COMBINATORICS

Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

Combinatorics

Henry Liu, 15 August 2007

Combinatorics is an area of mathematics which is not easy to define. Loosely


speaking, it is the area of mathematics which studies discrete structures. It brings
together many subareas. One of these subareas is the theory of counting objects,
where many ideas occur. These ideas include permutations, combinations,
factorials, and the use of binomial coefficients. This subarea is more formally known
as enumerative combinatorics. Another subarea, which can be considered as an area
in its own right, is graph theory. Roughly speaking, a graph is a structure which
consists of a set of points, and a set of line segments joining some pairs of the points.
We can say a lot of things about such a structure.
Although there are many more subareas of combinatorics, in this set of
notes we shall just focus on these two, since they occur the most often in olympiad
mathematics. We shall also have another section here, which discusses a very useful
idea: we shall see how to prove the impossibility of certain statements, by means
of taking a colouring, with finitely many colours, of a certain structure. This is
equivalent to partitioning our structure into finitely many parts, and we can deduce
results from such partitions.
Combinatorics is an area which is hardly visible at schools. Some of the ideas
which are well exposed are the simplest ones, such as factorials, some of the ideas
involving the binomial coefficient, and maybe some basic graph theory at A-level.
It is an area which must be learnt outside the classroom.
So, we shall prove many basic theorems here. We will only leave the proof of one
theorem: Turán’s Theorem, as an exercise. This is because a proof of this result is
rather technical. The completion of a proof is denoted by the symbol .

1. Enumerative Combinatorics: Counting Objects


Many problems in combinatorics involve counting objects. In this section, we
discuss many ideas related to such problems.

1.1 Permutations and Combinations


Definition 1 Let n ∈ N. Then, we define n!, read n factorial, to be the number
n! = 1 × 2 × 3 × · · · × (n − 1) × n.
We also define 0! = 1.
Now, given n ∈ Z, n ≥ 0 and 0 ≤ r ≤ n, we would like to know: “How many
ways can we choose r objects out of n objects, subject to certain conditions?”. The
conditions include whether we care about the order in which we pick the objects,
whether the objects are distinct, and whether we can choose the objects more than
once. We first distinguish the cases whether ordering matters or not.

1
Definition 2 Let n ∈ Z, n ≥ 0 and 0 ≤ r ≤ n. If we have n objects, then a
subset of r objects where the ordering matters is an r-permutation of the objects. If
ordering does not matter, then the subset is an r-combination.
In particular, when r = n, an n-permutation is simply called a permutation.
Remark. For r = 0 in either case, we choose none of the objects. We define
a 0-permutation or a 0-combination simply to be the empty set. This must be the
case when n = 0.

We first consider r-permutations. We would like to count how many of these


there are from n objects, under certain conditions. Firstly, we consider the case
when the objects are distinct, and that we can replace the objects after picking
them.
Theorem 1 Suppose that we have n ≥ 1 distinct objects. Let 0 ≤ r ≤ n. We pick
r of the objects, one at a time, allowing replacement. Then, the number of different
r-permutations is nr . Also, when n = 0, there is exactly one 0-permutation: the
empty set.
Remark. Note that, for n ≥ 1 and r = 0 the number of ways of picking nothing
is 1, the “empty pick”. This agrees with Theorem 1; when r = 0, we have n0 = 1
way of picking nothing. From now on, this “empty pick” will always be counted,
unless otherwise stated.

Proof of Theorem 1. There are n ways of picking the first object. Then there
are n ways of picking the second object, since replacement is allowed. Repeating
this r times, we get nr as the total number of possibilities. 

Example 1. There are 64 = 1296 4-digit numbers which contain the digits 1, 2,
3, 4, 5 and 6.

Next, we consider the same question, but not allowing replacement.


Theorem 2 Suppose that we have n ≥ 0 distinct objects. Let 0 ≤ r ≤ n. We pick
r of the objects, one at a time, without replacement. Then, the number of different
r-permutations is
n!
n × (n − 1) × · · · × (n − r + 1) = .
(n − r)!
Remark. In particular, when r = n, the total number of permutations of n
n! n!
distinct objects is (n−n)! = n!. Also, when r = 0, we have (n−0)! = 1 0-permutation,
as expected.

Proof of Theorem 2. The result is true for n = 0. For n ≥ 1, we argue as


in Theorem 1. We have n choices of picking the first object. After that, we have
n − 1 choices, then n − 2 choices, and so on, until we have n − r + 1 choices of

2
picking the rth object, since we are not allowing replacement. So the total number
n!
of possibilities is n × (n − 1) × · · · × (n − r + 1) = (n−r)! . 
n!
Definition 3 The expression (n−r)!
is usually denoted by nPr .
Example 2. If n = 5 and r = 2, we have the following 2-permutations of
{1, 2, 3, 4, 5}: {1, 2}, {1, 3}, {1, 4}, {1, 5}, {2, 1}, {2, 3}, {2, 4}, {2, 5}, {3, 1}, {3, 2},
{3, 4}, {3, 5}, {4, 1}, {4, 2}, {4, 3}, {4, 5}, {5, 1}, {5, 2}, {5, 3} and {5, 4}, giving a
5!
total of 20. This agrees with Theorem 2: (5−2)! = 20.

Before we proceed, we would like to divert and make a shorthand notation. The
above listing of r-permutations looks rather cumbersome. We can make such a listing
look a bit more pleasant simply by ignoring the chain brackets, and the commas
within the r-permutations. This shorthand notation also works for r-combinations.
Definition 4 For an r-permutation or an r-combination {x1 , x2 , . . . , xr }, we may
write it as x1 x2 · · · xr
So, the above listing can simply be written as 12, 13, 14, 15, 21, 23, 24, 25, 31,
32, 34, 35, 41, 42, 43, 45, 51, 52, 53, 54.
Next, we consider the same question, but when the objects may not be distinct.
In this case, we only consider the number of permutations.
Theorem 3 Suppose that we have n ≥ 0 objects, where n1 of them are of type 1,
n2 of them are of type 2, . . . , nk of them are of type k, so that n = n1 + · · · + nk .
Then, the number of permutations of the objects is
n!
.
n1 ! × n2 ! × · · · × nk !
Proof. If we disregard that some of the objects are the same, then we would
have n! permutations. Now, in each permutation, within the objects of type 1, they
can be permuted in n1 ! ways; within the objects of type 2, they can be permuted in
n2 ! ways, and so on. So, such a permutation has been counted n1 ! × n2 ! × · · · × nk !
times, and we must divide n! by n1 ! × n2 ! × · · · × nk ! to get the correct answer. 

Example 3. Another way of thinking about the situation of Theorem 3 is just


the number of anagrams of a word. For example, there are
11!
= 34650
1! × 4! × 4! × 2!
anagrams of the word ‘Mississippi’.

Remark. It is not easy to get a precise expression for the number of


r-permutations in the situation of Theorem 3.

So far, we have been counting the number of r-permutations from n objects,

3
where we care about the ordering in which the chosen objects appear. We now want
to count the number of r-combinations from n objects, so that we do not care about
the ordering of the chosen objects. Firstly, Theorem 4 below considers the case when
we cannot replace an object after picking it. This is the analogue of Theorem 2.

Theorem 4 Suppose that we have n ≥ 0 distinct objects. Let 0 ≤ r ≤ n. We pick


r of the objects, without replacement. Then, the number of different r-combinations
is
n!
.
r!(n − r)!
Remark. If r = n, there is only one possibility: to choose all n objects. This
n!
agrees with the expression in Theorem 4: n!(n−n)! = 1. Also, for r = 0, there is one
n!
possibility; to choose nothing. This again agrees with the expression: 0!(n−0)! = 1.

Proof of Theorem 4. Simply, if we fix an r-permutation of the objects, we


may permute the r objects in r! ways. Each such r-permutation represents the same
n!
r-combination. So the answer is just to divide nPr by r!, giving r!(n−r)! possibilities.

n!
is usually denoted by nCr , or by nr . It is

Definition 5 The expression r!(n−r)!
called a binomial coefficient and we read it as “n choose r”.

Example 4. There are 10 ways of choosing two numbers from the set {1, 2, 3, 4, 5},
the subsets being 12, 13, 14, 15, 23, 24, 25, 34, 35, and 45. This agrees with Theorem
5!
4: 2!(5−2)! = 10.

Finally, we want to count the number of r-combinations of n distinct objects,


but allowing replacement.

Theorem 5 Suppose that we have n ≥ 0 distinct objects. Let 0 ≤ r ≤ n. We pick r


of the objects, allowing replacement. Then, the number of different r-combinations
is  
(n + r − 1)! n+r−1
= .
r!(n − 1)! r
Proof. Given such an r-combination, we can code it by a sequence of dots
and dashes as follows. Suppose that the r-combination has r1 objects of type 1,
r2 objects of type 2, . . . , rn objects of type n, where 0 ≤ ri ≤ r for each i, and
r1 + · · · + rn = r. Now, starting with n − 1 dashes, we place r1 dots to the left of
the first dash, then r2 dots between the first and second dashes, r3 dots between
the second and third dashes, and so on, and rn dots to the right of the last dash.
Clearly, this gives a direct correspondence between the r-combinations and all the
possible permutations of r dots and n − 1 dashes. So, we can apply Theorem 3: we
have r + n − 1 objects, r of one type, and n − 1 of another type. The total number
of permutations is therefore (r+n−1)!
r!(n−1)!
= n+r−1
r
. 

4
Example 5. A bakery sells ten types of cakes. The number of ways of choosing
four cakes is 13
4
= 715 (provided that there are at least four of each type of cake
available).

1.2 More on the Binomial Coefficient


In Definition 5, we defined the binomial coefficient nr . In this section, we briefly


digress and discuss some results about these numbers, before we return to discuss
more on counting objects.
Definition 6 We define Pascal’s Triangle as follows. We start with a 1, and this
is row 0 of the triangle. A subsequent row is formed as follows. To get an entry in
row i, we add two adjacent entries in row i − 1, and place the sum in the space in
row i which is between and below the two row i − 1 entries. If one of the entries in
row i − 1 is missing, we assign a 0 to the missing spot.
Also, in each row, the entry on the left is the 0th entry of the row, and the entries
to its right are, in order, the 1st entry, the 2nd entry, and so on.
Below are row 0 to row 5 of Pascal’s Triangle.
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
It is well-known that the entries of Pascal’s Triangle are precisely the binomial
coefficients.
Theorem 6 For n ∈ Z, n ≥ 0 and 0 ≤ r ≤ n, the rth entry in row n of Pascal’s
triangle is precisely nr .
Proof. We use induction on n. The result is true for n = 0. We only have r = 0,
0

and 0 = 1, the only entry in row 0 of the triangle.
Now, let i ≥ 1, and suppose that the result holds for every entry in row i − 1 of
the triangle. Let 0 ≤ r ≤ i. If 1 ≤ r ≤ i − 1, then the rth entry in row i is
   
i−1 i−1 (i − 1)! (i − 1)!
+ = +
r−1 r (r − 1)!(i − r)! r!(i − 1 − r)!
 
(i − 1)! 1 1
= +
(r − 1)!(i − 1 − r)! i − r r
(i − 1)! i
= ·
(r − 1)!(i − 1 − r)! (i − r)r
 
i! i
= = .
r!(i − r)! r

5
If r = 0, then the 0th entry in row i of the triangle is
   
i−1 i
0+ =1= .
0 0

If r = i, then the ith entry in row i of the triangle is


  
i−1 i
+0=1= .
i−1 i

We have now shown that the result holds for the entries in row i. Hence the
result follows by induction. 

Next, we have the most well-known result involving binomial coefficients, the
binomial theorem.

Theorem 7 (Binomial Theorem) Let x, y ∈ C, and n ∈ Z, n ≥ 0. Then,


n  
n
X n n−r r
(x + y) = x y .
r=0
r

Proof. There are many proofs of this result. We give probably the simplest.
When we expand out (x + y)n , each term we obtain consists of some x’s and
some y’s multiplied together, the total number of x’s and y’s being n (this is before
we add any equal terms together). So, it suffices to show that, for each 0 ≤ r ≤ n,
n n−r r
there are precisely r terms x y . But this is clear by Theorem 4. We have n
y’s, and we have nr ways of picking r y’s from them. Whenever  we choose r y’s,
n n−r r
we are automatically left with n − r x’s. So we have precisely r terms x y . 

The binomial coefficient also satisfies many properties, and Theorem 8 below
shows some of the most well-known ones.

Theorem 8 Let n ∈ Z, n ≥ 0. We have the following.

(a) nr = n−r n n!
 
, if 0 ≤ r ≤ n. Both are equal to r!(n−r)! .

(b) n0 = nn = 1 if n ≥ 0, and n1 = n−1 n


   
= n if n ≥ 1.

(c) nr + r+1 n
= n+1
  
r+1
, if 0 ≤ r ≤ n − 1.

(d) n0 + n1 + n2 + · · · + nn = 2n .
   

(e) n0 − n1 + n2 − · · · + (−1)n nn = 0.
   

(f ) 1 n1 + 2 n2 + 3 n3 + · · · + n nn = n2n−1 .
   

2 2 2 2
(g) n0 + n1 + n2 + · · · + nn = 2n

n
.

6
Proof. Properties (a) and (b) are obvious from the definition of nr .


Property (c) has already been proved in the main calculation in the proof of
Theorem 6.
Properties (d) and (e) follow from Theorem 7 by letting x = y = 1 and x = 1,
y = −1, respectively.
= nr=0 nr y r .

To get property (f), we let x = 1 in Theorem 7 to get (1 + y)n P
P
n n r−1
Differentiating both sides with respect to y gives n(1 + y)n−1 = r=1 r r y .
Then, letting y = 1 gives the result.
n 2n
To get property (g), we
 consider the coefficient of x in the expansion of (1+x) .
2n
On one hand, this is n , by Theorem 7. On the other hand, by Theorem 7 again,
we have

n  
! n  
!
2n n n
X n r X n s
(1 + x) = (1 + x) (1 + x) = x x .
r=0
r s=0
s

In the product on the right, when we expand, we get a term involving xn whenever
we multiply a term which involves xt in the first sum by a term which involves xn−t
in the second
n n
 sum, for some 0 ≤ t ≤ n. For such a t, the corresponding coefficient
is t n−t . Summing these coefficients as t varies, and using property (a), we
get that the overall coefficient of xn is also n0 nn + n1 n−1
 n 
+ · · · + nn n0 =
   
n 2
2 2
+ n1 + · · · + nn . This gives property (g).

0


In addition to the properties of binomial coefficients given in Theorem 8, there


are many more which are a little bit more complicated.

1.3 Other Ideas on Counting


In this section, we consider some more ideas about counting objects.

1.3.1 Counting Objects in Two Ways


Suppose that we have a certain number of objects. If we are able to count the
objects to sum to the total in two ways, then we have an equation linking two
expressions, each of which equals the total number of objects.
We have already seen one instance of this idea: when we derived property (g) of
Theorem 8. There, we counted the number of terms involving xn in the expansion
of (1 + x)2n , in two ways.
We give another example below. There will be some more examples on this idea
later.

Example 6. n real numbers whose sum is n are placed on a circle. Show that
for any 1 ≤ k ≤ n, some k successive numbers must have a sum at least k.
Let a1 , a2 , . . . , an be the numbers. We can use the numbers to form an array as

7
follows.
a1 a2 a3 ··· an
a2 a3 a4 ··· a1
a3 a4 a5 ··· a2
.. .. .. ..
. . . .
ak ak+1 ak+2 · · · an+k−1
Note that the indices are taken modulo n. Now, the total sum of all the numbers
in the array is kn, since there are k rows, each with sum n. So, the sum of the
columns is also kn. Since there are n columns, some column must have a sum of
at least kn
n
= k (if every column has sum less than k, then the total sum of the
columns, hence the sum of everything in the array, is less than kn, a contradiction).
Moreover, each column represents successive terms on the circle. So we are done.

1.3.2 Bijection Proofs


We now describe a useful technique which can prove statements of the form:

“The number of objects with property A is equal to the number of objects


with property B”.

The idea is that we do not actually try to find the common total number (which
may be difficult to compute), but we try to ‘match’ the objects with property A to
the objects with property B in a ‘one-on-one’ manner. Such a matching is formally
known as a bijection.

Definition 7 Let A and B be sets. A function f : A → B is a bijection if

• for every x, y ∈ A, if x 6= y, then f (x) 6= f (y);

• for every y ∈ B, there exists an x ∈ A such that f (x) = y.

So, going back to the scenario, let A be the set of objects with property A, and
define B likewise. If we can find a bijection from A to B, then it easily follows that
|A| = |B|, provided that the common number is finite. For a problem of such a
form, a proof which exhibits a bijection is a ‘bijection proof’.

We will given an example which considers partitions of a number n ∈ N. We


need to define what a partition is first.

Definition 8 Let n ∈ N. A partition of n is a representation of n as a sum of


positive integers. Two partitions of n are distinct if the two sets of terms involved
are not equal as multisets. For example, 1 + 1 + 3 and 1 + 3 + 1 represent the same
partition of 5, while 5, 2 + 3 and 4 + 1 represent pairwise distinct partitions of 5.

Remark. So, given n ∈ N, every partition of n is equal to a unique partition


where the terms are listed in decreasing order (decreases of zero are possible). So, if

8
we want to list out all the distinct partitions of n, it is a good idea to assume that
the terms are decreasing in each partition.

Example 7. We prove the following result. For n, r ∈ N with 1 ≤ r ≤ n, the


number of partitions of n into at most r terms is equal to the number of partitions
of n into any number of terms, each of which is at most r.
Throughout, we assume that all partitions have their terms decreasing. As an
example, for n = 5 and r = 3, the partitions satisfying the former are 3 + 1 + 1,
2+2+1, 4+1, 3+2 and 5, while the partitions satisfying the latter are 1+1+1+1+1,
2 + 1 + 1 + 1, 3 + 1 + 1, 2 + 2 + 1 and 3 + 2.
We will use a bijection proof to prove the result. For any partition of n, we may
represent it by means of a diagram, the Young diagram, or the Ferrers diagram.
Figure 1 shows an example of such a diagram, for the partition 5 = 2 + 2 + 1.

.............................................................................
.... .... ....
... ... ...
... ... ...
..................................................................................................................
.... .... .... ....
.... .... .... ....
.............................................................................................................

Figure 1

In a Young diagram, the number of boxes is n. The columns are always decreasing
in height from left to right (possibly by zero), and the number of boxes in the columns
are the terms of the partition.
We use this idea to produce a bijection proof. Firstly, it is easy to check that
the correspondence between the partitions and the Young diagrams of n is bijective
(clearly, two different partitions of n map to different Young diagrams, and every
Young diagram of n is formed from some partition of n). Now, a partition of n with
at most r terms has at most r columns in its diagram, while a partition of n, each of
whose terms is at most r, has a diagram such that every column has height at most
r. So, it is enough to find a bijection between these two classes of Young diagrams.
But this is fairly obvious. For a Young diagram of the first type, if we reflect it about
the diagonal line at 45◦ , we get another Young diagram whose columns have heights
at most r, and so is of the second type. The converse holds as well. Obviously, two
distinct Young diagrams of the first type, when reflected, give two distinct Young
diagrams of the second type. From these, it is obvious that this ‘reflection’ is the
required bijection, and we are done.

9
For the example where n = 5 and r = 3, we get the bijection

3+1+1 ←→ 3+1+1
2+2+1 ←→ 3+2
4+1 ←→ 2+1+1+1
3+2 ←→ 2+2+1
5 ←→ 1+1+1+1+1

Remark. It is easily conceivable that, for general n and r, the above common
number of partitions is difficult to find exactly. This indeed is the case.

1.3.3 Inclusion-Exclusion Principle


When one attempts to count the number of objects, a very common mistake
is usually that some objects are counted too many times. The inclusion-exclusion
principle is one of the results which can help us to avoid this mistake. Let us look
at a simple example first.

Example 8. A class of students sat a maths exam and an English exam. 18


students got a grade A in maths only, 13 students got a grade A in English only,
and 5 students got grade A’s in both. How many students got at least one grade A?

If we were just to add the 18 to the 13, we get 31, but we will have counted those
students who got grade A’s in both exams twice. So we need to subtract the number
of such students, which is 5. The answer is therefore 18 + 13 − 5 = 26 students.
To generalise this, we first define some well-known set operations.
Definition 9 Let A and B be any two sets.
(a) We call the elements which belong to both A and B the intersection of A and
B. This set is denoted by A ∩ B.

(b) We call the elements which belong to A or B (including both) the union of A
and B. This set is denoted by A ∪ B.

(c) We call the elements which belong to A, but not to B, the difference of B from
A. This set is denoted by A \ B, or A − B.
Remark. These three operations can obviously extend to more than two sets.
But when this happens, if we have at least two of ∩, ∪ and \, or at least two of \
involved, we must use brackets to show which operation(s) must be done first. For
example, ‘A∩B ∪C’ does not make sense, since in general, (A∩B)∪C 6= A∩(B ∪C).
‘A \ B \ C’ makes no sense, since (A \ B) \ C 6= A \ (B \ C), in general.
However, if we just have ∩ or ∪ on its own, we do not need any brackets, because
in these cases, the set that we get at the end will be the same, no matter which order
we use for the operations. For example, A ∩ B ∩ C makes sense, since (A ∩ B) ∩ C =

10
A ∩ (B ∩ C) for any three sets A, B, C, and we use A ∩ B ∩ C for both of these.
Similarly for ∪ on its own, and for more than three sets.

So, we can interpret Example 8 in this way. Let A be the set of the students
who got a grade A in maths, and B be the set of the students who got a grade A
in English. So, we have |A| = 18, |B| = 13 and |A ∩ B| = 5. We want to find the
value of |A ∪ B|. With the same argument as given in the example, we get

|A ∪ B| = |A| + |B| − |A ∩ B| = 18 + 13 − 5 = 26.

We have now proved the formula

|A ∪ B| = |A| + |B| − |A ∩ B|.

Can we extend this formula to three sets? Suppose that we have three sets A, B
and C, and we know the values of |A|, |B|, |C|, |A∩B|, |A∩C|, |B∩C| and |A∩B∩C|.
Can we find the value of |A ∪ B ∪ C|?
We can indeed apply the same argument. If we add |A|, |B| and |C|, we will
have counted the elements of (A ∩ B) \ (A ∩ B ∩ C), (A ∩ C) \ (A ∩ B ∩ C) and
(B ∩ C) \ (A ∩ B ∩ C) twice, and the elements of A ∩ B ∩ C three times. Then, if
we subtract |A ∩ B|, |A ∩ C| and |B ∩ C|, we will then have counted the elements of
(A ∩ B) \ (A ∩ B ∩ C), (A ∩ C) \ (A ∩ B ∩ C) and (B ∩ C) \ (A ∩ B ∩ C) correctly,
but we also will have counted the elements of A ∩ B ∩ C 3 − 3 = 0 times. So, adding
|A ∩ B ∩ C|, we get the correct formula for |A ∪ B ∪ C|:

|A ∪ B ∪ C| = |A| + |B| + |C| − (|A ∩ B| + |A ∩ C| + |B ∩ C|) + |A ∩ B ∩ C|.

This, of course, generalises further to n sets. The result is the inclusion-exclusion


principle.
Theorem 9 (Inclusion-Exclusion Principle) Let A1 , A2 , . . . , An be finite sets.
Then,

|A1 ∪ A2 ∪ · · · ∪ An | = (|A1 | + |A2 | + · · · + |An |)


−(|A1 ∩ A2 | + |A1 ∩ A3 | + · · · + |An−1 ∩ An |)
+(|A1 ∩ A2 ∩ A3 | + |A1 ∩ A2 ∩ A4 | + · · · )
−···
+(−1)n−1 |A1 ∩ A2 ∩ · · · ∩ An |,

where, in the sum on the right, we have the size of every k-wise intersection for
every 1 ≤ k ≤ n, where a k-wise intersection is assigned with the sign (−1)k−1 .
Before we prove this, we shall prove a well-known identity about sets.
Lemma 10 Let n ∈ N, n ≥ 3. For sets B1 , B2 , . . . , Bn , we have

(B1 ∪ B2 ∪ · · · ∪ Bn−1 ) ∩ Bn = (B1 ∩ Bn ) ∪ · · · ∪ (Bn−1 ∩ Bn ).

11
Proof. We first prove that, for three sets A, B, C, we have (A ∪ B) ∩ C =
(A ∩ C) ∪ (B ∩ C). We have
x ∈ (A ∪ B) ∩ C ⇐⇒ x ∈ A ∪ B and x ∈ C
⇐⇒ x ∈ A or x ∈ B, and x ∈ C
⇐⇒ x ∈ A ∩ C or x ∈ B ∩ C
⇐⇒ x ∈ (A ∩ C) ∪ (B ∩ C),
so the identity holds. Now, using this identity repeatedly, we have
(B1 ∪ B2 ∪ · · · ∪ Bn−1 ) ∩ Bn = [(B1 ∪ B2 ∪ · · · ∪ Bn−2 ) ∩ Bn ] ∪ (Bn−1 ∩ Bn )
= [(B1 ∪ B2 ∪ · · · ∪ Bn−3 ) ∩ Bn ]
∪(Bn−2 ∩ Bn ) ∪ (Bn−1 ∩ Bn )
= ···
= (B1 ∩ Bn ) ∪ · · · ∪ (Bn−1 ∩ Bn ).


Proof of Theorem 9. We use induction on n. The assertion is trivial if n = 1,


and the case n = 2 has already been proved in the above discussion concerning
Example 8.
Now, let A1 , A2 , . . . , An be finite sets, where n ≥ 3. Assume that the statement
holds for any family of less than n finite sets. We have
|A1 ∪ A2 ∪ · · · ∪ An | = |A1 ∪ A2 ∪ · · · ∪ An−1 | + |An |
−|(A1 ∪ A2 ∪ · · · ∪ An−1 ) ∩ An |.
Now, in the expression on the right, we can expand |A1 ∪ A2 ∪ · · · ∪ An−1 | (by
induction), and when we do so, we get the required terms (with the correct signs)
which involve all the k-wise intersections of A1 , . . . , An−1 , for every 1 ≤ k ≤ n − 1.
We need to show that the other two terms will account for the missing required
terms. That is, all the k-wise intersections of A1 , . . . , An which involve An , for all
1 ≤ k ≤ n. We already have |An |. Now, using Lemma 10 and induction again, we
have
−|(A1 ∪ A2 ∪ · · · ∪ An−1 ) ∩ An | = −|(A1 ∩ An ) ∪ · · · ∪ (An−1 ∩ An )|
= −(|A1 ∩ An | + |A2 ∩ An | + · · · + |An−1 ∩ An |)
+(|A1 ∩ A2 ∩ An | + · · ·
+|An−2 ∩ An−1 ∩ An |)
−···
+(−1)n−2 (|A1 ∩ · · · ∩ An−2 ∩ An |
+|A1 ∩ · · · ∩ An−3 ∩ An−1 ∩ An | + · · ·
+|A2 ∩ · · · ∩ An−1 ∩ An |)
+(−1)n−1 |A1 ∩ A2 ∩ · · · ∩ An−1 ∩ An |.

12
The final expression consists exactly the missing terms that we needed, with the
correct signs. 

Example 9. How many positive integers less than 3000 are coprime to 3000?
The prime factorisation of 3000 is 3000 = 23 × 3 × 53 . Let S be the positive
integers less than 3000, that is, S = {1, 2, . . . , 2999}. We count the number of
elements of S which are not coprime to 3000. The highest common factor of such a
number with 3000 is divisible by at least one of 2, 3 and 5, so such a number must be
divisible by at least one of 2, 3 and 5. Let A be the elements of S which are divisible
by 2, B be the elements of S which are divisible by 3, and C be the elements of S
which are divisible by 5. We would like to find |A ∪ B ∪ C|, the number of elements
of S which are divisible by at least one of 2, 3 and 5.
We have |A| = |{2, 4, 6, . . . , 2998}| = 1499, |B| = |{3, 6, 9, . . . , 2997}| = 999, and
|C| = |{5, 10, 15, . . . , 2995}| = 599. Also, A ∩ B are the elements of S which are
divisible by 6, so |A ∩ B| = |{6, 12, 18, . . . , 2994}| = 499. Similarly,

|A ∩ C| = |{10, 20, 30, . . . , 2990}| = 299,


|B ∩ C| = |{15, 30, 45, . . . , 2985}| = 199, and
|A ∩ B ∩ C| = |{30, 60, 90, . . . , 2970}| = 99.

So, by the inclusion-exclusion principle, we have

|A ∪ B ∪ C| = (1499 + 999 + 599) − (499 + 299 + 199) + 99 = 2199.

This is the number of elements of S which are not coprime to 3000. So the
number of elements of S which are coprime to 3000 is 2999 − 2199 = 800.

1.3.4 Pigeonhole Principle


The simplest case of the pigeonhole principle is that, if we have n boxes and we
place at least n + 1 objects into them, then some box must contain at least two
objects. This easily generalises to the following result.
Theorem 11 (Pigeonhole Principle) Let n, k ∈ N. Suppose that we place at
least kn + 1 objects into n boxes. Then some box must contain at least k + 1 objects.
Proof. Assume not. Then every box has at most k objects, so that the total
number of objects is at most kn. This is a contradiction, since we have at least
kn + 1 objects. 

Theorem 11 is extremely useful, not just in combinatorics, but in just about


every branch of mathematics as well.

Example 10. Suppose that we have 10 points within a square of side length 3
(including on
√ the edges of the square). Prove that some two points must be within a
distance of 2 of each other.

13
Divide the square into nine 1×1 squares. Then by the pigeonhole principle, some
two points must lie within one of the squares (including the sides of that square).
So
√ we are done,
√ since the longest possible distance between any two such points is
2 2
1 + 1 = 2.

2. Graph Theory
2.1 Graphs: Definitions
Graphs are very important objects in combinatorics. They are so important that
there is actually an area of mathematics devoted to them: graph theory (which is
a rather large area). Graphs interact very well with other areas of combinatorics.
Roughly speaking, a graph is just a set of points, with some pairs of them being
joined by some lines. Below, we give a more precise definition.
Definition 10 A (simple) graph is a pair G = (V, E), where V is a set, and E is
a family of subsets of V , each is of size 2 (and unordered).
An element of V is called a vertex of G, and an element of E is called an edge
of G. V is the vertex set of G, and E is the edge set of G.
Sometimes, we may want to write V (G) = V and E(G) = E, to emphasize the
dependence of V and E on G.
We write |G| = |V (G)| and e(G) = |E(G)|, and call these the order and the size
of the graph G, respectively. Note that each of these can be infinite (we shall see
some examples soon).
The best way to think of a (simple) graph is by a picture. V is a set of points
(namely, vertices). E is a set of line or curve segments (namely, edges) joining pairs
of vertices such that, a subset of V of size 2 is in E if and only if there is an edge
joining the corresponding two vertices of V .
In this way, we usually use letters like v and u for the vertices of a graph. For
edges, if the edge with end-vertices u and v is in the graph, we just write uv to
denote this edge, rather than {u, v}. Note that uv = vu. We may also use letters
like e and f to denote edges.

Example 11. If V = {v1 , v2 , v3 , v4 , v5 } and E = {v1 v2 , v1 v4 , v1 v5 , v2 v4 , v2 v5 , v3 v5 },


then we get the following graph.

. . . .... .
. v1
.
....................
........ ...... ...........
......
. .. ..
.... .. ...
... ......................................... ...........
. .
. ..
v5 ..........
.
..................................
...... ....
...... ...
.
..
.
. ..

....
.
.................................
......
.......
...... v2
....... .... .
. .... .....
... ........... ..........
.... .... ... .
. ...........
... ...... ...........
.... ..........
v4
.
...... .......
....
v3

Figure 2

14
Remark. We need to be rather careful. We do not allow the following here.

• No ‘loops’ are allowed. That is, a vertex cannot be joined to itself. If we are
allowed to have ‘loops’, then we are talking about ‘graphs with loops’.

• No pair of distinct vertices can have more than one edge joining them. If we
allow this, then we have ‘multiedges’, and we are talking about ‘multigraphs’.

• The edges are ‘unordered pairs of vertices’. If we care about the ordering of
an edge, we have ‘directed edges’, and we are talking about ‘directed graphs’.

Hence, sometimes we use the word ‘simple’ to emphasize that we do not allow
these things. But here, we will drop this word, since we will only be talking about
simple graphs.
However, the edges do not necessarily have to be straight line segments. We can
use curves for edges. But it is absolutely essential to use large dots to show the
vertices, because for certain graphs, it may be the case that no matter how we draw
them, we may be forced to have that some edges must cross (this ought to happen if
we have lots of edges), so that we must declare if there is a vertex at such a position
or not.
Also, the set V can be an infinite set, in which case E can also be infinite. Here
are some examples.

Example 12. We may have V = Z, and E consisting of the edges which join
distinct pairs of even numbers. So, such a graph can only be partially drawn.
Indeed, depending on the vertex set, some graphs may be even harder to draw,
maybe impossible. For example, if V = R, and E consists of the edges which join
pairs of real numbers with unit distance between them. We can only barely draw a
few vertices and edges of this graph. We can only define it in an abstract way.

Graphs can model many real life situations. For example, a situation where a
‘network scenario’ is present. The vertices can represent people and edges can join
two people if they are related. Or, the vertices can represent computers and the
edges can join two computers if they are related in some way. And so on. In this
way, graphs present very nice ways of looking at such a situation.
Now, we define several important classes of graphs, with their notations.

Definition 11 Let n, m ∈ N.

(a) The complete graph on n vertices is the graph with n vertices, where each pair
of vertices is joined by an edge. It is denoted by Kn . We have e(Kn ) = n2 .


(b) The empty graph on n vertices is the graph with n vertices, but with no edges.
It is denoted by En . We have e(En ) = 0.

15
(c) For n ≥ 3, the cycle on n vertices is the graph with n vertices, v1 , v2 , . . . , vn
say, with edges v1 v2 , v2 v3 , . . . , vn−1 vn , vn v1 . It is denoted by Cn . We have
e(Cn ) = n.

(d) The path of length n − 1 is the graph with n vertices, v1 , v2 , . . . , vn say, with
edges v1 v2 , v2 v3 , . . . , vn−1 vn . It is denoted by Pn−1 . We have e(Pn−1 ) = n − 1.

(e) The complete bipartite graph on m + n vertices is the graph with m + n


vertices, u1 , u2 , . . . , um , v1 , v2 , . . . , vn say, with edges ui vj for every 1 ≤ i ≤ m
and 1 ≤ j ≤ n. It is denoted by Km,n . The sets {u1 , . . . , um } and {v1 , . . . , vn }
are the classes of Km,n . We have e(Km,n ) = mn.
Figure 3 below shows examples of these graphs.
. .
.................................................................. . .
........... .
..... . . .......... .
........
.... ......... ........ .... .
.............
..... ..........
.
...... . .....
.....
....
... .. ..... ..... ... ..
.... .... ............... .... ....
.... .........
.......
.... ....... ......... ....
.... ...
... ....
.
.
..
. ..
.
......
..
... .. ....... ........ ...
... ....... ... ... ....... ...
............. .... ..... ............
. ............
. ..... ..
.
................................................. . . . .
... .
...................................... . .....
.....
. . . .
..... .... .......

K4 E3 C5 P3 K2,3

Figure 3
Next, we consider some properties of a vertex of a graph.
Definition 12 Let G = (V, E) be a graph, and let v ∈ V be a vertex of G.
(a) The vertex u ∈ V is adjacent to v, or is a neighbour of v, if uv ∈ E.

(b) The set of neighbours of v is the neighbourhood of v. This set is denoted by


Γ(v). Note that this may be finite or infinite.

(c) The number of neighbours of v is the degree of v, and is denoted by d(v). Note
that d(v) = |Γ(v)|. Again, this may be finite or infinite.

(d) If d(v) = 0, then v is an isolated vertex.

(e) The smallest of all the degrees is the minimum degree of G, and the largest is
the maximum degree of G. These are denoted by δ(G) and ∆(G), respectively.
Sometimes, we may want to write ΓG (v) for Γ(v) and dG (v) for d(v), if we want
to emphasize the dependence of neighbourhood and degree on G.
Remark. If δ(G) and ∆(G) are both finite, then obviously, δ(G) ≤ ∆(G). Also,
these quantities are always well-defined, even if the graph G is infinite. It is possible
to have δ(G) = ∞ and/or ∆(G) = ∞.

We have an elementary result about the sum of the degrees of a graph with finite
order.

16
Theorem 12 Let G = (V, E) be a graph, with V finite. Let V = {v1 , v2 , . . . , vn }.
Then,
X n
d(vi ) = 2e(G).
i=1

So in particular, the sum of the degrees of the vertices of G is always even.

Proof. This is another good example illustrating counting objects in two ways.
We count the number of edges of G in two ways. On one hand, this is e(G). On
the other hand, if we consider summing the degrees of G, then each edge has been
counted exactly twice;
Pn an edge uv is counted in the terms d(u) and d(v) of the degree
1
sum. So, e(G) = 2 i=1 d(vi ), and we are done. 

Now, we define some more properties about graphs themselves. We can talk
about a graph containing other graphs.

Definition 13 Let G = (V, E) be a graph.

(a) A graph H = (V 0 , E 0 ) is a subgraph of G if V 0 ⊆ V and E 0 ⊆ E. We write


H ⊆ G.

(b) Let A ⊆ V . Then the subgraph of G induced by A is the subgraph with vertex
set A, and the edges of E whose end-vertices are in A. We write G[A] for this
subgraph.

(c) Let x ∈ V (G). Then the graph G − x is the graph formed by removing x from
G, along with the edges which touch x.

(d) Let e ∈ E(G). Then the graph G − e is the graph (V, E \ {e}). That is, we
just remove the edge e from G.

The above ideas are often useful for proving statements about graphs, when we
want to apply induction on the number of vertices, or on the number of edges (or
some other similar quantity).
Finally, given a graph, it has a ‘complement graph’.

Definition 14 Let G = (V, E) be a graph. The complementary graph of G, or the


complement of G, is the graph G = (V, E), where E is the complement of E in the
set of all possible edges on V .

In other words, to get G from G, we just remove the edges of G, and add back
the edges between vertices of V that were not present in G.

2.2 Bipartite and r-partite Graphs


We shall consider another, rather special, class of graphs. These graphs are a
huge generalisation of the complete bipartite graphs which we have already seen.

17
Definition 15 Let r ∈ N, r ≥ 2. A graph G = (V, E) is an r-partite graph if there
is a partition of V into r parts, V = V1 ∪ V2 ∪ · · · ∪ Vr say, such that for every edge
uv ∈ E, we have u ∈ Vi and v ∈ Vj for some i 6= j. The sets V1 , . . . , Vr are the
classes of G.
If G contains every edge between every two classes, then G is a complete r-partite
graph. In the case where all the classes are finite, G is denoted by Kk1 ,k2 ,...,kr , where
k
P i = |Vi | for every 1 ≤ i ≤ r, so that k1 + · · · + kr = |V |. We have e(Kk1 ,k2 ,...,kr ) =

i<j ki kj .
Finally, a 2-partite graph is more commonly called a bipartite graph.
Note that in an r-partite graph, two vertices in the same class cannot be joined
by an edge.
Figure 4 below shows a 4-partite graph, and the graph K4,2,3 . The classes in each
case are shown in the ovals.

... ....
........ ...
..... . ................
.........
..... ...................................................... ....... ..............
.. ..
.............................. .....
....
........................ ........................
.......

..
... . . . ... ... ..... ..... ......... ........ ........ ........ .....
... .............
.... .... .......... ..... ...... .... ..... .. ............................ ............. ............ ...............
..... ............................................. ... ..... ............... ..... .. ...............................................
..
.. . . . ..
... ... ... ...
. . ...
. . ...... ........................... ....... ............
................. .
. ..... ... ..
.. ......................................
. . .. ..... ...................... ........... ...... ...
.
.... .. . ... ..
... ...
... ... ... ......................... ....................................................................
.
.... ............................... . . ... .... ........... ... .. ...... ... ....................
.... .......... ..... ................ ..... .....
. . .. .............. .. ... ....................................................................... ...........
. . ..........................................
........ ....
..
... ......
...... ........................ .
..... .............. ................
..... ................. ..
..... ................ .. .......................... .. . .. ...
.
... .. .. .......... ... ...... .. .
... ... ................. ............... ...
..... ............................................................................................................................................................................. ...
.... .
. ..
..... .............................. ............. .. ....... ....
.
..
... ............................................................................................................................. ......
..
..................... ....
... .................................................................................................................. .. .......... ...
...
. ................. .......... .
......................... ..
... .........................................................................................................................................................
......
..................
. ....
.
.... .........
...............
.....
..
.... ................................................................................ ..... .....
..... ..
.........
. ....
.......... ..... ....
... .....
..........
.

K4,2,3

Figure 4
In certain problems, r-partite graphs, and in particular, bipartite graphs, are
very useful graphs to consider. Here is an example.

Example 13. A group of 13 mathematicians and 6 physicists met at a


conference. Many handshakes, all of which involved a mathematician and a
physicist, were made. It is known that everyone has made at least one handshake,
that every mathematician made the same number of handshakes, and that the number
of handshakes that every physicist made is a multiple of 3. How many handshakes
did each mathematician make?
Also, describe how the handshakes can be made so that the above scenario can be
realised.
This example can be modelled perfectly by a bipartite graph. Let G be such a
bipartite graph. The classes of G are M and P , where M are the mathematicians
and P are the physicists, so that |M | = 13 and |P | = 6. An edge joins two people
if they shook hands. For some a ∈ N, we have d(x) = a for every x ∈ M ; note
that our task is to find a. Let P = {p1 , . . . , p6 }. For some b1 , . . . , b6 ∈ N, we have
d(pi ) = 3bi for every 1 ≤ i ≤ 6. We count the number of edges of G in two P6 ways.
On one hand, we have e(G) = 13a. On the other hand, we have e(G) = i=1 3bi .

18
So, 13a = 6i=1 3bi , from which it follows that a is a multiple of 3. Now, since a
P

P6 at most 13 times, we have 1 ≤ bi ≤ 4 for each 1 ≤ i ≤ 6.


physicist can shake hands
It follows that 18 ≤ i=1 3bi ≤ 72, so that 18 ≤ 13a ≤ 72. Since a is a multiple of
3, the only possible value of a satisfying 18 ≤ 13a ≤ 72 is a = 3.
It remains to construct a bipartite graph G, with one class M of size 13 and the
other class P of size 6, such that d(x) = 3 for all x ∈ M , and d(y) is a multiple of
3 for every y ∈ P . Such a graph G is shown in Figure 5. Triple lines between two
classes means that we have a complete bipartite subgraph between the classes.

M ........... ......
...........................
.........................
...................................................
......
......
.....................................................
...
.................
..................................
........ ...... ....... ...... ........
........................... .........................
..
.. .....
..
... ........ ... ... ..... . .
.............
.....................
....................
..
........................................

..
........ ..
... .
. ..
.................. ..
.....
......
.. ... ............ ....... ..
...............
............... .. .... ...........................................
. ... ... ...
...............
. ..... ...... .....
...............
............... .... ... .....
. ...... .............. .... .......................
............... .. .. .. . .......... .............. .... .... ....
.. .. .. .. .. ... ............ ................ ... ... ...
............ ... ..........................
P ........ ...
...............
...............
................. .
.
...................
... .......
.....................................
. ... .. .
........ ... ..... ..... ........
...............
.
............... ..... ......................
.....................................
.

Figure 5

2.3 Connectedness
In this section, we consider yet another important property about graphs, the
concept of ‘connectedness’.

Definition 16 A graph G = (V, E) is connected if for every u, v ∈ V , there are


distinct vertices x1 , x2 , . . . , xk ∈ V such that ux1 , x1 x2 , x2 x3 , . . . , xk−1 xk , xk v ∈ E.
That is, every pair of vertices are joined by some path which is a subgraph of G.

In other words, a connected graph is a graph ‘in one piece’.


So, we can think about connectedness in this way. If a graph is not connected,
then it consists of several ‘pieces’. These ‘pieces’ are the components of the graph.
We give a more precise definition.

Definition 17 For a graph G = (V, E), a subgraph H ⊆ G, H = (V 0 , E 0 ), is a


component of G if the following hold.

• H is connected.

• For any u ∈ V 0 , there is no vertex v ∈ V \ V 0 such that uv ∈ E.

• Whenever x, y ∈ V 0 and xy ∈ E, we must have xy ∈ E 0 .

So in some sense, a component H of a graph G is a ‘maximal, connected


subgraph’; ‘maximal’ in the sense that we cannot extend H further to a connected
subgraph with a larger order or a larger size.
There are plenty of results in graph theory which concern connectedness. Here,
we present one of them.

19
Theorem 13 Let G = (V, E) be a graph, with V finite. Then either G or G is
connected.
Proof. If G is connected, we are done, so suppose not. Then G is made up
of at least two components, say, H1 , H2 , . . . , Hr , where r ≥ 2. Consider any two
components Hi and Hj . For any u ∈ V (Hi ) and v ∈ V (Hj ), we have uv ∈ E(G), so
that any two vertices of V (Hi ) ∪ V (Hj ) is connected by a path of length at most 2.
Since this holds for every i 6= j, we have, in fact, that every pair of vertices in G is
connected by a path of length at most 2. Hence G is connected. 

2.4 Mantel’s Theorem and Turán’s Theorem


In this section, we shall talk about two theorems (one being a special case of
the other) which, although are fairly advanced and are results one is not expected
to know, they are fairly easy to remember. There has been some olympiad level
problems where they can be extremely useful.
Consider the following question. “Given n ∈ N, n ≥ 2, how many edges on n
vertices can a graph G have, so that G does not contain K3 as a subgraph?”
In other words, if G is ‘triangle free’, then how large can its size be?
It seems that complete bipartite graphs are good candidates for G, since they do
not contain triangles. If we just restrict our consideration to these graphs, then we
can quite easily see (say, by the AM-GM inequality) that we should make the class
sizes as equal as possible, namely, one class with size b n2 c and the other with size
d n2 e. Such a graph does indeed attain the maximum number of edges for G (out of
all graphs on n vertices which do not contain a K3 ), and moreover, this graph is
the only graph which attains the maximum. This is a theorem of Mantel.
Theorem 14 (Mantel’s Theorem) Let n ∈ N, n ≥ 2. Let G be a graph on
n vertices. If G does not contain K3 as a subgraph, then e(G) ≤ e(Kr,s ), where
r = b n2 c and s = d n2 e. So, e(G) ≤ b n2 cd n2 e.
Moreover, equality holds if and only if G = Kr,s .
For example, if G is a graph on 7 vertices and is triangle free, then e(G) ≤ 3×4 =
12, with equality if and only if G = K3,4 .

Proof of Theorem 14. If xy is an edge of G, then they may not have a common
neighbour (otherwise we will get a triangle). So d(x) + d(y) ≤ n. Summing over all
edges, we have X
ne(G) ≥ (d(x) + d(y))
xy∈E(G)

In the sum on the right, we want to turn it into a sum over the vertices of G.
For each vertex v ∈ V (G),
P the number of occurrences of the term d(v) is d(v). So
2
the sum is the same as v∈V (G) d(v) . So we have
X
ne(G) ≥ d(v)2
v∈V (G)

20
Now,
P by the Cauchy-Schwarz Pinequality, if x1 , . . . , xn are non-negative reals,
P we
have ( x2i )(12 + · · · + 12 ) ≥ ( xi )2 , where we have n 1s. So xi ≥ n1 ( xi )2 .
P 2
Using Theorem 12 as well, we have
 X 2
X
2 1 1
ne(G) ≥ d(v) ≥ d(v) = (2e(G))2 ,
n n
v∈V (G) v∈V (G)

n2
which gives e(G) ≤ 4
. Since e(G) is integral, we have, in fact,
j n2 k
e(G) ≤ .
4
2
It is easy to verify that b n4 c = b n2 cd n2 e, by considering the parity of n. Indeed,
2 2
they are both equal to n4 is n is even, and n 4−1 if n is odd. So we have the first part
of the theorem.
Now we prove the second part. We first consider the case when n is even. Assume
2
that we have equality in the theorem, so that e(G) = n4 . We must have equality in
Cauchy-Schwarz, so that all the degrees are equal. We must also have d(x)+d(y) = n
for every distinct x, y ∈ V (G), so we must have d(v) = n2 for every v ∈ V (G). Now,
take an edge uw ∈ E(G). We have |Γ(u)| = |Γ(w)| = n2 , Γ(u) ∩ Γ(w) = ∅ (the empty
set), and Γ(u) ∪ Γ(w) = V (G). Now, every z ∈ Γ(u) \ {w} can only be adjacent
to vertices of Γ(w), otherwise a K3 will be created. Since d(z) = n2 = |Γ(w)|, it
follows that Γ(z) = Γ(w). Similarly, Γ(z 0 ) = Γ(u) if z 0 ∈ Γ(w) \ {u}. It follows that
G = Kn/2,n/2 , as required (with classes Γ(u) and Γ(w)).
2
Finally, let n ≥ 3 be odd, say, n = 2m + 1. This time, we have e(G) = n 4−1 =
m2 + m. So, v d(v) = 2e(G) = 2m2 + 2m. So, there exists a vertex v ∈ V (G) with
P

j 2m2 + 2m k j m k
d(v) ≤ = m+ = m,
2m + 1 2m + 1
since 0 < 2m+1m
< 1. Now, let G0 = G − v. Then, e(G0 ) ≥ m2 + m − m = m2 .
Moreover, |V (G0 )| = 2m, and G0 also does not contain a K3 . So by the first part of
2
the theorem, we have e(G0 ) ≤ b (2m) 4
c = m2 . Hence, e(G0 ) = m2 , so by the result
0
for even n, we must have G = Km,m . Also, we now know that dG (v) = m as well
(since e(G0 ) = m2 ), so, replacing v back in to reform G, v must be joined to all the
vertices of one of the classes of G0 . It follows that G = Km,m+1 , which is exactly
what we want. 

Can we generalise Mantel’s Theorem? Yes, we can, in the following way. We can
ask the more general question with Kk in place of K3 , for any prescribed k ∈ N.
That is, we want to know: “Given n, k ∈ N, where k ≥ 3, how many edges on n
vertices can a graph G have, so that G does not contain a Kk as a subgraph?”
In this case, it is conceivable that complete, (k − 1)-partite graphs are good
candidates, since they do not contain Kk as a subgraph. Moreover, the sizes of the
classes should be ‘as equal as possible’.
So, we introduce a little notation.

21
Definition 18 For n, r ∈ N, where n, r ≥ 2, the r-partite Turán graph of order n
is the complete r-partite graph on n vertices, where each class has size b nr c or d nr e.
It is denoted by Tr (n).

Remark.

(a) If r ≥ n, then Tr (n) is the complete graph Kn .

(b) Note that, given n, r ∈ N, where 2 ≤ r ≤ n, there is only one way to partition
n into r terms, each of which is either b nr c or d nr e. Namely, we can think of the
formation of the unique partition as follows. Deal a pack of n cards to r people
in a cyclic manner. The distribution of the cards is then clearly unique, (up
to permutation), and the first s people get d nr e cards, with the others getting
b nr c cards, where s is the remainder when n is divided by r.

So, from (a) and (b) above, the graph Tr (n) is always well-defined.
Also, note that δ(Tr (n)) = n − d nr e, and ∆(Tr (n)) = n − b nr c.

Figure 6 below shows the graph T3 (8).

...
.............
............................................
...... .......... ............ ........... ......
........................................................................
.... ..... .. ........ ... .
.. .. ......................... .............
.. . ... .. ........... ......... ...... ...
... ..............................................................................
.
.. ..... ...... ... ..... ...............................................
.. ..... ...... ...
........................ ..... ... ............ .. .............. ............

.
. ..
................................ ...............
... .. ..... ... .. ............ ...
... ... ............ ... ....... ..
. .
.......... ....
.... ...................................................................................................................................................... ....

. . .
.. .
... . . . .
... ....... ............................................................................ . . ..
... ............................................................................................................................................... ...
... ..
.... ...................................................... ..................... .........
.... ... ........................................................................... .....
.............. . .... .......
..........
...

T3 (8)

Figure 6

Now we can state Turán’s Theorem, which generalises Mantel’s Theorem, and it
says that these graphs Tr (n) are the best graphs.

Theorem 15 (Turán’s Theorem) Let n, k ∈ N, where n ≥ 2 and k ≥ 3. Let


G be a graph on n vertices. If G does not contain Kk as a subgraph, then e(G) ≤
e(Tk−1 (n)).
Moreover, equality holds if and only if G = Tk−1 (n).

This result is one of the most famous results in graph theory. It has many proofs.
However, all the known proofs are fairly technical. As a result, we will leave one of
these proofs as an exercise.
We will have at least one problem at the end where either Mantel’s Theorem or
Turán’s Theorem can be used.

2.5 Graph Ramsey Theory

22
The starting point of Ramsey theory is usually the following problem.

Example 14. Among a party of at least six people, there are either three people
who all know each other, or there are three people, none of whom knows the other
two (assuming that ‘knowing’ is a symmetric relationship).
Moreover, the assertion is false for fewer than six people. That is, ‘six’ is the
minimum integer for which this assertion holds.
To prove the first part, it suffices to prove it for six people, since if we have more
than six people, we can just take any subset of six people and consider whether they
know each other or not.
So, consider six people. Represent each of them by a vertex in a graph, and join
two vertices with a blue edge if the two people they represent know each other, and
by a red edge if the two people do not know each other. So, we have a K6 , whose
edges are coloured either blue or red. We are done if we can show that, no matter
which colouring we use, there is always either a blue triangle or a red triangle.
So, take any such colouring. Choose any vertex x. Since x has five neighbours,
by the pigeonhole principle, some three neighbours y1 , y2 , y3 are joined to x with the
same colour. Suppose that this colour is blue. If some yi yj is blue, then x, yi and
yj form a blue triangle. Otherwise, y1 y2 , y1 y3 and y2 y3 are all red, so y1 , y2 and y3
form a red triangle. A similar argument holds if xy1 , xy2 and xy3 are all red; we just
switch the colours and use the same argument. Since our colouring was arbitrary,
we have now proved the first part.
For the second part, it suffices to show that there is a colouring of the edges of
K5 , using blue and red, so that there is no triangle in one colour. For, if we have
such a colouring of the edges of K5 , then for each n ≤ 4, we can get a colouring of
the edges of Kn with no triangle in one colour, simply by taking any subset of n
vertices, along with the corresponding colours of the edges.
So, for K5 itself, we take the colouring as shown in Figure 7. The solid lines
are the blue edges, and the dotted lines are the red edges. We see that there is no
triangle in either colour in this colouring.

.
. .. . .
. . ............ . . .
.. .. .. . .
. . . ..... ....... . . .
. .. .......
..
.. ........... .... ..
.. ..
...
... .......... ..
..
.. .
.
...........................................................................................................................

. . ... ...... .
.. .. . .. . .
...... ......
.. .... ........................ ....... ...
..
.. ... ........ ......... .... .. . .. .. .
................ ...... ... .
. ...... . . . . . . . . . . ..........
.... .
.
Figure 7

There is a bit of terminology which is useful to introduce.

Definition 19 Let G be a graph.

23
(a) For r ∈ N, an (edge) r-colouring of G is a function f : E(G) → {1, 2, . . . , r}.
In other words, such a function is just a colouring of the edges of G, using at
most r colours.

(b) If G is given an (edge) r-colouring, then a subgraph is monochromatic if all


of its edges have the same colour.
We usually ignore the word ‘edge’, if it is clear that we are colouring the edges
(we may get this confused with colourings of the vertices). We will drop this word
for the rest of this section.
Remark. Note that an r-colouring is also an s-colouring for all s ≥ r. For
example, if we have a 3-colouring of a graph, we may not necessarily need to have
all three colours present.

So in our example above, we have proved that, in any 2-colouring of Kn , where


n ≥ 6, there is always a monochromatic triangle. Also, the assertion is false if n ≤ 5.
In loose terms, “Ramsey theory” is the area of combinatorics which deals with
questions of the following type: “Within a sufficiently large region of disorder, there
is always a small region of order”. In our example, our “sufficiently large region of
disorder” is a complete graph with at least six vertices, whose edges are coloured
(disorderly) by two colours. Our “small region of order” is a monochromatic trian-
gle.
We may ask questions of this type when we have colourings of other objects as
well. When we are colouring graphs, we are talking about “Graph Ramsey theory”.
We return and continue to discuss Example 14. We can extend it by asking the
following question. “Given integers s, t ≥ 2, what is the least integer n such that,
whenever we have a 2-colouring of the edges of Kn , using blue and red say, we can
always find a blue Ks or a red Kt ?”

Definition 20 For the above question, the least such integer, if it exists, is denoted
by R(s, t). It is called the Ramsey number of Kr versus Ks .

So, Example 14 shows that R(3, 3) = 6.


Can we say more about R(s, t)? Obviously, R(s, t) = R(t, s). Also, it is easy to
see that R(s, 2) = s for all s ≥ 2, so these Ramsey numbers are exactly determined.
What about for s, t ≥ 3? Apart from s = t = 3, only a few more of these Ramsey
numbers are known exactly. It is not hard to show that R(3, 4) = 9, R(3, 5) = 14 and
R(4, 4) = 18; we leave the determination of these Ramsey numbers as an exercise.
The other known Ramsey numbers are R(3, 6) = 18, R(3, 7) = 23, R(3, 8) = 28,
R(3, 9) = 36, and R(4, 5) = 25.
Also, going back to Definition 20, is R(s, t) actually well-defined for every
s, t ≥ 2? Obviously, R(s, t) grows rapidly when s and t increase. Can it be the
case that there are s and t so large that, no matter how large an integer n is, any
2-colouring of the edges of Kn using blue and red can never have a blue Ks or a

24
red Kt ? The answer is “no”; it turns out that R(s, t) does have an upper bound,
for every s and t. We have R(s, t) ≤ s+t−2s−1
, and we will leave a proof of this as
another exercise.
To round off this section, we end by mentioning some more extensions. We may,
of course, consider using r colours (where r ≥ 3) instead of 2 colours. The resulting
Ramsey numbers are also known to exist, but almost nothing is known about their
exact values. We may also consider finding monochromatic copies of some other
graphs, rather than complete subgraphs. For example, R(P3 , C4 ) denotes the least
integer n such that, whenever Kn is 2-coloured in blue and red, there is either a blue
P3 or a red C4 .

3. Colouring Proofs
We shall have a very short section here about another very useful technique for
combinatorics problems: the idea of colouring structures.

3.1 Colouring Proofs


There are many combinatorics problems which can be tackled by a “colouring
proof”. What this essentially means is that we just take a partition of a certain
structure into finitely many parts (each of which is assigned with a colour), and try
to make an argument from such a partition.
Let us look at an example.

Example 15. Suppose that we have a standard 8×8 chessboard. It is well-known


that, if we have a knight, and place it on any square on the board, then it is possible
to visit every square exactly once, and then return to the starting square.
Can we say the same thing for a 9 × 9 chessboard?
(As usual, a knight moves from one square to another either by moving two
squares left or right, followed by one square up or down, or by moving one square
left or right, followed by two squares up or down.
We claim that the answer is “no”. Consider colouring the squares black and
white, in the usual chequered manner. Note that, when the knight moves, it goes
from a square of a particular colour to a square of the opposite colour. Now, if the
task was possible, then the knight will have made 81 moves. Since 81 is odd, this
means that, after 81 moves, the knight will land on a square whose colour is different
from that of the starting square. So the final square cannot be the starting square,
a contradiction. Hence the task is impossible.

This idea of colouring a structure is extremely useful, and often, it is not easy
to notice in a certain problem that the idea can be applied. So one has to be very
wary about the possibility of using a colouring to tackle a problem.

4. Problems

25
We divide the problems into two parts. The first set consists of a few exercises,
which complete the missing proofs in the notes. The second set consists of the
challenges.

4.1 Exercises
1. Use the following guidelines to prove Turán’s Theorem. (This proof, when
k = 3, does not coincide with the proof of Theorem 14. Hence it is another
proof of Mantel’s Theorem.)
Let n, k ∈ N with n ≥ 2 and k ≥ 3. Suppose that G is a graph on n vertices
which does not contain Kk . We shall prove Theorem 15 by induction on n.
The base case n = 2 is trivial (why?). So, assume that n ≥ 3.

• Assume firstly that e(G) = e(Tk−1 (n)). We claim that G = Tk−1 (n).
– Explain why we have δ(G) ≤ δ(Tk−1 (n)).
– Then, show that there exists a vertex x ∈ V (G) such that e(G − x) ≥
e(Tk−1 (n)). Hence, considering induction on n, what can you say
about G − x?
– Now, replace x back. Where in G − x ( = ?) must x be joined to?
Deduce that G = Tk−1 (n).
• Now, assume that e(G) > e(Tk−1 (n)). By considering the removal of a
suitable number of edges from G, explain why we cannot have this case.

Finally, explain why this is sufficient for a proof of the theorem.

2. Prove that R(s, t) ≤ R(s − 1, t) + R(s, t − 1) for s, t > 2 (Hint: Mimic the
argument used in Example 14).
Deduce that  
s+t−2
R(s, t) ≤ ,
s−1
so that the Ramsey numbers R(s, t) are actually well-defined (Hint: Use
induction).

3. Prove that R(3, 4) = 9, R(3, 5) = 14, and R(4, 4) = 18.

4.2. Olympiad Style Problems


Here are the more challenging problems. Almost every problem here has a fairly
short solution. Problems 1 to 14 are fairly standard, while Problems 15 to 18
(marked with a +) are a little bit more trickier.

1. Suppose that we have a standard 8×8 chessboard, and we remove two opposite
corners. Is it possible to cover the mutated chessboard with 31 1×2 dominoes,
where each domino covers exactly two abutting squares of the chessboard?

26
2. We have a 4 × 7 chessboard. Prove that, no matter how we colour each square
with black or white, we can always find four distinct squares with the same
colour which form the four corners of a rectangle.
Is the same result true for a 4 × 6 chessboard?

3. Let A be a subset of {1, 2, . . . , 99, 100} with 10 elements. Prove that there are
two disjoint, non-empty subsets of A such that the sum of the elements of one
set equals the sum of the elements of the other set.

4. Consider the 2-dimensional integer lattice (that is, the collection of all points
in the plane where each point has integer coordinates). Show that, no matter
how we choose five distinct lattice points, some two of these points have the
property that their midpoint is also a lattice point.
bn/2c   
X n n − k n−2k
5. Simplify 2 .
k=0
k k

6. How many permutations are there of the letters

A, A, A, A, A, B, B, B, B, B, C, C, C, C, C

such that, in each permutation, every letter is adjacent to an identical letter?

7. Let x1 , x2 , . . . , x101 be a sequence of 101 real numbers. Prove that we can find
11 indices 1 ≤ i1 < i2 < · · · < i11 ≤ 101 such that xi1 , xi2 , . . . , xi11 is either an
increasing sequence or a decreasing sequence.
(A sequence of real numbers y1 , y2 , y3 , . . . is increasing if yi ≤ yj whenever
i < j, and decreasing if yi ≥ yj whenever i < j.)

8. 20 points on the circumference of a circle are evenly spaced. They are joined
up in disjoint pairs by 10 chords. Prove that, no matter how these 10 chords
are drawn, some two of them have the same length.

9. An unordered pair of 9-digit numbers, each of which consists of all the


digits from 1 to 9, is called friendly if their sum is 987654321. For example,
{123456789, 864197532} is friendly.
Prove that the number of these friendly unordered pairs is odd.

10. In a certain country, there are at least 101 towns. A railway network connects
the towns, and it satisfies the following properties.

• The capital is connected directly to 100 other towns.


• Every town, apart from the capital, is connected directly to 10 other
towns.

27
• From any town, it is possible to get to any other town, possibly via some
other towns.

Prove that it is possible to close down 50 of the lines connected to the capital,
and preserve the property that, from any town, we can get to any other town
(again, possibly via some other towns).
(Note that “directly connected” is a symmetric relationship. That is, if we
can get directly from town A to town B, then we can get directly from town
B to town A.)

11. The circumference of a circle has length 3. We mark 21 points on the circum-
ference. Prove that, among the arcs which join two of the points, we can find
at least 100 of them with length less than 1.
Can ‘100’ be replaced by a larger integer?

12. Suppose that n is a positive integer not divisible by 2 or 5. Show that there
exists a multiple of n which consists entirely of 1s.

13. 11 students tackled a set of 18 problems. Each problem was solved by at least
7 students. Prove that we can find three students who, between them, solved
all the problems.

14. Suppose that we have 100 fruits, 10 each of 10 types. They are arranged in a
10 × 10 formation. Prove that there is a column or a row which contains at
least four different types of fruit.

+15. Is it possible to have a subset A ⊂ {1, 2, 3, . . . , 3000} with |A| = 2000, and
with the property that whenever x ∈ A, we have 2x 6∈ A?

+16. Let n be a positive integer. Prove that the number of partitions of n into
distinct terms is equal to the number of partitions of n into terms which are
all odd. For example, for n = 5, the partitions satisfying the former are 1 + 4,
2 + 3 and 5, while the partitions satisfying the latter are 1 + 1 + 1 + 1 + 1,
1 + 1 + 3 and 5.

+17. A rectangular board has m rows and n columns of unit squares, where m < n.
We mark some of the squares so that each column has at least one marked
square. Prove that we can find a marked square such that the row containing
it has more marked squares than the column containing it.

+18. Prove that in a group of 50 people, we can find two people who commonly know
an even number (possibly zero) of the other 48 people. (Here, “knowing” is a
symmetric relationship. That is, if A knows B, then B knows A.)

28

You might also like