furstenberg, marks the beginning of
the subject of Ergodic Ramsey Theory. Here Furstenburg derived
(through a correspondence principle) Szemeredi’s theorem on arith¬
metic progressions from the following multiple recurrence result:
Let (Ar, B, p) be a probability measure space and T an invertible
measure-preserving transformation on (AT, B, p). Then for any pos¬
itive integer k and any A £ B with p(A) > 0, there exists a positive
integer n such that

p (a n T~nA n T~2nA n ••• n T~knA) > o.

Density version of Griinwald’s Theorem was deduced by Fursten¬

berg and Katznelson [68] from a generalization of the above theo¬
rem, where the set {T, T2, • • • Tk} is replaced by any set of commut¬
ing measure-preserving transformations. We should mention that
for this higher dimensional analogue of Szemeredi’s theorem, this is
the only proof available so far.
Bergelson and Leibman [26] established a joint extension of the
above result of Furstenberg and Katznelson and of a theorem of
Furstenberg-Sarkozv ([66], [156]) which states that given any poly¬
nomial f(x) £ Q[x] taking integer values on the integers and sat¬
isfying /(0) = 0, if S is a subset of Z of positive upper Banach
density, then there exists a solution to the equation x - y = f(z),
x,y £ S, x ^ y, z G Z.
Here, upper Banach density of S is the number

BD(S) = lim sup -———

F|->oc HI
 4. NOTES 59

where / ranges over intervals of Z.

Similarly, S C Zk, is said to have positive upper Banach density

if for a sequence of parallelopipeds Pn = [an}^n }] X • • • X [a^}, bffl] C
Z*, n € Z+, with &W - ->• oo for i = 1, • • • , k, one has

\SnPn\
limsup - > 0.
n-yoo IPn|

The polynomial extensions of Szemeredi’s theorem due to Bergel-

son and Leibman [26] is as follows:
Let k, l and t be positive integers. Let S C Zl be a set of positive
upper Banach density and fiffix),--- ,/M(x), f2>i(x),-“
■ • • , fk,i{x), • • • , fk,tix) be polynomials unth rational coefficients tak¬
ing integer values on the integers and satisfying /M (0) = 0 for i =
1, • • • , k, j = 1, • • • , t. Then for any finite set {vx, • • • , vt} C Zl,
there exists a positive integer n and ueZ1, such that
t
u + L Aj(n)vj £ s,
j=l
for 1 < i < k.
Later, in a very important paper [27], polynomial extension of
Hales-Jewett theorem was also obtained by Bergelson and Leibman.
An expository account of this result is available in [22],
The paper [69] of Furstenberg and Katznelson, some results
from which were presented in Sections 2 and 3 of this chapter, con¬
tains some essential ingredients for the proof of density version of
the Hales-Jewett theorem by Furstenberg and Katznelson [70].
We would like to mention the following Ramsey-tvpe result which
was conjectured by Kneser and proved by Lovasz [112].
If the n-subsets of a (2n + k)-element set are partitioned into
(k + 1) classes, then some class will contain a pair of disjoint sets.
The method employed by Lovasz is topological. One of the main
ingredients of this proof is the following well-known (Ramsey-type)
theorem of Borsuk:
If the k-sphere Sk is covered by k + 1 subsets, all closed or all
open, then one of these must contain a pair of antipodal points.
For more information regarding further developments in the di¬
rection of Kneser’s conjecture and other discrete applications of
60 4. TOPOLOGICAL METHODS

Borsuk’s theorem, we recommend the nice expository article [30] of

Bjorner on applications of topological methods in combinatorics.
McCutchen [115] refers to the application of the above men¬
tioned theorem of Lovasz by Kriz and Rusza in establishing ‘chro¬
matic intersectivity’ of certain constructions.
We have already mentioned the ultrafilter-proof of Hindman’s
theorem due to Glazer. Later, in a significant development [25]
it was shown that the Stone-Cech compactification of the set of
positive integers can be used to deduce van der Waerden’s theorem
as well. This was also reported in [102]. We recommend [101],
[102] and [22] for gaining an insight into applications of the Stone-
Cech compactification of the discrete set Z+ of positive integers to
Ramsey-type results.
Apart from the many original papers mentioned in this section,
once again we recommend the beautiful book of H. Furstenberg [66]
and a recent monograph of R. McCutcheon [115] for the study of
the developments in ergodic Ramsey theory.
 CHAPTER 5

Euclidean Ramsey theory

1. Introduction

In this chapter, we give an introduction to a very interesting

area of Combinatorics.

We have already encountered results related to the existence of

monochromatic substructures of a specific type, for finite colourings
of the n-dimensional Euclidean space En (or its subsets, like the
set of integer lattice points). For example, Griinwald’s Theorem,
the generalization of van der Waerden’s Theorem (Theorem 4.1 of
Chapter 1) to higher dimensions, tells us that given any finite set
S C (Z+)d, we have a monochromatic homothetic copy of 5 for any
finite colouring of (Z+)d.
Again, in Exercise 4.1 of Chapter 1, we see (see solutions to
selected exercises) that whichever way we 3-colour the points of the
Euclidean plane E2, there will be two points at a distance 1 apart,
receiving the same colour.
In general, given a group H of transformations of En, a finite
subset K of En and a positive integer r, one can consider the prob¬
lem of determining whether or not for any r-colouring of En there
is a monochromatic set K' which is the image of K under some
element of H. In Euclidean Ramsey theory, one takes the group H
to be the one consisting of the Euclidean motions of En.

The subject was developed in the pioneering papers ([53], [54],

[55]) of Erdos, Graham, Montgomery, Rothschild, Spencer and
Straus. In what follows, we-present some of the important results
of this subject. For further information, we would like to refer to
[78] and [79].
61
62 5. EUCLIDEAN RAMSEY THEORY

2. Preliminaries

Definition 2.1. Given a finite set C C En, we shall say that

7Z(C, n, r) holds, if for any r-colouring of En, there exists a monochro¬
matic set C' C En such that C' is congruent to C. Here, by saying
that C' is congruent to C, one means that C' is the image of C
under some element in the group of Euclidean motions of En.
For example, let P denote a set of two points at a distance 1
apart. Then it is easy to see that 7Z(P, 1, 2) does not hold. Whereas,
in dimension 2, considering 2-colourings of the vertices of an equi¬
lateral triangle with sides of length 1, we can observe that 7Z(P, 2, 2)
holds. In fact, in dimension 2, we have stronger results. Because,
Exercise 4.1 of Chapter 1, which we have already mentioned in the
introduction above, says that 7Z(P, 2,3) holds. On the other hand,
if one considers the standard tiling of E2 by regular hexagons of side
9/10, then it is easy to see that we can 7-colour E2 such that there
is no monochromatic set congruent to P. In other words, 1Z(P, 2, 7)
does not hold.
Definition 2.2. Given a finite subset C in some Euclidean
space, one says that C is Ramsey if for every positive integer r,
there is a positive integer n = n(C, r) such that 7Z(C, n, r) holds.
We observe that a two-point set is Ramsey. This observation,
together with the first theorem in the next section, will provide us
with a large class of Ramsey sets.
Exercise 2.1. Show that there is a (2n)-colouring \ of the set
of real numbers such that the equation

Y(xi - x’i) = 1
i= 1
has no solution with x(Ti) = x(xi), 1 < * < n.

Exercise 2.2. Given real numbers Ci,-- - ,cn and a non-zero

real number 6, show that there is a (2n)n-colouring x* of the set of
real numbers such that the equation
n

Y Ci(Vi ~ Vi) = b
2—1

has no solution with X*(Vi) = X*(Vi), 1 < i < n.

Definition 2.3. We shall call a set S spherical if it lies on a

sphere in some En.
 2. PRELIMINARIES 63

Remark 2.1. For ao, ai, a2, • • • , ar in some En, linear indepen¬
dence of the vectors az — ao-, i — 1, • • • , r can be taken as the def¬
inition of geometric independence of the points az, 0 < i < r in
En.

Eventhough, a simplex spanned by a geometrically independent

set a,, 0 < i < r in En is the set of all points x in En with
n n

X = 52 where ]T tt = 1,
i=0 i =0
here by a simplex we shall refer to its set of vertices. For more
informations about simplices, one may look into [120] for instance.

The next lemma gives a characterization of spherical sets.

Lemma 2.1. ( Erdos-Graham-Montgomery- Rothschild-
Spencer-Straus) If a set A = {a0, ai5 a2, • • • , ar} is spherical,
that is, the points a* ’s lie on a sphere in some En, then there can not
exist real numbers Ci, C2, • • • , cr not all zero, satisfying the following
two equations.

52 ^(a, - ao) = 0, (19)

l—l

5]ct(a2 - a£) = b / 0. (20)

i=i
Here, for x = (xux2, • ■•!„), x2 = x • x = x\ + x\ ^-h x2n.
Conversely, if A is not spherical, then there exist real numbers
Ci, c2, • • • ,cr not all zero, satisfying equations (19) and (20).

Proof. Let us assume that A is spherical. Let the points of A

lie on a sphere with centre y and of radius d in En.
We have, (a* - y)2 = d2 for i = 0,1, • • • r.
Therefore,
a2 - ag = (al — y)2 — (a0 - y)2 + 2(a; — a0) • y = 2(a, - a0) • y,

which implies that

52 c(a- - a2) = 2y • 5^ c{(a, - a0).

i= 1 i= 1
The last equality shows that equations (19) and (20) can not
hold simultaneously.
64 5. EUCLIDEAN RAMSEY THEORY

Now, we suppose that A is not spherical. We can assume that

A is a minimal non-spherical set. That is, all proper subsets of A
are spherical. Now, the vectors a* — ao, i — l,--- ,r are linearly
dependent. For, otherwise, at, 0 < i < r will form a simplex and a
simplex is spherical.
Therefore, there exist real numbers Ci,c2,-*- ,cr not all zero,
such that
r
^ct(a* - ao) = 0.
1=1

Assume that cr is non-zero. By the minimality assumption, the

points a*, 0 < i < r - 1 lie on a sphere with centre y, say.

Then

£<*(»?-«3) = E c, ((a, - y)2 - (ao - y)2)

1=1 1=1

= cr ((ar - y)2 - (ao - y)2) / 0.

This proves the lemma. □

3. Fundamental Theorems

In this section, we shall present two fundamental theorems in

this area. Whereas the first theorem in this section will provide us
with a large class of Ramsey sets, the content of second one will pro¬
vide us with informations in the opposite direction. More precisely,
the second will say that a Ramsey set is necessarily spherical.
We shall need the following lemma, which is nothing but a ver¬
sion of the Compactness Principle suited for our purpose. In the
solution of Exercise 4.2 of Chapter 1, in the proof of the fact that
the statement V(k,r) implies the existence of W(k,r), we have al¬
ready seen the method to establish the compactness principle for
the countable case. For several versions of this principle, one may
look into [83].

Lemma 3.1. Consider a finite set C C Em. Let C be Ramsey.

Given a positive integer r, we choose a positive integer n = n(C, r)
such that 7Z(C,n,r) holds. Then, there exists a finite set S C En,
such that for any r-colouring of S, there is a monochromatic C' C S,
such that C' is congruent to C.
 3. FUNDAMENTAL THEOREMS 65

Proof. If possible, let the result be false. Let E be the set of all
functions / : En —» [r]. In other words, E is the set of r-colourings
of En. Giving [r] the discrete topology, by Tychonoff’s theorem E
is compact with the induced function space topology.
For every finite subset S of En, let Es be the set of functions
/ G E such that there is no monochromatic C' C 5, congruent to
C. Now, Es is closed and by our assumption it is non-empty.
Considering finitely many finite subsets Si, S2, • ■ • , S/ of En, we
have

^SiUSa-USj c ESl n ES.2 n • • ■ ESr

Since S1US2 • • -USi is a finite set, by our assumption, Esx\js2--us1

is non-empty and hence Esx flEs2 H • • • Esl is non-empty. Therefore,

by finite intersection property, there is a function / : En -a [r] such
that / G Es for all finite S C En.
Therefore, we have found an r-colouring / of En such that there
is no monochromatic C' C En, congruent to C, contradicting our
hypothesis. □

Theorem 3.2. (Erdos-Graham-Montgomery-Rothschild-

Spencer-Straus) If C\ C Eni and C2 C En'2 are two finite sets
which are Ramsey, then so is C\ x C2.

Proof. Let r be a positive integer. We choose mi such that

IZ{C\,m\,r) holds.
By Lemma 3.1 there exists a finite set S C E1711, such that
for any r-colouring of S, there is a monochromatic set C[ C S,
congruent to C\.
Let S = {x1? x2, • • • ,xs}.
Now, choose m2 such that IZ(C2, m2,rs) holds. Again, there
exists a finite set T C Em'2, such that for any rs-colouring of T,
there is a monochromatic set C'2 C T, congruent to C2.
We consider the set 5 x T C Emi+m'2 and let \ t:>e an r-colouring
of S x T.
Now, we define a colouring y* on T as follows. For y1? y2 in T,
we say that x*(yi) = X*(y2), if and only if x((x,y!)) = x((x,y2))
for all x € 5. Clearly, x* is an recolouring of T. Therefore, there
is C'2 C T, congruent to C2 such that C2 is monochromatic for this
colouring. In other words, x* is constant on C2.
66 5. EUCLIDEAN RAMSEY THEORY

We choose a point y0 G C2 and define a colouring x** on S as

follows. For x G 5, we write x**(x) = x((x7 Yo))- Now, x** is an
r-colouring of S and therefore, there is C[ C S, congruent to C\
such that C[ is monochromatic under x**- Thus, according to the
construction of x**,
X is constant on C[ x {y0}- Therefore, by the
choice of x*) X is constant on C[ x C'2 and hence the theorem. □

Corollary. Since any two-point set is Ramsey, by the above

theorem so is the Cartesian product of finitely many two-point sets.
Remark 3.1. A product of finitely many two-point sets is called
bricks in the literature. Therefore, the above corollary can be re¬
stated by saying that all bricks (and hence all subsets of a brick)
are Ramsey.

Theorem 3.3. (Erdos-Graham-Montgomery-Rothschild-

Spencer-Straus) If a set S is not spherical, then it can not be
Ramsey.

Proof. Let the set

A = {ao, ai, a2, • • • , ar}

be a non-spherical.

Therefore, by Lemma 2.1, there exist real numbers c\, c2, • • • ,cr
not all zero, satisfying equations (19) and (20).

By Exercise 2.2, there is a (2r)r-colouring x* of the set of real

numbers, such that the equation

^2ci{yi ~ y[) = b
i= 1

has no solution with x*{Vi) = 1 < i < r, where the numbers

Cj and b are as in equation (20).

Now, we define a colouring x on En as follows:

x(y) = x*(y -y)-‘

We observe that equations (19) and (20) are satisfied if A is
replaced by a set congruent to A. For equation (19), it is clear.
Equation (20) is invariant under isometries fixing the origin, as by
these afs remain unchanged. Now we observe that (20) remains
unchanged by translations as well. For, if we translate by z, we
have
 4. FURTHER RAMSEY CONFIGURATIONS 67

((at + z)2 - (a0 + z)2) = - aj) + 2z • J]cj(az - a0)

*=i i=i i=i

i=i
Thus, if x assigns a single colour to a set

A' = {a'0, a'l5 a'2, • • • , a'r},

which is congruent to A, then x* assigns a single colour to the set

{U,0 < i < r}, where U = a',2. Since A' satisfies Equation (20), U's
satisfies
r

- *<>) = 6,
t=l
contradicting the choice of x*- □

Remark 3.2. The above theorem implies in particular that a

configuration consisting of three collinear points is not Ramsey.
From the proof of the theorem, we see that there is a 16-colouring
of En such that there does not exist any monochromatic copy of
this configuration.

4. Further Ramsey configurations

In connection with Remark 3.1, the question which arises is to

decide whether there are Ramsey sets other than subsets of bricks.
In particular, it had been often asked (see Graham [78] ) whether
the set of vertices of an obtuse triangle is Ramsey. Frankl and Rodl
[63] proved a result in this direction which says that all triangles
are Ramsey. Here, by a triangle one means the set of the vertices
of a triangle. In this section, following the original paper [63] of
Frankl and Rodl, we give the proofs of some weaker results, which
are strong enough to provide us with obtuse triangles which are
Ramsey.

The first theorem says that isosceles triangles of a particular

type are Ramsey.
THEOREM 4.1. (Frankl and Rodl) For all t > 2, the triangle
with side lengths s/2t, \/21 and \/8t — 6 is Ramsey.
68 5. EUCLIDEAN RAMSEY THEORY

Proof. Suppose we are given an integer t > 2. By Ramsey’s

Theorem (Theorem 3.1 of Chapter 1), there exists a positive integer
n = n(2t - 1, r, 2t + 1) such that for any r-colouring of the collection
of (21 — 1)- subsets of the set [n] = {1, 2, • • • n}, there is a (21 T 1)-
subset of [n] all of whose (21 - l)-subsets are of the same colour.
For a (21 - l)-subset B = {zi,22, ■ ■ • ,i2t-i} of where ix <
i2 < • • • < i2t-1, we associate a point P{B) € En as follows:

P{B) = {xux2, ■ • • ,xn),

where we put Xj =0, for j <£ B and

v if 1 < v < t
x,
21 — v \i t + \ < v < 2t — 1.

For example, if n = 8 and B = {1,5,6} (we have t = 2 here),

then P(B) = (xx,x2,-■• ,x8), where x2 = x3 = x4 = x7 = x8 — 0,
xx = 1, x5 = 2 and x6 = 1, that is, P(B) = (1,0,0,0, 2,1,0, 0).

Now, suppose an r-colouring \ °f Pn is given. Then {P{B) :

B is a (21 - l)-subset of [n]} C En receives an r-colouring. Now,
associating the same colour to a (21 — l)-subset B as that of P{B),
we get an r-colouring on the collection of (21 — l)-subsets of [nj.

Therefore, by our choice of n, there exists a (21 + l)-subset A =

{ji, j2, • • • , j2t+i} of [n] such that all the (21 — l)-subsets of A are
of the same colour.

In particular, considering the subsets A\ = {ji, j2, • • • ,j2t-i},

A2 = {j2,j3,--- ,j2t}, and As = (j3,j:4, • • • ,j2f+i} of T, the three
points P{A\), P{A2) and P(A8) have the same colour.

Now, the points P(AX), P{A2) and P(A8) can differ from each
other only in the jr-th positions for r = 1, • • • , 2t +1. We have zeros
in all the remaining positions. Thus the points P{AX), P{A2) and
P{As) will respectively be of the following forms (where for these
n-tuples, we show the values at the positions j2, j3, jt, jt+l, j2t_i,
ht and j2t+1 ):

(,1, ■ ,2, • ,3, •• • ,t- 1, • ,1, ■ ,0, ■ .0,),

(.0, ■■ ,1, • ,2, • ,t 1, ■ , t, • ,2, •• ,1, •■ .0.),
(.0, ■ ,0,' • ,1, ■ A-2. • :t-1, • ,3, ' ,2, ■ ,1,).
 4. FURTHER RAMSEY CONFIGURATIONS 69

We can now easily calculate the distances between the pairs of

points in the set (P(Ti), P{A2), P{A3)} and obtain that

d(P(A1),P(A2)) = d(P(A2),P(A3)) = VTt

and

d(P(A1),P(A3)) = Vst^e.

□
Theorem 4.2. (Frankl and Rodl) All isosceles triangles are
Ramsey.

Proof. Let ABC be an isosceles triangle with sides a, a and b

where the side BC is of length b and angle at A is a.

In Theorem 4.1 above, taking t very large, we can make the

angle P(Ai)P(A2)P(A3) very close to 180°. Therefore, we take t
large enough so that

a < angle P(Ai)P(A2)P(A3) < 180°.

Now, we rotate the triangle ABC around the side BC and let
A0BC be the projection of it on the ABC-plane after it is rotated
through an angle /3.

Since after a rotation by an angle 90°, the projection coincides

with the side BC, there exists such that the angle at A0 of the pro¬
jection triangle A13 BC is equal to angle P(Ai)P(A2)P(A3). Since
the triangle A13BC is similar to the triangle P(Ai)P{A2)P(A^) and
the later is Ramsey, by suitable scaling of the coordinate system, it
will follow that A13BC is also Ramsey.

We introduce the notation A'BC to denote the position of the

triangle after it is rotated through an angle (3.

By Theorem 3.2, the prism {A0,B,C} x {A13, A'} is Ramsey.

Since {A', B, C} is a copy of {A, B, C} and the first one is a subset
of the prism {A0,B,C} x {A0, A'}, we conclude that {A,B,C} is
Ramsey. O
70 5. EUCLIDEAN RAMSEY THEORY

5. A theorem of Graham

In this section, we talk about another type of Euclidean Ram¬

sey theorems, namely when a fixed Euclidean space En is finitely
coloured and for a finite subset S one looks for monochromatic con¬
gruence copies of S in En. We shall present a proof of a result of
this type (some other results of this type will be mentioned in the
notes, in the last section of this chapter) which is one of the most
interesting results in the subject of Euclidean Ramsey theory.
Gurevich had conjectured that for any finite colouring of E2,
there are always three points having the same colour and forming a
triangle of area 1. Graham ([76], [78]) proved the following result,
from which a strengthened form of Gurevich’s conjecture follows at
once as a corollary.

Theorem 5.1. (Graham) For a positive integer r, there is

a positive integer Tr such that given any r-colouring of the set of
integer lattice points in E2, there is always a right triangle of area Tr
such that its all three vertices are lattice points of the same colour.

Here, following [4], we shall give a shorter proof of Theorem 5.1

of Graham. It should be noted that while our proof will be shorter,
we shall be still exploiting the main idea of Graham.

Remark 5.1. We shall consider the Euclidean plane E2 en¬

dowed with rectangular Cartesian coordinates. From the proof, we
shall see that the triangle in Theorem 1 can be made to satisfy fur¬
ther constraints of having two of its sides parallel to the axes. It will
be easy to see that the arguments apply to non-rectangular case as
well. By suitable scaling, the following corollary is easily obtained.

Corollary. For any a > 0 and any pair of non-parallel lines

L\ and L2, in any partition of the plane into finitely many classes,
some class contains the vertices of a triangle which has area a and
two sides parallel to the lines Lt.

Notations. In what follows, for positive integers k and r, we

shall use the notation W(k,r) which was defined in Section 4 of
Chapter 1, in the statement of the classical theorem of van der
Waerden. That is, IV(k,r) denotes a positive integer such that for
any r-colouring of the set {1,2, • • • ,\V(k, r)}, there is a monochro¬
matic arithmetic progression of k terms. The theorem of van der
Waerden guarantees the existence of such a number W(k,r) (and
 5. A THEOREM OF GRAHAM 71

hence the existence of the smallest positive integer with this prop¬
erty).

Now, we proceed to prove Theorem 5.1.

PROOF. Our proof is by induction on r.

When r = 1, considering the triangle with vertices (0,0), (1,0)

and (0, 2), we get a right triangle with two sides parallel to the axes
and area 1.

For r > 2, we assume that there are positive integers Cr_i,

Dr-1(> 2) and Tr_x such that for any (r — l)-colouring of the set of
integer lattice points in the rectangle {(a, b) : 0 < a < Cr-X — 1, 0 <
b < jDr_i — 1}, there is a right triangle of area Tr_x with vertices in
this set and two sides parallel to the axes such that all its vertices
have the same colour.
Let

TV-i = Tr_i! Tr_i,

wr = W(Er-i(Cr-i + 1), r),
vr = wr\ Er-1-

Now, suppose an r-colouring of the set of integer lattice points in

E2 is given. Considering the points {(nTT_x,0) : n = 1,2,3, •••},
there is a monochromatic set (say with colour r)

S = {(ar + igrTr-i, 0) : 1 < i < TV-^CV-i + 1)},

such that S C {(nTr_l5 0), 1 < n < wr}. Obviously, gr <wr.

Now, consider the lattice points

Qij — ^ar + igrTr-1, — — j^j , 1 < i < Tr_iCr_i, 1 < j < Dr-\.

Case I. (One of the Qlj,s have colour r) If a QtJ from the above
collection have colour r, then we consider the triangle with vertices

2v
Qij — ( ar + WrTr-1, yy j 1 J Pi — iar + igrTr-1, 0)
\ Er^xgr J

and Pij — (Qir T 1grPi—1 "F gr ^'v — 1 . j 0j ,

where the last two points belong to *5.

72 5. EUCLIDEAN RAMSEY THEORY

Now, Qij, Pi and PlJ are of the same colour and vertices of a
right triangle with two sides parallel to the axes and area

1 T ^r-l 2vr
_
j — Er — \Vr.
2 ' 9r r_l j ' Er.igr

Case II. (None of the Q^s have colour r)

In this case, the collection of QlJ,s receive an (r — l)-colouring.
We consider the sub-collection
Er-i . rri 2vr
T T'Qr-’-r— li
2Tr_i J Er-\gr
By the induction hypothesis, considering the gaps between the
points in this rectangular array of points, there is a right triangle
with two sides parallel to the axes and with vertices at three Qtj's
of the same colour and area
bj i Qy 2 x) ^
t;r— 1 ET—\VT.
2 Er—\gr

Therefore, from Case I and Case II, the result holds in the case
of r-colouring , with
2vr
Tr — Tr_ivr, CT - (Cr_i + l)Er-iWr and Dr = ——Dr_i.
JT— 1
□
6. Notes

Regarding nonhomogeneous equation, if Q is a field of charac¬

teristic zero, then the linear non-homogeneous equation

CqXq + C\X\ + • • • + cnxn = b, Ci, (0 b 6 Q

is regular on Q if and only if ^ 0 . For a proof of this, one

may look into [83].
In connection with Remark 3.2, Graham [79] gives a simple and
nice proof of the particular case that a configuration consisting of
three collinear points x, y, z with

distance (x, y) = distance (y, z)

is not Ramsey.
In various ways the real line can be 2-coloured so that points
which are unit distance apart have different colours. A natural
problem can be posed to determine the minimum number of colours
needed to colour the real line so as to avoid assigning the same colour
 6. NOTES 73

to points whose distance apart lies between 1 — e and 1 + e, where

0 < e < 1. This and several related problems are considered in [43].
We now mention some results which are similar in nature to
Theorem 5.1 where for finite colourings of a fixed Euclidean space
En, one looks for monochromatic congruent copies of finite sets in
En. In fact, one of the early papers [55] in the area had already
results which say that vertices of some particular type of triangles
are among those finite sets which have monochromatic congruent
copies for any 2-colourings of E2. For instance, isosceles triangles
with vertical angles 30°, 72°, 108°, 120° or 150° are among those
triangles. Shader [159] proved that vertices of any right triangle
have monochromatic congruent copies for finite colourings of E2.
Before that, Shader and Erdos had established (see [159] ) the result
for certain types of right triangles.
'
 CHAPTER 6

Additive number Theory and related questions

1. Introduction

In this chapter, in Section 2, we shall see some Ramsey type

theorems in additive number theory. In Section 3, after discussing
some problems in additive number theory, we shall see an applica¬
tion of the infinite version of Ramsey’s theorem (Theorem 3.4) in
dealing with a corresponding multiplicative problem in combinato¬
rial number theory.

2. Zero-sum problems

We start with the following theorem due to Erdos, Ginzburg and

Ziv [52]. We point out that this result in combinatorial number
theory is a Ramsey type theorem; it talks about some unavoidable
regularity. This result also motivated study of certain Ramsey type
problems for Graphs (see [29], for instance).

Theorem 2.1. (Erdos-Ginzburg-Ziv Theorem) If for some

integer n> 1, we are given a sequence ai, a2, •• • , a2n-i of not neces¬
sarily distinct integers, then there exists a set I C {1, 2, • • • , 2n — 1}
with \I\ — n such that

^ at = 0 (mod n).
iEl
There are many proofs of the above theorem available in the
literature (see [10] or [121] , for instance). Here we present a proof
due to Zimmerman which was presented by Alon and Dubiner in
[10]. In Remarks 2.4 and 2.6 we shall sketch two other proofs.
75
76 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

PROOF. We first prove the theorem in the case that n — p, is

prime.
Let J = {1,2,--* ,2p — 1}. We now consider the sum

s= E fe-T.
ICJ, |/|=p V*e/ /

After expanding, 5 is a sum of monomials of the form C ElzeJ af‘

such that Y1 h — P ~ 1- The number j of distinct at's appearing in
one such monomial does not exceed p — 1. One observes that the p-
subsets I of J that contribute to the coefficient of this monomial are
those which contain these particular a^s appearing in the monomial.
Therefore, the number of distinct p-subsets I of this type is •
This number is clearly congruent to 0 modulo p. Again, it is clear
that each such I contributes equally to the coefficient and therefore
the coefficient is congruent to 0 modulo p. Since this is true for the
coefficient of each of the terms, we obtain that S is congruent to 0
modulo p.
Now, if possible, assume that there is no subset / C J of cardi¬
nality p such that Yliei ai — 0(mod p).
Then, by Fermat’s little theorem, for each such subset we have

- 1 (mod p).

Since there are ^2p~1 j such subsets, we have

5 = M = 1 (mod p).

This contradicts the already established fact that S is congruent to

0 modulo p.

Hence, there must be a subset I C J of cardinality p such that

Y.iaai = O(modp). Thus the result is established when n is a
prime.

For the case n — 1, there is nothing to prove. For n > 1, we

proceed to prove the result by induction on the number of prime
factors (counted with multiplicity) of n. Therefore, if n > 1 is not
a prime, we write n = mp where p is prime.

By the result we have just established, each subsequence of 2p-1

members of the sequence ■ ,a2n_i has a subsequence of p
 2. ZERO-SUM PROBLEMS 77

elements whose sum is 0 modulo p. From the original sequence we

go on repeatedly omitting such subsequences of p elements having
sum equal to 0 modulo p. Even after 2m — 2 such sequences are
omitted, we are left with 2pm - 1 - (2m - 2)p = 2p - 1 elements
and we can have at least one more subsequence of p elements with
the property that sum of its elements is equal to 0 modulo p.

Thus we have found 2m — 1 pairwise disjoint subsets I\, /2, • • • ,

12m-i of {1, 2, • • • , 2mp - 1} with \It\ = p and Gj = 0(mod p)
for each i.

We now consider the sequence &i,i>2,--* , 62m_ i where for i E

{1, 2, • • • , 2m - 1}, bt is the integer £ £ie/. ar

By the induction hypothesis, this new sequence has a subse¬

quence of m elements whose sum is divisible by m. The union
of the corresponding sets /, will supply the desired subsequence of
mp — n elements of the original sequence such that the sum of the
elements of this subsequence is divisible by n. □

Remark 2.1. Let us now observe that in Theorem 2.1, the num¬
ber 2n — 1 is the smallest positive integer for which the theorem
holds. In other words, if f(n) denotes the smallest positive integer
such that given a sequence ai,a2,-- - , a/(n) of not necessarily dis¬
tinct integers, there exists a set I C {1,2,-•• ,/(n)} with \I\ — n
such that J2iei ai = 0(mod n), then f(n) = 2n — 1. This can be seen
as follows. From Theorem 2.1, it follows that f(n) < 2n— 1. On the
other hand, if we take a sequence of 2n — 2 integers such that n — 1

among them are 0 modulo n and the remaining n — 1 are 1 modulo

n, then clearly, we do not have any subsequence of n elements sum
of whose elements is 0 modulo n.

Definition 2.1. In general, for any positive integer d, one de¬

fines f{n,d) to be the smallest positive integer such that given a
sequence of f{n,d) number of not necessarily distinct elements of
Z+d, there exists a subsequence xix, xi2, • • • , xln of length n such
that its centroid (xtl + xl2 H-f xin)/n also belongs to Z+d.

Remark 2.2. The number of elements of Z+d having coordi¬

nates 0 or 1 is 2d. Considering a sequence where each of these
elements are repeated (n - 1) times, we observe that

1 + 2d(n - 1) < f{n,d).

78 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

Again, observing that in any sequence of 1 + nd(n — 1) elements

of (Z/nZ)d there will be at-least one vector appearing at least n
times, we have
1 + nd(n — 1) > /(n, d).

From Remark 2.1 above, /(n, 1) = f(n) = 2n — 1.

For the two dimensional case, it is conjectured by Kemnitz [105]
and others that the lower bound is sharp, that is /(n, 2) = 4n — 3.
Harborth [88] and Kemnitz [105] have established this conjecture
for few particular values of n.

Next we present a result due to Alon and Dubiner [10] which

says /(n, 2) < 6n — 5. As in the case of Theorem 2.1, one observes
that it is enough to prove the result when n = p, a prime.
We mention that the upper bound for /(p, 2) has been improved
since then. Gao [73] proved that for all primes p, /(p, 2) < 5p — 1.
Before that Alon and Dubiner [10] could show that f(p, 2) < 5p — 2
when the prime p is sufficiently large. Recently Ronyai [150] went
very close to proving the conjecture by establishing that f(p, 2) <
4p — 2 for all primes p. From the result that f(p, 2) < 4p — 2 for
primes p, using an inequality of Harborth [88] (see [150]) it can be
deduced that /(n, 2) < (41/10)n, for general n. We give a proof
of the result of Alon and Dubiner, since it involves some very nice
combinatorial ideas and is not very lengthy at the same time. To
make our presentation complete, we shall state and prove certain
necessary results which are interesting in themselves. After proving
the result of Alon and Dubiner, we shall give a proof of the stronger
result by Ronyai [150]. It should be mentioned that the proof of
Ronyai uses certain results and techniques due to Alon and his co¬
authors.

Theorem 2.2. (Warning’s theorem) Let Fq be the finite field

with q-elements where q = pr, p being a prime and r > 1 any integer.
Let f(xi,x2, • • • ,xn) E Fq[xi,x2, •••,£„] be of degree less than n.
Then the number of solutions of the equation f(xi, x2, • ■ ■ , xn) = 0
in Fq = Fq x • • ■ x Fq is divisible by p.
s -''
n times

Proof. Let

g{x\,x2, ■ ■ ■ ,xn) = 1 - {f{xux2,--- ,xn))q-\

Then the degree of g is less than n(q - 1).

 2. ZERO-SUM PROBLEMS 79

Since for {aua2, • • • , an) 6 Fgn,

p(o:i,Of2, ■ • • ,an) = 1 if f{aua2, • • • ,an) = 0

and

g{<* i> «2, • • • , An) = 0 if f(aua2, ,an) ^ 0,

the number of solutions of the equation f(xi,x2,--- ,xn) =0 in

Fq is E g{oc\,a2, • • • , atn) where the sum is over all (aq, a2, • • • , an)
€ Fq and proving E <?(ai, <a2, • • • ,an) = 0 in Fq would prove our
assertion.

Considering a monomial Cx^ • • • xl£ appearing in the expansion

of g,

£
ai,',an€E9
“i‘= (£
\a\£Fq
«i‘)j ■■■{ £ <"V
\an£Fq )

If one of the zr’s, say z: is 0, then YaieFq ^i1 = q = 0 € Fq. On

the other hand, if all the zr’s are non-zero, then at least one of the
zr’s (say o) is less than q — 1 because of the degree consideration.
Now, there is a € F* such that a11 ^ 1. Since multiplication by a
permutes the elements of Fq, YQleFq ^i1 = YaieFq R11^!1 and since
a11 # 1, EQ x£Fq aY — 0- Therefore, in any case,

( £ a'l) " ‘ ( £ an ) — 0-
Ql £Fq Qn£Fq

The result follows trivially as E g{&i , o2, • • • , otn) would be sum

of these sums corresponding to each of the monomials appearing in
the expansion of g. □

Theorem 2.3. ( Chevalley’s Theorem) Let f(x i,x2,--- ,xn)

e Fq[xi,x2, ■ ■ ■ ,xn] be of degree less than n satisfying /(0, 0, • • • ,0)
= 0. Then there exists (c^, a2, ■ ■ ■ , an) € with not all at’s zero,
such that /(oq, a2, ■ ■ • , an) — 0.

Proof. Since /(0,0, • • • ,0) = 0, there is at least one solution.

Now, by Theorem 2.2, the number of solutions of the equation
f(xi,x 2, ••• ,xn) — 0 in Fff is a multiple of p and hence at least
p > 2. This proves the theorem. □
80 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

Remark 2.3. In the case of simultaneous solutions for any finite

number of polynomials in Fq[x\,x2, • • • ,Tn], if the sum of their de¬
grees is less than n, to obtain similar result as in Theorem 2.3, one
can modify the construction of the polynomial g in Theorem 2.2.
Remark 2.4. As shown in [9], [121] and [10] etc., the theorem
of Erdos, Ginzburg and Ziv (Theorem 2.1) is an easy consequence
of Chevalley’s theorem. We have already seen that it is enough to
establish the case when n = p is a prime.
Given a sequence ai, a2, • • • , «2p-i of elements in Fp, we consider
the following system of two polynomials in (2p — 1) variables over
the finite field Fp:

2p—1
Y arf-1 = 0,
i=i
2p-l
Y xrl = o.
i=i

Since 2(p - 1) < 2p — 1 and x\ = x2 = • • • = x2p-\ = 0

is a solution, by the above remark, there is a nontrivial solution
(yiu“ ,V2P-i) °f the above system. By Fermat’s little theorem,
writing I = {i : yl / 0}, from the first equation it follows that
52i£i ax — 0 and from the second equation we have |7| = p.
Remark 2.5. We give one more application of Chevalley’s the¬
orem in establishing a Ramsey type result. We need some more
definitions related to graphs.
Let the vertices of a graph G be denoted by iq, v2, • • ■ , vn and its
edges by ei,e2, • • • , em. Then the incidence matrix of G is defined
to be the matrix M(G) — [my], where is the number of times
(0 or 1) the vertex vt is incident to the edge er
For an integer k > 1, a graph G is said to be k-regular if each
vertex v of G has degree k. A regular graph is a graph which is k-
regular for some positive integer k.
Suppose we have a 4-regular graph and to it we add one more
edge having two of the vertices of the original graph as its ends.
Let the final graph G be such that it has n vertices and 2n + 1
edges. Then Alon, Friedland and Kalai [12] have applied Cheval¬
ley’s theorem to show that G contains a 3-regular subgraph. As
Alon, Friedland and Kalai point out, if we consider a graph on 3
vertices where each pair of vertices happens to be incident to two
 n
2. ZERO-SUM PROBLEMS 81

edges, then that graph does not have a 3-regular subgraph. There¬
fore ‘adding one more edge’ is necessary. The above mentioned
result is related to Berge’s conjecture; for a proof of this conjecture
and related references, one should look into [170].
We now give the proof of the result as given by Alon, Friedland
and Kalai. Let [dij] be the incidence matrix of the given graph G.
Consider the system of congruences:

Y, dijX2j = 0 (mod 3) (i = 1, • • • ,n).

j=i
Since 2 n < m and (ci,C2,** - , cm), where Ci = 0(mod 3), is
a solution, by Chevalley’s theorem we have a nontrivial solution
(cVl, 0:2 , ‘ ' ' i Oc-ra).
Let (<f> 7^) J C {1, 2, • • • , m} be the set of those indices j such
that oj ^ 0 (mod 3).
Now, for j G J,
Q!j = 1 (mod 3).
This implies that

y dij = 0 (mod 3), (i = 1, • • • ,n).

j€J

Now, for each j £ J there are two values of i G {1,2, ••• , n}

such that dij ± 0. Let I C {1, 2, • • • , n} be defined by

/ = {{ g {1, 2, • • • , n} : 3j G J such that ± 0}.

Then, for i G /, 1 < ai,j < 5 and this together with the
fact that EjgjOij = 0(mod 3), would imply

y ai,j — 3.
jeJ
Thus the edges e3 with j G J and vertices vt with i G / form a
3-regular subgraph of the given graph.
LEMMA 2.4. Let p be a prime and bx — (014,01,2), b2 — (o2,i, 02,2),
... , 63p = (a3pii, a3pi2) be a sequence of 3p elements of Fp © Fp such
that
b\ + b2 + • • • + b%p = (0,0), in Fp Fp.
Then there is a p-subset I of {1, 2, • • • , 3p} such that

ybt = (0,0), in Fp® Fp.

iei
82 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

Proof. We consider the following system of equations in 3p — 1

variables xx over Fp:

3p—1

Y
i= 1
= °’

3p—1

Y oitf-1 = °’
i— 1
3p—1

Y tf"1 = o.
i=i

Since 3(p - 1) < 3p — 1 and X\ = x2 = ■ • • = t3p_i = 0 is a

solution, by Chevalley’s Theorem (see Remark 2.3), there is another
solution. Let J C {1, 2, • • • ,3p— 1} be the set of all indices of the
non-zero entries of such a solution.
By Fermat’s little theorem, from the first two equations it follows
that
Ybi = (°»0)» in FP®FP
ieJ

and from the third equation we have

| J\ = p or 2p.

If | J\ — p, we take I = J and if | J\ = 2p, we take I = {1, 2, • • • , 3p}\

J satisfying the assertion of the lemma. □

Definition 2.2. For integers 1 < k < n, a k-arrangement a of

a set D, with \D\ = n, is an injective map from [k] into D. This
is equivalent to giving an ordered /c-tuple (c*(l), q(2), • • • ,a{k)) of
elements of D.
Definition 2.3. Let A = [atJ] be a matrix with m rows and n
columns, where the entries a,j belong to a commutative ring R and
m < n.
The permanent of A, denoted by per A, is defined by

per A ^ ) Ri,q(1)®2,q(2) ' * * ^m,o(m) 1

where the summation is taken over all m-arrangements a of [n].

Thus, if
al,l «1,2
A =
a2,l ^2,2
then per A = a1)1a2,2 + ai,202,1-
 2. ZERO-SUM PROBLEMS 83

Lemma 2.5. Let

P{x 1, • • • ,Xm) = 51 bu n z*
t/C[m] i£U

be a multilinear polynomial over a commutative ring with identity.

If P(xi, • • • , xm) = 0 for each (xx, • • • , xm) E {0, l}m, then P is
identically zero, that is, bu = 0 for all U C [mj.

Proof. We proceed by induction on m.

For m = 1, there is nothing to prove. Suppose the result is
proved for m < M — 1 where M > 1.
Given that P(xi, • • • ,xm) — 0 for each (x\, ■ ■ • , iM) E {0,1}A/,
we put xm — 0 and by induction hypothesis, it follows that co¬
efficients of all terms not containing xm are equal to zero. Now,
putting Xm — 1) it follows that the remaining coefficients are all
zero. □

Lemma 2.6. Let A = [a^] be.a m x m matrix over Fp such that

per A / 0.
Then for any C\, • • • , cm E Fp, there are t\, e2, • • • ,cm E {0,1}
such that Yff1-1 ej°i,j 7^ ci for all \ < i < m.

Proof. If possible let the lemma be false. Then there are no

Ci, 62, • * • , cm as above.
We consider the polynomial
m i m
P(x 1, , Xm) = PJ I 55 ai,jxj ~ ci
i=1 V=1
By our assumption, P(xi, • • • , xm) = 0 for each (xi, • • • , xm) E
{0,1}-.
Let P'(xi, • • • , xm) be the multilinear polynomial obtained from
P(x!, • • • , Xm) by first writing P in the standard form as a sum of
' £.
monomials and then replacing each monomial of the form bv n*et/ xi >
with 5i > 1, by buUieuxi-
Clearly, P'{xi, • • • ,xm) = P{xu • • • ,xm) = 0 for each {xu- • • ,xm
E {0, l}m. Therefore, by Lemma 2.5, P' is identically zero.

But, the coefficient of n™i xi in P\x ,xm) (which is the

same as the coefficient of n™i xi in P{xir * • ,^m)) is per A ^ Q
and we arrive at a contradiction.

Hence the lemma. □

84 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

Remark 2.6. Following the treatment in [10], we shall now

see how to derive Erdos-Ginzburg-Ziv Theorem (Theorem 2.1) as a
consequence of Lemma 2.6.
For a prime p, we are given a sequence ai,a2,--- , a2p_i, of
elements in Fp. We identify them with the integers in the set
0,1, • • • ,p — 1 representing them (mod p) and by rearranging the
elements, if necessary, we assume that a\ < a2 < • • • < a2p_i.
If there is an i < p— 1, such that a, = a,-+p_ i (which means that
a, = ai+1 = • • • = a,+p_i), then taking I = {z, i + 1, • - • , z + p — 1},
the theorem is established.
If there is no such z, we define

bt = di - al+p_!(/ 0) V z, 1 < i < p — 1.

Let ci,c2, • • • ,Cp_i be the set of all elements in Fp besides the

element Y?yFp aj) ■
Let A = [ajj] be the (p — 1) x (p — 1) matrix over Fp defined by
altj = bj for all 1 < z, j < p — 1.
Now,
p-1
per A — (p - 1)! • J^[ bj ^ 0.
j=i

Therefore, by Lemma 2.6, there are ti, e2, • • • , ep_l E {0,1} such
that €jai,j 7^ C{ for all 1 < z < p — 1.
Therefore,
p-l 2p—1
^2ejai,j — ~ ^2 aj-
3=1 j=P

Rearranging,
p-i
«2P-i + (aj+p-1 + ei{aj ~ aj+P-1)) = 0.

3=1

Since each term ai+p^ -ai+p-i) is either al+p_! or au this

gives a p-subset of the sequence {a(} the sum of whose elements is
0, as required.

Exercise 2.1. Let p > 2 be a prime and al5a2,--- ,as is a

sequence of integers each prime to p satisfying al a2 (mod p)
where 2 < s < p.
Then the set {E^=i eiai : 6 = 0 or 1} contains at least (s + 1)
distinct congruence classes modulo p.
 2. ZERO-SUM PROBLEMS 85

Exercise 2.2. Show that starting with arguments as in Remark

2.6 and then applying the above exercise one can get a proof of the
Erdos-Ginzburg-Ziv Theorem (Theorem 2.1). (In fact, this is how
the original proof of Erdos, Ginzburg and Ziv [52] goes.)

Before stating the next lemma, we fix some notations. For a

vector v = (d,e) in Fp 0 Fp, let v* = v*(p) denote the (column)
vector in 02ip1 ^ Fp whose first (p — 1) coordinates are d and last
(p — 1) coordinates are e.
Lemma2.7. Let v\,v2,--- , p2p-2 be 2p — 2 vectors in Fp0Fp
and A be the (2p — 2) x (2p — 2) matrix whose columns are the
vectors v*, v2, • • • , v2p_2. U m Fp> Per A ^ 0, then for any vector
v = (d,e) m Fp 0 Fp there are c1,e2,-- - , e2p_2 G {0,1} such that
v = Ei=l tiVi-

Proof. Let c = (ci, c2, • • • , c2p_2) be a vector whose first (p— 1)

coordinates are all the elements of Fp different from d and the last
(p — 1) coordinates are all the elements of Fp different from e.
By Lemma 2.6 there are ex, e2, ■ ■ ■ , e2p_2 G {0,1} such that for
every 1 < i < p — 1, the z-th coordinate of differs from
Ci-

Therefore, by our choice of c, the first and the second coordinates

of Zl;=72 tjvj are d an^ e respectively. □

Definition 2.4. A line in Fp0Fp is a subset L of FPQ)FP

consisting of all the elements of the form x + ty, t G Fp, where x
and y are two fixed elements of Fp 0 Fp. If a subset S of Fp 0 Fp is
not contained in a line, then clearly S contains three elements u, v
and w such that u — w and v — w span Fp 0 Fp.
Lemma 2.8. Let S be a sequence ofQp—7 vectors in Fp 0 Fp such
that no line in Fp 0 Fp contains more than 2p 2 members of S.
—

Then there is a subsequence of ip—A members of S, which we denote

by ai, a2, • • • , a4p_4 such that if b{ — a2i - a2i_i for 1 < i < 2p - 2
and A denotes the (2p - 2) x (2p — 2) matrix whose columns are the
vectors b*, then in Fp, per A ^ 0.

Proof. We consider the elements ex — (1,0) and e2 = (0,1) of

Fp 0 Fp.
Let A0 be the (2p - 2) x (2p - 2) matrix each of whose first p - 1
columns are e* and each of whose last p - 1 columns are e*2.
86 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

Clearly

per A0 = ((p - l)!)2 ^ 0 in Fp.

Now, \S\ = 6p - 7 > p. Therefore, S is not contained in a line

and hence it contains three elements it, v and w such that it — w
and v — w span Fp ® Fp. In particular, the first column of the
matrix A0 is a linear combination of the two column vectors (it — w)*
and (v — w)*. Therefore, per A0 is a linear combination of the
permanents of the two matrices obtained from it by replacing its
first column by (u—w)* and (v — w)* respectively. Hence at least one
of these matrices, say the first, has permanent non-zero in Fp. We
define a\ — u and a2 = w. Then writing A\ for the matrix obtained
from Ao by replacing its first column by b\ where b\ — ai — a2, we
have per A\ ^ 0.
We proceed in this way defining the elements <21,02, ••• , &\p-\
sequentially.
Suppose we have defined ai, • • • , a2i for 1 < i < 2p — 2. Let
bj = a2j — a2j-1, 1 < j < i. If A, is the matrix obtained from
Ao by replacing its first i columns by the columns &*,••• ,6* such
that in Fp, per At ^ 0 then we proceed to construct a2l+i, a2l+2, bl+\
and Ai+i as above. We observe that writing S' = 5 \ {a!, • • • , a2l},
|5'| > 6p — 7 — (2p -3)2 — 2p— 1 > p and hence S' is not contained
in a line. The rest is the same as in the case of constructing A\. □

Theorem 2.9. (Alon and Dubiner)

/(n, 2) < 6n — 5.

Proof. As we have mentioned earlier, by an argument similar

to the one in Theorem 2.1, it follows that it is enough to prove the
result when n = p, a prime.
Let 5 be a sequence of 6p - 5 elements in Fp ® Fp.

Case I. (There are 2p — 1 members of S on a line in Fp ® Fp)

Let the 2p-1 elements x + Uy, 1 < i < 2p- 1 of S lie on the line
{x + ty : t € Fp} for some x, y € ® Fp. By Erdos-Ginzburg-Ziv
Theorem (Theorem 2.1), there is an / C {1,2, • • • , 2p — 1} with
|/| — p such that Yhei U = 0 (mod p).
Therefore, in Fp® Fp, Eia(x + Uy) = 0. This proves the theo¬
rem in this case.
 2. ZERO-SUM PROBLEMS 87

Case II. (No line in Fp 0 Fp contains more than 2p — 2 elements of

S)
By Lemma 2.8, we can renumber the elements of S so as to have
the first 4p — 4 elements a1? a2, • • • , a4p_4 satisfy the following:

If bt — a2l — a2i-i for all 1 with 1 < i < 2p — 2 and A denotes the
(2p — 2) x (2p — 2) matrix whose columns are the vectors 6*, then
in Fp, per A t).

Therefore, by Lemma 2.7, every vector in Fp 0 Fp is a {0,1}- lin¬

ear combination of the vectors bt. In particular, there are ej, e2, • • • , e2p
G {0,1} such that

2p—2 2p—2 5p—2

X] y: a2i~i X) o»-
i= 1 i=i i=4p—3

Hence in Fp 0 Fp,

5p—2 2p—2

X" ai + X] (®2i—1 + fx(fl2i — Cl2i—1)) = 0.

i=4p—3 i=l

Thus we have obtained a 3p-subset of S the sum of whose ele¬

ments is 0 in Fp 0 Fp and hence by Lemma 2.4, S contains ap-subset
the sum of whose elements is 0 in Fp 0 Fp. □

We now proceed to prove the sharper result due to Ronyai.

We shall need the following fact about polynomial functions on
Boolean hypercubes.

Lemma 2.10. Let F be a field and m a positive integer. Then

the monomials Yll£iXl,I C {1,2,*-- ,m} constitute a basis of the
F-linear space of all functions from {0, l}m to F.

PROOF. It is easy to observe that the dimension of the space

spanned by the monomials LIz<e/ xt, I C {1, 2, • ■ • , m} over F is 2m
which is the same as that of the F-linear space of all functions from
{0, l}m to F.

We proceed to prove the lemma.

For U C {1,2,--- ,ra}, writing Uc = (1,2,■■■ ,m}\U, we con¬

sider the function fv : {0, l}m F defined by

fu{x 1, X2, , Xm) {{ Xj {{ (1 — Xj).

jeu jeuc
88 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

Since the functions fu, U C {1, 2, • • • , m} clearly span the linear

space of functions from {0, l}m to F, our lemma follows from the
above expression. D

Theorem 2.11. (Ronyai)

f(p, 2) < 4p - 2.

Proof. Since the assertion is obvious for p = 2, we assume that

p is an odd prime.

Let m = 4p - 2 and V\ — (ai,&i), v2 = (a2,M>‘ "■> vm —

(am, bm), be a sequence of vectors in Fp ® Fp.

We consider

a(x1,x2, • • • ,xm) = IIxi’

/C{1,2,-- ,m},\I\—p i&I

the p-th elementary symmetric polynomial of the variables X\, x2,

i %m •

If we prove that there is a subset J of {1, 2, • • • , m} with | J\ = p

or | J\ = 3p such that YljeJ vj ~ (0, 0), then by Lemma 2.4 we shall
be through.
If possible, let this statement be false.

Let us consider the following polynomial in Fp[xi, x2, • • • , xm\:

def
P(xux2, ■■■ ,xm) - 1

(a(xux2, ■ • • ,xm) - 2).

Given C = (ci,c2, ••• ,cm) in {0, l}m, if the Hamming weight

(the number of ones) of C is 2p, then cr(C) = (2pp) = 2 G Fp and
therefore the last factor in P vanishes for (xi,x2, • • • ,xm) = C.

If the Hamming weight of C is p or 3p, then ((E^i aJxz)p_1 — l)

((E^i biXt)p 1 - l) = 0 for (xf, x2, • • • ,xm) = C, by our assump¬

tion.

Finally the third factor in P vanishes unless the Hamming weight

of C is divisible by p.
 2. ZERO-SUM PROBLEMS 89

Therefore, P vanishes on all vectors in {0, l}m except at 0 =

(0,0, •• • ,0). We have P(0) = 2. Thus, with the notations used in
m times
the proof of Lemma 2.10. we have P = 2/$. We observe that deg
P < 3(p — 1) + p = 4p — 3.

We now reduce P into a linear combination of multilinear mono¬

mials by replacing x\ by X{ for each i and let Q denote the resulting
expression. We note that as functions on {0, l}m, P and Q are the
same. Therefore, as a function on {0, l}m, Q = 2/$. Also, since
reduction can not increase the degree, we have deg Q < Ap — 3. But,
because of the uniqueness part of Lemma 2.10, Q has to be identi¬
cal with 2(1 — Xi)(l — x2) • • • (1 — xm). This leads to a contradiction
since the later has degree m = Ap — 2. □

We shall now present a result due to W. D. Gao [72] which, in

a particular case (when n is prime), gives more precise information
than the theorem (Theorem 2.1) of Erdos, Ginzburg and Ziv.
THEOREM 2.12. (Gao) For a cyclic group G (which we shall
identify with the additive group of positive integers modulo p) of
prime order p and any element a in it, given an arbitrary sequence
S = {gi, • • • , g2p— i} of 2p — 1 elements of G, if r(S, a) denotes the
number of subsequences of S of length exactly p whose sum is a,
then r(S,a) is 1 or 0 modulo p according as a is zero or not.

PROOF. For a positive integer m we have

J^amr(5, a) = (au ^-h av) • (21)

a—0 l<zi<—<»P<2p-l

Now,

/ \m \t\ Ak
\an +-L aiP) = zl t \. ..t.\an ‘ ’ ’ aJk■
V t1+-+tk=m O- Lk-
til’ ' ,!p}

Therefore, from (21),

(2p — \ — k\ m!
y>"V(S,a)= £ a
a=0 tl+-+tk=m
\ p—k )ti\---tk\
{j 1.—

Since for all k, 1 < k < p — 1, (2pplkk) = °(mod P)i we have

90 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

amr(S, a) - O'G Fp for m = 1, 2, • • -p - 1.

a=l

Since in Fv,

1 2 - • - (p - 1)

1 22

1 2”"1 •••(p-1)'’-1
we have

r(S, a) = 0 (mod p), for 1 < a < p - 1.

Finally,

r(S, 0) = I)-Ijr(S,a)

= j (mod p)

= 1 (mod p).

3. A question of Sidon and some related problems

We start this section with the following questions of Sidon in

additive number theory. Let

A : ai < a2 < • ■ •

be an infinite sequence of non-negative integers.

Let f(A,n) denote the number of solutions of n = at -f a; with
i < j-
Sidon asked the following questions.
Question 1 . How ‘dense’ can A be if f(A,n) < 1 for all n?
Question 2. Does there exist an A for which f(A,n) > 0 for
all n > n0, and, for every e > 0,
f{A,n)
-A 0?
ne
Before going to discuss about the status of these questions, we
define certain terminologies. A sequence of non-negative integers
 3. A QUESTION OF SIDON AND SOME RELATED PROBLEMS 91

A : ax < a2 < ■ ■ ■ is called a Sidon set if f(A,n) < 1 for all n. In

other words, A is a Sidon set if all the sums a* + a, are distinct.
Similarly, a finite set B of non-negative integers < b2 < ■ ■ ■ < bn,
is called a Sidon set if the sums bx + bj are distinct. A finite Sidon
set A C [TV] is said to be maximal with respect to TV, if there is no
Sidon set A' such that A C A' C [TV].

The greedy algorithm provides a simple way ( see [133], for

instance), to show that there is a finite Sidon set B C {1,2, - - ,7V}
with |J5| is not less than the integral part of TV1/3.

By similar arguments, one can show that there is a Sidon set A

such that | A fl [n]| is not less than the integral part of n1/3 for all
positive integers n.

In the other direction, Ruzsa [154] has proved that there is a

maximal Sidon set A in [TV] such that

\A\ — O ((TV log TV)1/3) .

Regarding the second question, by probabilistic method, Erdos

[45] showed the existence of a sequence of positive integers A for
which
ci log n < f (A, n) < c2 log n,

for large n, where C\ and c2 are positive constants.

But constructing an A satisfying even the weaker condition given
in Question 2 is open.
The following question of Erdos and Turan is also open.

Question 3. Does f{A,n) > 0 for all positive integers n imply

limsup/(A,n) = oo?
n —y oo

The corresponding multiplicative problem was answered affir¬

matively by Erdos.
We say that X C Z+ is a multiplicative base if for any n G Z+,
there are X, such that n = xy.

Erdos [48] proved:

Theorem 3.1. If X is a multiplicative base, then, for every
positive integer l, there exists a positive integer n = n(l) such that
n can be expressed as a product of two elements of A in at least l
different ways.
92 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

Nesetril and Rodl [123] gave a simpler proof of Theorem 3.1.

We present their proof now.-
Let S denote the subset of Z+ consisting of the square-free inte¬
gers, that is, the integers which are product of distinct primes. Let
S' be the collection of all finite subsets of primes.
We consider the map <j>: S —¥ S' defined by

(J>{PlP2 ' • ■ Ps) = {Pl,P2,"’ ,Pa}-

Clearly, 0 gives a one-to-one correspondence between S and S'.
With this observation, it is clear that to prove Theorem 3.1, it
is enough to establish the following theorem.
Theorem 3.2. Let Aj denote the set of all finite (possibly empty)
subsets of an infinite set A = {ai, a2, • • • , }. Let A C Af be such
that, for any P £ Af, there are Q,Q' £ A such that

Q U Q' — P and Q D Q' — <L.

Then, for every integer l, there exists a set P £ Af, such that
P can be expressed in at least l different ways as a union of two
disjoint elements of A.

Proof. (Ne§etfil and Rodl)

Let l > 2 be a given integer.
For every 2 = 1, 2, •••,/ — 1, we consider a partition

p)=c;uc;

which is defined as follows:

For any Q £ Q will be in C\ if Q € A\ Q € C\, otherwise.
By Theorem 3.4 of Chapter 1, there is an infinite set Bx C A
such that all 1-subsets of B\ belong to the same part corresponding
to the partition = C\ U C\ of

Applying the same theorem once again, there is an infinite set

B2 C Bx such that all 2-subsets of B2 belong to the same part
corresponding to the partition = C\ U C\ of .
We iterate Ramsey’s theorem in this way to obtain an infinite
set B C A such that for any i £ {1, 2, •••,/- 1}, all 2-subsets of B
belong to the same part corresponding to the partition ( ^ = CjuQ

of(T
 4. NOTES 93

If there is an element T G (f) such that B £ A, then T can be

written as T = 7\ U T2 where 7\ and T2 are two nonempty sets such
that TuT2eA and T, n T2 = <F.
Now, if |7\| = j, then |T2| = l — j. Since, 7j,T2 € *4, by the
construction of B we have

(j) c C{ and (t - ;) c ^

This implies that every S G can be represented as a union

of two disjoint elements of A in at least (f) > / ways.

On the other hand, if for every T G we have T e A, then

by an argument as above, any element of can be represented as
a union of two disjoint elements of A in at least ways. □

4. Notes

For a proof of Chevalley’s theorem (Theorem 2.3) due to Ax

and some information relevant to Theorem 2.3 and Theorem 2.2,
one may look into [103],
For a cyclic group G of prime order p and a sequence S =
{<7i,' • • , <?2P-i} of 2p — 1 elements of G, with the notations as in
Theorem 2.12, a result of Mann [113] says that r(S,a) > 1 for
all a G G, provided no element is repeated more than p times in
the sequence S. This result of Mann, along with Theorem 2.12
would imply ([72]) that r(S, a) > p for all a(7^ 0) G G, provided
no element is repeated more than p times in S. In [72], it was also
derived that r(5,0) > p + 1 unless only two elements x and y occur
in 5, x occurs p times and y occurs p — 1 times.
For general d > 1, it was proved by Alon and Dubiner [11] that
f{n,d) < c(d)n, where c(d) is a constant independent of n.
For further studies of combinatorial problems in additive num¬
ber theory we would like to suggest the books of Freiman [65], Mann
[114], Nathanson [121] and Halberstam and Roth [86]. An expos¬
itory article of R. Thangadurai [173] gives a quick introduction to
several results in this area.
We should mention that a quantitative version of Freiman’s the¬
orem [65], obtained by Ruzsa [153], regarding the structure of a
94 6. ADDITIVE NUMBER THEORY AND RELATED QUESTIONS

finite set of integers with small sumset, was one of the ingredients
in the very important work of Gowers which we mentioned in the
notes in Chapter 1. Ruzsa’s proof of Freiman’s result as well as
the Balog-Szemeredi theorem [18], which is also an ingredient in
Gowers’ work, is available in [121].
For some conjectures and results due to J. E. Olson related
to Erdos-Ginzburg-Ziv Theorem (Theorem 2.1) in more general
groups, one may look into a paper of Olson [127] and a recent
paper of B. Sury [165]. Caro [38] has written a nice survey on
results around zero-sum problems obtained until 1994.
Before the paper [123] of Nesetril and Rodl, Erdos [49] had
already applied Ramsey’s theorem to additive number theory. It
should be mentioned that apart from a new proof of Theorem 3.1,
the paper of Nesetril and Rodl [123] has other results using Ramsey
type theorems.
For more information about Sidon sets and related questions,
one may look into [86], [157] and [133]. The article of Sarkozy
and Sos [157], contains a remark regarding the way the problem of
Sidon sets is related to ‘anti-Ramsey type’ problems.
 ■VI

CHAPTER 7

Partitions of integers

1. Introduction

In the Chapter 6, we dealt with certain problems in additive

number theory.
The subject-matter of the present chapter also belongs to the
same area of number theory. More precisely, we shall deal with
certain properties of the partition function in this chapter.
Definition 1.1. A partition of a positive integer n is a repre¬
sentation of n as a sum of any number of positive integers, called
parts of the partition. Here, in a partition, the parts are not nec¬
essarily distinct and the order in which the parts are arranged is
irrelevant.
Since the order is irrelevant, we shall assume that the parts are
arranged in descending order of magnitude. Thus the integer 5 has
the following partitions:

5, 4 + 1, 3 + 2, 3+1 + 1, 2 + 2 + 1, 2 + 1 + 1 + 1 and l + l + l + l + l.

Remark 1.1. It is easy to observe that a partition of a positive

integer n is nothing but a solution in integers xz > 0 of the equation

X\ + 2X2 + ' ‘ ‘ + 712+ — n.

Definition 1.2. The partition function p(n) is the number of
partitions of n.
The domain of definition of the function p(n) is often extended
to the set Z of all integers by defining

p{0) = T p{~ 1) = P{~2) = 1 • • = 0.

95
96 7. PARTITIONS OF INTEGERS

2. Preliminaries

Some results on partitions can be derived by a graphical repre¬

sentation of partitions, where a partition is represented by horizon¬
tal rows of dots, the dots in a row corresponding to a part.
Thus the first two partitions of 5 mentioned after definition 1.1
above have respectively the following graphical representations.

• •••• • • • •
Now, consider the graphical representation of the partition 2 +
1 + 1 + 1 of the number 5 :

• •

If we read the above representation by columns, then it would

represent the partition 4 + 1.
Definition 2.1. Given a partition, a partition obtained from
it by reading its graphical representation by columns is called the
conjugate partition of the former. Clearly, the relation of being
conjugate is symmetric. Thus the partitions 4+1 and 2 + 1 + 1 + 1
of the number 5 are conjugate to each other.

Definition 2.2. A partition of a positive integer n is called self¬

conjugate if it is the same as its conjugate partition. In other words,
a self-conjugate partition is one whose graphical representation is
such that considering the first dot as the origin (0, 0), it is symmetric
with respect to the line x = — y on the xy-plane with rectangular
Cartesian co-ordinates.
For instance, the partition 3 + 2 + 1 of the integer 6 is self¬
conjugate. We can have a look at its graphical representation which
is given below.
 2. PRELIMINARIES 97

The following two theorems can be easily proved by using graph¬

ical representations of partitions.
THEOREM 2.1. Given a positive integer n, the number of par¬
titions of n into m parts is equal to the number of partitions of n
into parts the largest of which is m.

Proof. If we consider a partition of n in which at most m parts

appear, then in its graphical representation, there could be at most
m dots in any column.
Therefore, reading the graphical representation by columns, the
conjugate of such a partition will have no part larger than m. Thus
we have a one-to-one correspondence between the two types of
partitions mentioned in the statement of the theorem and we are
through. □

Theorem 2.2. Given a positive integer n, the number of par¬

titions of n into unequal odd parts is equal to the number of self¬
conjugate partitions of n.

PROOF. With the partition (2mi -f 1) + (2m2 + 1) + • • • of n

with mi > m2 > • • •, we associate a partition which is constructed
as follows. First, we consider the first row of 2m! + 1 dots and
after folding it in the middle, we place it in such away that keeping
the middle point at the North-West corner, it forms the sides (one
kept horizontally and other vertically) of an isosceles right-angled
triangle. Thus a row of seven dots

• ••••••
takes the following shape.

We continue doing the same with the other rows of the graphical
representation of the original partition and place the folded rows
one below the other so that keeping the middle point of the first
folded row at the origin (0,0), the successive middle points are all
98 7. PARTITIONS OF INTEGERS

on the line x = -y on the xy-plane with rectangular Cartesian

co-ordinates.
Thus, considering the graphical representation

• ••••••
• • • • •
• • •

of the partition 7 + 5 + 3 of 15, the partition associated to it in the

above way will have the following graphical representation.

• • • •
• • • •
• • • •
• • •

It is clear that for a given integer n, this method establishes a

one-to-one correspondence between the partitions into unequal odd
parts and the ones which are self-conjugate. □

Exercise 2.1. For a positive integer n, let Pdo(n) denote the

number of partitions of n with distinct odd parts.
Show that
p(n) = Pdo{n) (mod 2).

Exercise 2.2. For a given partition T of a positive integer n,

let gr,n denote the number of distinct parts in T and hr^n denote
the number of l’s appearing as parts in the partition T. Thus, for
the partition T given by6 = 3+ l + l + l of the number 6, we have
gr 6 = 2 and hr,e — 3. Let

Pl(n) =f Y, hT,n > Pp{n) =

where both the sums extend over all the partitions T of n.

With the above notations, show that for a positive integer n,

Pi{n) = PP{n).
 3. GENERATING FUNCTIONS 99

3. Generating functions

While working on problems related to an arithmetical function

(complex valued functions on Z+), one often utilises its generat¬
ing function. For an arithmetical function f(n) appearing in addi¬
tive number theory, it is convenient to use a power series F(x) =
£]/(n)xn, with general coefficient /(n), as its generating function.
In the case of the partition function p(n), the generating function
is described in the following theorem. For our requirements, which
are essentially formal, we establish the relation only for real x with
0 < x < 1.
Theorem 3.1. (Euler) For 0 < x < 1, we have
OO 1 oo

II
m= 1 1 x
= EM")*"-
n=0

PROOF. For x with 0 < x < 1, the infinite product on the left
hand side converges absolutely.
We write
OO 1

F(I) = n,
We introduce the functions

Fr{x) = n i—~—,
v ’ J-1,
m— 1
1 - xm

for integers r > 1.

Let pr(n) denote the number of partitions of n into parts not
exceeding r. In other words, pr{n) is the number of solutions of the
equation
n — X\ -F 2^2 T • • • T rxr.

Now, being a product of a finite number of absolutely convergent

series. Fr(x) is also an absolutely convergent series and we can write
OO

Fr(x) = 1 + Y,Pr(n)xn-
n= 1

This is obtained by expanding each factor into a power series,

multiplying them together and rearranging the resulting terms. For
instance, if we take r = 3, then writing

—3—— = (1 + x + x2 H-) (1 + x2 -I- x4 H-) (1 + x3 -I- x6 -I-),

771=1 ^ ^
100 7. PARTITIONS OF INTEGERS

the partition
3+2+2+1
of the number 8, such that no part exceeds 3, corresponds to the
product of x3 in the third factor, x4 = x2.x2 in the second and x in
the first.
Since pr(n) < p(n) with equality holding for r > n, we have
r OC

Fr{x) = 1 + Y P{n)xn + Y Pr{n)xn.

n=1 n=r+1

Since for 0 < x < 1, Fr(x) < F(x), we have

1 + Yp(n)xTl < Fr{x) < F(x) for all r.

n—1

Therefore, the series p(n)xTl is convergent. Since pr(n) <

p(n), we have
OO OC

1+ Y PM)xn < Y P(n)xn-

71=1 71=0

Therefore, for any fixed x in 0 < x < 1, the series Y^=\Pr{ri)xn

converges uniformly in r.
Since pr(n) -> p{n) as r —> oo, for every n, we have
OO OO / OO

1+ Y P(n)xn = 1+ Y rlim pr{n)xn = Um (1 + Y PrHxn

71=1 71=1 \ 71=1

= lim Fr(x) = Fix).

r_v ' ' ' '

□
Remark 3.1. One can easily find the following generating func¬
tions (all of which are absolutely convergent for jx| < 1) which
enumerate various restricted partitions of n.

i)
(1 + x)(l + t2)(1 + x3) • • •

enumerates partitions into unequal parts.

ii)
1
(1 — x)(l — x3)(l — X5) • • •
enumerates partitions into odd parts.
 3. GENERATING FUNCTIONS 101

iii)
(1 + x)(l + t3)(1 + t5)•••

enumerates partitions into unequal odd parts. In other

words, this is the generating function of the restricted par¬
tition function Pdo{n) defined in Exercise 2.1 in the previous
section.

iv)

(1 — x2)(l — X4) • • • (1 — x2m)

enumerates partitions of (n — t) into even parts not exceed¬

ing 2m. We observe that this is the same as enumerating
the partitions of \ {n — t) into parts not exceeding m, which
in turn, by Theorem 2.1, is the same as enumerating the
partitions of \{n — t) into at most m parts.

Generating functions for many other restricted partition func¬

tions can be found in [91] and various text books on the subject.
Some properties of partitions can be observed (see [91] for example)
from the forms of these generating functions.

REMARK 3.2. Given a self-conjugate partition of a positive

integer n, from its graphical representation one observes that the
graph consists of a square of m2 points and two tails representing
2
partitions of the integer nz™ into at most m parts. For example,
in the last graphical representation in Theorem 2.2 in Section 2, we
have m = 3.
Conversely, given a positive integer n and a square m2 not ex¬
ceeding n, there is a class of self-conjugate partition of n based on
a square of m2 points. Now, by Part (iv) of the last remark,

_x^__
(1 — x2)(l - X4) • • • (1 — x2m)

enumerates the partitions of \(n-m2) into at most m parts. There¬

fore,
oo y.m2

enumerates all self-conjugate partitions of n.

102 7. PARTITIONS OF INTEGERS

Hence, a glance at Part (iv) of the last remark, enables one to

see that Theorem 2.2 corresponds to the following identity of Euler.

OO
x
(1 + t)(1 + x3)(l + t5) • • • - 1 +
(1 — x2)(\ — xA) •••(! — x2m)

Exercise 3.1. Using combinatorial reasoning like the one em¬

ployed in the above remark, prove the following theorem of Euler.
1
(1 — t)(1 — x2)(l — x3)

x XA X“
= 1-
(1 — x)2 (1 — x)2(l — x2)2 (1 — x)2(l — x2)2(l — X3)2

4. An identity of Euler

We shall need the following identity of Euler in the Section 5. For

a more general identity (Jacobi's triple product identity) one may
consult [91], [108], [15] or [14]. The Rogers-Ramanujan identities
(which resemble the identity of Euler stated at the end of the last
section) and their combinatorial interpretations can be found in [91]
and [14].

Theorem 4.1. (Euler) For |x| < 1, we have

OO

JJ (1 — xn) = 1 — x — x2 + x5 + x7 * * — x12 — xl0 * + • • •

n— 1
OO

= 1 + + rH<3t+l>\
k= 1

Proof. The elementary proof which we give here is due to F.

Franklin. Let us write

(l-i)(l-a;2)(l-2;3)-.. = ^7(n)x”.

Clearly, every partition of n into unequal parts contributes to

the coefficient 7(71) of xn. For instance, the partition 6 = 34-2 + 1
contributes (-1)3 to the coefficient of x6. More generally, a partition
of n into m unequal parts contributes (-l)m to y(n). Therefore,

y(n) = pe(n) - p0{n),

 4. AN IDENTITY OF EULER 103

where pe(n) is the number of partitions of n into an even number

of unequal parts and p0{n) is the number of partitions of n into an
odd number of unequal parts.

Considering the graphical representation Gj of a partition of a

• • • • • C
• • • • D

• • • E

.4 B

G!

positive integer n into unequal parts we call the lowest row (AB in
the graph Gi) the base 0 of the the graphical representation. We
observe that the base 0 may just contain a single vertex represented
by a dot.

Again, starting from the last dot in the first row (that is, the
extreme north-east dot), the longest south-westerly line possible in
the graph will be called the slope a of the the graphical represen¬
tation. Once again, we observe that the slope a may just contain a
single dot.

When there are more dots in a than in 0 (as in our graph Gj),
we write 0 < a.
Similarly we use the notations 0 = o and 0 > a to denote
respectively the cases when 0 has the same number of dots as in a
or more number of dots than in a.

We consider all the possibilities one after another.

104 7. PARTITIONS OF INTEGERS

Possibility I : (ft < a)

This case is represented by the graph Gi above. We move the

base (3 into a parallel position next to a as is shown in G2.

.. A

..

• • • • • E
• • •

G2
This gives a new partition into distinct parts. Since the number
of parts go down by one, the parity of the number of parts in this
newly obtained partition is just the opposite to that of the one
corresponding to the graph Gi. We call this operation O. If we
consider the converse operation (Q, say), that is the operation of
removing the slope and putting it below the base, we observe that
is not possible in this case as it will not obey our convention of
writing every partition in the descending order of magnitude.

Possibility II : (/3 = a)

In this case, for partitions with graphical representations as in

G3, the operation O is possible.

• • • • •
• • • •
• •

G3

Operation O will not be possible exactly, for those partitions

having graphical representations such that (3 meets a as in the graph
G4, while operation Q will not be possible for any partition with
(3 = a.

• • • • •
• • • •
• • •

G4
 4. AN IDENTITY OF EULER 105

Possibility III : (/3 > a)

In this case, operation O is always impossible. Operation ft will

be possible (as in the graph G5) unless /3 meets a and the number
of dots in /3 is exactly one more than that in a. The graph G6 below
represents one such case.

• •••••
• • • • •
• • •

G5

• •••••
• • • • •
• • • •
G6

Thus, considering the two types of partitions of an integer n into

unequal parts, namely the ones having even number of parts and
the ones having odd number of parts, we have observed that there is
a 1-1 correspondence between partitions belonging these two classes
except those exemplified by the graphs G4 and G6.

In the cases exemplified by G4, the integer n is of the form

k + (k + 1) + • • • + (2k — 1) = ^k(3k - 1)

and we shall have an extra partition into an even number of terms

or an extra partition into an odd number of terms according as k is
even or odd.

In the other class of partitions exemplified by G6, n is of the

form

(k + 1) -f- {k + 2) + • • • + 2k — —k(3k -I- 1)

and again we shall have an extra partition into an even number of

terms or an extra partition into an odd number of terms according
as k is even or odd.
106 7. PARTITIONS OF INTEGERS

Therefore,

0, if n ^ ^k(3k ± 1)
7 (n)
(—l)fc, if n = \k(3k ± 1).

□
Remark 4.1. From Theorem 3.1 and Theorem 4.1 above, we
have
OO
1
i= n m= 1
1 - Xm
no
= (p{n)x) • (1 - x - x2 + x5 + x7 - x12 - x15 H-),
\n=0 /
from which, expanding and collecting terms, we get

p(n) + 5Z(-1)/c (p(n - sk) + p(n - tk)) = 0, (22)

>t> i

where

sk = ^k(3k - 1), tk = ^k(3k + 1) (23)

and the summation extends over all non-negative arguments of the

partition function.

MacMahon used (22) to calculate values of p(n) for 1 < n < 200.
In particular, he obtained p(200) = 3972999029388.
Remark 4.2. The partial sums of the terms of the arithmetic
progression

1,4, 7,10, • • • , 3d + 1, • • •

are related to the pentagons shown in the figure in the next page.
Writing p(n) for the sum of the first n terms of this progression,
we have

p(n) = J^(3d + 1) = in(3n - 1).

d-0 Z

The numbers p(n) and p(—n) — \n(3n + 1) are known as pen¬

tagonal numbers. For this reason, Theorem 4.1 is known as Euler’s
pentagonal number theorem in the literature.
 5. PARITY OF PARTITION FUNCTION 107

5. Parity of partition function

The arithmetic of partition function congruences was initiated

by Ramanujan. From MacMahon’s table for p(n) for 1 < n < 200,
Ramanujan [144] conjectured certain congruence properties of p(n)
and proved the following particular cases:
p(5n + 4) = 0 (mod 5), (24)

p(7n + 5) = 0 (mod 7). (25)

Ramnujan’s manuscript [147] contains more results on partition
function congruences. Later Hardy extracted some more material
from Ramnujan’s manuscript [147] and this appeared in a posthu¬
mous paper [146], A recent article [28] of BerndtBerndt, Bruce
C. and Ono, which provides many clarifications and corrections in
the above mentioned manuscript of Ramanujan, is a very valuable
commentary on it.
The general conjecture of Ramanujan as modified by Watson is
as follows.
//24m = l(mod 5a7bllc), then
p(m) = 0 (mod 5a7'jllc), (26)

where (3 is the integral part of Ap.

108 7. PARTITIONS OF INTEGERS

Watson [180] proved (26) for arbitrary powers of 5 and 7; and it

was Atkin [17] who first proved (26) for lln, for all positive integers
n.

Apart from these original papers, for some of these results, one
may look into the books [91] and [108].

Also, in the relevant sections in the ‘Commentary on Ramanu¬

jan’s collected papers’ by Bruce C. Berndt, which appears at the
end of the third printing of ‘Collected papers of Ramanujan’ [145],
one will find many references as well as recent developments in the
area.

Further, we recommend the above mentioned article [28] of

Berndt and Ono as it contains interesting information related to
partition function congruences.

The expository article [5] gives information and references to the

recent developments in the area of partition function congruences
and related questions.

Here we shall be mainly concerned with the question of parity

of p{n). Calculations support the ‘Folklore Conjecture’ [129] that
the number of n < x for which p(n) is even is ~ 1x. In other words,
it is believed that the partition function takes odd and even values
‘equally often’.

We start with the following result of Kolberg [109] which is a

pioneering result in this direction.

Theorem 5.1. (Kolberg) For n > 1, p{n) assumes both even

and odd values infinitely often.

Proof. Kolberg’s proof [109] of Theorem 5.1 is based on Eu¬

ler’s identity (22) mentioned in Remark 4.1 in the last section.

Reading modulo 2, from the above mentioned identity (22), we

have

p(n) + 5Z (p(n - sk) + p{n - 4)) =o (mod 2). (27)

k> 1

Now, suppose that for some positive integer a, p(n) is even for
all n> a. Taking n = f (3a - 1), from (27) we have

f a(3a — l)\ f a(3a — 1) \

P (-g-“)+M 2-" “ 1 ]+---+7>(2a-l)+p(0) = 0 (mod 2).
 5. PARITY OF PARTITION FUNCTION 109

Since p(0) = 1 and all the remaining terms in the above are even
by our assumption, we get a contradiction. Therefore, p(n) takes
odd values infinitely often.

Similarly, supposing that for some positive integer 6, p(n) is odd

for all n > 6, taking n = |6(36 +1), from (27) we see that the left
hand side contains an odd number of odd terms and we arrive at a
contradiction. □

Later, different proofs for Kolberg’s result were given by several

mathematicians. We here give the proof due to Newman [124]
which is very interesting.

Newman observes that

Lemma 5.2. If a sequence of integers T = {tn}, n > 0 is

ultimately periodic modulo a positive integer m (that is, there is
a positive integer N such that for integers p,q > N with p = q
(mod m), tp = tq), then the formal power series f(x) —
is congruent modulo m to a quotient of polynomials with integral
coefficients, the denominator having constant term 1.
Here by congruence of f(x) modulo m, it is understood that the
coefficients of the corresponding powers of x are congruent modulo
m.

Proof. If T is ultimately periodic modulo m, then there ex¬

ist two polynomials a (a;) and /3(x) with integral coefficients and a
positive integer d such that

f(x) = a{x) + fi(x) + xd3{x) + x2dffix) + • • • (mod m).

Therefore,

. . a(x)(l - xd) + /3(x) ,

no = ~ (mod m)-

□
The next lemma is quite interesting on its own.

Lemma 5.3. Let

oo
no = E o < co < (, < c2 < • • ■.
71=0
110 7. PARTITIONS OF INTEGERS

be a power series with integral coefficients and exponents such that

dn+1 — dn -o- cx) (28)

and

gcd(cn, cn+i, • • •) = 1, for all n > 0. (29)

Then there do not exist two polynomials a(x) and /3(x) with integral
coefficients such that o;(0) = 1 and

Pjx) (mod m) for some integer m > 1.

fix)
a{x)

Proof. Let
r

<*{x) = 55
n=0
a0 -

and

PiX) = 51
n—0
bnXH-

Now, for an integer m > 1,

a(x)f(x) = (3(x) (mod m)

implies that
55 ckan~dk = bn (mod m).
dk <n

Replacing n by dn in the above, we have

n n— 1

55 ckCidn-dk = Cn + ckadn~dk = bdn (mod m). (30)

k=0 k=0
Bv the condition given by Equation (28), we can choose n0 so
large such that for all n > no, dn — > r, dn > s. Then (30)
implies that for all n > n0,

cn = 0 (mod m).

This contradicts (29), since m > 1. □

We shall now see how Theorem 5.1 can be derived from Lemma 5.3
and Lemma 5.2. From Lemma 5.3, we observe that in particular,
oo oc

F(x) = I] (1 - xn) - 1 + 55(-l)* + Xik{:ik+1))

n=0 k—\
(and hence 1 /F{x), which is the generating function of p(n)) is
never congruent modulo m to a quotient of the type fi(x)/ot{x),
 5. PARITY OF PARTITION FUNCTION 111

where q;(t) and j3(x) are polynomials with integral coefficients and
Qf(0) = 1.

Therefore, by Lemma 5.2, it can not happen that for some inte¬
ger m0, p(n) is even for all n > ra0 or odd for all n > ra0.

Later, by more effective exploitations of Kolberg’s idea, first

Mirsky [118] and then Nicolas and Sarkozy [126] obtained better
results. We shall go directly to discuss further improvements made
by Nicolas, Ruzsa and Sarkozy [125].

Theorem 5.4. (Nicolas, Ruzsa and Sarkozy) There is a

constant Ci > 0 and a positive integer N0 such that for N > No,
there are at least C\N1^2 integers n < N such that p(n) is even.

Proof. We start with Euler’s identity (22).

Consider the set

M.n =f {ft} U {n — Sk : 1 < s*; < n) U {n — t* : 1 < < n).

From (22),

y: p(m) = 0 (mod 2).

m£Mn

Therefore, if \Mn\ is odd, then there is at least one m G Mn for

which p[m) is even.

Now, observing the sequence n > n — s\ > n — t\ > n — s2 > • • •,

\Mn\ is odd if and only if n belongs to an interval of type [tj, sJ+1).
Therefore,

|{(m, n) : n < N,m G Mn,p(m) = 0 (mod 2)}| > dN

for some positive constant c'.

For a fixed m, the number of integers n of the form m + tj or

m + Sj is at most the number of f s satisfying tj < N or Sj < N
which is, clearly, < c"Nl/2 for some positive constant c".

Thus there are at least c'N/c"N1/2 = CxNl/2 distinct values of

m < N for which p(m) is even. 0

The corresponding result for number of integers n < N such that

p(n) is odd obtained by Nicolas, Ruzsa and Sarkozy is as follows.
112 7. PARTITIONS OF INTEGERS

Theorem 5.5. (Nicolas, Ruzsa and Sarkozy) For each e >

0, there is a positive integer N\ = N\(e) such that if N > N\
then there are at least iV1//2exp( — (log 2 + e) log N/ log log N) integers
n < N such that p(n) is odd.

Proof. First we make the following claim.

Claim. For each e > 0, there is a positive integer Nx = Nx(e)

such that if N > Nx then there are at least Arl//2exp(—(log 2 +
e) log N/ log log N) integers n < N such that

p(n) ^ p(n — 1) (mod 2).

We define
1, if p(n) ^ p(n — 1) (mod 2)
9(n) =
0, if p(n) = p(n — 1) (mod 2).

Further, let

G(N) ^ jf
71=1
g(n).

Once the above claim is established, we have

G(N) > N1/2exp(-(log2 + e/2)logN/loglogiV) (31)

for N > Ni.

Now, g(n) = 1 if and only if one of the integers p(n) and p(n— 1)
is odd and the other one is even and this is so if and only if Eli2{n) -
Ei}2(n — 2) = 1 for both 2 = 0 and 1, where ErtQ(N) denotes the
number of positive integers n < N such that p(n) = r(mod q).
Therefore, for both i = 0 and 1 we have,
[JV/2] l(N+l)/2]
G{N) = J]^(2m)+ Y, g{2m-l)
m— 1 m= 1

IN/2]
< Y (Ei,2(2m) - Eit2(2m - 2))
71= 1

t(N+l)/2]
+ Y (Eii2(2m - 1) - Eit2(2m - 3))
71=1

= (B,l2(2[JV/2]) - £i,s(0)) + (£i,2(2[(JV + l)/2]) - £,,2(-l))

< 2Eit2(N).

This together with (31) implies Theorem 5.5.

Now, we proceed to prove our claim.

 5. PARITY OF PARTITION FUNCTION 113

Proof of the claim. Defining f(n) = p(n) - p(n - 1), as in the

proof of Theorem 5.4, from Euler’s identity (22) we obtain

Y f{m) = 0 (mod 2),

m£Mn

for any integer n > 1, where, as before,

M-n =f {n} U {n — Sk : 1 < Sk < n} U {n — tk : 1 < 4 < n}.

Now, if we take n to be of the form Sk or tk, then 0 E Ain and

since /(0) — 1, the set {m : m E Ain,f{m) = l(rnod 2)} must have
at least one further element.
But there are at least C2W1/2 numbers < N, which are of the
form tk or where C2 is a positive constant. To each of these,
there corresponds a non-zero integer where the partition function
takes odd value. Now, since putting l — —k, sk can be written as
^k(3k — 1) = ^1(31 + 1), elements of the set

{tk — Sj : 1 < Sj < tk < n} U {sk — tj : 1 < t3 < sk < n)

are of the form

u(3u+l) v(3v + 1) (u — v)(3u + 3v + 1)
m ~ 2 2 ~ 2
(32)

with certain integers u and v.

Since from (32), we see that for a particular m, the value of

(u — v) determines the value of (u + v) and the values of (u — v)
and (if + v) together determine u and v, the number of distinct
expressions of a particular m can be at most twice the number of
divisors of 2m.
Therefore, by Wigert’s theorem [182] (the reader may also
refer to [91], for instance) on the order of magnitude of the divisor
function, for N > N(e), the number of distinct expressions of a
particular m is less than

exp((log2 + e/2)^gL).

Therefore, for N > N{t), the total number of distinct m values

counted is at least

a>^/exp((log2 + (/2)j^L)
114 7. PARTITIONS OF INTEGERS

log N
> iV1/2exp -(log 2 + e) (for N > Ni(e))
log log N
□

6. Notes

Important properties of the partition function started appearing

in the literature with the work of Euler. The interested reader may
look into the relevant chapter of a recent book of Tattersall [169] for
some historical titbits related to some of the early partition theoretic
results.
Hardy and Ramanujan [90] first proved that p(n) is asymptotic
to
(eK^)/(4nV3),
where K = 7r(2/3)’^2, as n —> oo.
In fact, Hardy and Ramanujan found an asymptotic formula for
pin), which however was found to be divergent. Later, Rademacher
[136] made some changes in the analysis and with slightly different
terms obtained an expression involving a convergent series. For
these, interested readers may look into the texts [16], [108] or [137],
A simple proof of the fact p(n) < eK^, with K — 7r(2/3)1//2, can
be found in [15].
In the above mentioned paper of Hardy and Ramanujan [90],
circle method was introduced, which became a very important tool
in additive number theory. For an account of many applications of
this method, one may look into [178].
The converse of Lemma 5.2 is also true and Newman’s paper
[124] supplies a simple proof of this converse as well.
In an Appendix to the paper of Nicolas, Ruzsa and Sarkozy
[125] , J. -P. Serre proves some general results in the direction of
Theorem 5.4 using the theory of modular forms. In the context of
the partition function, his result would say that for every positive
integer m, if we consider an arithmetic progression r(mod m), 0 <
r < m - 1, the number of integers n < N in that arithmetic pro¬
gression for which p(n) is even is > c.\/N for any positive constant
c , if TV is large enough.
Regarding parity of p(n) on arithmetic progressions, Subbarao
[164] had conjectured that for every positive integer m, on every
 6. NOTES 115

arithmetic progression r(mod m), 0 < r < m — 1, p(n) assumes

both even and odd values infinitely often.
In an major achievement in this direction, two years before the
above mentioned result of Serre, Ono [128] proved that for every
positive integer m, on every arithmetic progression r(mod m), 0 <
r < m — 1, p(n) assumes even values infinitely often. Also, if p(n)
assumes odd value for a single n in an arithmetical progression, then
p{n) assumes odd values for infinitely many n in that arithmetical
progression.
Regarding odd values of p(n), further sharpening due to Ahlgren
[7] says that given an arithmetic progression r(mod m), for some
r, 0 < r < m, if there is at least one integer M in that arithmetic
progression such that p(M) is odd, then there is a positive constant
C — C(r,m) such that the number of integers n < X in that
arithmetic progression such that p(n) is odd is at least C\/X.
-
 CHAPTER 8

Ramsey-type results in posets

1. Introduction

Definition 1.1. We recall that a partially ordered set is a pair

(A, -<) where X is a set and -< is a binary relation on X satisfying
the following properties:
(i) the relation is reflexive, that is, x -< x for all x G X.
(ii) the relation is antisymmetric, that is, x -< y and y -< x will
imply that x — y.
(iii) the relation is transitive, that is, x -< y and y -< z will imply
that x < z.

Usually, one uses the term poset to mean a partially ordered set.
For the early results of the present theme, X is the collection
V(S) of all non-empty finite subsets of a set S and is the usual
set inclusion ‘C’.
In Section 2, we set up some terminologies. In Section 3, we
shall prove a theorem of Harzheim which is about some unavoidable
regularity for certain maps / : V(S) —> S, when S is a sufficiently
large finite set; in Section 4, certain generalizations of this theorem
are considered.
Finally, in the notes in Section 5, we shall supply references for
further generalizations and state an interesting result belonging to
the present theme for posets of bounded width.

117
118 8. RAMSEY-TYPE RESULTS IN POSETS

2. Preliminaries

Definition 2.1. Given a partially ordered set (X, -<) and x, y G

X, t and y are said to be comparable if at least one of the relations
x -< y, y -< x holds.

A chain of (X, -<) is a subset C of X such that for any two

elements x,y G C, x and y are comparable.

An antichain of (X, -<) is a subset A of X such that for any two

distinct elements x,y G A, x and y are not comparable.

Let (X, -<) be a finite poset. The height of (X, -<) is the maxi¬
mum of the cardinalities of its chains and its width is the maximum
of the cardinalities of its antichains.

We call a map / : (A, -<) -» (Ar, -<) a regression or a kernel

function if f(x) ~< x for every x G X.
In the present chapter, we shall discuss some Ramsey-type re¬
sults in posets. The result of Harzheim, which will be taken up in
the next section, says that given a positive integer n, if S is a suffi¬
ciently large finite set then given any map / : V(S) —> S, with the
property that f{M) G M for all M G V(S), there exists a chain C
of length n + 1 (that is, consisting of n + 1 distinct elements) such
that / is constant on C.

A function / : V(S) 5, with the property that f(M) G M

for all M G V(S) is called a choice function.

We remark that if an element s G S is identified with the

singleton {s}, then a choice function / on V(S), that is, a map
/ : V(S) —> 5, satisfying f(M) G M for all M G V(S) can be con¬
sidered as a regression / : V(S) -* V(S) according to the definition
above (since the singleton {/(M)} will be a subset of the set M).

3. A theorem of Harzheim

In 1964 Harzheim proved (see [92]) the following theorem:

Theorem 3.1. (Harzheim) Let n be a positive integer. Then

there exists a positive integer H(n) such that for any set S of H(n)
elements and any mapping f : 'P(S) —» S, with /(A/) G M for
all M G V(S) (that is, f is a kernel function on (V(S), c) ) there
exists a chain
 3. A THEOREM OF HARZHEIM 119

of length n + 1 such that /(Mo) = /(MJ = • • • = f{Mn).

( Considering the choice function / as a regression from 'P(S)
to V(S), the above assertion is sometimes expressed by saying that
there exists a chain of length n + 1 on which / is constant or by
saying that there exists a constant chain of f of length n + 1.)

In [131], Perry shows that H(n) can always be taken as 2n. We

here present a proof of Perry’s result which was given by Kleitman
and Lewin [107].
Theorem 3.2. (Perry) Let n be a positive integer and N > 2”.
Let S = {1,2, ■ ■ • N} and f be a choice function on the set V(S)
of non-empty subsets of S. If for 1 < * < N, ti(S) denotes the
maximum of the lengths of the chains formed of those subsets whose
image under f is i, then for some i we have

U(S) > n + 1.

Clearly, the above result follows from the following result.

Theorem 3.3. ( Kleitman and Lewin) Let n be a positive

integer and N = 2n+j for some j, 0 < j < 2n. Let S = {1, 2, • • • N}
and f be a choice function on the set V(S) of non-empty subsets of
S. If for 1 < i < N, ti{S) denotes the maximum of the lengths of
the chains formed of those subsets whose image under f is i, then

't(U(S)-n)> 2j + l. (33)
1=1

PROOF. Before proceeding, we introduce the following notation.

For a subset Si C 5, for 1 < i < TV, tt(Si) will denote the
maximum of the lengths of the chains formed of those subsets of S
which are contained in Si and whose image under / is i.
We observe that (33) can be written as:

'tti(S)-Nn>2j + l. (34)
i=i

First, we observe that (34) is true for n = 1 and j = 0. For, in

this case N = 2. Now, the number of non-empty subsets of S =
120 8. RAMSEY-TYPE RESULTS IN POSETS

{1,2} is 3. In other words, |P(S)| = 3. If f(S) = 1, then we have

/({!}) = /({1, 2}) = 1 and hence ti(S) = 2 in this case. Clearly,
<2(-S') = 1 and we have Y?i=\ U(S) = 3. Similarly, if f(S) = 2, then
t\(S) — 1 and t2(S) = 2 and again we have U(S) = 3. Since
Nn = 2, (34) holds in this case.
We proceed as follows. First, assuming that (34) is true for
n = m where m is a positive integer and j = r with 0 < r < 2m, we
show that it holds for n — m and j = r + 1.
Let n = m and j — r -f 1 so that N = 2m + r + 1.
We consider the subset

S1 = {l,2,.*.,JV-l}c5.
By our hypothesis,

N-1
52 ti(5i) - (TV - \)m > 2r + 1 > 0. (35)
i=1
Then for some i0 with 1 < iq < N — 1, we have

tio(Si) > m. (36)

Now, we consider the subset

52 = {1,2,...,JV}\{»0}cS.
Again, by our hypothesis,

52U(S2) ~ {N - l)m > 2r + l. (37)

i^io

Now, we observe that U(Si) < t^S) and L(52) < t,(5) for every
i. Again the set 5 will contribute in increasing the length of a
maximal chain with image f(S).
Therefore,
N

52 ti{S) - Nm > 1 + tio(Si) + 52 U{S2) - (N - 1 )m - m.

i=1 ^l0 (38)

From (38), using (36) and (37) we get

N

52ti{S)-Nm > 1 + 1 + (2r+ 1) = 2(r + 1) + 1,

2=1

as required.
 4. GENERALIZATIONS 121

Now, once we have (34) for n = m and j - 2m, we have N =

2m+1 and
N
Y/tl{S)-N{m+ 1) > 1. (39)
i=i

But (39) is nothing but (34) with n = m + 1 and j = 0. Hence

the theorem. □

4. Generalizations

The following generalization of Theorem 3.1 is due to Rado

[142],

Theorem 4.1. ( Rado) Given a positive integer n, there is a

positive integer r*(n) such that if S is a set of r*(n) elements and
f : V(S) —A V(S) is such that f(X) C X for all X G V{S), then
there are subsets Xq, Xi, ■ • • , Xn of S such that

and
vs
xcxc Xn

f(Xo)Cf(Xl)C---Cf(Xn).
We are going to present a further generalization due to Rado
[143]. However, we do not present this result of Rado in its full
form.
Now we are ready to state the following generalization of Theo¬
rem 4.1.

Theorem 4.2. ( Rado) Let k and n be non-negative integers.

Then there is a positive integer r**(n) such that if |S| > r**(n),
then given any k+ 1 kernel functions /o, f\, ■ • • , fk on the collection
V(S) of the non-empty finite subsets of S, the following statement
is true.
C(n, S, /o, /i, • • • , fk)•' There are sets .40, Ai, • ■ • , An such that
either

(i) An^fiiAn-i) C An-^-'-A^fiiAo) C T0 C 5,

for some i, 0 < i < k, or

122 8. RAMSEY-TYPE RESULTS IN POSETS

(ii) f.(A„) = ■ ■ ■ = fi(Ac) C An^A„., ^ ^A0 C S,

for some i, 0 < i < k.

Proof. Suppose the result is not true. That is, given non¬
negative integers k and n, there is no such positive integer r**(n)
as claimed in the theorem.
Therefore, for any positive integer m, we have a set Sm such
that \Sm\ = rn and kernel functions fm,o, fm,ir ” , fm,k on V(Sm),
such that the statement C(n, Sm, fm,o, fm,i, • ■ • , fm,k) is false.
Without loss of generality, we can assume Sm to be the set [m] =
{1,2, * -- , m}.
We proceed to define functions g0, gi, ■ ■ ■ , gk on the collection of
non-empty finite subsets of Z+ as follows.
Let r be a fixed positive integer. Since by all the maps fm,o, =
r+1, r+2, • • •, a subset of [r] = {1,2, • • • , r} is mapped to a subset of
[r], as all the maps involved are kernel functions, we have an infinite
sequence rii < n2 < ■ ■ ■ such that all the maps /n,,o, i = 1,2,-*-
agree on ?([r]). So, for all subsets D of Z+ with D C [r], we define

go{D) — fni0(D)(= fn2}o(D) = fn3to(D) = •

Now, an infinite subsequence of the functions fn,,o, rij > r + 1,

will have to agree on the finitely many subsets of [r -I- 1] of the form
{r+l}ufi, D C [rj. For such subsets we define <?o({p + 1}UD)
to be this common value.
Proceeding in this way, we define g0 on the collection of non¬
empty finite subsets of Z+. We observe that the construction of g0
consists of nothing but employing the compactness principle for the
countable case and hence the argument is similar to the solution to
Exercise 4.2 of Chapter 1. *
The other functions gi, • • • , gk are defined in the same way.
Clearly, C{n, Z+,g0,gu ■■ ■ ,gk) is false.
We consider the following cases.

Case I. (Given a positive integer a, there is a positive integer b

such that whenever \B\ > b, there is a proper subset A of B such
that |<7i(.A)| > a for some i, 0 < i < k.)
 :n
4. GENERALIZATIONS 123

Let

t — (/c ,-f- 1) (n -I- 1) “I- 1 •

By repeated application of our condition, we find sets 5, Ar0,

• • •, Xt and z0, • • • , it < k such that

9u(Xt) C Xt ••• gtl(Ah) c X\ Oj0(A^o) c Ar0 C S.

I / /

By the pigeonhole principle, there are numbers a0 < ai <

•' • an < t such that iao — iai — ■ • ■ — ian — i say, such that

9i(Xt) Cly-g,(A'O C JfAji(A'„) C A'„ c S,

But, this means that the statement C(n, Z+, g0, gx, ■ ■ ■ ,gk) is
true, which is a contradiction.

Case II. (There is positive integer a0 such that, for every posi¬
tive integer 6, there is a set B with |L?| > b, such that \gi{A)\ < a0,
for all proper subsets A of B and 0 < i < k.)

For any positive integer m, we have a set Bm C Z+ such that

\Bm\ — m and |^(yl)| < a0 for all proper subsets A of Bm and
0 < i < k.

This gives kernel functions hm,(b ^m,i, • • • , hm k (they are ob¬

tained from the g^s via some one-to-one correspondence of Bm and
[m] — {1,2,*** ,m} ) on V([m]) such that \hm^{A)\ < a0 for all
proper subsets A of [m] and i < k.
Now, employing the compactness principle argument as before,
we obtain kernel functions ho,hi,--- ,hk on the collection V(Z+)
of non-empty finite subsets of Z+ such that |fil(A)| < ao for all
A G V(Z+) and 0 < i < k and C(n, Z+, h0, h\, ■ ■ ■ , hk) is false.
For an infinite subset A' of Z+ and a positive integer r, using
(A")r to denote the collection of subsets of A' of cardinality r written
in a sequence ordered by ‘<’, by the infinite version of Ramsey’s
theorem (Theorem 3.4 of Chapter 1), there is a sequence of infinite
sets
Z+ D Ni D N2 ■ ■ ■

such that for any positive integer r, for sequences

A : {ax < ■ • • < ar}, B : {bx < ■ ■ ■ < br} in (iVr)r,

124 8. RAMSEY-TYPE RESULTS IN POSETS

by the order preserving bijection

A -> B,

hi(A) is mapped onto hz(B) for each z, 0 < i < k.

For applying Ramsey’s theorem, we had only to observe that cor-
responding to a kernel function ht and a finite sequence A E {Nr)r,
there are finitely many possibilities for ht(A) for each z, and hence
finitely many equivalence classes corresponding to the equivalence
relation given by the requirement of having an order preserving bi¬
jection as above.

Now, by the fact that \ht(A)\ < a0, for all A E P(Z+) and
0 < z < k, once again by the infinite version of Ramsey’s theorem,
there is an infinite set R C Z+ such that, if r, s E R and A E (Nr)r,
B E (Ns)s, there is an order preserving bijection

ulMA) U*=0ft.(S)

where, for each z, hl(A) is mapped onto h^B).

We write

h(A) — ho(A) U hi (A) U • • • h^(A).

Then the number t = \h(A)\ is independent of r and A as long

as A E (Nr)r for some r E R.
Now, for a subset

P — {Up! ^ O-p-2 ^ ‘ ^ Rp( }

of a set
A = {<2i < • • • < ar} C Z+,

we define v(A, P), to be the vector

(Pi - 1 >P2 - Pi - 1,P3 -P2 - -Pt-1 - l,r -pt).

We remark that /r — 1 is the number of elements of .4 which are

before the first element aPl of P. Similarly, r - pt is the number of
elements of A which are after the last element aPt of P. The other
coordinates in the vector v(A, P) measure the ‘gaps’ between the
successive elements of P as a subset of A.
With the above notation, we observe that the vector v(A, h(A))
depends only on r and not on A as long as A e (Nr)r for some fixed
r E R.
 4. GENERALIZATIONS 125

Therefore, for an r € R we write

■y(j4,/l(^4)) (Xj-fi, Xr\, >Tr f).

Now, choosing the least i/0 G i? C Z+ such that

Zt/o.o = minja^o : u e Rj

and then the least vx > u0 in R such that

x„li0 = minjx^o : u > u0}

and so on, we get a sequence vQ < u\ < v2 ■ ■ ■ in R such that

Xj>0fi ^ Xvj o ^ X^fi •

Next, we take a subsequence /u0 < Mi < M2 • • • of the sequence

^0 < ^1 < M2 • • * such that

Proceeding in this way, we get an infinite set R' C R such that

whenever r, s e R' with r < s, then

xt,t < for all r, 0 < r < t.

Now, let r, s G R' with r < s. We choose B G (Ns)s. Then

v{B,h(B)) = (xS)0, ts>1, • • • ,x8ft).

Since xr,T < xSiT for all r, 0 < r < t, there is A C B with
h(B) C A such that

v(A,h(B)) = (xrfi, xT>i, • • ■ ,xTtt).

Now, for any A! G (Nr)r, we have

v(A',h(A!)) = (xrfi,Xrtl, ■■ ■ ,Xrtt).

Hence

\A\ — \h(B)\ = |xr)o| + \xr,\| + • • • + \xr,t\ = \A'\ — |h(A )|.

But since r, s G R! C R, we know that \h(B)\ = \h(A')\ and

therefore we have |T| = \A'\ — r.

Since Nr D Ns, we have

v(A, h(A)) — (Xrfii Xfti, 1 xT t).

Thus v(A,h(A)) = v(A,h(B)) and hence h(A) = h(B).

126 8. RAMSEY-TYPE RESULTS IN POSETS

Once again, since r, s G R!, B G (Ns)s and A G {Nr)r, there is

an order preserving bijection

uf=ohi(A) y Ui=0hi(B)

such that for each i, hz(A) is mapped onto ht{B).

Therefore, we have established the following statement:
Statement R: If s G R' and B G (iVs)s, then for any r G R' with
r < s, there is A G (B)r C (Ns)r such that hr{B) = hT(A) for all
t, 0 < t < k.

We now choose ro, ri, • • • , rn in R' such that r0 > > • • • > rn.
Let A0 £ (iVro)r° • By repeated application of Statement R, obtained
above, we find

Ai ^ ■ ■ • ^An

such that
hT{Ao) = hT(Ai) = • • • = hT(An),

for all r, 0 < r < k.

But, this means that the statement C(n, Z+, h0, hx, ■ ■ ■ ,hk) is
true, which is a contradiction. □

5. Notes

It was proved in [92] that in Theorem 3.1, the smallest value

which H(n) can have is > 2n. Therefore, from the theorem (Theo¬
rem 3.2) of Perry, it follows that H(n) = 2n.
Some generalizations of Theorem 3.1, other than the ones given
in Section 4, are obtained in [181] and [95] and for the infinite case
in [93] and [94] .
We are now going to state a sharp result of this type for posets
of bounded width. We shall need the following terminology.
If (A', -<) is a poset and f : X X is a regression, then given a
/c-element chain
C- .X\ 2',2 * * 1 Xk

in A', one says that F is monotone on C if

}{x 1) -< f{x2) -<-< f{xk).

 5. NOTES 127

Thus, Theorem 4.1 says that given a positive integer n, there is

a positive integer r*(n) such .that if 5 is a set of r*(n) elements and
/ is a regression on the subset lattice of S, then / is monotone on
some (n + l)-element chain.
G. W. Peck, P. Schor, W. T. Trotter and D. West [130] proved
that if w and k are positive integers and (A-, -<) is a poset of width
at most w and |A| > (w + l)fc_1, then every regression on X is
monotone on some k-element chain. We mention that the inequality
|Ar| > (w + l)^-1 in this result is best possible. In this connection
we mention a very nice article of Trotter [175], which, apart from
mentioning this result on posets of bounded width and supplying a
proof of it, refers to some density results of ‘Szemeredi-type’ towards
the existence of ‘arithmetic progressions of finite lengths’ in posets
of bounded width.
'
 ■n

CHAPTER 9

Solutions to selected exercises

Exercises from Chapter 1

2.1 Let (a) denote the fractional part of the given real number
a. Here, by the fractional part of a, one means the number
obtained by subtracting the greatest integer less than or
equal to a from a .
Now, the n + 1 numbers 0, (a), (2a), • • • , (na) will lie
in the interval 0 < x < 1. Therefore, by the pigeonhole
principle, one of the n intervals 0<x<^,^<x<
••• , < x < 1 will contain at least two numbers
(<7ia), (<72a) for integers <71, <72 with 0 < q\ < q2 < n. Writ¬
ing q = q2 — <71, we have 0 < q < n and \q2a — q^a — p\ < ^
for some integer p.
Hence we have,
p, 1 1
a - - < — < —.
q qn qz

2.4 First Method (By induction):

For n — 2, let the vertices be a, 6, c and d. Now, if a
particular pair of vertices, say c and d, are not adjacent to
each other, then all other pairs have to be adjacent. This
is because the number of edges is five which is just one less
than the number of pairs Q). Hence the simple graph G
has the triangle formed by the vertices a, b and c. Thus
the result is established for the case n = 2.
Let the result be true when n — m > 2.
Assume that G is of order 2(m + 1) and it contains
(m + l)2 + 1 = m2 + (2m + 1) + 1 edges.
Let x and y be two vertices which are adjacent. We
denote the set of remaining vertices by S. We have \S\ =
129
130 9. SOLUTIONS TO SELECTED EXERCISES

2m and if the vertices in S have at least m2 + 1 edges

among themselves, that is, at least m2 +1 pairs of them are
adjacent, then we are through by the induction hypothesis.
Therefore, we assume that not more than m2 pairs of
the remaining vertices are adjacent. If x and y are respec¬
tively adjacent to nx and ny number of vertices belonging
to 5, then nx + ny > 2m + 1. By the pigeonhole principle,
at least one vertex, say 2, belonging to 5 will have to be
adjacent to both x and y. In other words, G will have a
triangle formed by the vertices x, y and 2.

Second Method :
Let G be a simple graph of order 2n such that G does
not contain any triangle.
Let the vertices be Vi, ■ ■ ■ , u2n. With every vertex vl we
associate a real number (called weight) > 0 such that
= 1.

Let S be the sum Y,wiwji where the sum is taken over

each pair of indices i,j such that vt and Vj are adjacent.
For a vertex Uj, let iV* be the sum of the weights at¬
tached to the vertices adjacent to ly. That is, TV* = Hie.?,
where the sum is taken over all j's such that vt is adjacent
to Vj. Let vk be such that Nk is the maximum among all
the NSs.
If there is a vertex vr which is not adjacent to 14, then
we claim that we do not decrease the value of 5, if we shift
some of the weight attached to vr to 14. To verify the claim,
we observe that for a positive number e < wr,

{wk + e)Nk + (wr - e)Nr > wkNk + wrNr.

We shift the weights attached to all such vertices to

vk without decreasing the value of S and assume that all
the non-zero weights are attached to the vertices which are
adjacent to vk. Again, since the given graph is assumed to
be triangle free, if vs and vt are two vertices adjacent to vk,
then vs and vt are not adjacent to each other. Therefore,
if Ns > Nt, once again we can shift the weight wt to vs
without decreasing the value of S. Hence S is maximum
when all the weight is concentrated on an edge, that is,
 9. SOLUTIONS TO SELECTED EXERCISES 131

all the non-zero weights are attached to just two vertices

which are the ends of an edge. Therefore S is maximum
when both the vertices have equal weights | and hence the
maximum possible value of S is On the other hand,
taking all the weights wt's to be equal to we have S =
^2 where s is the number of edges in G.

From 4^2 < |, we obtain s < n2.

2.5 Let S = {al5 • • • , amn+1} be a sequence of mn + 1 distinct

real numbers.

If possible, let there be neither a monotone increasing

subsequence of length m + 1, nor a monotone decreasing
subsequence of length n + 1.

We consider the map / : S —> (Z/mZ) x (Z/nZ) defined

by f(cLi) = (a;*, yi) where xx is the length of a monotone in¬
creasing subsequence of 5 of largest possible length starting
at a,i and yl is the length of a monotone decreasing subse¬
quence of 5 of largest possible length with its last term
a*-

By the pigeonhole principle, there will be ax and d3

with i < j such that f(ax) = f(aj). Therefore, with our
notations,

(xi,Vi) = (xj,yj). (40)

Now, if at < dj, then considering a monotone increasing

subsequence of length x3 starting at aj, if we add the term
Oj in the beginning, we get a monotone increasing subse¬
quence of length Xj + 1 starting at ax. Therefore xt > Xj,
which contradicts equation (40).

Similarly, if ax > dj, then considering a monotone de¬

creasing subsequence of length yt ending at a2, if we add
the term d3 at the end, we get a monotone decreasing sub¬
sequence of length yi + 1 ending at dj. Therefore yt < y3,
and once again we get a contradiction to equation (40).

3.1 Let |Sj = r and N — n(2, r, 3), where n(fc, r, /) is as defined

in Theorem 3.1. We then consider an arbitrary sequence
{ai,'a2, • • • , (In} of elements dt G S and the r-colouring x of
132 9. SOLUTIONS TO SELECTED EXERCISES

The Moser graph

the collection of 2-element subsets of [N] defined as follows.

If 1 < i < j < N, then we define

X(}) 1 ' ‘ ‘ —1 ^ S.
By the choice of N, 3 a 3-element subset {i,j, k} of [N]
with i < j < k such that

x({°o aj}) = x(K>afc}) = X{{ai,ak}).

This means

n «i= n = n a«=°
i<t<] j<t<k
at
i<t<k

Since

n = (n
i<t<k \y<t< j J
(n
\j<t<k J

we have a2 = a.
4.1 Suppose a 3-colouring of the Euclidean plane is given. The
colours involved are red, blue and green, say.
We consider the figure of the Moser graph above. It is
obtained as follows. DEG and AGE are equilateral trian¬
gles having length of their sides equal to 1. Now keeping D
fixed, we apply the rigid motion of swinging DEAG clock¬
wise to the position DFBC such that the length of AB is
1. Now, the points- A, B, C, D, E, F, and G are such that
DEG, AGE, BCE and DFC are all equilateral triangles
and the length of AB is 1.
 9. SOLUTIONS TO SELECTED EXERCISES 133

If possible, let there be no two points at a distance 1

receiving the same colour in the given colouring. Suppose
D has received the colour red. Then the colour of the point
E can not be red. If its colour is happens to be blue, then
G will be green. Since the colours of E and G will have
to be blue and green, in some order, the colour of A will
have to be red. Similarly, B will have to be red. But then,
A and B are two points of the same colour, situated at a
distance 1, which is a contradiction.
4.2 That Theorem 4.1 implies the statement V(k,r) is easy to
see.
Now, let us assume that the statement V(k,r) holds. If
possible, suppose that the integer W(k,r) does not exist.
Therefore, for every positive integer n, we have an r-
colouring \n '■ [ft] —> {1, • • - , r} such that there is no
monochromatic arithmetic progression of k terms.
There is cx G {1, • • • , r}, such that Xn(l) = G for infin¬
itely many values of n.
Among these Xn s with Xn(l) = G, there will be infin¬
itely many with \n{2) = c2, for some c2 G {1, • • • ,r}. We
proceed in this way and obtain an infinite sequence

ci,c2,c3,

with C{ G {1, • • • , r}, Mi.

Now, for the colouring x : Z+ —> {1, • • • , r}, defined
by x(m) — cm, we see that if there is a monochromatic
arithmetic progression a1,a2,---afc of k terms, then writ¬
ing ak = N, for some N\ > N there is a monochromatic
arithmetic progression of k-terms for the restriction of the
colouring X/Vj on [Ah], contradicting our assumption.

5.1 We need to prove the result for m > 3.

By the proof of Schur’s theorem (Theorem 5.1 of Chap¬
ter 1), taking prime p such that (p - 1) > N = n(2, m, 3),
where n(k,r,l) is as defined in Theorem 3.1 of Chapter 1,
for any m-colouring of 1, 2, • • • ,p - 1, 3 a monochromatic
subset {t, y, zj such that x + y = z.
Now, given a positive integer m > 3, we define a finite
colouring on F* (which we identify with the set 1, 2, • • • , p—
1) as follows.
134 9. SOLUTIONS TO SELECTED EXERCISES

We define our finite colouring by the requirement that

x,y £ F* will have the same colour if and only if x~ly = Sm
for some 5 £ F*. In other words, our colour corresponds
to the partition of F* into distinct cosets of its subgroup
H consisting of m-th powers. The number of colours is the
index of H in F*, and hence it is equal ton = g.c.d.(m,p-
1) < m.

Therefore, for primes p > N + 1, we have x, y, z of the

same colour such that x + y = 2, which means in F*,

1 + x~ly = x_12:,

where 1 ,x~ly and x~lz are m-th powers in F*.

5.2 Hints : See Remark 2.1 of Chapter 3.

Exercises from Chapter 2

2.1 Suppose that for d,r £ Z+, we are given a finite set 5 C
(Z+)d, and an r-colouring of (Z+)d.
Let |5| = t. We consider Cfn where n = HJ(r,t) and
Ctn is the collection of words of length n over the alphabet
of t symbols; we take 5 as our alphabet, that is, C(n consists
of words Xi • • • xn with xt- £ S.
One can choose integers ki,--- , kn such that for two
distinct elements xx • • • xn and yi • • • yn of Cf, we have
n n

2>xi yt-
i=l z=l

This is possible because one can choose k^s such that

they do not satisfy the finitely many equations
n

EMx*-yt) = o>
i=i

corresponding to pairs of distinct elements xi ■ • • xn and

yi • • -yn of cp.
After such a set of integers ku - ■ ■ , kn is chosen, we r-
colour Cp so that the colour of xx ■ ■ ■ xn is that of £"=1 kjXj.
By Hales-Jewett theorem, there exists a monochromatic
combinatorial line in Cp for this colouring.
 9. SOLUTIONS TO SELECTED EXERCISES 135

Under the map

n
Xi • • -Xn ^
i— 1

the words belonging to this combinatorial line corresponds

to a translated homothety of S in (Z+)d which is monochro¬
matic in the given colouring, as desired. In the above,
the homothety comes from the hi s at the moving positions
and the translation, possibly zero, from the remaining ones.
The requirement that the map x\ ■ ■ ■ xn i-> kixl is in¬
jective is necessary so as to assure that distinct points on
the monochromatic combinatorial line do not map to the
same point.

3.1 We fix r and demonstrate the existence of n(r, k) by induc¬

tion on k. The case k = 1 is trivial.
We assume that n(r, k) exists and then proceed to es¬
tablish that n(r, k + 1) exists.
We claim that the number

t d= 2W(r, n(r, k))

can be taken as n(r, A; + 1).

Let an r-colouring

X : [A] —>• {ci, • • • , cr},

of [£] be given. By the definition of t, the set {t/2 + 1, • • • t}

contains a monochromatic (say of colour cr) arithmetic pro¬
gression, {a + md : 0 < m < n(r, A;)} for the colouring x-
We observe that n(r,/c) < W(r, n(r, k)) < a.
Now, consider the set

{d, 2d, 3d, • • • , n(r, k)d}.

By the induction hypothesis, this set contains positive
integers da\ < da,2 < • ■ ■ < dak satisfying Xu<i<A:ai —
dn(r, k) such that for any subset / of {1,2,-•• , k}, the
colour of the element £ze/ dax depends only on max(/).
Now, we define the integers by, b2 • • • , bk+\ as follows:

bx = da-L, for 1 < i < k, 6^+1 = a.

136 9. SOLUTIONS TO SELECTED EXERCISES

If for a subset I of {1, 2, • • • , k +1}, max(/) < k+1, then

Xag/ bi = Yliei dax depends only on max(/). If max(/) =
k+1, then Yhei h — bk+i+sd, where 0 < s < n(r, k). Hence
for any subset I of {1,2,•••,&}, with max(/) = k + 1,
X4e/ bt is of colour cr.

Exercises from Chapter 3

3.1 Let r and k be two positive integers, and an r-colouring

X : Z+ -> • , cr}, be given.
By Exercise 3.1 of Chapter 2, there is a set

A = {ai < <22 < • • • < ar(fc_i)+i}

of positive integers such that for any subset / of {1, 2, • • * ,

r(k — 1) + 1}, the colour of the element Yhei ai depends
only on max(7).
By the pigeonhole principle, there is a colour cl5 for
some i with 1 < i < r, such that for k different integers
,4 in {1, 2, • • • , r(k — 1) + 1}, elements ^Z,€/ at with
max(7) G {4, • • ■ , 4}, have the colour ct.
Clearly, the set {a^, • • • ,al/c} satisfies our requirement.

3.2 We assume the finite union theorem (Theorem 3.2 of Chap¬

ter 3) and deduce Folkman’s theorem from it.
We claim that

n{r,k) < 2u(r’*} - 1. (41)

If V = V([u(r, /)]) denotes the collection of non-empty

subsets of [w(r, k)], then

r/ : V [2u(ri/c) - 1

defined bv r](J) = 2Z_1 is one-one and onto.

Now, given an r-colouring on 2u^r,/c) - 1 , through t] we
get an r-colouring on V and hence by the finite union the¬
orem there exists a collection V — {Di}ieI C P of disjoint
sets with |/| = k such that FU(V) is monochromatic.
Writing

at = Y, 2l_1, for t G /,
i<EDt
 9. SOLUTIONS TO SELECTED EXERCISES 137

through 77, £f€T at, ($ ^)T C /, are identically coloured

in \2u(r’k) - 1

This establishes the claim (41).

Exercises from Chapter 4

2.1 Let J C S be a minimal left ideal of a compact semigroup

S and s G S.
Now, if possible, suppose that Js is not a minimal left
ideal. Then there is a left ideal / of S such that / is a
proper subset of Js.
Let
I\ = {x € S\xs € /}.

Clearly, Ix is a left ideal of S. Therefore,

is a left ideal.
If I2 — J, then it would imply that J C I\ and there¬
fore, by the definition of Ix, it would mean that Js C /,
contradicting the fact that I is a proper subset of Js.
Therefore, /2 is a proper subset of J, contradicting the
fact that J is a minimal left ideal.

Exercises from Chapter 5

2.1 We define a (2n)-colouring x : R- ~> [2n] as follows.

Given any real number x,

x G [2m -I- -, 2m + —),

n n
for some even integer 2m and an integer j, 1 < j < 2n. We
define

x(t) = J-
Now, xixi) — x{x'i) will imply xl -x\ — 2ml+9i, where
6: is a real number with |0,| < 1 jn.
138 9. SOLUTIONS TO SELECTED EXERCISES

Therefore, if there is a solution of the given equation in

Xi,x[ with x(xi) ~ 1 < i < n, we shall have
n n

- i>
i= 1
which is impossible, as — 1 < @i < 1-

2.2 We observe that

YsCiivi-y\) = h (42)
i=l

holds if and only if

Y1 b~lCi(yi — y'j) = 1. (43)

i=l

We consider the colouring x* on R which is induced

by the colouring x defined in the solution of the previous
exercise, as follows, x* is defined by the requirement that
for a pair of real numbers a and /?,

X*{a) = X*{P)
if and only if

X(6_1cta) = x(&-1Ci/3), Vi.

Clearly, x* is a (2n)n-colouring.

If possible, let (42) hold with x*{Vi) — X'iVi), 1 < i < n.

Then, x{b~lcJyl) = x(&-1c,y')> for 1 < i,j < n. which
would imply that

Y2 - y'x) = + ^),
1=1 Z=1
where is a real number with \6t\ < 1/n.

Since, — 1 < YJi=\ < 1, the above can not hold.

 •»»
9. SOLUTIONS TO SELECTED EXERCISES 139

Exercises from Chapter 6

2.1 We proceed by induction on s.

Let s — 2. It is given that ax ^ a2(modp). Again,
since ax ^ 0(mod p) and a2 ^ 0(mod p), neither ax nor a2
can be congruent to ax + a2. Thus, the numbers ai, a2 and
ai +a2 are all incongruent and the result holds in this case.
Now, for some s > 2, we assume the result for s — 1
and prove it for s. Let 61,62,--- , bk, 0 < bt < p be the
representatives of all distinct congruence classes modulo p
to which the numbers in the set {€iai '■ O = 0 or 1}
belong. By assumption, k > s. If A; > s + 1, there is
nothing to prove. Therefore we assume that k = s < p.
We consider the integers bt + as, 1 < i < k. Since as 0
(mod p), bt + as ^ bt (mod p) for all i. In case each integer
bt + as is congruent to one of the numbers &i,&2)''‘
then considering the fact that bt ^ bj (mod p) for i / j, the
congruence classes of the numbers bt + as are represented
exactly by the set of numbers &i, &2, • • • , b^.
Adding we have

b\ + &2 + ■ ■ ■ T b^ + kas = + 62 + • • • + b^ (mod p),

which implies kas = 0(mod p).

Since 2 < k < p and as ^ 0(mod p), this leads to a
contradiction.
Thus at least one of the integers bt + as, 1 < i < k. is
incongruent to all the s. Hence the numbers in the set
{ELi eiai : et = 0 or 1} belong to at least s + 1 distinct
congruence classes modulo p.
2.2 We start arguing as in the proof of Theorem 2.1 given in
Remark 2.6 in Chapter 5.
For a prime p, we are given a sequence cp, a2, • • • , a2p_i,
of elements in Fp. We identify them with the integers in
the set 0,1, • • • ,p - 1 representing them (mod p) and by
rearranging the elements, if necessary, we assume that ax <
a2 < • • • < CL2p-l-

If there is an i < p - 1, such that at = al+p_i (which

means that at = al+i = ••• = fli+p-i), then taking I =
{?, i'+ 1, • • • , i + p — 1}, the theorem is established.
140 9. SOLUTIONS TO SELECTED EXERCISES

If there is no such i, we define

bt = ap+i - ai+1(7^ 0) V i, 1 < i < p - 1.

We can assume that X^f=iai ^ 0(mod p). Let c with

0 < c < p be the representative of the congruence class of
YJl-1 ca modulo p.
We claim that
p-1
-c = ^
i= 1
is solvable in e* E {0,1}.
If Vs are not all congruent, then our claim follows from
Exercise 2.1 above as the set {Hf=i •' O = 0 or 1}
will contain representatives of all the p distinct congruence
classes modulo p.
If all the Vs are congruent, let d with 0 < d < p be the
representative of the congruence class of all the Ws modulo
p. If d' with 0 < d! < p is the integer such that dd' = 1
(mod p), then (—d'c)d = — c(mod p). If e with 0 < e < p
is the representative of (—d'c) modulo p, then taking e* = 1
for i E / C {1,2, •••} and = 0 for i E {1, 2, • • •} \ I, for
any I with |/| = e,
p-1
y, €ibi — ed = —c (mod p).
i=l

Thus,
p p-i
+ Y^ti(aP+i ~ flz+0 = 0 (mod p)
i=1 i=l
is solvable in e, E {0,1}. Since the left hand side is the
sum of exactly p of the ads, we are through.

Exercises from Chapter 7

2.1 By Theorem 2.2 of Chapter 7, pdo(n) is equal to the number

of self-conjugate partitions of n. Thus, p(n) -pd0(n) counts
the partitions of n which are not self-conjugate. If S de¬
notes the set of partitions of n which are not self-conjugate,
then the elements of S can be paired where each pair con¬
sists of partitions conjugate to each other. This implies
that |Sj = p(n) — Pdo{n) is even.
 9. SOLUTIONS TO SELECTED EXERCISES 141

2.2 If we add a 1 to a partition T of a positive integer r, then

we get a partition of the integer r + 1 with at least one
1 appearing in it. In this process, distinct partitions of
r will give rise to distinct partitions of r + 1. Again, all
the partitions of r + 1 with at least one 1, are obtained
by adding a 1 to a partition of r. Thus, the number of
partitions of r + 1 with at least one 1 is just p(r).

Similarly, the number of partitions of r + 2 with at least

two l’s is p(r).

Therefore,

Pi(n) = Y. ,LT,n — p{n — 1) 4- p(n — 2) H-1- p( 1) + 1,

(44)

the T' at the end corresponding to the number of l’s in the

only partition of 1.

Now, the integer n appears in just one partition of n,

n — 1 appears in p{ 1) partitions of n etc. In general, for
n > k > 1, the integer k appears in n — k partitions of
n. Since the sum of the distinct parts in the partitions of
n is nothing but the sum of the number of partitions each
k, n > k > 1 appears in, we have

PP{n) = 1 + p(l) + p(2) + • • • + p(n - 1). (45)

From (44) and (45), we get that p\{n) = pp{n).

3.1 Given a partition of a positive integer n, we consider the

largest square of dots in the north-west corner of its graph¬
ical representation. For instance, the graphical representa¬
tion given below for the partition 7-F6-F4-F2-F1 of the
number 20, the square has 32 dots.

In general, for a positive integer n, if the square men¬

tioned above has r2 dots, then the tails on the right of the
square and below the square will respectively correspond
to partitions of integers p and q respectively such that the
first one will not have more than r parts and the second
one will not have any part exceeding r. Also, we shall have
n — r2 + p + q. In our figure H, n — 20, r = 3, p — 8 and
q = 3.
142 9. SOLUTIONS TO SELECTED EXERCISES

• ••••••
• •••••
• • • •
• •

Now, the number of partitions of the integers p into not

more than r parts is (as seen during the proof of Theo¬
rem 3.1) the coefficient of xp in
1
(1 — x)(l — x2) • • • (1 — xr)
Similarly, by Theorem 2.1, the number of partitions of
the number q into parts not exceeding r is the coefficient
of xq in the same expression.
Therefore, the number of possible pairs of tails in the
graphical representation of partitions of the integer n such
the largest square of dots in its north-west corner has r2
dots is the coefficient of xn~r2 in

(_1_
\(1 — x)(l — x2) •••(!— xT)
which is the same as the coefficient of xn in

((1 -x)(l -x2)---(l -x'))2

Hence the total number of partitions of n is the coeffi¬

cient of xn in
x x4
(1 — x)2 (1 — x)2(l — X2)2

x9
+ (l-x)2(l-x2)2(l-x3)2 + " ' '
 ■n

Bibliography

[1] H. L. Abbott and D. Hanson, A problem of Schur and its generalizations,

Acta Arith. 20, 175-187 (1972).
[2] S. D. Adhikari, Colours in the fields coming in handy: a walk in the garden
of Ramsey-type theorems, Bona Mathematica, 6 No. 4, 7-14, (1995).
[3] S. D. Adhikari, A note on a question of Erdos, Exposition. Math. 15, no.
4, 367-371 (1997).
[4] S. D. Adhikari, On a theorem of Graham, Exposition. Math., to appear.
[5] S. D. Adhikari and Anirban Mukhopadhyay, Partition function congru¬
ences: Some flowers and seeds from Ramanujan’s garden, Exposition.
Math., to appear.
[6] S. D. Adhikari and R. Thangadurai, Monochromatic solutions of Diophan-
tine equations, Algebraic Number Theory and Diophantine Analysis: The
Proceedings of the International Conference held in Graz, Austria, from
August 30 to September 5, 1998, ( Eds. F. Halter-Koch and R.F.Tichy),
Walter de Gruyter GambH & Co., Berlin, New York, 1-9 (2000).
[7] S. Ahlgren, The distribution of parity of the partition function in arithmetic
progressions. Indagationes Math., to appear.
[8] M. Ajtai, J. Kolmos and E. Szemeredi, A note on Ramsey numbers, J.
Combin. Theory Ser. A, 29, 354-360 (1980).
[9] N. Alon, Tools from higher algebra, Handbook in Combinatorics, Vol. II,
(Eds. Graham, Grotschel and Lovasz), North Holland, 1749-1783 (1995).
[10] N. Alon and M. Dubiner, Zero-sum sets of prescribed size. Combinatorics,
Paul Erdos is eighty (Volume 1), Keszthely (Hungary), 33-50, (1993).
[11] N. Alon and M. Dubiner, A lattice point problem and additive number
theory, Combinatorica 15, 301-309 (1995).
[12] N. Alon, S. Friedland and G. Kalai, Every 4-regular graph plus an edge
contains a 3-regular subgraph, J. Combinatorial Theory, Ser B 37, 92-93
(1984).
[13] P. G. Anderson, A generalization of Baudet’s conjecture (van der Waer-
den’s Theorem), Amer. Math. Monthly, 83, 359-361, (1976).

143
144 BIBLIOGRAPHY

[14] G. E. Andrews, Number Theory, Hindustan Publishing Corporation (In¬

dia), Delhi, 1992.
[15] T. M. Apostol, Introduction to Analytic Number Theory, Undergraduate
Texts in Mathematics, Springer-Verlag, 1976.
[16] T. M. Apostol, Modular functions and Dirichlet series in Number Theory,
Springer-Verlag, 1976.
[17] A. 0. L. Atkin, Proof of a conjecture of Ramanujan. Glasgow Math. J., 8,
14-32 (1967).
[18] A. Balog and E. Szemeredi, A statistical theorem of set addition, Combi-
natorica, 14, 263-268 (1994).
[19] J. Baumgartner, A short proof of Hindman’s theorem, J. Combinatorial
Theory, Ser A 17, 384-396 (1974).
[20] F. A. Behrend, On sets of integers which contain no three terms in arith¬
metic progressions, Proc. Nat. Acad. Sci. USA 32, 331-332 (1946).
[21] Vitaly Bergelson, A density statement generalizing Schur’s theorem, J.
Combinatorial Theory, Ser A 43, 338-343 (1986).
[22] Vitaly Bergelson, Ergodic Ramsey Theory - an Update, London Mathemat¬
ical Society Lecture Note Series 228, Cambridge University Press, 1996.
[23] V. Bergelson, W. Deuber and N. Hindman, Rado’s theorem for finite fields,
Proceedings of the conference on sets, graphs and numbers, Budapest,
1991, Colloq. Math. Soc. Janos Bolyai, 60, 77-88 (1992).
[24] V. Bergelson, W. Deuber, N. Hindman and H. Lefmann, Rado’s theorem
for commutative rings, J. Comb. Theory, Series A 66, 68-92 (1994).
[25] V. Bergelson, H. Furstenberg, N. Hindman and Y. Katznelson, An algebraic
proof of van der Waerden’s theorem, L’Enseignement Mathematique, 35,
209-215 (1989).
[26] V. Bergelson and A. Leibman, Polynomial extensions of van der Waerden’s
and Szemeredi’s theorems, J. Amer. Math. Soc., 9, no. 3, 725-753 (1996).
[27] V. Bergelson and A. Leibman, Set-polynomials and polynomial extension
of the Hales-Jewett theorem, Ann. of Math. (2) 150, no. 1, 33-75 (1999).
[28] Bruce C. Berndt and Ken Ono, Ramanujan’s unpublished manuscript on
the partition and tau functions with proofs and commentary. The Andrews
Festschrift (Maratea, 1998), Sem. Lothar. Combin., 42 (1999).
[29] A. Bialostocki and P. Dierker, On the Erdos Ginzburg Ziv theorem and the
Ramsey numbers for stars and matchings, Discrete Math. 110. 1-8 (1992).
[30] A. Bjorner, Topological methods, Handbook in Combinatorics, Vol. II, (Eds.
Graham, Grotschel and Lovasz), North Holland, 1819-1872 (1995).
[31] J. Bourgain, On triples in arithmetic progression, Geom. Funct. Anal., 9,
no. 5, 968-984 (1999).
[32] J. Bourgain, Harmonic analysis and combinatorics: how much may they
contribute to each other? Mathematics: frontiers and perspectives, Amer.
Math. Soc., Providence, RI, 13-32, 2000.
 BIBLIOGRAPHY 145

[33] T. C. Brown, Variations on van der Waerden’s and Ramsey’s theorems,

Amer. Math. Monthly, 82, no. 10, 993-995 (1975).
[34] T. C. Brown, P. Erdos, F. R. K. Chung and R. L. Graham, Quantitative
forms of a theorem of Hilbert, J. Comb. Theory, Series A 38, 210-216
(1985).
[35] T. C. Brown and V. Rodl, Monochromatic solutions to equations with unit
fractions, Bull. Austral. Math. Soc., 43, 387-392 (1991).
[36] S. A. Burr and P. Erdos, Generalizations of a Ramsey-theoretic result of
Chvatal, J. Graph Theory, 7, no. 1, 39-51. (1983).
[37] Peter J. Cameron,Combinatorics: Topics, Techniques, Algorithms, Cam¬
bridge University Press, 1994.
[38] Yair Caro, Zero-sum problems - A survey, Discrete Math. 152, 93-113
(1996).
[39] F. R. K. Chung, A note on constructive methods for Ramsey numbers, J.
Graph Theory 5, no. 1, 109-113 (1981).
[40] F. R. K. Chung and John L. Goldwasser, Integer sets containing no solution
to x + y = 3z, The mathematics of Paul Erds, I, Algorithms Combin., 13,
Springer, Berlin, 218-227, 1997.
[41] F. R. K. Chung and C. M. Grinstead, A survey of bounds for classical
Ramsey numbers, J. Graph Theory, 7, no. 1, 25-37 (1983).
[42] W. Deuber, Partition theorems for Abelian groups, J. Comb. Theory, Series
A 19, 95-108 (1975).
[43] R. B. Eggleton, P. Erdos and D. K. Skilton, Colouring the real line, J.
Comb. Theory, Series B 39, 86-100 (1985).
[44] R. Ellis, Lectures on topological dynamics, Benjamin, New York, 1969.
[45] P. Erdos, Problems and results in additive number theory, Colloque sur la
theorie des nombres (CBRM) (Bruxelles), 127-137, 1956.
[46] P. Erdos, Some remarks on the theory of graphs, Bull. Amer. Math. Soc.,53,
292-294 (1947).
[47] P. Erdos, Graph theory and probability, II, Canad. M. Math. 13, 346-352
(1961).
[48] P. Erdos, On the multiplicative representation of integers, Israel J. Math.,
2, 251-261 (1964).
[49] P. Erdos, Some applications of Ramsey’s theorem to additive number the¬
ory, European J. Combin., 1, 43-46 (1980).
[50] P. Erdos, Some problems and results in number theory, Number Theory
and Combinatorics (edited by J. Akiyama et ah), 65 - 87 (1985).
[51] P. Erdos, Some of my favorite problems and results, The mathematics of
Paul Erds, I, Algorithms Combin., 13, Springer, Berlin, 47-67, 1997.
[52] P. Erdos, A. Ginzburg and A. Ziv, Theorem in the additive number theory,
Bull. Research Council Israel, 10F, 41-43, (1961).
146 BIBLIOGRAPHY

[53] P. Erdos, R. L. Graham, P. Montgomery, B. L. Rothschild, J. H. Spencer

and E. G. Straus, Euclidean Ramsey Theorems I, J. Comb. Theory, Series
A 14, 341-363 (1973).
[54] P. Erdos, R. L. Graham, P. Montgomery, B. L. Rothschild, J. H. Spencer
and E. G. Straus, Euclidean Ramsey Theorems II, Colloq. Math. Soc. Janos
Bolyai, vol. 10, Infinite and finite sets, Keszthely, Hungary and North-
Holland, Amsterdam, 520-557 (1973).
[55] P. Erdos, R. L. Graham, P. Montgomery, B. L. Rothschild, J. H. Spencer
and E. G. Straus, Euclidean Ramsey Theorems III, Colloq. Math. Soc.
Janos Bolyai, vol. 10, Infinite and finite sets, Keszthely, Hungary and
North-Holland, Amsterdam, 559-583 (1973).
[56] P. Erdos, A. Hajnal, A. Mate and R. Rado, Combinatorial Set Theory:
partition relations for cardinals. North Holland, Amsterdam, 1984.
[57] P. Erdos and R. Rado, A combinatorial Theorem, J. London Math. Soc.
25, 249-255 (1950).
[58] P. Erdos and G. Szekeres, A Combinatorial Problem in Geometry, Com-
positio Math. 2, 464-470 (1935).
o
[59] P. Erdos and P. Turan, On some sequences of integers, J. London Math.
Soc. 11, 261-264 (1936).
[60] Martin J. Erickson,Introduction to Combinatorics, John Wiley & Sons,
1996.
[61] P. Frankl, A constructive lower bound for some Ramsey numbers, Ars Com-
binatorica, 3, 297-302 (1977).
[62] P. Frankl, R. L. Graham and V. Rodl, On the distribution of monochro¬
matic configurations, Irregularities of partitions (Fertd, 1986), Algorithms
Combin. Study Res. Texts, 8, Springer, Berlin, 71-87, 1989.
[63] P. Frankl and V. Rodl, All triangles are Ramsey, Trans. Amer. Math. Soc.
297, No. 2, 777-779 (1986).
[64] H. Fredericksen, Five sum-free sets, Proc. 6th Ann. S. E. Conf. Graph The¬
ory, Combin. &; Comput. Congressus Numerantium XIV, Utilitas Math.
Pub. Inc, 309-314 (1975).
[65] G. A. Freiman, Foundations of a Structural Theory of Set addition, Trans.
Math. Monographs, A.M.S. 37, 1973.
[66] H. Furstenberg, Recurrence in Ergodic Theory and Combinatorial Number
Theory, Princeton University Press, 1983.
[67] H. Furstenberg, Ergodic behaviour of diagonal measures and a theorem
of Szemeredi on arithmetic progressions, J. d’Analyse Math. 31, 204-256
(1977).
[68] H. Furstenberg and Y. Katznelson, An ergodic Szemeredi theorem for com¬
muting transformations, J. d’Analyse Math. 34, 275-291 (1978).
[69] H. Furstenberg and Y. Katznelson, Idempotents in compact semigroups and
Ramsey theory, Israel J. Math., 68, No. 3, 257-270 (1989).
 BIBLIOGRAPHY 147

[70] H. Furstenberg and Y. Katznelson, A density version of the Hales-Jewett

theorem, J. Analyse Math., 57, 64-119 (1991).
[71] H. Furstenberg and B. Weiss, Topological dynamics and combinatorial
number theory, J. Analyse Math., 34, 61-85 (1978).
[72] W. D. Gao, Two addition theorems on groups of prime order, J. Number
Theory, '56, 211-213 (1996).
[73] W. D. Gao, Addition theorems and group rings, J. Comb. Theory, Series
A 77, 98-109 (1997).
[74] W. T..Gowers, A new proof of Szemeredi’s theorem for arithmetic progres¬
sions of length four, GAFA, 8, 529-551 (1998).
[75] W. T. Gowers, Fourier analysis and Szemerdi’s theorem, Proceedings of
the International Congress of Mathematicians, Vol. I (Berlin, 1998). Doc.
Math., Extra Vol. I, 617-629 (1998).
[76] R. L. Graham, On partitions of En , J. Comb. Th. (A) 28, 89-97 (1980).
[77] R. L. Graham, Recent developments in Ramsey theory, Proc. Interna¬
tional Congress of Mathematicians, Warsaw, 1983,(Eds. Z. Ciesielski and
C. Olech), 2, North Holland, Amsterdam, 1555-1569 (1984).
[78] R. L. Graham, Rudiments of Ramsey Theory, American Math. Soc., 1981.
[79] R. L. Graham, Topics in Euclidean Ramsey Theory, Mathematics of Ram¬
sey Theory (Eds. J. Nesetfil and V. Rodl), Springer-Verlag, 200-213 (1990).
[80] R. L. Graham, K. Leeb and B. L. Rothschild, Ramsey’s theorem for a class
of categories, Advances in Math., 8, 417-433 (1972); errata, 10, 326-327
(1973).
[81] R. L. Graham and J. Nesetfil, Ramsey theory in the work of Paul Erdos,
The mathematics of Paul Erds, II, Algorithms Combin., 14, Springer,
Berlin, 193-209, 1997.
[82] R. L. Graham and B. L. Rothschild, Ramsey’s theorem for n-parameter
sets, Trans. Amer. Math. Soc., 159, 257-292 (1971).
[83] R. L. Graham, B. L. Rothschild and J. H. Spencer, Ramsey Theory, John
Wiley &; Sons, 1980.
[84] J. E. Graver and J. Yackel, Some graph theoretic results associated with
Ramsey’s theorem, J. Combinatorial Theory, 4, 125-175 (1968).
[85] R. E. Greenwood and A. M. Gleason, Combinatorial relations and chro¬
matic graphs, Canad. J. Math., 7, 1-7 (1955).
[86] H. Halberstam and K. F. Roth, Sequences, 2nd Ed., Springer, 1983.
[87] Frank Harary, A tribute to Frank P. Ramsey, 1903-1930, J. Graph Theory
7, 1-7, (1983).
[88] H. Harborth, Ein Extremalproblem fur Gitterpunkte, J. Reine Angew.
Math. 262/263, 356-360 (1973).
[89] A. W. Hales and R. I. Jewett,Regularity and positional games, Trans. Amer.
Math. Soc. 106, 222-229 (1963).
[90] G. H. Hardy and S. Ramanujan, Asymptotic formulae in combinatory anal¬
ysis, Proc, London Math. Soc. (2), 17, 75-115 (1918).
148 BIBLIOGRAPHY

[91] G.H. Hardy &: E. M. Wright, An introduction to the Theory of Numbers,

5th edition, Oxford University Press, 1981.
[92] E. Harzheim, Ein kombinatorisches Problem iiber Auswahlfunktionen,
Publ. Math. Debrecen, 15, 19-22 (1968).
[93] E. Harzheim, Kombinatorische betra chtungen iiber die Struktur der potenz-
menge, Math. Nachr., 34, 123-141 (1967).
[94] E. Harzheim, Ein satz der kombinatorischen mengenlehre, Fund. Math.,
61, 283-294 (1968).
[95] E. Harzheim, Combinatorial theorems on contractive mappings in power
sets, Discrete Math. 40, 193-201 (1982).
[96] D. R. Heath-Brown, Three primes and an almost-prime in arithmetic pro¬
gressions, J. London Math. Soc. (2) 23, 396-414 (1981).
[97] D. R. Heath-Brown, Integer sets containing no arithmetic progressions, J.
London Math. Soc. (2) 35, 385-394 (1987).
[98] I. N. Herstein, Topics in Algebra, 2nd edition, Wiley, New York, 1975.
[99] D. Hilbert, Uber die Irreduzibilitat ganzer rationaler Funktionen mit gan-
zahligen Koeffizienten, J. Math. 110 , 104-129 (1892).
[100] N. Hindman, Finite sums from sequences within cells of a partition of N,
J. Combinatorial Theory, Ser A 17, 1-11 (1974).
[101] N. Hindman, Ultrafilters and combinatorial number theory, Number The¬
ory Carbondale, (M. Nathanson ed.), Lecture notes in Math., 751, 119-184
(1979).
[102] N. Hindman, The semigroup /?N and its applications to number theory,
The Analytical and Topological Theory of Semigroups, (K. H. Hofmann,
J. D. Lawson and J. S. Pym, eds.), de Gruyter, 1990.
[103] Kenneth Ireland and Michael Rosen, A Classical Introduction to Modern
Number Theory, 2nd edition, Springer-Verlag, (1990).
[104] Scott Johnson, A new proof of the Erdos-Szekeres convex k-gon result, J.
Comb. Th. (A) 42, 318-319 (1986).
[105] A. Kemnitz, On a lattice point problem, Ars Combinatorica 16b, 151-160
(1983).
[106] A. Y. Khinchin, Three Pearls of Number Theory, Graylock Press, 1952.
[107] D. J. Kleitman and M. Lewin, Another proof of a result of Perry on chains
of finite sets, Amer. Math. Monthly, 79, 152-154 (1972).
[108] M. Knopp, Modular functions in analytic number theory, Chelsea, 1993.
[109] O. Kolberg, Note on the parity of the partition function. Math. Scand. 7,
377-378 (1959).
[110] H. Lefmann, On partition regular systems of equations, J. Combin. Theory
Ser. A, 58, 35-53 (1991).
[111] M. Lothaire, Combinatorics on Wards, Addison - Wesley, 1983.
[112] L. Lovasz, Kneser’s conjecture, chromatic number and homotopy,
J.Combin. Theory Ser. A, 25, 319-324 (1978).
 BIBLIOGRAPHY 149

[113] H. B. Mann, Two addition theorems, J. Combinatorial Theory 3, 233-235

(1967).
[114] H. B. Mann, Addition Theorems in Group Theory and Number Theory,
R. E. Krieger Publishing Company, Huntington, New York, 1976.
[115] R. McCutcheon, Elemental methods in ergodic Ramsey theory, Springer,
Lect. Notes Math., Vol. 1722, 1999.
[116] D. H. Mellor, The Eponymous F. P. Ramsey, J. Graph Theory 7, 9-13,
(1983).
[117] K. R. Milliken, Ramsey’s theorem with sums and unions, Combin. Theory
Ser. A, 18, 276-290 (1975).
[118] L. Mirsky, The distribution of values of the partition function in residue
classes. J. Math. Anal. Appl., 93, 593-598 (1983).
[119] J. R. Munkres, Topology, A first course, Prentice-Hall, 1975.
[120] J. R. Munkres, Elements of algebraic topology, The Benjamin/Cummings
Publishing Company, Inc., 1984.
[121] Melvyn B. Nathanson, Additive Number Theory: Inverse Problems and
the Geometry of Sumsets, Springer, 1996.
[122] J. Nesetril, Ramsey Theory, Handbook in Combinatorics, Vol. II, (Eds.
Graham, Grotschel and Lovasz), North Holland, 1331-1403 (1995).
[123] J. Nesetril and V. Rodl, Two proofs in combinatorial number theory, Proc.
American Math. Soc., 93, no. 1, 185-188 (1985).
[124] M. Newman, Periodicity modulo m and divisibility properties of the par¬
tition function. Trans. Amer. Math. Soc., 97, 225-236 (1960).
[125] J. -L. Nicolas, I. Z. Ruzsa and A. Sarkozy, On the parity of additive
representation functions. (With an appendix by J. -P. Serre). J. Number
Theory, 73, 292-317 (1998).
[126] J. -L. Nicolas and A. Sarkozy, On the parity of partition functions. Illinois
J. Math., 39, 586-597 (1995).
[127] J. E. Olson, On a combinatorial problem of Erdos, Ginzburg and Zw, J.
Number Theory 8, 52-57 (1976).
[128] K. Ono, Parity of the partition function in arithmetic progressions. J.
Reine Angew. Math., 472, 1-15 (1996).
[129] T. R. Parkin and D. Shanks, On the distribution of parity in the partition
function. Math. Comp. 21, 466-480 (1967).
[130] G. W. Peck, P. Schor, W. T. Trotter and D. West, Regressions and mono¬
tone chains, Combinatorica, 4, 117-119 (1984).
[131] R. L. Perry, Representatives of subsets, J. Combin. Theory, 3, 302-304
(1967).
[132] Karl Petersen, Ergodic theory, Cambridge studies in advance mathematics
2, Cambridge University Press, 1983.
[133] C. Pomeranceand A. Sarkozy, Combinatorial Number Theory, Handbook
in Combinatorics, Vol. I. (Eds. Graham, Grotschel and Lovasz), North
Holland, 967-1018, 1995.
150 BIBLIOGRAPHY

[134] Paul A. Pritchard, Andrew Moran and Anthony Thyssen, Twenty-two

primes in arithmetic progression, Math. Comp. 64, no. 211, 1337-1339
(1995).
[135] Hans J. Promel and B. Voigt, Graham-Rothschild Parameter Sets, Math¬
ematics of Ramsey Theory (Eds. J. Nesetril and V. Rddl), Springer-Verlag,
113-149, 1990.
[136] H. Rademacher, On the partition function p{n), Proc. London Math. Soc.
(2), 43, 241-254 (1937).
[137] H. Rademacher, Topics in Analytic Number Theory, Springer-Verlag,
1973.
[138] R. Rado, Verallgemeinerung eines Satzes von van der Waerden mit
Anwendungen auf ein problem der Zalentheorie, Sonderausgabe aus den
Sitzungsbericten der Preuss, Akad. der Wiss. Phys. -Math, klasse 17, 1-10
(1933).
[139] R. Rado, Studien zur Kombinatorik, Math. Z. 36, 424 -480 (1933).
[140] R. Rado, Some recent results in combinatorial analysis, Congres Interna¬
tional des Mathematiciens, Oslo, 1936.
[141] R. Rado, Note on.combinatorial analysis, Proc. London Math. Soc., 48,
122-160 (1943).
[142] R. Rado, A theorem on chains of finite sets, J. London Math. Soc. 42,
101-106 (1967).
[143] R. Rado, A theorem on chains of finite sets, II, Acta. Arith. 18, 257-261
(1971).
[144] S. Ramanujan, Some properties of p(n), the number of partitions of n.
Proc. Cambridge Phil. Soc., 19, 207-210 (1919).
[145] S. Ramanujan, Collected papers, Chelsea, New York, Third printing, 1999.
[146] S. Ramanujan, Congruence properties of partitions. Math. Z., 8, 147-153
(1921).
[147] S. Ramanujan, Unpublished manuscript on the partition and tau func¬
tions. The Lost Notebook and Other Unpublished Papers, Narosa, New
Delhi, (1988).
[148] F. P. Ramsey, On a problem of formal logic, Proc. London Math. Soc. (2)
30, 264-285 (1930).
[149] Paolo Ribenboim, My Numbers, My Friends, Springer-Verlag, 2000.
[150] L. Ronyai, On a conjecture of Kemnitz, Combinatorica 20 (4), 569-573
(2000).
[151] K. F. Roth, On certain sets of integers, J. London Math. Soc. 28. 104-109
(1953). •
[152] Walter Rudin, Functional Analysis, McGraw-Hill, New York, 1973 (Tata
McGraw-Hill, New Delhi, 1973).
[153] I. Z. Ruzsa, Generalized arithmetic progressions and sumsets, Acta Math.
Hungar., 65, 379-388 (1994).
 BIBLIOGRAPHY 151

[154] I. Z. Ruzsa, A small maxima' Sidon set, The Ramanujan Journal, 2, 55-58
(1998).

[155] R. Salem and D. C. Spencer, On sets of integers which contain no three

terms in arithmetic progressions, Proc. Nat. Acad. Sci. USA 28, 561-563
(1942).

[156] A. Sarkozy, On difference sets of integers III, Acta. Math. Acad. Sci.
Hungar., 31, 125-149 (1978).
[157] A. Sarkozy and V. T. Sos, On additive representation function, The math¬
ematics of Paul Erds, I, Algorithms Combin., 13, Springer, Berlin, 129-150,
1997.
[158] I. Schur, fiber die Kongruenz xm + ym = zm (modp), Jber. Deutsch.
Math. Verein. 25, 114-117 (1916).
[159] L. E. Shader, All right triangles are Ramsey in E2, J. Comb. Theory,
Series A 20, 385-389 (1976).
[160] S. Shelah, Primitive recursive bounds for van der Waerden numbers, J.
Amer Math. Soc. 1, 683-697 (1988).
[161] W. Sierpiriski, Elementary Theory of Numbers, North Holland, Amster¬
dam, 2nd edition, 1988.
[162] J. H. Spencer, Ramsey’s theorem for spaces, Trans. Amer. Math. Soc.
249, 363-371 (1979).
[163] J. H. Spencer, Ramsey Theory and Ramsey Theoreticians, J. Graph The¬
ory 7, 15-23,(1983).
[164] M. V. Subbarao, Some remarks on the partition function. Amer. Math.
Monthly, 73, 851-854 (1966).
[165] B. Sury, The Chevalley- Warning theorem and a combinatorial question
on finite groups, Proc. Amer. Math. Soc. 127, No. 4, 951-953 (1999).
[166] E. Szemeredi, On sets of integers containing no four elements in arith¬
metic progression, Acta. Math. Acad. Sci. Hungar. bf 20, 89-104 (1969).
[167] E. Szemeredi, On sets of integers containing no k elements in arithmetic
progression, Acta. Arith. bf 27, 199-245 (1975).
[168] E. Szemeredi, Integer sets containing no arithmetic progressions, Acta.
Math. Hungar.,56, 155-158 (1990).
[169] James J. Tattersall, Elementary Number Theory in Nine Chapters, Cam¬
bridge University Press, 1999.
[170] V. A. Taskinov, Regular subgraphs of regular graphs, Soviet Math. Dokl.
26, 37-38 (1982).
[171] Alan D. Taylor, Bounds for the Disjoint Unions Theorem, J. Comb. The¬
ory, Series A 30, 339-344 (1981).
[172] Alan D. Taylor, A note on van der Waerden’s Theorem, J. Comb. Theory,
Series A 33, 215-219 (1982).
[173] R. Thangadurai, On some direct and inverse problems in additive number
theory, Bull. Allahabad Math. Society, 12-13, 37-55 (1997-98).
152 BIBLIOGRAPHY

[174] Andrew Thomason, An upper bound for some Ramsey numbers, J. Graph
Theory 12, 509-517, (1988).
[175] W. T. Trotter, Partially ordered sets, Handbook in Combinatorics, Vol.
I, (Eds. Graham, Grotschel and Lovasz), North Holland, 433-480 (1995).
[176] B. L. van der Waerden, Beweis einer Baudetschen Vermutang, Nieuw
Arch. Wisk. 15, 212-216 (1927).
[177] B. L. van der Waerden, How the proof of Baudet’s conjecture was found,
Studies in Pure Mathematics (Edited by L. Mirsky), Academic Press, 251—
260 (1971).
[178] R. C. Vaughan, The Hardy-Littlewood circle method, 2nd ed., Cambridge
University Press, 1997.
[179] Honghui Wan, Upper Bounds for Ramsey Numbers R(3,3,..., 3) and Schur
Numbers , J. Graph Theory 26, 119-122, (1997).
[180] G. N. Watson, Ramanujans Vermutung iiber Zerfallungsanzahlen. J. Reine
Angew. Math. 179, 97-128 (1938).
[181] D. J. White, On finite nests of intervals of integers, J. London Math. Soc.
42, 501-503 (1967).
[182] S. Wigert, Sur I’ordre de grandeur du nombre de diviseurs d’un entier.
Arkiv for matematik, 3, 1-9 (1907).
 Index

(AT)r, 123 • Bergelson, V., 47, 58, 59

q\ 24 Berndt, Bruce C., 108
En, 61 Bjorner, A., 60
HJ(r,t), 28 Borsuk’s theorem, 59
Kn, 3 Bourgain, J., 35
R(k,r,l), 4 Brick, 66
R{k,r;li,--- ,/r), 5 Brown, T. C., 29, 48
R(m), 5
5 Canonical isomorphism, 24
5(r), 21 Caro, Y., 94
Sp, 38 Chain, 118
r), 9 Chevalley’s Theorem, 79
d(A), 33 Chevalley’s theorem, 93
/(n,d), 77 Chung, F. R. K., 35
A:- arrangement, 82 Clique, 3
^-parameter sets, 32 Columns condition, 41
/c-regular graph, 80 Combinatorial line, 27
p(n), 95 Compact semigroup, 49
per A, 82 Compactness Principle, 64
P(S), 117 Comparable, 118
7Z(C,n,r), 62 Complete graph, 3
p(n), 106 Congruent, 62
Conjugate partition, 96
Abbott, H. L., 21
Correspondence principle, 58
Adjacent, 3
Ahlgren, S., 115
Degree of a vertex, 3
Ajtai, M, 20
Deuber, W., 47
Alon, N., 78, 80, 86, 93
Dirichlet, 1, 2
Alphabet, 16
Dubiner, M., 78, 86, 93
Antichain, 118
Artin, Emil, 21
Edges of a graph, 3
Atkin, A. O. L., 108
Ends
Ax, J., 93
(of edges of a graph), 3
Balog, A., 94 Enveloping semigroup, 54
Base, 103 Erdos -Szekeres Theorem, 6
Baudet, 21 Erdos, P., 3, 8, 19-21, 33, 34, 61,
Behrend, F. A., 35 63, 65, 66, 73, 75, 91, 94

153
154 INDEX

Erdds-Ginzburg-Ziv Hajnal, A., 21

Theorem, 75, 84, 85, 89, 94 Halberstam, H., 93
Erdos-Straus’ equation, 48 Hales, A. W., 23, 32
Erickson M. J., 19 Hales-Jewett Theorem, 28, 57
Euclidean Ramsey theory, 61 Hales-Jewett theorem, 59
Euler, 99, 114 Hamming weight, 88
Euler’s identity, 102 Hanson, D., 21
Euler’s pentagonal Harary, Frank, 19
number theorem, 106 Harborth, H., 78
Hardy, 107, 114
Finite Union Theorem, 42 Harzheim, E., 118
Fliptop (colouring), 24 Heath-Brown, D. R., 35
Fliptop (Shelah s-space), 25 Height, 118
Folkman’s theorem, 41, 42, 46 Hilbert, 14
Folkman, J., 37 Hilbert’s theorem, 16, 42
Fractional part, 129 Hilbert’s theorem
Frankl, P., 67, 69 (original proof), 17
Franklin, 102 Hindman’s theorem, 42, 48, 58, 60
Fredericksen, H., 22 Hindman, N., 41
Freiman, G. A., 93
Friedland, S., 80 Ideal (left), 51

Furstenberg, H, 23, 57 Ideal (minimal left), 51

Furstenberg, H., 21, 33, 49, 51, 52, Ideal (right), 51

54-60 Ideal (two-sided), 51

Idempotent, 50

Gao, W. D., 78, 89 Incidence, 3

Generating function, 99 Incidence matrix, 80

Geometric independence, 63 Independent set, 3

Ginzburg, A., 75
Jewett, R. I., 23, 32
Glazer, S., 60
Johnson S., 6
Gleason, A. M., 20
Goldwasser, John L., 35
Kalai, G., 80
Gowers, W. T., 35, 94
Katznelson, Y., 49, 51, 52, 54-56,
Grunwald (Gallai), 10, 21
58, 59
Griinwald’s Theorem, 10, 61
Kemnitz, A., 78
Griinwald’s theorem, 28
Kernel function, 118
Graham, R. L., 32, 48, 58, 61, 63,
Kinchin, 9
65, 66, 70, 72
Klein, Esther, 19
Graph, 3
Kleitman, D. J., 119
Graph (finite), 3
Kneser, M., 59
Graph (simple), 3
Kolberg, O, 108
Graphical representation
Kolmos, J., 20
of partitions, 96 Kriz, I., 60
Graver, J. E., 20
Greenwood, R. E., 20 Leeb, K., 32
Gurevich’s conjecture, 70 Lefmann, H., 48
 INDEX 155

Left ideal, 51 Rado’s conjecture, 47

Leibman, A., 58, 59 Rado’s theorem, 40
Lewin, M., 119 Rado’s theorem
Line in Fp0Fp, 85 (abridged version), 39
Lovasz, L, 59, 60 Rado, R., 21, 37, 46, 47, 121
Ramanujan, 107, 114
MacMahon, 106
Ramsey set, 62
Mann, H. B., 93 Ramsey’s Theorem, 4
Mantel, 3
Ramsey’s theorem, 48
Mate, A., 21
Ramsey’s Theorem (generalized), 4
McCutcheon, M., 60 Ramsey’s theorem
Mellor, D. H., 19 (infinite version), 7
Milliken, K. R., 48 Recurrent point, 15
Minimal left ideal, 51 Regression, 118
Mirsky, L., Ill
Regular graph, 80
Monochromatic solution, 14 Regular
Montgomery, P., 61, 63, 65, 66 (system of linear equations), 40
Moving co-ordinates, 24 Ribenboim, P., 34
Multiplicative base, 91 Right ideal, 51
Roth, K. F., 33, 35, 93
Nathanson, M., 93
Rothschild, B. L., 32, 48, 58, 61, 63,
Nesetril, J., 32, 92, 94
65, 66
Newman, M., 109, 114
Ruzsa, E., 60, 91
Nicolas, J. -L., Ill, 112, 114
Ruzsa, I. Z., Ill, 112, 114

Olson, J. E., 94
Sarkozy, A., 94, 111, 112, 114
Ono, Ken, 107, 108, 115
Sos, V. T., 94
Order of a graph, 3
Salem, R., 35

Partial order, 50 Schor, P., 127

Partially ordered set, 117 Schreier, Otto, 21

Partition, 95 Schur Number, 21

Partition function, 95 Schur’s Theorem, 14

Peck, G. W., 127 Schur’s theorem, 29, 33, 37, 46, 47

Pentagonal numbers, 106 Schur, I., 14, 21

Permanent, 82 Self-conjugate partition, 96

Perry, R. L., 119 Semigroup, 5

Pigeonhole principle, 1 Serre, J. -P, 114

Pigeonhole principle Shader, L. E., 73

(generalized), 2 Shelah s-space, 24

Poset, 117 Shelah line, 24

Promel, Hans J., 32 Shelah point, 24

Probabilistic Method, 8 Shelah, S., 23, 32

Shift map, 16
Rodl, V., 20, 48, 67, 69, 92, 94 Sidon set, 91
Ronyai, L., 78, 88 Sierpinski’s equation, 48
Rademacher, H., 114 Simplex, 63
156 INDEX

Size of a graph, 3 Upper Banach density, 58

Slope, 103 Upper natural density, 33
Spencer, D. C., 35
van der Waerden, 21
Spencer, J. H., 32, 58, 61, 63, 65, 66
van der Waerden’s Theorem, 9, 61
Spherical set, 62
van der Waerden’s theorem, 23, 28,
Straus, E. G., 61, 63, 65, 66
31, 42, 46, 57, 60
Subbarao, M. V., 114
Vertices of a graph, 3
Subgraph, 3
Voigt, B., 32
Super modulo colour, 38
Sury, B., 94 Wan, Honghui, 21
Symbolic flow, 16 Warning’s theorem, 78
Szekeres, G., 3, 19 Watson, G. N., 107, 108
Szemeredi’s theorem, 58
Weiss, B., 21, 57
Szemeredi, E., 20, 23, 33-35, 57, 94
West, D., 127
Width, 118
Tarsy, M., 20
Wigert, S., 113
Tatter sail, James J., 114
Word, 16
Taylor, Alan D., 42, 48
Thangadurai, Ft., 93
Yackel, J., 20
Thomason, A., 20
Trotter, W. T., 127 Ziv, A., 75
Turan, P., 33, 91
Two-sided ideal, 51
DATE DUE
 TRENT UN VERSITY

0 1 64 0541420 6
Aspects of Combinatorics and
Combinatorial Number Theory

Sukumar Das Adhikari

Harish-Chandra Research Institute
Chhatnag Road, Jhusi
Allahabad-221 019, India

The present volume largely concerns Ramsey-type results in

combinatorial number theory. These results talk about
‘unavoidable regularities’. Attempt has been made to touch upon
all the classical results and, at the samejime, to give glimpses of
various techniques used to tackle these problems. However, most
of the proofs rely on combinatorial arguments. Starting from a
discussion on the pigeonhole principle (of which the classical
theorem of Ramsey can be thought of a generalization) and
the early results in the area of ‘Ramsey-type theorems in
combinatorial number theory’, later this book discusses the
theorem of Hales and Jewett, several variations of the van der
Waerden’s theorem, some generalizations of Schur’s theorem,
an introductiorvto Euclidean Ramsey theory, some Ramsey-type
theorems in additive number theory and application of Ramsey’s
theorem to number theoretic problems. Recent results on the
parity of the partition function and some Ramsey-type results in
partially ordered sets are also presented.

Preliminary knowledge of elementary number theory, linear

algebra, the rudiments of point-set topology and some basic
results in the theory of finite fields are the prerequisites for this
book.