Thanks to visit codestin.com
Credit goes to www.scribd.com

0% found this document useful (0 votes)
266 views15 pages

Combinatorial Number Theory Guide

This document summarizes Ramsey theory and van der Waerden's theorem on arithmetic progressions. It begins by describing combinatorial number theory and Ramsey's theorem for graphs colored with k colors. It proves Ramsey's theorem for subsets of size t, and the infinite version. It then discusses van der Waerden's theorem, which guarantees monochromatic arithmetic progressions of any length when the integers are colored with k colors, and proofs it using Ramsey's theorem.

Uploaded by

VVD 101
Copyright
© Public Domain
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
266 views15 pages

Combinatorial Number Theory Guide

This document summarizes Ramsey theory and van der Waerden's theorem on arithmetic progressions. It begins by describing combinatorial number theory and Ramsey's theorem for graphs colored with k colors. It proves Ramsey's theorem for subsets of size t, and the infinite version. It then discusses van der Waerden's theorem, which guarantees monochromatic arithmetic progressions of any length when the integers are colored with k colors, and proofs it using Ramsey's theorem.

Uploaded by

VVD 101
Copyright
© Public Domain
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 15

Combinatorics 3 - Combinatorial Number Theory

Henry Liu, 6 February 2012

“Combinatorial number theory”, in very loose terms, can be described as an


area of mathematics which is a cross between combinatorics and number theory.
More precisely, the area concerns structures of integers (or similar sets), with some
number theoretic properties, which can be studied mainly by combinatorial means,
rather than, for example, by algebraic methods.
This area really only truly emerged throughout the 20th century, but has become
extremely popular in the last four decades or so. In recent years, more and more
mathematical olympiad style problems related to the area have also appeared.
In this set of notes, we shall consider many of the most well-known theorems
in combinatorial number theory, and show some applications along the way. These
notes can be considered as a continuation of the set titled “Combinatorics”, and any
undefined graph theoretic terms here can be found in the preceding notes.

1. Ramsey’s Theorem Revisited


In Section 2.5 of the notes titled “Combinatorics”, we introduced the notion of
Ramsey Theory. In this section, we consider some further results in this topic. We
recall that the most basic version of Ramsey’s Theorem is the following.
Theorem 1 (Ramsey’s Theorem, 1930) Let m1 , m2 ≥ 2 be integers. Then,
there exists an integer N = N (m1 , m2 ) such that, for all n ≥ N , whenever the edges
of the complete graph Kn is coloured with two colours, there exists monochromatic
Km1 in the first colour, or a monochromatic Km2 in the second colour.
The proof of Theorem 1 was left as an exercise in the preceding notes. There are
many extensions to Theorem 1. To help us to describe these results, we make the
following definition.
Definition 1
(a) For n ∈ N, let [n] = {1, 2, . . . , n}.
(b) For a set X and t ∈ N, let X (t) = {A ⊂ X : |A| = t}. That is, X (t) is the
family which consists of all subsets of X with size t.
Note that [n](2) can trivially be identified with the complete graph Kn on n
vertices: the members of [n](2) are identified with the edges of Kn .
Our first extension of Theorem 1 is to consider using k colours, for any positive
integer k.
Theorem 2 (Ramsey’s Theorem, 1930) Let k ≥ 1 and m1 , . . . , mk ≥ 2 be
integers. Then, there exists an integer N = N (m1 , . . . , mk ) such that, for all n ≥ N ,
whenever [n](2) is coloured with colours 1, . . . , k, there exist 1 ≤ i ≤ k and a set
M ⊂ [n] with |M | = mi such that all members of the set M (2) have colour i.

1
In other words, in graph theoretic terms, whenever the edges of Kn are coloured
with colours 1, . . . , k, with n sufficiently large, we have either a Km1 in colour 1, or
a Km2 in colour 2, or . . . , or a Kmk in colour k.
Proof of Theorem 2. We refer the k colours used as colour 1, colour 2, . . . , and
colour k, and think of the elements of [n](2) as the edges of the complete graph Kn
on n vertices. We use induction on k. The case k = 1 is trivial, and the case k = 2
is Theorem 1. Now, let k ≥ 3, and suppose that the result holds for k − 1. Then
there exists an integer N 0 = N 0 (m2 , m3 , . . . , mk ) such that whenever n ≥ N 0 and the
edges of Kn are coloured with colours 2, . . . , k, we have a Kmi in colour i for some
2 ≤ i ≤ k. Now, there exists an integer N = N (m1 , N 0 ) such that, whenever n ≥ N
and the edges of Kn are coloured with colours 1, . . . , k, then either there exists a
Km1 in colour 1, or a KN 0 using only colours 2, . . . , k, and hence containing a Kmi
in colour i for some 2 ≤ i ≤ k. Hence this value of N works for the k colours, and
the result follows by induction. 

Definition 2 The least integer N = N (m1 , . . . , mk ) for which Theorem 2 holds is


the Ramsey number of Km1 , . . . , Kmk , and is denoted by R(m1 , . . . , mk ). If m1 =
· · · = mk = m, then we may write the Ramsey number as Rk (m).

We have already seen that very few exact values of the Ramsey numbers are
actually known. To date, for two colours, these are R(2, m) = m for any m ≥ 2,
R(3, 3) = 6, R(3, 4) = 9, R(3, 5) = 14, R(3, 6) = 18, R(3, 7) = 23, R(3, 8) = 28,
R(3, 9) = 36, R(4, 4) = 18, and R(4, 5) = 25. For more than two colours, the only
non-trivial (with ki ≥ 3 for all i) known Ramsey number is R(3, 3, 3) = R3 (3) = 17.
Our next result is an infinite version of Theorem 2, as follows.

Theorem 3 (Ramsey’s Theorem, Infinite Version, 1930) Let k ∈ N. Then,


whenever N(2) is coloured with k colours, there exists an infinite set M ⊂ N such
that M (2) is monochromatic.

Proof. As in the proof of Theorem 2, we refer the k colours used as colour 1, colour
2, . . . , and colour k, and think of the elements of N(2) as the edges of the infinite
complete graph on N. Take a1 ∈ N. Since we have used finitely many colours to
colour N(2) , there exists an infinite set S1 ⊂ N such that all edges from a1 to S1 have
the same colour, say colour c1 ∈ {1, . . . , k}. Next, take a2 ∈ S1 . Then as before,
there exists an infinite set S2 ⊂ S1 such that all edges from a2 to S2 have the same
colour, say colour c2 ∈ {1, . . . , k}. Repeating this inductively, we obtain a sequence
a1 , a2 , a3 , . . . in N and a sequence of colours c1 , c2 , c3 , . . . such that if i < j, then the
edge ai aj has colour ci . But then, we have ci1 = ci2 = ci3 = · · · for some infinite
subsequence i1 , i2 , i3 , . . . . Hence, the infinite set M = {ai1 , ai2 , ai3 , . . . } is such that
M (2) is monochromatic, as required. 
Example 1 (a) Suppose that we use two colours to colour N(2) such that {x, y}
is red if x + y is even, and blue otherwise. Then taking M to be the set of even
integers, or the set of odd integers, we have M (2) is an infinite monochromatic set.

2
(b) Now, suppose that we use two colours to colour N(2) such that {x, y} is red
if x + y has an even number of distinct prime divisors, and blue otherwise.
Then Theorem 3 says that we can find an infinite set M such that M (2) is a
monochromatic set. In other words, there exists an infinite set M = {x1 , x2 , x3 , . . . }
such that for all i < j, we have either xi + xj has an even number of distinct prime
divisors, or an odd number of distinct prime divisors. However, no actual example
of such an infinite set is known!
Although Theorems 2 and 3 are extremely useful and the colourings can be
viewed quite easily, namely, we are colouring the edges of a complete graph on
finitely many vertices, or on N, the theorems can be extended. It turns out that the
number ‘2’ plays no significant role in the theorems. We may extend the theorems
by considering colourings of subsets with size t.
Theorem 4 (Ramsey’s Theorem for t-sets, 1930) Let k, t ≥ 1 and m1 , . . . ,
mk ≥ t be integers. Then, there exists an integer N = N (t, m1 , . . . , mk ) such that,
for all n ≥ N , whenever [n](t) is coloured with colours 1, . . . , k, there exist 1 ≤ i ≤ k
and a set M ⊂ [n] with |M | = mi such that all members of the set M (t) have colour
i.
Theorem 5 (Ramsey’s Theorem for t-sets, Infinite Version, 1930) Let
k, t ∈ N. Then, whenever N(t) is coloured with k colours, there exists an infinite set
M ⊂ N such that M (t) is monochromatic.
Theorems 4 and 5 can be proved by induction on t. See Exercises 1 and 2 in
Section 4.

2. Monochromatic Sets in Nr and Rr


In this section, we want to consider finding monochromatic structures in X r , for
some r ∈ N, where X = N or X = R, when the members of X r are coloured.
Definition 3 Let m ∈ N. An arithmetic progression of length m is a sequence of
the form a, a + d, a + 2d, . . . , a + (m − 1)d, and an infinite arithmetic progression is
an infinite sequence of the form a, a + d, a + 2d, a + 3d, . . . , where in both cases, we
have a, d ∈ R. The numbers a and d are the first term and the common difference
of the arithmetic progression.
Throughout, we always assume that the common difference of an arithmetic
progression is positive. The first of our results is van der Waerden’s Theorem, which
states that we can find monochromatic arithmetic progressions of any given finite
length when enough initial members of N are coloured.
Theorem 6 (van der Waerden’s Theorem, 1927) Let m, k ∈ N. Then, there
exists an integer N = N (m, k) such that for all n ≥ N , whenever [n] is coloured
with k colours, there exists a monochromatic arithmetic progression in [n] of length
m.

3
Proof. We use induction on m. The theorem holds for m = 1. Now, let m ≥ 2,
and suppose that the theorem holds for m − 1. The idea of the proof goes as follows.
Provided that n is sufficiently large, then for any 1 ≤ s ≤ k, we show that we
can either find a monochromatic arithmetic progression of length m in [n], or we
can find s monochromatic arithmetic progressions in [n], with distinct colours, with
each having length m − 1, and such that the mth term of all of them is the same
integer f . For the latter case, we call such a collection of arithmetic progressions
colour-focussed, and the integer f the focus of the arithmetic progressions. Then we
are clearly done, because by taking s = k, then no matter what the colour of f is,
we will always get a monochromatic arithmetic progression with length m in [2n].
More precisely, we are done once we have proved the following claim.
Claim. For all 1 ≤ s ≤ k, there exists t = t(m, s, k) such that, whenever [t] is
k-coloured, either there exists a monochromatic arithmetic progression of length m
in [t], or there exist s colour-focussed arithmetic progressions of length m − 1 in [t].
Theorem 6 then follows by taking s = k and N = 2t.
We shall prove the claim by induction on s. For s = 1, by the induction
hypothesis for m, there exists N 0 = N 0 (m−1, k) such that, if [N 0 ] is k-coloured, then
there is a monochromatic arithmetic progression of length m − 1 in [N 0 ]. Hence, we
may take t = N 0 for s = 1.
Now, let s ≥ 2, and suppose that t0 = t0 (m, s − 1, k) is suitable for s − 1. Also
0
by the induction hypothesis for m, there exists N 00 = N 00 (m − 1, k 2t ) such that,
0
if [N 00 ] is k 2t -coloured, then there is a monochromatic arithmetic progression of
length m − 1 in [N 00 ]. We show that t = 2t0 N 00 is suitable for s. Take a k-colouring
of [t]. We are done if [t] contains a monochromatic arithmetic progression of length
m, so assume otherwise. Then, we divide [t] up into N 00 blocks of length 2t0 , say
B1 , B2 , . . . , BN 00 , where Bi = {2t0 (i − 1) + 1, . . . , 2t0 i} for every 1 ≤ i ≤ N 00 . Since
0
there are k 2t ways to colour each block, by the definition of N 00 , there are blocks
Bp , Bp+q , Bp+2q , . . . , Bp+(m−2)q which are identically coloured, for some p, q ≥ 1. For
the block Bp , since it has length 2t0 , by definition of t0 , there are s−1 colour-focussed
arithmetic progressions of length m − 1 which, together with their focus f , all lie in
Bp . Let the arithmetic progressions be A1 , . . . , As−1 , where

Aj = {aj , aj + dj , aj + 2dj , . . . , aj + (m − 2)dj } for 1 ≤ j ≤ s − 1.

Note that f = aj + (m − 1)dj for every j, and has a different colour from the
colours of A1 , . . . , As−1 . Now for each j, consider

A0j = {aj , aj + (dj + 2t0 q), aj + 2(dj + 2t0 q), . . . , aj + (m − 2)(dj + 2t0 q)}.

That is, for each j, A0j is formed by considering the copies of Aj in each of
Bp , Bp+q , Bp+2q , . . . , Bp+(m−2)q , and then taking the first term of the first copy, the
second term of the second copy, . . . , and the (m − 1)th term of the (m − 1)th copy.
Then, A01 , . . . , A0s−1 are colour-focussed at f + 2t0 q(m − 1), since f + 2t0 q(m − 1) =

4
aj + (m − 1)(dj + 2t0 q) for every j. But,

A0s = {f, f + 2t0 q, f + 2(2t0 q), . . . , f + (m − 2)(2t0 q)}

is a monochromatic arithmetic progression (consisting of the copies of f in Bp ,


Bp+q , Bp+2q , . . . , Bp+(m−2)q ), and has a different colour from A01 , . . . , A0s−1 . Hence,
A01 , . . . , A0s are s colour-focused arithmetic progressions, with focus f + 2t0 q(m − 1).
The claim follows by induction on s, and Theorem 6 is proved. 

Definition 4 The least integer N = N (m, k) for which Theorem 6 holds is denoted
by W (m, k). These are the van der Waerden numbers.

Similar to the Ramsey numbers, very few exact values of W (m, k) are known.
Trivially, we have W (m, 1) = m and W (2, k) = k + 1. For the other values of m
and k, it is known that W (3, 2) = 9, W (4, 2) = 35, W (5, 2) = 178, W (6, 2) = 1132,
W (3, 3) = 27 and W (4, 3) = 76. The exact value of W (6, 2) was determined by
Kouril and Paul in 2007.
We have the following stronger version of van der Waerden’s Theorem, which
states that not only can we find arbitrarily long arithmetic progressions when enough
initial members of N are coloured, but the common difference of the arithmetic
progression also uses the same colour.

Theorem 7 (Strengthened van der Waerden’s Theorem) Let m, k ∈ N.


Then, there exists an integer N = N (m, k) such that for all n ≥ N , whenever [n]
is coloured with k colours, there exists an arithmetic progression of length m such
that all of its terms, along with its common difference, are in [n] and have the same
colour. That is, there exist a, d ∈ N such that a, a + d, a + 2d, . . . , a + (m − 1)d and
d are all in [n] and have the same colour.

Theorem 7 can be proved by induction on k. See Exercise 6 in Section 4. We


have the following corollary by taking m = 2.

Corollary 8 (Schur’s Theorem, 1916) Let k ∈ N. Then, there exists an integer


N = N (k) such that for all n ≥ N , whenever [n] is coloured with k colours, there
exist x, y, z ∈ [n], all with the same colour, such that x + y = z. 

Alternatively, Corollary 8 can be proved by Ramsey’s Theorem.


Second proof of Corollary 8. We show that N = Rk (3) will work for the result.
Let n ≥ N = Rk (3). Then given a k-colouring c : [n] → {1, . . . , k}, define the
k-colouring c0 : [n](2) → {1, . . . , k} by c0 (ab) = c(b − a), for a < b. By Theorem 2
with m1 = · · · = mk = 3, there exist u, v, w ∈ [n] with u < v < w and c0 (uv) =
c0 (vw) = c0 (uw). So, c(v − u) = c(w − v) = c(w − u) and (v − u) + (w − v) = w − u,
and hence we can take x = v − u, y = w − v and z = w − u. 
Our next result is the Hales-Jewett Theorem, which considers colourings of a
much more abstract structure: the lattice [`]r for some `, r ∈ N. Recall that [`]r

5
consists of all ordered r-tuples (i.e., vectors) x = (x1 , . . . , xr ), where each entry xi
is in [`] = {1, . . . , `}. For each 1 ≤ i ≤ r, xi is the ith coordinate of x. We also
say that r is the dimension of [`]r . Then, roughly speaking, the theorem says that
if the dimension r is sufficiently large, whenever [`]r is coloured with k colours, we
can find a monochromatic ‘line’ of ` points in [`]r . We will see the importance of
this result, since for example, its proof is a generalisation of the proof of van der
Waerden’s Theorem.
Before we state and prove the result, we need to define exactly what the term
‘line’ means.

Definition 5 Let `, r ∈ N. A combinatorial line, or simply a line in [`]r , is a set


L ⊂ [`]r such that, for some non-empty set I = {i1 , . . . , it } ⊂ [`] and some aj ∈ [`]
for each j 6∈ I, we have

L = {x ∈ [`]r : xj = aj for j 6∈ I, and xi1 = · · · = xit }.

For ` ≥ 2, the set I is the set of active coordinates of L, and the aj , j 6∈ I, are
the inactive coordinates of L.
Let L− and L+ denote the points of L where L− +
i = 1 and Li = ` for any i ∈ I.
− −
L and L are the ‘end-points’ of L: L is the first point and L+ is the last point
+

of L. Note that we have |L| = `.

In other words, L is a ‘line’ of ` points in [`]r where certain coordinates, not


indexed by I, are some fixed numbers in [`] (these coordinates are ‘inactive’), while
the other coordinates, indexed by I, each range together from 1 up to ` (the ‘active’
coordinates). There must be at least one active coordinate.
Example 2 In [3]2 , examples of lines are

L = {(1, 3), (2, 3), (3, 3)}, I = {1};


L = {(2, 1), (2, 2), (2, 3)}, I = {2};
L = {(1, 1), (2, 2), (3, 3)}, I = {1, 2}.

Note that {(1, 3), (2, 2), (3, 1)} is not a line (even though it appears like one when
drawn!).
In [5]3 , examples of lines are

L = {(4, 1, 1), (4, 2, 1), (4, 3, 1), (4, 4, 1), (4, 5, 1)}, I = {2};
L = {(1, 5, 1), (2, 5, 2), (3, 5, 3), (4, 5, 4), (5, 5, 5)}, I = {1, 3};
L = {(1, 1, 1), (2, 2, 2), (3, 3, 3), (4, 4, 4), (5, 5, 5)}, I = {1, 2, 3}.

In each case, L− and L+ are the first and the last element in the corresponding
list for L.
We can now state and prove the Hales-Jewett Theorem.

6
Theorem 9 (Hales-Jewett Theorem, 1963) Let `, k ∈ N. Then, there exists
an integer N = N (`, k) such that for all r ≥ N , whenever [`]r is coloured with k
colours, there exists a monochromatic combinatorial line in [`]r .
An interesting consequence of the Hales-Jewett Theorem is that, if the game of
noughts and crosses with ‘`-in-a-row’ is played by any finite number of players, then
the game must end with a winner if it is played in a high enough dimension.
Proof of Theorem 9. The proof is generally quite similar to that of Theorem
6. Given a colouring of [`]r , we say that the lines L1 , . . . , Ls have focus f ∈ [`]r
if L+i = f for every i. L1 , . . . , Ls are colour-focussed (at f ) if in addition, each
Li \ {L+ + +
i } is monochromatic, and Li \ {Li } and Lj \ {Lj } have different colours for
all i 6= j.
We use induction on `. The theorem holds for ` = 1. Now let ` ≥ 2 and suppose
that the theorem holds for ` − 1, with any number of colours.
Claim. For all 1 ≤ s ≤ k, there exists r = r(`, s, k) such that, whenever [`]r is
k-coloured, then either there exists a monochromatic line in [`]r , or there exist s
colour-focussed lines in [`]r .
We are then clearly done once we have proved the claim, because by taking s = k,
we either have a monochromatic line, or k colour-focussed lines, which also gives us
a monochromatic line, no matter what colour the focus of the k lines has.
To prove the claim, we use induction on s. For s = 1, by the induction hypothesis
0
on `, there exists N 0 = N 0 (` − 1, k) such that, if [` − 1]N is k-coloured, then there
0
is a monochromatic line in [` − 1]N . The claim then clearly holds for s = 1 when
0
[`]N is k-coloured, and hence we may take r = N 0 .
Now, let s ≥ 2, and suppose that r0 = r0 (`, s − 1, k) is suitable for s − 1. Also
r0
by the induction hypothesis for `, there exists N 00 = N 00 (` − 1, k ` ) such that, if
00 r0 00
[` − 1]N is k ` -coloured, then there is a monochromatic line in [` − 1]N . We show
0 00
that r = r0 + N 00 is suitable for s. Take a k-colouring of [`]r = [`]r +N . We are
done if [`]r contains a monochromatic line, so assume otherwise. We consider [`]r as
0 00 0 00 00
[`]r = [`]r +N = [`]r ×[`]N . This means that we can think of [`]r as [`]N , with each
00 0
point of [`]N replaced by a copy of [`]r . Indeed, we may write each point (vector)
0 00
v ∈ [`]r as v = (v 0 , v 00 ), where v 0 ∈ [`]r and v 00 ∈ [`]N , so that v 00 indicates the
0 00
position of the copy of [`]r in [`]N , and v 0 consists of the coordinates of v within
0 r0 0
the copy of [`]r . Now, since there are k ` ways to colour a copy of [`]r , we can
r0 00 r0
think of the whole structure [`]r as a k ` -coloured [`]N , with each of the k ` colours
0
corresponding to a k-coloured configuration of [`]r . By the definition of N 00 , we have
0
` − 1 identically coloured copies of [`]r in [`]r which, when they are identified with
00
their corresponding points in [`]N , the points become L \ {L+ }, for some line L in
00
[`]N . Let I be the set of active coordinates of L. Also, by the definition of r0 , either
0
there is a monochromatic line in each of the ` − 1 identically coloured copies of [`]r ,
or there are s − 1 colour-focussed lines within each copy. If the former, then let L0
0
be the monochromatic line within the copy of [`]r corresponding to L− , with active

7
coordinates I 0 . But then, the line in [`]r , with active coordinates I ∪ I 0 , and whose
first and last points are (L0− , L− ) and (L0+ , L+ ), is monochromatic, a contradiction.
Hence, the latter assertion holds. Let L1 , . . . , Ls−1 be the colour-focussed lines in
0
the copy of [`]r corresponding to L− , say with active coordinates I1 , . . . , Is−1 , and
0
focus f in the same copy of [`]r (i.e., f = L+ +
1 = · · · = Ls−1 ). Note that f has
a different colour from L1 , . . . , Ls−1 . Now for each 1 ≤ i ≤ s − 1, consider the
line L0i in [`]r with first point (L− − + + +
i , L ), last point (Li , L ) = (f, L ), and active
coordinates I ∪ Ii . Then, the L0i are colour-focussed at (f, L+ ). Moreover, the line
L0s in [`]r , with first and last points (f, L− ) and (f, L+ ), and active coordinates I, is
0−
such that L0s \ {L0+ 0
s } is monochromatic, with a different colour from each Li \ {Li },
1 ≤ i ≤ s − 1. Hence, L01 , . . . , L0s is a set of s colour-focussed lines in [`]r , with focus
(f, L+ ). The claim follows by induction on s, and Theorem 9 follows. 

Definition 6 The least integer N = N (`, k) for which Theorem 9 holds is denoted
by HJ(`, k). These are the Hales-Jewett numbers.

As we have remarked earlier, we may deduce van der Waerden’s Theorem from
the Hales-Jewett Theorem.

Proof of Theorem 6, using Theorem 9. Given m and k, we claim that N =


m · HJ(m, k) will work. Let n ≥ N . Given a k-colouring c : [n] → {1, . . . , k},
0
define the k-colouring c0 : [m]n → {1, . . . , k} by c0 ((x1 , . . . , xn0 )) = c(x1 + · · · + xn0 ),
0
where n0 = HJ(m, k). By Theorem 9, [m]n contains a monochromatic line L in the
colouring c0 . The line L corresponds to a monochromatic arithmetic progression of
length m in [n], in the colouring c, where each term is the sum of the coordinates
of a point of L (the first term being the sum of the coordinates of L− ), and the
common difference is the size of the set of active coordinates of L. 

Another application of the Hales-Jewett Theorem is that we can use it to prove


Gallai’s Theorem, which is a very useful result itself and immediately implies van
der Waerden’s Theorem. Before we state and prove Gallai’s Theorem, we need a
definition.

Definition 7 Let X = N or R, and S ⊂ X r for some r ∈ N. A homothetic copy


of S is a set S 0 ⊂ X r of the form

S 0 = aS + b = {av + b : v ∈ S},

for some a ∈ X with a 6= 0, and b ∈ X r .

Theorem 10 (Gallai’s Theorem, 1943) Let X = N or R, r, k ∈ N, and let


S ⊂ X r be a finite set. Then, whenever X r is coloured with k colours, there exists
a monochromatic, homothetic copy of S in X r .

8
Gallai’s Theorem is sometimes known as Grünwald’s Theorem, or the Gallai-Witt
Theorem. We see that Gallai’s Theorem is a generalisation of van der Waerden’s
Theorem, by taking X = N, r = 1 and S = {1, . . . , m}, where m is the length of the
monochromatic arithmetic progression that we wish to find in van der Waerden’s
Theorem.

Proof of Theorem 10. Let S = {s(1), . . . , s(m)} for some m ∈ N, and n =


HJ(m, k). Given a k-colouring c : X r → {1, . . . , k}, define the k-colouring c0 :
[m]n → {1, . . . , k} by c0 (x) = c(s(x1 ) + · · · + s(xn )). By Theorem 9, [m]n contains a
monochromatic line L in the colouring c0 . It is easy to see that the line L corresponds
to a monochromatic, homothetic copy of S in X r , with respect to the k-colouring c.
Indeed, if I is the set of active coordinates of L, and aj ∈ [m], j 6∈ I are the inactive
coordinates of L, then with respect to the colouring c, the set
n X X X o
|I|s(1) + s(aj ), |I|s(2) + s(aj ), . . . , |I|s(m) + s(aj )
j6∈I j6∈I j6∈I

is a monochromatic, homothetic copy of S in X r . 

Example 3 To demonstrate the usefulness of Gallai’s Theorem and van der


Waerden’s Theorem, let us see an application to an olympiad problem which
appeared very recently.

The set of real numbers is split into two subsets which do not intersect. Prove
that for each pair (m, n) of positive integers, there are real numbers x < y < z all
in the same subset such that m(z − y) = n(y − x).
(British Mathematical Olympiad 2011/12, Round 2, Question 3)

We can quite easily solve this problem by using Gallai’s Theorem, or indeed, by
using van der Waerden’s Theorem. However, it would be a bit more difficult without
the knowledge of either theorem.
To solve the problem, observe that it suffices to prove the version with the word
‘real’ replaced by ‘natural’. Then, we may obviously think of the problem as taking
a 2-colouring of N. By Gallai’s Theorem with X = N, r = 1, k = 2, and S =
{1, m + 1, m + n + 1}, there is a colour class containing a homothetic copy of S; that
is, a set of the form {a + b, a(m + 1) + b, a(m + n + 1) + b} for some integers a and b
with a ≥ 1. We are done with x = a + b, y = a(m + 1) + b and z = a(m + n + 1) + b,
since then we have m(z − y) = amn and n(y − x) = amn.
Alternatively, using van der Waerden’s Theorem, there is a colour class
containing an arithmetic progression of length m + n + 1, say with first term a ∈ N
and common difference d ∈ N. Then the terms a, a + md and a + (m + n)d all belong
to this arithmetic progression. Hence, taking x = a, y = a+md and z = a+(m+n)d
gives m(z − y) = dmn and n(y − x) = dmn.

9
3. Density Theorems
In this short section, we shall present some rather advanced results, which were
proved more recently than all the results that we have seen so far. We shall only
state these results, since the proofs involved are highly advanced and are beyond our
scope here. These results may also be less useful for mathematical olympiads, since
one would not expect the necessity to apply such high-powered results to solve an
olympiad style problem. Nevertheless, the statements of the results are relatively
simple, and hence it is certainly not harmful to learn these beautiful results and also
see some exciting mathematical history.
We begin by observing that if the positive integers N are coloured with k colours,
then roughly speaking, some colour class must contain at least a proportion of k1 of
the positive integers. In other words, the ‘density’ of one of the colour classes in N
is at least k1 . To proceed, we must define more precisely what the word ‘density’
means.

Definition 8 Let A ⊂ N. The upper density of A in N is defined by

|A ∩ {1, 2, . . . , N }|
d(A) = lim sup .
N →∞ N

[For a sequence of real numbers (xn )∞


n=1 , its limit superior is defined by
 
lim sup xn = lim sup xm .
n→∞ n→∞ m≥n

Roughly speaking, if the sequence (xn )∞


n=1 does not converge, then it will have
several ‘limiting values’ as n → ∞. The limit superior is the supremum of these
limiting values.]

In this direction, the great mathematicians Erdős and Turán conjectured in 1936
a much stronger assertion than van der Waerden’s Theorem, which says that: Every
subset of N with positive upper density contains an arithmetic progression of any
given length. This of course implies van der Waerden’s Theorem, since whenever
the positive integers are coloured with k colours, then some colour class has upper
density at least k1 > 0, and hence contains an arithmetic progression of any given
length.
The first major result related to the conjecture appeared in 1953, when Roth
proved the case for 3-term arithmetic progressions. Later, in 1969, Szemerédi proved
the case for 4-term arithmetic progressions. Finally, in 1975, Szemerédi settled the
conjecture, using some intriguing combinatorial arguments. This beautiful result is
now named after him.

Theorem 11 (Szemerédi’s Theorem, 1975) Let m ∈ N. Then any set A ⊂ N


with positive upper density contains an arithmetic progression of length m.

10
However, the research into the Erdős and Turán’s problem did not end there.
Only two years after Szemerédi’s result was announced, Fürstenberg, in 1977, used
methods in ergodic theory (a rather advanced branch of mathematics) to give
another elegant proof of the result. Not only did Fürstenberg’s result revitalised
ergodic theory, but also, the methods involved led to many subsequent
generalisations of Szemerédi’s Theorem. For example, about a year later in 1978,
Fürstenberg and Katznelson proved a multi-dimensional version of Szemerédi’s
Theorem. In 1996, Bergelson and Leibman proved a ‘polynomial version’ which is
also a multi-dimensional result. The 1-dimensional case is as follows.

Theorem 12 (Bergelson and Leibman, 1996) Let m ∈ N, and A ⊂ N be a


set with positive upper density. Let p1 (x), . . . , pm (x) be polynomials with rational
coefficients such that for 1 ≤ j ≤ m, we have pj (0) = 0 and pj (z) ∈ Z for every
integer z. Then for every v1 , . . . , vm ∈ N, there exists integers n 6= 0 and u such
that u + pj (n)vj ∈ A for each j.

For example, setting pj (x) = jx2 and v1 = · · · = vm = 1 implies that, every


subset of N with positive upper density contains an arithmetic progression of any
given finite length, whose common difference is a perfect square.
Another proof of Szemerédi’s Theorem was given by Gowers in 2001, using ideas
from Fourier analysis and combinatorics.
Finally, another famous and related conjecture of Erdős, also proposed in 1936,
is the following.

Conjecture 13 (Erdős, 1936) Let A ⊂ N be a set of positive integers such that


X1
= ∞.
a∈A
a

Then for any m ∈ N, A contains an arithmetic progression of length m.

Conjecture 13 is still open, even for the case m = 3. If the conjecture holds,
then using the well-known fact that the sum of the reciprocals of the prime numbers
is divergent, we obtain the result that the sequence of prime numbers contains
arbitrarily long arithmetic progressions. This latter assertion itself was also a long-
standing conjecture, and was settled by Green and Tao in 2004.

Theorem 14 (Green-Tao Theorem, 2004) Let m ∈ N. Then the set of prime


numbers contains an arithmetic progression of length m.

Note that Szemerédi’s Theorem does not imply the Green-Tao Theorem, since
by the Prime Number Theorem, the number of prime numbers not exceeding n is
asymptotically equal to lnnn , which implies that the set of prime numbers has upper
density of zero in N. As of 2010, the longest arithmetic progression of primes where
the terms are actually known has 26 terms.

11
4. Problems
We divide the problems into two parts. The first set consists of some exercises,
which include the completion the missing proofs in the notes. The second set consists
of the olympiad style problems.

4.1 Exercises
1. Prove Theorem 4 as follows.
• Firstly, consider the case k = 2. Prove this case by using induction on t
as follows. The claim holds for t = 1 (trivial), so assume that t ≥ 2 and
the claim holds for t − 1. We want to find N = N (t, m1 , m2 ) which works
for the claim. To prove this we use induction on m1 + m2 . Explain why
we can find a suitable N when m1 = t, and for m2 = t. Now, let m1 , m2 ≥
t + 1, and assume that we can find suitable N1 = N1 (t, m1 − 1, m2 ) and
N2 = N2 (t, m1 , m2 − 1). By the induction hypothesis on t, we can find
N3 = N3 (t − 1, N1 (t, m1 − 1, m2 ), N2 (t, m1 , m2 − 1))
for which Theorem 4 is true. Show that we can then take N =
N (t, m1 , m2 ) = N3 + 1, hence completing the proof.
• Then, mimic the argument in the proof of Theorem 2 to deduce the result
for any k ≥ 1.
2. Prove Theorem 5 by using induction on t.
[Hint: If t ≥ 2 and the theorem holds for t − 1, then given a k-colouring of
N(t) , take a1 ∈ N and define the k-colouring c0 on (N \ {a1 })(t−1) by c0 (F ) =
c(F ∪ {a1 }). Then, mimic the idea of the proof of Theorem 3.]
3. Prove that W (3, 2) = 9.
4. Show that van der Waerden’s Theorem does not hold for infinite arithmetic
progressions when N is k-coloured, where k ≥ 2. That is, for any k ≥ 2, give
an example of a k-colouring of N such that, there do not exist a, d ∈ N where
a, a + d, a + 2d, a + 3d, . . . is monochromatic.
5. Prove the version of van der Waerden’s Theorem for geometric progressions.
That is, prove that: Given m, k ∈ N, there exists an integer N = N (m, k)
such that for all n ≥ N , whenever [n] is coloured with k colours, there exists
a geometric progression of length m.
[A geometric progression of length m is a sequence of the form a, ar, ar2 , . . . ,
arm−1 for some a, r ∈ R. Here, we assume that a, r 6= 0.]
6. Prove Theorem 7 by using induction on k.
[Hint: For k ≥ 2, show that, if N works for the theorem for k − 1, then
W (N m + 1, k) works for k.]

12
7. How many combinatorial lines are there in [`]r ?

8. Prove that HJ(2, k) = k for all k ∈ N.

4.2. Olympiad Style Problems


Here are some olympiad style problems. Some of these problems can be solved
with the help of a theorem in the notes, while others can be classed within the
combinatorial number theory area.

1. Given a sequence x1 , x2 , x3 , . . . of real numbers, a subsequence is a sequence


of the form xi1 , xi2 , xi3 , . . . , for some indices 1 ≤ i1 < i2 < i3 < · · · .

(a) Prove that an infinite sequence x1 , x2 , x3 , . . . of real numbers contains


either an infinite increasing subsequence, or an infinite decreasing
subsequence.
(b) A sequence y1 , y2 , y3 , . . . of real numbers is strictly convex if for every
i ≥ 2, we have yi < 12 (yi−1 + yi+1 ). The sequence y1 , y2 , y3 , . . . is strictly
concave if the sequence −y1 , −y2 , −y3 , . . . is strictly convex.
Prove that an infinite sequence x1 , x2 , x3 , . . . of real numbers contains
either an infinite strictly convex subsequence, or an infinite strictly
concave subsequence.

2. Define a square to be four points in the Cartesian plane with integer


coordinates such that they form the vertices of a square, with the sides parallel
to the x-axis and the y-axis.
Find the least positive integer n such that, whenever the points of

S = {(a, b) : a, b ∈ {1, 2, . . . , n}}

are coloured with two colours, then there exists a monochromatic square in S.

3. The points of the plane are coloured with two colours. Prove that there exists
a triangle in the same plane
√ such
√ that all of its vertices have the same colour,
and its sides have lengths 2, 6 and π.

4. Let S be a set of points in the plane with integer coordinates such that any
circle of radius 2012 in the same plane contains a point of S in its interior.
Prove that, given any positive integer n, there exist n points of S which are
concyclic.

5. Given a positive integer n ≥ 3, prove that there exists an integer f (n) such
that, for any f (n) points in the plane with no three points collinear, some n
points form the vertices of a convex polygon.

13
6. Each point of the plane is painted with one of three colours. Show that there
exists a triangle in the plane such that the following three conditions are
satisfied:

(a) The three vertices of the triangle have the same colour.
(b) The radius of the circumcircle of the triangle is 2009.
(c) One angle of the triangle is either two or three times larger than one of
the other two angles of the triangle.

7. (a) Each point of the 3-dimensional space is coloured with one of three
colours. Prove that some colour class realises all distances. That is,
prove that there is a colour class A such that, for all d ≥ 0, there exist
x, y ∈ A such that the distance from x to y is d.
(b) Does the same conclusion hold when we colour the 2-dimensional plane
with three colours?

8. Let k be a positive integer. Prove that, whenever the edges of the complete
graph on b 3k+1
2
c vertices are coloured with two colours, then there exists a
monochromatic path with length k.

9. Prove or disprove the following statement: Whenever the positive integers are
coloured with the colours red and blue, then either there exists a red arithmetic
progression of length 3, or there exists a blue arithmetic progression of infinite
length.

10. The positive integers are coloured with two colours. Prove that, given
a positive integer n, there exist distinct integers a, b > n such that the set
{a, b, a + b} is monochromatic.

11. Prove that for all m ∈ N, if the positive integers are partitioned into two
classes, then either one class contains m consecutive integers, or both classes
contain arithmetic progressions, each with m terms.

12. Each square of a 1000 × 1000 chessboard is given one of 500 possible colours.
Prove that there exist three squares with the same colour such that, the
triangle formed by their centres is a right-angled triangle, with two sides
parallel to the sides of the chessboard.

13. What is the minimum number of colours required to colour the elements of
{1, 2, . . . , 100} such that no colour class contains distinct integers x and y,
where x divides y?

14. The positive integers are coloured with finitely many colours. Show that there
exists a colour class S with the following property: For every positive integer
n, S contains infinitely many multiples of n.

14
15. The integers are coloured with four colours. Let x and y be odd integers
with |x| =
6 |y|. Show that there are two integers with the same colour whose
difference is one of x, y, x + y or x − y.

16. The positive integers are coloured with k colours (k is a positive integer).
Prove that there exists a colour class, say A, with the following property:
There exists a number M such that, for any positive integer n, we can find
a1 , a2 , . . . , an ∈ A such that 0 < aj+1 − aj ≤ M for all 1 ≤ j < n.

17. The positive integers are coloured with two colours. Prove that there exists
an infinite strictly increasing sequence of positive integers a1 , a2 , a3 , . . . such
that the infinite sequence of positive integers
a1 + a2 a2 + a3 a3 + a4
a1 , , a2 , , a3 , ,...
2 2 2
is monochromatic.

18. Given a positive integer k, find the smallest integer N = N (k) such that,
whenever the set {1, 2, . . . , N (k)} is coloured with k colours, then some colour
class contains three terms of the form a + x, a + y, a + x + y, for some integers
a, x, y with a ≥ 0 and 1 ≤ x < y.

19. Prove or disprove the following statement: There exists an integer k ≥ 2 such
that, whenever the positive integers are coloured with k colours, then there exist
integers x1 , x2 , . . . , xk , one from each colour class, such that x1 + x2 + · · · +
xk−1 = xk .

20. The positive integers are coloured with three colours. Prove that we cannot
have integers x and y such that x, y and x2 − xy + y 2 have distinct colours.

21. The elements of the set S = {1, 2, . . . , 3n} are coloured with three colours,
with n numbers in each colour. Is it always possible to choose integers a, b, c
from S, one in each colour, such that a + b = c?

15

You might also like