Introduction to Combinatorics

University of Toronto Scarborough

Lecture Notes
Stefanos Aretakis

July 4, 2018

Contents
1 Introduction 2
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The prisoners’ Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Solution to the prisoners’ Problem . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Combinatorial Principles: Contradiction, Reduction and Induction . . . . . . . 5

2 The Pigeonhole Principle 5

2.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Systematic approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Solved Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Ramsey Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Erdös–Szekeres Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Diophantine Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.7 Practice Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 The Principle of Extremals 19

3.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Solved Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Practice Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 The Principle of Invariants 23

4.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Solved Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.3 Semi-invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.4 Practice Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5 Permutations and Combinations 27

5.1 Additive and Multiplicative Principle . . . . . . . . . . . . . . . . . . . . . . . . 27
5.2 Solved Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.3 Permutations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.4 Combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

1
 5.5 Identities of the binomial coefficients . . . . . . . . . . . . . . . . . . . . . . . . 31
5.6 Practice Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

6 Combinations with Repetition 33

6.1 Solutions to linear equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2 The Path Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.3 Practice Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

7 Inclusion–Exclusion principle 37
7.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
7.2 Solved Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
7.3 Practice Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

8 Recurrence Relations 40
8.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
8.2 Solved Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.3 Practice Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

9 Generating Functions 46
9.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9.2 Solved Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
9.3 Applications in recurrence relations . . . . . . . . . . . . . . . . . . . . . . . . . 56
9.4 Practice Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

10 Partitions of Natural Numbers 59

10.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
10.2 Partitions with restrictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.3 Euler’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

1 Introduction
1.1 Introduction
Combinatotics is about...
... counting without really counting all possible cases one by one.

More broadly:
Combinatorics is about...
... derivining properties of structures satisfying given conditions without analyzing each
and every possible case separately.
Analyzing, deriving and counting common properties of structures satisfying given con-
ditions can in principle be quite challenging and require a non trivial amount of focus and
concentration. This is the challenging part of our course.
The fun part of our course is that the structures we will be considering are very elementary
(no involved defitions, symbols etc.). Hence, for most of the time we will not be obstracted
by general and abstract nonsense.
In other words, we should view our course as a fun and challenging way to learn how to
learn. (Question: what is the Greek word for ”learning how to learn”: Answer: Mathematics!).

2
1.2 The prisoners’ Problem
Let’s consider the so-called ”prisoners’ problem” as a way to see a few Combinatorial principles
in action:
We consider an island full of male prisoners such that the following conditions hold:
1. there are 100 prisoners,
2. all have green eyes,
3. they can all see that all other prisoners have green eyes,
4. they, however, do not know that they themselves have green eyes,
5. no communication is allowed between the prisoners,
6. right of the prisoners: any prisoner can go at any midnight to the guards and claim that
they have green eyes. If the prisoner has indeed green eyes, then the prisoner is freed
otherwise the prisoner is killed.
The prisoners are aware of their right but unfortunately they do not know if they have
green eyes and also there is no way of communicating with any other prisoner. Hence, as
is, there is no way the prisoners can find out that they have green eyes and hence escape by
making use of their only right (condition 6 above).
However, the situation changes if we add one more condition. Specifically, the prisoners
are all gathered together and are told the following statement:
• (additional condition) at least one of you have has green eyes...
Even though at first glance the above statement seems useless, since everyone can see that
all the remaining 99 prisoners have green eyes1 , the prisoners manage to realize that they
themselves have green eyes and hence make use of their right and escape.
How is this possible? How did the additional condition above shake the dynamics of our
problem?

1.3 Solution to the prisoners’ Problem

For simplicity, let’s assume that there are less than 100 prisoners to make the problem more
manageable (REDUCTION PRINCIPLE). Let’s assume that we have only 2 prisoners: P1 , P2 .
Let’s consider the point of view of one of them, for example P1 :
• P1 ’s reasoning on the first day: If I have green eyes, then I can make use of my right and
escape. hence, I would like somehow to prove to myself that I do have green eyes. How
can I do this? What if I assume that I do not have green eyes and reach a conclusion
that is obviously wrong (CONTRADICTION ASSUMPTION)? Then my assumption
that I do not have green eyes must have been wrong, and hence I will be able to safely
conclude that I have green eyes and make use of my right.
If I do not have green eyes then P2 must be able to see that I do not have green eyes.
However, since we were told that at least one of us has green eyes (additional condition)
then P2 must be able to conclude that it is him who has green eyes and hence make use
of his right the first midnight.
1
and everyone can see that everyone else sees at least someone–in fact at least 98 prisoners– with green eyes

3
 However,P2 does not make use of his right to escape since he cannot conclude if he has
green eyes, since he see that P1 has green eyes and therefore cannot conclude, using the
additional condition, that he is the one who has green eyes. Therefore, at the beginning of
the second day, P1 makes the following realization:

• P1 ’s reasoning on the second day: Since P2 is still here it must mean that he is not
certain if he has green eyes or not which means that my assumption that I do not have
green eyes (which would mean that P2 would realize that he has green eyes) is wrong.
Therefore, I must have green eyes and therefore P2 must be watching me.

Therefore, P1 realizes on the second day that he has green eyes and makes use of his right
the second midnight and escapes.
Exactly, the same reasoning applies to P2 (symmetrically) and so P2 also escapes the
second night. Therefore, they both realize that they have green eyes on the second day and
make use of their right the same night.
This concludes the solution to the prisoners’ problem in the case of n = 2 prisoners P1 , P2 .
Let’s consider next the case of n = 3 prisoners.
Let’s consider the point of view of one of them, for example P1 :

• P1 ’s reasoning on the first day: If I do not have green eyes then P2 , P3 are not looking
at me and and are only looking at each other. Hence, the problem should be reduced to
the case of two prisoners (INDUCTION PRINCIPLE) and thefore P2 , P3 should both
escape on the second day.

However,P2 , P3 do not make use of their right to escape since they cannot conclude if they
have green eyes, since they are also observing P1 since they can see that he has green eyes.

• P1 ’s reasoning on the third day: Since P2 , P3 are still here it must mean that they are
not certain if they have green eyes or not which means that my assumption that I do
not have green eyes (which would mean that P2 , P3 would realize that they have green
eyes on the second day) is wrong. Therefore, I must have green eyes.

Therefore, P1 realizes on the third day that he has green eyes and makes use of his right
the second midnight and escapes.
Exactly, the same reasoning applies to P2 , P3 (symmetrically) and so P2 and P3 also escape
the third night. Therefore, they all realize that they have green eyes on the third day and
make use of their right the same night.
This concludes the solution to the prisoners’ problem in the case of n = 3 prisoners
P1 , P2 , P3 .
The general problem is solved similarly, or more precisely inductively. Each prisoners
assumes that he does not have green eyes and therefore the problem is reduced to the case
of 99 prisoners with by induction (INDUCTION PRINCIPLE) should terminate on the 99th
day. But this does not happen, and hence every prisoner realizes on the 100th day that they
have green eyes and all escape on the same night.

• Question: How can we change the additional condition, slightly, so that all 100 prisoners
escape on the second day, instead of the 100th day?

Please also see the following link for an animated explanation of the problem:
https://www.youtube.com/watch?v=98TQv5IAtY8

4
1.4 Combinatorial Principles: Contradiction, Reduction and Induction
The following three principles were used and were of fundamental importance in solving the
prisoners’ problem:

• Contradiction: Assume that what we want to show does not hold and use this extra
assumption to reach an obviously wrong conclusion (such as 0 = 1). This is one of the
most powerful techniques in mathematics, precisely because it allows for the introduction
of one additional assumption (the contradiction assumption) which can be freely used.
A nice application of contradiction is the standard proof for the fact that there are
infinitely many prime numbers.

• Reduction Principle: Address the problem by first lowering its complexity, for example
by lowering the number of variables in the problem.

• Induction Principle: Solve the general problem by using the solution for the reduced
problem.

2 The Pigeonhole Principle

2.1 Theory
We next study the very simple but surprisingly important and with far reaching applications
Pigeonhole Principle.

• If n + 1 pigeons are put in n pigeonholes then there is a pigeonhole with at least 2

pigeons.

Consider for example the case where n = 4 with 5 pigeons in 4 pigeonholes. Clearly, there
are many different ways to put 5 pigeons in 4 pigeonhole and we will certainly not consider
each of these cases separately. We need to show that all of these cases have a common property,
namely that there is always a pigeonhole that contains at least two pigeons. In order to prove
this, without going through all possible cases separately, we need to argue by contradiction.
Specifically we assume that every pigeonhole contains at most one pigeon. Since we have 4
pigeonholes and each of them contains at most one pigeon, we must have at most 4 pigeons
in total. This is however contradiction since we have 5 pigeons in total. The proof for the
general case with n pigeonholes is done similarly (exercise).

If we increase the number of pigeons while keeping the number of pigeonholes the same,
then we can guarantee the existence of a pigeonhole with even more pigeons:

• If m · n + 1 pigeons are put in n pigeonholes then there is a pigeonhole with at least

m + 1 pigeons.

Proof: Exercise.

Problem 1. Among 13 people there are always 2 who were born on the same month.
Among 37 people there are always 4 that were born on the same month.

Proof. Indeed, 13 = 12 + 1 and 37 = 12 · 3 + 1.

5
 The above problem has 13 variables (people) and we want to show that a small subset
of them (consisting of 2 variables) has much greater structure (namely that they were born
on the same month) compared to the orginal set of 13 variables (for which we can in general
make no specific statement about their date of birth).

2.2 Systematic approach

Let v be the number of variables in our problem (e.g. people). If we want to show that at
least l of them has a property P then we set:

• p = the number of all possible cases that the variables might satisfy (e.g. p = 12 in
Problem 1 above since the DOB can be in any of the 12 months),
• v = m · p + 1,
• l = m + 1.

Then, by pigeonhole principle, we have that, if the initial v variables (pigeons) satisfy any of
the p conditions (pigeonholes), then there are at least l variables (pigeons) which satisfy one
condition in common (ie. in one pigeonhole). Note that
v−1
p= .
l−1
Hence, we need to make sure that there are in total p = v−1
l−1 different cases (properties) for
the v variables.
Methodology: If we are given n classes of objects such that
• objects belonging to the same class are different, but,
• objects belonging to different classes might coincide,
and we want to show that there is the same object in at least l different classes then we
1. we count all objects from all classes, and say we compute that this number is v, and
2. we count all possible values that each object can take, and say that this number is m.
If
v = m · (l − 1) + 1,
then we have m · (l − 1) + 1 objects (pigeons) taking at most m values (pigeonholes) and
hence there must be at least l objects in all our classes which take the same value (i.e. they
coincide). Hence, one object will belong to at least l classes simultaneously.

2.3 Solved Problems

We will next study problems and applications of the Pigeonhole Principle.

Problem 1. Consider the set

A = {1, 2, 3, · · · , 18}
and randomly choose four mutually distinct numbers x1 , x2 , x3 , x4 ∈ A. Prove that there are
always two xi , xj such that
|xi − xj | ≤ 5.

6
Proof. Of course, in general the difference between two numbers in A is at most 18-1=17. We
need to show that if we have four numbers in the set A then the difference of two of them is
at most 5 (which is greatly smaller than the general bound 17).
Following our systematic approach, we have v = 4 variables x1 , x2 , x3 , x4 and we want to
show that at least l = 2 of them satisfy a more refined property. Hence we need to show that
4−1
our variables satisfy one of p = 2−1 = 3 properties. This motivates us to decompose the set
A in three sets as follows

A = {1, 2, 3, 4, 5, 6} ∪ {7, 8, 9, 10, 11, 12} ∪ {13, 14, 15, 16, 17, 18} .

Since we have 4 variables in A, each of our variables must lie in one of the above three sets.
Therefore, by pigeonhole principle, at least 2 of our variables lie in the same set! But the
difference of any two numbers belonging to the same set is at most 5. This finishes the
proof.

Remarks:

1. Informally speaking, the pigeonhole principle is used when we want to show that in a
given set of variables there is a smaller subset of variables which satisfy a more refined
property and hence has more structure than the original set of all variables. This is
useful when one wants to partition a given the set of variables into smaller sets such
that at least one of them has more structure (divide and conquer principle).

2. The pigeonhole principle is used in order to show only the existence of variables satisfying
a given (refined) property. There is no explicit construction of these variables. No
algorithm is provided on how to get these variables.

Problem 2. Among n consecutive natural numbers there is always one of them which is
divisible by n.

Proof. The special feature of this problem is that given our n variables we want to show that
there is at least one (and not two or more as is typical in pigeonhole principle) which satisfies
an extra refined property.
The trick is to consider the remainders of these n numbers when divided by n. We want to
show that one of these remainders is zero. We argue by contradiction. Assume that they all
give remainders different from 0. Then there are n − 1 possibilities for the remainders to be:
1, 2, · · · , n − 1. But we have n numbers, and so we have n remainders, each of which can only
take one of n − 1 values. Hence, by the pigeonhole principle, there are two numbers which
leave the same remainder when divided by n. But then this implies that their difference is
divisible by n (exercise: why?). This however is impossible, since the difference of any two
number in a any set of n consecutive numbers is always between 1 and n − 1 and hence not
divisible by n. This is contradiction and it shows that the desired result indeed holds.

Problem 3. We are given 33 people with the following property among any 9 of them
there are always 2 with the same height. Show that there are at least 5 people in the original
set of 33 people with exactly the same height.

7
Proof. Following our systematic approach, we have v = 33 variables and we want to show
that at least l = 5 of them satisfy a refined property. Then we need to show that the 33
people satisfy at most
33 − 1
p= =8
5−1
conditions.
In other words, in order to show that there are at least 5 people with the same height, we
need to show that there are 8 possible heights for the original 33 people. This is however true
since in view of the assumption of the problem, if we had 9 possible heights then picking one
person with each of these heights would yield a group of 9 people with no two people sharing
the same height.

Remark: Why is the following reasoning wrong: Consider 9 people out of these 33 people.
Two of them have the same height. We are left with 33-9=24 people. Choose 9 people out of
these 33 people. By assumption, two of them have the same height. We have two pairs so far
and we are left with 24-9=15 people. Choose 9 people out of these 15 people. Two of them
have the same height. Then we have 3 pairs of people with the same height. This gives us
6 people with the same height, which in particular gives us 5 people with the same height.
What is wrong?

Problem 4. Consider the real numbers a1 , a2 , · · · , an such that

0 ≤ ai ≤ n

for all i = 1, 2, · · · , n. Show that there are two distinct ai , aj such that
n
|ai − aj | ≤ .
n−1
Proof. We divide the interval [0, n] in n − 1 equal sub-intervals with end points:

n 2n 3n (n − 2)n
0, , , , ..., , n.
n−1 n−1 n−1 n−1
n
The size of each of these intervals is exactly n−1 . By pigeonhole principle, since we have
n numbers in n − 1 sub-intervals, there must be two numbers ai , aj which lie in the same
sub-interval. The difference of these numbers must be bounded by the size of the sub-interval
n
which is equal to n−1 . Hence this two numbers satisfy the desired property.

Problem 5. Consider six mutually distinct numbers xi ∈ {1, 2, 3, 4, 5, 6, 7, 8, 9} , i =

1, 2, 3, 4, 5, 6. Show that there are four xi1 , xi2 , xi3 , xi4 of these numbers such that

xi1 + xi2 = xi3 + xi4 .

Proof. We want to show a refined property for the sums of all pairs, namely that there are
two such sums which coincide in value. We therefore want to apply pigeonhole principle for
the sums of all pairs.
We have in total 15 possible sums that we can form. For example

x1 + x2 ,

8
 x1 + x3 ,
x1 + x4 ,
x1 + x5 ,
x1 + x6 ,
x2 + x3 ,
are 6 of these 15 possible sums. The values that these 15 sums can take are between 3 and
17 (why?). There are exactly 15 numbers between 3 and 17. Hence we have 15 sums taking
15 possible values. We, therefore, cannot yet apply the pigeonhole principle. For this reason,
we consider the following two cases:
Case I: The 15 sums take in fact at most 14 values. Then by pigeonhole principle two of
the sum must take the same value and hence must be equal.
Case II: The 15 sums take exaclty 15 values. Then our sums must take all values between
3 and 17. Hence, there is a pair (a, b) whose sum a + b = 3 and there is a pair (c, d) whose
sum is c + d = 17. But then we must have

a = 1, b = 2, c = 8, d = 9

and hence, obviously,

a+c=b+d
which shows the desired result.

Problem 6. Show that in any collection of n + 1 numbers from the set {1, 2, 3, · · · , 2n},
where n is a natural number, there are two which are consecutive.

Proof. We are given n + 1 numbers in the set

{1, 2, 3, · · · , 2n}

or in other words we are given n + 1 numbers in the n sets

{1, 2} ∪ {3, 4} ∪ {5, 6} ∪ · · · ∪ {2n − 3, 2n − 2} ∪ {2n − 1, 2n} .

By pigeonhole principle, two of the given numbers must belong to the same set and hence
must be consecutive.

Problem 7. Show that in any collection of n + 1 numbers from the set {1, 2, 3, · · · , 2n},
where n is a natural number, there are two such that one is a multiple of the other.

Proof. Every natural number m can be written as follows:

m = 2k · o

where o is an odd number known as the greatest odd divisor of m.

Now for each of the given n + 1 numbers of our exercise we consider the associated greatest
odd divisors. Since we have n + 1 numbers we also must have n + 1 odd divisors. But in the
set {1, · · · , 2n} we have exactly n odd numbers. Hence, by pigeonhole principle, there must
exist two numbers whose greatest odd divisors must be equal. These numbers must have the
property that one if a multiple of the other (why?).

9
 Problem 8. Consider n natural numbers: x1 , · · · , xn . Show that there is always a
sequence of them such that their sum

xi + xi+1 + · · · + xj

is divisible by n.

Proof. Consider the n sums

S1 = x1 ,
S2 = x1 + x2 ,
S3 = x1 + x2 + x3 ,
···
Sn = x1 + x2 + x3 + · · · + xn .

If one of the above sums is a multiple of n then we are done. If no sum above is a multiple
of n then when divided by n they must all leave remainders between 1 and n − 1. Since we
have n sums and n − 1 possible remainders, by pigeonhole principle, there must be two sums
which leave the same remainder. It is easy to see (exercise) that the difference of these two
sums is now a multiple of n. But this difference is a sum of the form of the statement of the
problem and hence we have proved that such a sum (of consecutive terms) is a multiple of n.

Problem 9. A student studied for a period of 37 days according to the following rules
1. every day he studied for at least 1 hour,

2. every day he studied for an integer number of hours without exceeding 12 hours,

3. he had to study for 60 hours in total.

Show that there was a period of a few consecutive days when he studied exactly for 13 hours.

Proof. Let Ai denote the number of hours that the student studied during the first i days.
Then we have the following
1. Ai+1 ≥ Ai + 1 (why?),

2. Ai+1 ≤ Ai + 12 (why?),

3. Ai 6= Aj for i 6= j (why?),

4. A37 = 60.
We want to show that there are i, j with i ≥ j + 2 such that

Ai = Aj + 13.

Following our methodology, we realize that we have the following two classes

Class 1: {A1 , A2 , · · · , A37 }

and
Class 2: {A1 + 13, A2 + 13, · · · , A37 + 13} .

10
We want to show that there are two objects from the two classes above which coincide. Clearly
we have 37 + 37 = 74 objects in total. But what are the values that these 74 numbers can
take? Clearly we have
1 ≤ Ai ≤ 60 < 73
and
1 ≤ Ai + 13 ≤ 73.
Hence the given 74 numbers can only take at most 73 values. Hence, by pigeonhole principle,
there must be two numbers which take the same value (i.e. they are equal). Hence, for some
i and j , with i ≥ j + 2 (why?), we must have

Ai = Aj + 13

which shows the required result.

Problem 10. Let A1 , · · · , A2000 be subsets of a set M such that each set Ai contains at
least two thirds of the elements of M . Show that there is an element of M which belongs to
at least 1334 of the 2000 subsets Ai .

Proof. We have 2000 subsets, that is 2000 classes of objects and we want to show that at least
1334 of their elements coincide.
Following our methodology, we compute that the total number of all objects in all the
2000 subsets is at least
2
2000 · · |M |,
3
where |M | is the number of elements of the set M . But all the elements in these subsets are
simply elements of the set M , and hence can be at most |M | elements.
Therefore, we have 2000· 23 ·|M | elements (pigeons) taking at most |M | values (pigeonholes),
and hence by the pigeonhole principle, there must be at least 2000 · 23 elements taking the
same value. That is at least 2000 · 32 are the same. That is, there is an element belonging to
at least 2000 · 23 = 1333.33333 sets. Hence, we obtain the desired result.

Problem 11. Let p be a prime number different from 2 and 5. Show that among the
numbers
1, 11, 111, . . . , 11 · · · 1(p 10 s)
there is always one divisible by p.

Proof. We are given p numbers and we want to show that at least one of them is a multiple
of p. We argue by contradiction. Assume that none of them is a multiple of p. Then arguing
as above, by pigeonhole principle, there must exist two numbers out of the p numbers whose
difference is a multiple of p. But the difference of these numbers is always of the form

(11 · · · 1) · 10k .

Hence, p must divide (11 · · · 1) · 10k . But p does not divide 10k and hence p must divide
11 · · · 1, which contradicts our assumption that p does not divide any number with unit digits!
Hence, p must divide of one these numbers.

11
2.4 Ramsey Theory
In this section we will study problems that are part of a very important area in Combinatorics
known as Ramsey theory. The problems we will consider are applications of the pigeonhole
principle.
Clearly, if we say consider all people on earth (about 7 billion people) then there must
three of them who are mutually friends or three of them who are unknown to each other. We
would like to consider a smaller group of people such that they will always have this property
(namely that there will always be three of them who are mutually friends or three of them
who are unknown to each other). The smallest number of (random) people that we need to
consider so that there will always be three of them who are friends or three of them who are
unknown to each other is in fact only six. We therefore need to prove the following problem:

Problem of six people: Among any six people there are always three who are mutually
friends or three who are mutually unknown.

Proof. Consider one person (call him P0 ) of the six people. Then this person is either friend
or unknown to each of the remaining five people. By pigeonhole principle, there must be at
least three of the remaining five people, call them P1 , P2 , P3 , such that P0 is either a friend of
P1 , P2 , P3 or P0 is unknown to P1 , P2 , P3 .
Case I. Assume that P0 is a friend of P1 , P2 , P3 .

Sub-case 1. If there is a pair of friends, call them Pi , Pj (with i, j ∈ {1, 2, 3}), among
these three people P1 , P2 , P3 then P0 , Pi , Pj are mutally friends, and so we are done (since we
found three people out of the six people who are mutually friends).

Sub-case 2. If there is NO a pair of friends among these three people P1 , P2 , P3 then

P1 , P2 , P3 are mutally unknown to each other, and so we are done (since we found three people
out of the six people who are mutually unknown to each other).

Case II. Assume that P0 is unknown to P1 , P2 , P3 .

Sub-case 1. If there is a pair of unknown people, call them Pi , Pj (with i, j ∈ {1, 2, 3}),
among these three people P1 , P2 , P3 then P0 , Pi , Pj are mutally unknown, and so we are done
(since we found three people out of the six people who are mutually unknown to each other).

Sub-case 2. If there is NO a pair of unknown people among these three people P1 , P2 , P3

then P1 , P2 , P3 are mutally friends with each other, and so we are done (since we found three
people out of the six people who are mutually friends with each other).

So, no matter how the initial six people are related to each other, there will always be a
group of three people which consists of friends or a group of three people which consists of
people unknown to each other.

It is convienient to represent people by points on the plane and relations of people by

colored edges connecting the points.
For example, let the points A, B represent two people P1 , P2 . Then we join A, B with a

• blue edge if P1 .P2 are friends or,

12
 • red edge if P1 .P2 are unknown to each other.

Then the above problem can be restated as follows:

Version 2 of Problem of six people: Consider any six points on the plane and color
all edges which connect these six points either red or blue. Show that there must always be
a red triangle or a blue triangle.

Proof. Consider one of these points, call it A. Then A is connected to five more points and
so there are five edges which terminate at A. Each of these five edges is either red or blue.
We have five edges (pigeons) and two colors (pigeonhole), and so by pigeonhole principle, we
have at there must be three edges which terminate at A which are either red or blue. Assume
that there are three edges which are red (similarly we deal with the case if they are blue).
Let B, C, D be the other endpoints of these three red edges. In other words, we have the
red segments AB, AC, AD. If either of the segments BC, BD, CD is red, then we have a red
triangle (why?). If none of the edges BC, BD, CD is red then they are all blue, and hence the
triangle BCD is blue. Hence, in all cases, there must always exist a red or a blue triangle.

Note that the graph of six points contains exactly 15 colored edges. Since we use only two
colors, we have in total 215 = 32k different possibilities for the colorings of the graph. Hence,
the above theorem, says that every graph out of the 32k possible graphs must contain a red
or a blue triangle. It is remarkable that we were able to show a result for 32k different graphs
WITHOUT considering each one of them, but instead by using the pigeonhole principle.
What if we consider five points, instead of six. Then can we always find a monochro-
matic triangle? The answer is no. See the following picture for a case where there is no
monochromatic triangle in a doubly colored graph of five points.

Hence, six is the smallest number that has the property that we can always find blue or a
red triangle always.
As we see, the above problem is an example where we seek regularity (e.g. a monochromatic
triangle) amid disorder (e.g. a randomly colored graph with two colors).
This is part of a combinatorial theory called Ramsey theory, that, in general, seeks regu-
larity amid disorder. We will not spend more time on Ramsey theory in this course. However,
it is interesting to include a few more definitions and provide a few known results (without
any proofs).

Definition of Ramsey numbers R(m, n). The natural R(m, n) is defined as the smallest
natural number that has the following property:

13
 • Consider R(m, n) points on the plane, and all edges connecting all pairs of these R(m, n)
points. Color each of these edges either red or blue (two possible colors). Then there is
always a blue m-gon, such that all sides and diagonals are blue, or there is a red n-gon
such that all sides and diagonals are red.
The theorem we showed above gives us that

R(3, 3) = 6.

Results that are known (no need to memorize them):

R(2, n) = n,
R(3, 4) = 9,
R(3, 5) = 14,
R(3, 6) = 18,
R(3, 7) = 23,
R(3, 8) = 28,
42 ≤R(5, 5) ≤ 55,
102 ≤R(6, 6) ≤ 169.

Note that the numbers R(5, 5), R(6, 6) are not known! Note that computationally it would
almost take infinite time for a computer to determine the exact value, of say R(5, 5), between
42 and 55 (why?).
Exercise: Define R(m, n, k) and show that R(3, 3, 3) = 17.

2.5 Erdös–Szekeres Theorem

The Erdös–Szekeres theorem concerns maximally ordered numbers in a random list of num-
bers. This theorem has applications to computer science.

Theorem (Erdös–Szekeres) Consider n2 + 1 real numbers a1 , a2 , · · · , an2 +1 . Show that

there is always an increasing or decreasing sub-sequence consisting of n + 1 numbers.
Exercise: Does the theorem hold if we replace n2 + 1 with n2 .

Proof. Let’s first consider a few special cases:

• n = 1: Then we are given 2 = 12 + 1 numbers a1 , a2 . Clearly we will have a1 ≤ a2 or
a2 ≤ a1 . So there will always be an increasing or decreasing sub-sequence with 2 = 1 + 1
numbers.

• n = 2: Then we are given 5 = 22 + 1 numbers a1 , a2 , a3 , a4 , a5 . We want to find 3 = 2 + 1

of them which form an increasing or a decreasing chain. Consider as a special case, just
to gain intuition, the following 5 (randomly selected) numbers:

2, 3, 1, 1.5, 4

Note that we can consider the following increasing sub-list: 2, 3, 4. If we replace the
final 4 by something less than 3 (so that we break the above increasing list) then we
would get this list of numbers:
2, 3, 1, 1.5, 2.5

14
in which case we still have the increasing list 1, 1.5, 2.5 (which simply does not start
from 2, but from 1. If we replace 2.5 with 0, so that we break again this increasing list,
then we would have the numbers
2, 3, 1, 1.5, 0
in which case there is no increasing sequence with 3 numbers! But then we can find a
decreasing sequence of three numbers: 2,3,1 or 3,1, 0.
Hence, we see that in all of the above cases, there is always an increasing or decreasing
sub-list of 3 numbers. How can we show this property for all possible lists of 5 numbers?
Let’s consider the (general) numbers a1 , a2 , a3 , a4 , a5 and let’s assume that there is no
list of three of them which is increasing, as was the case for the numbers 2, 3, 1, 1.5, 0.
We will then show that there are three of them which form a decreasing list.

– Consider now the longest decreasing lists starting with the number a1 . If this lists
has (at least) three numbers then we are done. Assume then that this list has one
or two numbers.
– Consider now the longest decreasing lists starting with the number a2 . If this lists
has (at least) three numbers then we are done. Assume then that this list has one
or two numbers.
– Consider now the longest decreasing lists starting with the number a3 . If this lists
has (at least) three numbers then we are done. Assume then that this list has one
or two numbers.
– Consider now the longest decreasing lists starting with the number a4 . This list
has at most two numbers a4 , a5 .
– Consider now the longest decreasing lists starting with the number a5 . This list
has exaclty only number: a5 .

Hence, we have five numbers and for each of them we consider the length of the longest
decreasing sequence starting with them. The lenghts are either 1, 2. Hence we have five
numbers and for each of them we have a list of either 1 or 2 numbers. By pigeonhole
principle, there must be three numbers with the property that three, call them x1 , x2 , x3 ,
of the five numbers must have maximal associated decreasing sequences of the same
length (either 1 or 2). Say that this length is 2.
We claim that
x1 ≤ x2 ≤ x3 .
Let’s show this. Let’s assume that we have x1 > x2 . Then by considering the longest
decreasing list of 2 numbers starting with x2 and joining to this list the number x1 we
would obtain a decreasing list of 3 numbers which starts from x1 . But then the maximal
decreasing sequence starting from x1 would contain 3 numbers and not 2, as is assumed
above. Hence, contradiction. Hence, we must not have x1 > x2 and hence we must have
x1 ≤ x2 . Similarly we show that x2 ≤ x3 . Hence, we have

x1 ≤ x2 ≤ x3 .

But this gives us a list of three numbers which is increasing.

So there is always a list of three numbers which is either increasing or decreasing.

15
 General case n ∈ N: Consider for each number ai the longest decreasing sequence
starting from that number ai (here i = 1, 2, · · · , n2 + 1). This list will contain say li
numbers.

– If li ≥ n + 1, for some i, then we are done!

– If li < n + 1, for all i, then we have

1 ≤ li ≤ n.

Hence, we have n2 + 1 numbers li which take at most n values. By pigeonhole

principle, there must be n + 1 numbers li which take the same value, that is which
are equal. Let these numbers be:

lk1 , lk2 , · · · , lkn+1 .

Consider the associated numbers from our list

ak1 , ak2 , · · · , akn+1 .

Then all these numbers have the property that the longest decreasing list starting
with each of them has the same number of numbers for all of them!
This implies that these numbers must form an increasing sequence

ak1 ≤ ak2 ≤ · · · ≤ akn+1 .

(why?–argue as above by contradiction assuming that for instance ak1 > ak2 ) Hence
we have a list of n + 1 numbers which is increasing.

The above result shows that if we are given sufficiently many numbers (that is n2 + 1
numbers) then we can find a subset (with n + 1 numbers) which has much more structure,
that is, it is either increasing or decreasing.

2.6 Diophantine Approximations

We next consider an application of pigeonhole principle in number theory. We start with the
following
Theorem (Diophantine approximations). Let a be a positive irrational number and let
 > 0 be any positive number. Then there are natural numbers n, m such that

|n · a − m| ≤ . (1)

As a corollary, there are natural numbers n, m such that

 m  
a −  | ≤ ≤ .

n n
Proof. First of all, it is easy to find infinitely many m, n such that
 m  1
a − ≤ .

n n


16
Indeed, the above inequality is equivalent to

|n · a − m| ≤ 1.

Consider now any natural number n > a1 and consider the number n · a. Then clearly this
number must be between two consecutive natural numbers m, m + 1. Then for this m we have

m≤n·a≤m+1

which implies that

|n · a − m| ≤ 1.
If  > 1 then we are done (why?). If  < 1, then we consider the multiples of :

0 <  < 2 < 3 < · · · < (k − 1) < 1 < k.

Hence, the multiplies of  divide the interval [0, 1] in k intervals:

[0, ],

[, 2],
[2, 3],
[3, 4],
···
[(k − 1), k].
Let’s now consider k + 1 different numbers of the form n · a − m which are less than 1 (by
the above). Since we have k + 1 numbers in a set of k intervals (the ones with endpoints the
multiples of ) two of them must be in the same set! But then their difference is bounded by
, since  is the size of the interval. The difference of numbers of the form n · a − m is of the
same form. Hence, we proved that there exists numbers n, m such that

|n · a − m| ≤ .

Hence,  m  
a −  ≤ ≤ .

n n

Theorem (Application in number theory). There is a power of 2 which starts with 2017,
that is there is n ∈ N such that
2n = 2017 · · · .

Proof. We want to show that there are natural numbers n, m such that

2017 · 10m ≤ 2n ≤ 2018 · 10m

or equivalently, if a = log 2 then

log 2017 + m ≤ n log 2 ≤ log 2018 + m

17
or
log 2017 ≤ n log 2 − m ≤ log 2018.
If we let a = log 2, then by above problem, see equation (1), we have that there are natural
numbers n0 , m0 such that

|n0 log 2 − m0 | ≤ (log 2018 − log 2017)/2.

Then an appropriate multiple of the number n0 log 2−m0 will lie between the numbers log 2018
and log 2017 (why?). This multiple gives us the desired result.

2.7 Practice Problems

Note: Problems with a (*) are slightly harder.

1. Given any 5 points in a unit square, show that two of these points must be within √1
2
of each other.

2. How many people must there be in a stadium to ensure that at least 2 people have the
same first and last initial?

3. A bag contains 100 apples, 100 oranges, 100 bananas and 100 pears. Every minute you
choose one fruit from the bag. How long will it take to ensure that you have at least 10
fruits of the same kind?

4. Show that given n + 2 natural numbers there are always two whose sum or difference is
divisible by 2n + 1.

5. Let m, n be natural numbers such that their greatest common divisor gcd(m, n) = 1.
Prove that there is a natural number k such that n divides mk − 1.

6. A group of friends meet at a restaurant and some of them shake hands. Show that at
least two of them shook the same number of hands.

7. We place n cards with the names of n students on a circular table in a uniform way.
The students sit around the table randomly, such that in fact none of them faces their
card. Prove that we can rotate the table such that at least two students face their own
card.

8. Prove that if we randomly pick 39 numbers then there will always be 3 of them with
the property that their pairwise differences are divisible by 19.

9. If we pick 85 numbers randomly from the set {1, 2, 3, ..., 150}, then there will always be
two where their difference is 19.

10. (*) If we pick 77 numbers randomly from the set {1, 2, 3, ..., 150}, then there will always
be two where their difference is 19.

11. Prove that for any 7-element subset A of the set T23 = {1, 2, 3, ..., 23} we can always
find two non-empty subsets of A with equal sums.

12. Does there exist an n × n matrix, with n ≥ 7, such that 1) each entry is a natural
number between 0 and n, and 2) the sums of the entries of any 2 × 2 sub-matrix are
pairwise distinct?

18
3 The Principle of Extremals
3.1 Theory
We next consider a class of combinatorial problems the solution of which makes use of the
so-called principle of extremals. According to this principle:

• if we want to show that a certain construction exists, then we consider the largest or the
smallest among a specific class of structures. The largest one, or the smallest, might be
the one that satisfies the given properties.

• if we want to show that a certain construction cannot possible exist, then, arguing by
contradiction, we assume that at leaste one such construction exists. We next, consider
the largest (or the smallest) such construction and using it and the assumptions of the
problem we deduce that there is an even larger (or smaller respectively) construction.
But this contradicts the fact that we started with the largest (or smallest respectively)
construction. Hence, no such construction can exist.

As we shall see, the principle of extremals is an extremely strong principle and can provide
very elegant answers to surprisingly complex problems.

3.2 Solved Problems

We next present a few applications of the principle of extremals.

Problem 1. Consider 100 numbers a1 , a2 , · · · , a100 , each number being on the vertex of
a regular 100-gon. We assume that any number is the average of its two neighbors, that is
ai−1 + ai+1
ai = .
2
Show that all these numbers are equal, that is a1 = a2 = · · · = a100 .

Proof. Consider the largest number, say aj , of these 100 numbers. By assumption aj is the
average of its two neighbors aj−1 , aj+1 . However since aj is the largest number, we necessarily
have
aj ≥ aj+1 ,
aj ≥ aj−1 .
This is possible only if
aj−1 = aj = aj+1 .
Therefore, aj−1 , aj+1 are also the greatest numbers. So by the above reasoning they are also
equal to their neighbors. We proceed inductively.

Problem 2. Consider n points on the plane. Color each of these points either red or
blue. We assume that on any edge with endpoints of the same color there is a point of different
color (that is, there is always a red point between two blue points, and a blue point between
two red points). Show that all the points must lie on the same line.

19
Proof. Assume that not all points like on the same line. Then we can form triangles using our
points. We consider that triangle with the smallest area. Let is be ABC. Two of its vertices,
say B, C, will have the same color. Hence, by assumption, there will be another point, call
it D, of different color on the side BC. But then clearly the triangle ABD has smaller area,
contradiction. Hence there cannot be any triangles and hence all points lie on the same line.

√
Problem 3. Show that 2 is irrational.
√ √
Proof. Assume that 2 is rational. There is a multiple of 2 which is a natural number.
Consider the following set
n √ o
A = n ∈ N : n 2 is a natural number

By assumption the set A is a non-empty subset of the natural numbers. Hence, it must have
a smallest element. Let this smallest number in A be n0 . Consider the number
√
n1 = n0 2 − n0 .

Clearly by our assumptions n1 is a natural number (why?). Furthermore

√ √
n1 2 = 2n0 − 2n0

is also a natural number (why?). Hence, n1 must be a member of the set A. But n1 < n0
(why?) which contradicts the fact that n0 is the smallest element of A.√Hence there cannot be
a smallest element in A. Hence A must be empty. Which means that 2 is indeed irrational.

Problem 4. Consider a solar system with 2017 planets such that all planets have
mutually distinct distances from each other. Consider an astronomer on each of these planets
(so 2017 astronomers in total) who observe the closest planet to them. Show that there must
be a planet which is not observed by any astronomer.

Proof. Consider first for simplicity (reduction principle) the case where we have three planets
P1 , P2 , P3 . Arguing using the principle of extremals, we consider the two planets with the
smallest distance. Let’s assume that these planets are P1 , P2 . Clearly, the astronomer on P1
observes the planet P2 and the astronomer on P2 observes P1 . The observer on P3 observes
either P1 or P2 and hence nobody observes P3 . So we proved the desired result for the case
of three planets.
What about the case of 2017 planets P1 , P2 , · · · , P2017 ?
First note that it suffices to show that there are at least two astronomers who observe
the same planet. Indeed, there are exactly 2017 planets and exactly 2017 astronomers each
of which observes one and only one planet. So, if two astronomers observe the same planet,
there must be a planet which is not observed.
Arguing using the principle of extremals, we consider the two planets, say P1 , P2 , with the
smallest distance. As before, the astronomers on these planets observe each other. We next
consider two cases:
• Case I: If there is a third astronomer that observes either P1 or P2 then either P1 or P2
is observed by at least two astronomers. Hence we are done.

20
 • Case II: No other astronomer observes P1 or P2 . Then, the reduced planetary system
P3 , P4 , ...P2017 satisfies the same property as the bigger one. We proceed by induction.
We argue for the reduced planetary system P3 , P4 , ...P2017 as above. We consider the
pair of planets with the smallest distance and consider cases I, II as above. If case I
holds then we are done. If case II holds then we are able to remove two more planets and
end up with 2013 planets. We keep doing the same thing until we reach three planets.
The case of three planets was however addressed at the beginning.

Problem 5. Consider a collection C of points on the plane such every point of C is the
midpoint of two other points of C. Show that C must contain infinitely many points.

Proof. Assume that we have finitely many points. Consider the longest segment AB defined
by these (finitely many) points. By assumption there is segment CD whose midpoint is B.
But then AB is the median of the triangle ACD and hence (by a well-known result in plane
geometry)
AB < (AC + AD)/2
which implies that either AC or AD is longer than AB, which contradicts the maximality
assumption for the length of AB. Hence, no such longest segment can exist and so there must
infinitely many points (because for any collection of finitely many points there are always two
with the longest distance).

Problem 5. (Sylvester–Gallai theorem) Consider a collection D of n points on the

plane with the property that for any two points in D there is a third point in D which lies on
the line defined by the two points. Show that all points in D must lie on the same line (that
is, they must be all co-linear).

Proof. Assume that not all points lie on the same line. So there must be triangles that are
formed with vertices being points in the collection D.
Since we assume that such triangles exist, we can consider that triangle CAB with the
smallest height AL (where L is the projection of A on BC). By assumption, there is a point
K of the collection D on the line AB. Let’s assume that C lies between the points K and B.
Then the height from B of the triangle BCK is smaller than the height from A of the
triangle CAB (why?). But we have assumed that the triangle CAB has the smallest height.
This is contradiction. Hence, there cannot be any triangle with the smallest height, hence
there cannot be any triangles to begin with. Hence all points must lie on the same line (in
which case they indeed do not form any triangle).

Problem 7. Consider n points on the plane such that every point is connected via
edges with at least three other points. Show that there must exist a closed path with an even
number of edges. (equivalently, using terms from graph theory, the problem says that for any
finite graph with the property that every vertex has degree at least three, there is always an
even cycle).

21
Proof. Consider the longest path in our finite graph:

A1 → A2 → · · · Ak−1 → Ak .

By assumption, the point A1 must be connected with at least three points. One of them is A2 .
The other two points must necessarily be points of that path (otherwise we would be able to
obtain an even larger path contradicting the fact that we considered the longest path). Let’s
assume that A1 is connected with Ai and Aj with i < j. Then we have three cycles:

cycle 1. A1 → A2 → · · · Ai → · · · → Aj → A1

cycle 2. A1 → A2 → · · · Ai → A1 .
cycle 3. A1 → Ai → · · · Aj → A1 .
If i is even then cycle 2 is even. If j is even, then cycle 3 is even. If i, j are both odd, then
cycle 1 is even, since its length is j − i. Hence, in all cases, there is always an even cycle.

3.3 Practice Problems

1. Decide if there is a minimum or a maximum for each of the following sets:

(0, 1), (0, 1], N, {x ∈ R : x2 ≥ 21}.

√
2. Prove that the number 21 is irrational.

3. In a TA room there are 20 students. Each of them has at least two friends who are also
in the TA room. Prove that at least three students can sit around a (round) table such
that each student sits between friends.

4. 2019 students are standing in a field such that the distance between each pair is distinct.
Each student is holding a ball, and when the teacher blows a whistle, each student throws
their ball to the nearest student. Prove that there is a student who receives no ball.

5. (*) The integers from 1 to 64 are placed in the cells of an 8x8 chessboard. Show that
there is a pair of horizontally, vertically, or diagonally adjacent cells whose values differ
by at least 9.

6. (*) Every vertex of a graph has degree exactly 3. Prove that we can partition the
vertices of the graph into two sets and such that each vertex is connected with at least
two vertices from the other set.

7. (*) A finite set of points in the plane has the property that the triangle formed by any
of them has area less than 1. Prove that there is a triangle of area less than 4 that
contains all the points.

22
4 The Principle of Invariants
4.1 Theory
The principle of invariants is yet another very powerful principle that is applicable to problems
which involve a process and the goal is to see what the possible final states of that process
are. A general statement of this principle is the following:

• Assume that we have a process which takes place in discrete steps S1 , S2 , S3 , ..., Sk where
Sk is the final step of the process.

• Assume also that the variables of the problem take different values at different steps
(that is, the system changes from step to step).

• According to the principle of invariants, for each step Si we need to obtain an appropriate
quantity Qi that depends on the values variables at the Si step. The quantity Qi should
be such that
Q1 = Q2 = · · · = Qk .
In other words, we would like the quantity Qi not to change from step to step meaning
that it is indeed a genuine invariant of the process.

• Using the initial state of the system we determine the value of the invariant.

• By the invariant property, the value we found above should also coincide with the value
at the final state.

• Finally, knowing the value of the invariant at the final state should allow us to obtain
significant information about the possible final states of the process.

The most challenging part in applying this principle is finding what the correct invariant
quantity is. There is no general rule as to how to obtain an invariant quantity. It depends on
the specific problem under consideration. We next present a few applications of the principle
of invariants.
Some problems will require a generalized version of an invariant. That is, we will need
to construct quantities Q for each step, which although its value changes from step to step,
its parity (for example) does not change. That is Q will always be an even number, or will
always be an odd number, or will always be a number of the form 3n + 1 etc. It is the latter
property of Q that will serve as our invariant.

4.2 Solved Problems

Problem 1. The numbers 1, 2, 3, ·, 200 are written on the blackboard. At each step

• We select two numbers, say x, y, that are written on the blackboard

• We delete the numbers x, y from the blackboard

• We add the number x + y on the blackboard.

What are the possible final numbers on the blackboard.

23
Proof. Let Q be the sum of all numbers written on the blackboard at each step. Clearly Q
is an invariant, that is it does not change from step to step (why?). Initially Q is equal to
20100. Hence, the final number on the blackboard must be 20100 .

Problem 2. We write the numbers 1,0,1,0,0,0 (in this order) on a circle. At each step

• We select two consecutive such numbers on the circle.

• We add 1 to both of the numbers we selected.

Is it possible to perform this process in such a way so that we obtain equal numbers after a
finite number of steps?

Proof. Let x1 , x2 , x3 , x4 , x5 , x6 be the numbers on the circle at each step. We consider the
quantity
Q = x1 − x2 + x3 − x4 + x5 − x6 .
Note that this quantity is initially equal to 2. This quantity is an invariant of our process
(why?). Hence, we can never reach a state where wll the numbers are equal since such a state
would have Q = 0. But we showed that Q = 2 always.

Problem 3. Can we put the numbers 1,2,3,..., 2019 in a row to form one number which
is a perfect square?

Proof. What all the numbers that we can form have in common? All the possible numbers
we can form have equal sum of digits
1
1 + 2 + 3 + · · · + 2019 = 2019 ∗ 2020 = 2039190.
2
This number is a multiple of 3, but not a multiple for 9. However any perfect square which is
a multiple for three is also a multiple of 9. So we cannot form a perfect square.

The next problem is a very impressive application of the principle of invariants.

Problem 4. The numbers 1,2,3,..., 10 are written on the blackboard. At each step:

• We select two numbers, say x, y that are written on the blackboard

• We delete the numbers x, y from the blackboard

• We add the number x + y + xy on the blackboard.

What are the possible final numbers on the blackboard.

24
Proof. In the first problem, the invariant quantity was simply the sum of all numbers written
on the blackboard. However, the sume is not an invariant for this problem and hence is not
useful here. On the other hand, we observe that

x + y + xy + 1 = (x + 1)(y + 1).

This means that if we add 1 to all numbers that are written on the blackboard and then take
the product of the resulting numbers, then this product is an invariant (why?)!! Initially, this
product is equal to 11!. Hence, the final number must be 11! − 1 (why?).

Problem 5. Let n be an odd natural number. The numbers 1, 2, 3, · · · , 2n are written

on the blackboard. We follow the following process. At each step:

• We select two, say x > y, of the numbers that are written on the blackboard

• We delete them from the blackboard

• We add the difference number |x − y| on the blackboard.

is it possible to perform the process such that 2 is the final number written on the blackboard?

Proof. Let Q be the sum of the numbers written on the blackboard at each step. How does
Q change from step to step? Since we replace x, y by |x − y| we have that

Q(i + 1-step) = Q(i-step) − x − y + (x − y) = Q(i-step) − 2y.

Hence Q changes by an even number. Since Q is initially odd (why?) it must always be odd.
Hence, the final number cannot be even!

Problem 6. There are 100 0’s and 100 1’s written on the blackboard. We follow the
following process:

• At each step we choose two of the numbers written on the blackboard.

• If the numbers we chose are equal, then we delete them from the blackboard and add a
0 on the blackboard.

• If the numbers we chose are not equal, then we delete them from the blackboard and
add a 1 on the blackboard.

Is it possible to perform the steps appropriately such that the final (unique) number that is
written on the blackboard is 1?

Proof. Let Q be the sum of all numbers on the blackboard. Clearly, Q is initially even. At each
step, Q either is unchanged or is decreased by 2. Hence Q remains even always. Therefore,
the final number must be even, that is 0. Hence it cannot be 1.

Problem 7. We have three tubes containing 8, 9, 10 coins respectively. We follow the

following process:

25
 • At each step we select two of the three tubes.

• We remove one coin from each of the two tubes we selected.

• We add these two coins to the third tube.

Can we perform this process such that all coins are transferred to the same tube?

Proof. At each step, the difference of the number of coins in any two tubes is either unchanged
or is changed by 3 or -3. Suppose we are able to put all coins in one of the tubes. Then the
other two tubes will have zero coins (and hence the difference will also be zero). But this
difference is initially 1. So it is impossible to transfer all coins to the same tube.

4.3 Semi-invariants
For a type of problems we will not be able to find an invariant quantity; however, we might be
able to find what is know as a semi-invariant quantity. A semi-invariant quantity is a quantity
that is always increasing or always decreasing from step to step. Using this monotonicity
property of the semi-invariant quantity might yield important properties of the final states of
the process under consideration.
A spectacular application of the use of semi-invariants is the so-called Conway’s army.

Problem (Conway’s army). Please see the following link:

https://en.wikipedia.org/wiki/Conway’s_Soldiers

4.4 Practice Problems

1. A dragon has 10 heads. A knight can cut off 5 or 6 heads with one blow of his sword. If
5 heads are cut then 2 new heads grow on its shoulders. If 6 heads are cut then 3 new
heads grow on its shoulders. If all heads are blown off, the dragon dies. Can the dragon
ever die?

2. In a basket we have 300 balls each of which has a random color. These balls then then
places in 100 smaller baskets, each of which contains only 3 balls. If there is a black
balls in one of these small baskets, then all the remaining two balls in this basket are
also colored black. In this possible to distribute the 300 balls in the small baskets such
that at the end we have exactly 151 black balls?

3. Two opposite corners are removed from a standard 8x8 chessboard. Is it possible to tile
the resulting board with 2x1 dominoes?

4. In a chess tournament there are n players participating. Every player has to play.
Initially, every player has 0 points. Winners get 3 points , losers get -1 point and the
players who tie get 1 point each. If every player has to play with every other player,
what is the sum of points of all players at the end of the tournament?

5. Two players A and B play the following game: They have 21 cups on a table and each
player alternatively removes 1 or 2 or 3 cups. The player who removes the last cup on
the table wins. Which player has a winning strategy?

26
 6. The Euler characteristic is equal to the number of vertices (V) minus the number of
edges (E) plus the number of faces (F) of a triangulation of some shape. That is, it
is V − E + F . Explain why this is an invariant. Compute the Euler characteristic for
triangulations of the plane, the sphere and the torus.

7. In a classroom, each students has at most three enemies. Prove that the classroom can
be separated into two classrooms, so that each student has at most one enemy in his
own classroom.

8. In a zoo there are 13 red, 15 green and 17 blue chameleons. If two chameleons of different
colors meet, they both simultaneously change color to the third color (e.g. if a red and
a green chameleon meet each other, they both change to blue).

• Is it possible that they will eventually all be the same color?

• Is it possible that there will eventually be the same numbers of gray, brown, and
crimson chameleons?

9. A group of boys is seated around a table. They play a game with 12 cookies. Initially
a boy has all the 12 cookies with him. Every minute, if any boy has 2 or more cookies
with him, he passes a cookie to the boy on the left, and a cookie to the boy on the right.
The game ends when each and every boy has 1 and only 1 card with him. Will the game
ever end, if there are

• 4 boys around the table?

• 11 boys around the table?

5 Permutations and Combinations

5.1 Additive and Multiplicative Principle
We spent the first half of the course in analyzing structures and finding ways to obtain
properties of them. We will spend the second half of the course in counting structures. We
will need to find ways to count all possible configurations that satisfy given properties without
considering each and every possible configuration. In other words, we need to find ways to
count without really counting.
We start we some basic combinatorial principles.

• Additive principle: If there are n ways to perform process A and m ways to

perform process B then there are n + m ways to perform process A or the
process B.

It is very important to emphasize that n + m refers to the number of all possible ways of
performing either of the two processes and not both of them.
A trivial, but instructive, example is the following: A school has two classes. The first
class has 20 students and the second class has 30 students. The total number of students of
the school is computed using the additive principle since any student belongs to either classes
but not in both of them. Hence the total number is 20+30=50.
Multiplicative principle:
The multiplicative principle concerns processes happening simultaneously, supplementing
therefore the additive principle.

27
 • If there are n ways to perform process A and m ways to perform process B
then there are n · m ways to perform process A and process B.

For example, in the case of the school above, there are 20 · 30 ways to choose a student from
the first class and a student from the second class (forming thus a pair of students). Indeed,
choosing a student from the first class can be done in 20 ways. Choosing a student from the
second class can be done in 30 ways. Hence, according to the multiplicative principle, choose
a pair from both classes can be done in 20 · 30 = 600 ways.
The multiplicative principle (also known as the rule of product) is of fundamental impor-
tance in combinatorics and for this reason is known as the fundamental principle of counting.
Principle of bijections
The principle of bijections allows to show that two different processes can be performed
in the same number of ways. We first need to recall a few basic definitions.
Let A, B be two sets. A function f : A → B is called

• injective if for all x1 6= x2 in A we have f (x1 ) 6= f (x2 ) in B,

• surjective if for all y ∈ B there is an x ∈ A such that f (x) = y,

• bijective if it is both injective and surjective.

Principle of bijections: Two (finite) sets A, B have the same number of elements
if there is a bijection function from A to B.
The case of infinite sets is more involved. For example, there is a bijection between all
even natural numbers and all natural numbers, but there is no bijection between all natural
numbers and all real numbers.

5.2 Solved Problems

We next present two problems that appeared in mathematical olympiads. These problems
are beautiful applications of the additive and the multiplicative principles.

Problem 1. Consider 3 lines on the plane passing through the point O creating 6
different sectors. We put 5 different points in each of these sectors (so 30 points in total).
Show that there are at least 1000 triangles with vertices 3 of these 30 points and such that
the point O is either in the interior or on the boundary of these triangles.

Proof. Let’s enumerate the sectors by 1,2,3,4,5,6. If the vertices of a triangle belong in the
sectors (1,3,5) or (2,4,6) then they must contain O. How many such triangles are there?
There are 5 points in sector 1, 5 points in sector 3 and 5 points in sector 5. Hence, by the
multiplicative principle, there are 5 · 5 · 5 triangle with vertices in the sectors (1,3,5). Similarly,
there are 5 · 5 · 5 triangles with vertices in the sectors (2,4,6). This gives us 250 triangles with
the desired property. We need to find 750 more such triangles.
Consider an pair of points from opposite sectors, that is from sectors (1,4), (2,5), (3,6).
Consider the case where the points are taken from the sectors (1,4). Consider the line that
connects these points. Then the point O must lie in one of the two parts of the plane that are
created by this line. Either the sectors (2,3) or the sectors (5,6) also line in the same part.
Suppose that the sectors (2,3) line on that part of the plane. Then any triangle with vertices
from the sectors (1,4,2) or (1,4,3) contain the point O. How many such triangles are there?
By the multiplicative principle there are 5 · 5 · 5 triangles with vertices on the sectors (1,4,2)

28
and 5 · 5 · 5 triangles with vertices on the sectors (1,4,3) giving in total 250 triangles. However
we have in total 3 pairs of opposite sectors to begin with, hence by the additive principle,
there are 750 more triangles that contain O giving a total of 1000 such triangles.

Problem 2: Compute the number of all squares on the plane with vertices of the form
(x, y) with x, y ∈ N such that 1 ≤ x ≤ n, 1 ≤ y ≤ n.
Answer: We will first compute the number of all squares with sides parallel to the x and
the y axes and of type k × k. The bottom left vertex (a, b) of any such square must satisfy
(why?)
1 ≤ a ≤ n − k, 1 ≤ b ≤ n − k.
By the multiplicative principle, there are (n−k)2 such points and hence (n−k)2 such triangles.
There are squares whose sides are not parallel to the horizontal and the vertical axes.
Every such square can be inscribed, however, in a square with horizontal and vertical sides
(why?). Hence, in order to compute the number of all squares, it suffices to compute all
parallel squares (that is with horizontal and vertical sides) and all the squares inscribed in
them. Given any k × k parallel square, we can find k − 1 inscribed squares (why??). Hence,
any k × k parallel squares gives rise to k squares in total, 1 is the parallel square itself and
k − 1 are the inscribed squares. Hence, by the multiplicative principle, there are k · (n − k)2
squares that are created by the k × k paralles squares. By the additive principle we have in
total
Xn
S(n) = k · (n − k)2
k=1

squares. The number S(n) can be computed easily using

n
X
k = n(n + 1)/2,
k=1
Xn
k 2 = n(n + 1)(2n + 1)/6,
k=1
Xn
k 3 = n2 (n + 1)2 /4.
k=1

It follows (why?)
S(n) = n2 (n + 1)(n − 1)/12.

5.3 Permutations
Consider n objects. For example, consider the first n natural numbers

Tn = {1, 2, 3, · · · , n}

A permutation of Tn is a re-arrangment of Tn , that is it is a new ordered list where

all the natural numbers from 1 to n appear exactly once in some specific order.
For example, the following are permutations of T3 :

(1, 2, 3), (1, 3, 2), (2, 1, 3), (2, 3, 1), (3, 1, 2), (3, 2, 1).

29
In fact these 6 permutations are all permutations of T3 .
More generally, there are
n! = 1 · 1 · 2 · 3 · · · n
(n factorial) permutations of Tn and hence n! permutations of any n distinct objects. This fol-
lows immediately by the multiplicative principle. Indeed, any permutation can be determined
by first determining its first entry (there are n possible ways for this), then determining its
second entry (there are n−1 ways for this) etc... until we reach the final entry (which can only
be determined in 1 way, since all the other n − 1 entries have been already selected). Hence,
by the multiplicative principle the entire process can be completed in 1 · 1 · 2 · 3 · · · n = n!
ways.

Ordered subsets

An ordered subset with k elements of Tn is a list which consists of k elements taken from
Tn such that each element in the list appears not more than once. For example, the following
lists
(1, 2), (1, 3), (2, 1), (2, 3), (3, 1), (3, 2)
are ordered subsets of T3 with k = 2 elements.
By the multiplicative principle, and arguing as above, there are
n!
n · (n − 1) · (n − 2) · · · (n − (k − 1)) =
(n − k)!

ordered subsets of Tn with k elements. Indeed, there are n ways to select the first entry, n − 1
ways to select the second entry, ...., n − (k − 1) ways to select the k th entry.

5.4 Combinations
A combination of Tn with k elements is a subset of Tn with k elements for which
every element appears at most once and for which order does not matter. In
other words, the order of the elements does not play a role in combinations and, moreover,
no repetition of elements is allowed.
For example, the following

{1, 2, 3} , {3, 2, 1} , {1, 3, 2}

represent the same combination of T4 with 3 elements. On the other hand, the following

{1, 2, 2} , {3, 3, 1} , {1, 1, 2}

are not combinations since we have elements which are appear more than once.
Question: How many combinations of Tn with k elements are there?
Answer: The only difference between combinations with k elements and ordered-subsets
with k elements is that order does not play a role in combinations whereas order plays a role in
ordered subsets. Since we have k elements, there are in total k! different re-orderings (permu-
n!
tations) of any ordered subset with k elements. On the other hand, there are (n−k)! ordered

30
subsets with k elements. However, since k! ordered subsets represent the same combination,
we have that the total number of all combinations with k elements is
n!  
(n−k)! n! n
= := .
k! k! · (n − k)! k
n


The number k is called the binomial coefficient because of the following property:
n          
X n n n n 2 n n
(1 + x)n = · xk = + x+ x + ··· + x . (2)
k 0 1 2 n
k=0

This identity follows easily by observing the following: for each k the term xk is formed by
choosing k of the n factors of the product:

(1 + x)n = (1 + x) · (1 + x) · (1 + x) · · · (1 + x).

Clearly there are nk to choose k factors and hence there are nk terms of the form xk in the

expanded sum.

5.5 Identities of the binomial coefficients

The following identities hold

1. Special cases: n0 =1, n1 =n, n2 = n(n − 1)/2, n

n = 1.

2. Symmetry identity: nk = n−k n

.

3. Pascal identity: nk = n−1 + n−1

k k−1 .

These identities are easily proved algebraically using the formula for the binomial coefficients.
However, we can also provide combinatorial proofs which require no computations.
Combinatorial proofs of identities: There are two generals methods to prove identities
in a combinatorial way:

1. The first method makes use of the bijection principle. That is, we need to find a
bijection between the elements counted on the left hand side of the identity and the
elements counted on the right hand side of the identity.

2. The second method computes the elements of the same set in two different ways. Since
the same number is computed in two different ways, the final results from these two
ways must coincide leading to the desired identity.

Combinatorial proof of the symmetry identity:

The symmetry identity easily follows from the bijection principle. Indeed, there is a
bijection between combinations with k elements and combinations with n−k elements. Indeed,
if we choose a combination with k elements then we have left out n − k elements
which
clearly form an n − k combination. Hence, by the bijection
principle, the number nk of all
n
combinations with k elements are equal to the number n−k of all combinations with n − k
elements.

31
 Combinatorial proof of the Pascal identity:
Pascal’s identity follows from counting the same set in two different ways. The LHS gives
the number of all k combinations. We will prove that the RHS gives the numbers of all
k combinations as well. Consider the element {n} of the set
Tn . A k combination either
n−1
contains {n} or does not contain {n}. There are in total k−1 combinations with k elements
which contain {n}, since any such k combination is determined by the choice of the first k − 1
elements from the set Tn−1 (since the k th element is {n}). On the other hand, there are
n−1


k combinations with k elements which do not contain {n} since each such combination is
completely determined by the choice of k elements from the set Tn−1 (since the element {n}
is now excluded). The identity follows immediately from the additive principle.

More identities of the binomial coefficients

The following identities hold:
n          
X n n n n n
= + + + ··· + = 2n ,
k 0 1 2 n
k=0
n          
X n n n n n
k =0· +1· +2· + ··· + n · = n2n−1 ,
k 0 1 2 n
k=0
n  2  2  2  2  2  
X n n n n n 2n
= + + + ··· + = ,
k 0 1 2 n n
k=0

Proof. The first identity can be proved algebraically by setting x = 1 in the the binomial
identity (2). However, we can also provide a combinatorial proof of this identity. The left hand
side, by the additive principle, counts all possible combinations that we can form. The right
hand side counts the same thing. Indeed, for each of the n elements there are 2 possibilities:
either the elements is a member of the combination or not. Since we have n elements, and
for each of them we have 2 possibilities, by the multiplicative principle, there are in total 2n
possibilities and hence 2n combinations in total. Note that at the last step we had to use the
multiplicative principle and not the additive principle, since each combinations is determined
by knowing if every element is part of it or not. That is we have to know the state of all
elements simultaneously and hence
The second identity can be proved analytically by first differentiating the binomial iden-
tity (2) and then setting x = 1 in the resulting identity. However, we can again provide a
combinatorial proof of this identity. The term k nk counts all possible teams


(combinations)
with k players (elements) with an assigned leader. Indeed, there are nk teams with k play-

ers and for each of these teamas there k ways to choose the leader. Hence, by the additive
principle, the LHS counts the number of all teams with an assigned leader. We will show
that the RHS counts the same thing. Indeed, since we have n players to begin with, there
are n possibilities for the assigned leader. For each of the remaining n − 1 players we have
2 possibilities: either they are a member of the team or they are not a member of the team.
Hence, by the multiplicative principle we have in total n2n−1 different ways to make a team
with a leader.
The third identity can also be proved combinatorially. In fact, using the symmetry identity,
it suffices to prove that
n      
X n n 2n
· = .
k n−k n
k=0

32
The RHS counts all combinations with n elements from 2n objects. We can divide these
2n objects in two sets S1 , S2 each of which contains n sets. Hence, the RHS counts all
combinations with n elements that we can form using the elements in the sets S1 , S2 . We
will show that the LHS counts the same thing. Indeed, any n combination of the 2n elements
of the set S1 ∪ S2 will contain k
elements from S1 and n − k elements from S2 , for
some k
n n
between 0 and n. There are k ways to choose the k elements from S1 and n−k ways to
choose the n − k elements from S2 . By the multiplicative principle, there are nk · n−k
n



such
combinations. The result follows from the additive principle where we sum all possibilities for
k.

5.6 Practice Problems

1. Prove that        
2n 2n 2n 2n
+ + + ··· + = 22n−1 .
0 2 4 2n
2. Prove that            
n n n n n n
+ + + ··· = + + + ···
0 2 4 1 3 5
3. Prove combinatorially that
       
n m n n−k
· = · .
m k k m−k

4. Prove that          
1 2 3 n+1 n+2
+ + + ··· + = .
0 1 2 n n
5. Prove in two ways, combinatorially and algebraically, that
       
n 0 n 1 n 2 n
·2 + ·2 + · 2 + ··· + · 2n = 3n .
0 1 2 n

6. Relate the powers of 11: 11, 112 , 113 , 114 with the Pascal triangle.

6 Combinations with Repetition

A combination with repetition with k elements of n objects is an un-ordered collection of k
elements from n given objects where each element might appear repeatedly up to k times.
For example, there are 3 combinations of 2 elements from 3 objects, namely

{1, 2} , {1, 3} , {2, 3}

but there are 6 combinations with repetition of 2 elements from 3 objects, namely

{1, 2} , {1, 3} , {2, 3} , {1, 1} , {2, 2} , {3, 3} .

We will next show that there are  

n−1+k
k

33
combinations with repetition with k elements from n objects.
We observe that any combination with repetition with k elements from the set {1, 2, 3, · · · , n}
gives rise to a sequence of (n − 1) + k objects such that k of them are the k objects of the
combination with repetition and (n − 1) are arrows → that are used to declare that we done
with using one number in our combination and we move to the next number. For example,
the combination
{1, 2}
gives rise to the sequence with n − 1 + k = 2 + 2 = 4 entries:

(1, →, 2, →).

The above list basically provides a way to exhaust the set {1, 2, 3} in the following way:

1. we write 1 as the first entry, since 1 is a member of the combination

2. we are done with 1, so we use the arrow → to declare that we need to move to 2

3. we write 2 as the next entry, since 2 is a member of the combination

4. we are done with 2, so we use the arrow → to declare that we need to move to 3 where
the process terminates since we cannot use 3 anymore because we have already chosen
2 members.

The above imply that every combination with repetition with 2 members gives rise to an
exhaustion of the set {1, 2, 3} that consists of 4 in total steps, as above. An another example,
consider the case of the combination with repetition

(→, 2, 2, →)

of {1, 2, 3} . The above provides a way to exhaust the set {1, 2, 3} in the following way:

1. 1 is not a member of the combination, so we need to use → to move to 2

2. we write 2 as the next entry, since 2 is a member of the combination

3. we write 2 as the next entry, since 2 is again member of the combination

4. we are done with 2, so we use the arrow → to declare that we need to move to 3 where
the process terminates since we cannot use 3 anymore because we have already chosen
2 members.

This produces the following sequence:

(→, 2, 2, →).

More generally, every combination with repetition with k elements produces a unique sequence
with n − 1 + k entries (n − 1 arrows and k objects). The main observation is that every such
sequence is uniquely determined by the location of the n − 1 arrows. Hence, it suffices to know
in how many ways we can place n −
1 arrows in n − 1 of the n − 1 + k entries of the sequence.
This can clearly be done in n−1+l
n−1 ways.

Permutations with repetition

34
 Consider the triplet (1, 1, 2). How many different re-arrangments (permutations) of this
triplet are there? Clearly we have the following 3 permutations
(1, 1, 2), (1, 2, 1), (2, 1, 1).
Note that we only obtained 3 permutations instead of 6, because it does not make any sense
to flip 1 and 1. In other words, we have 6! permutations modulo the ones that flip 1 and
1 we are 2! and hence we obtain 6!/2! = 3 permutations where two of the three objects are
repeated.
More generally, if we have n such that k1 are of the same type (that is the same repeated
object), k2 are of the same type, ..., kl are of the same type (hence l different types of objects),
such that
n = k1 + k2 + · · · + kl
then there are
n!
k1 ! · k2 ! · · · kl !
permutations with repetition.

6.1 Solutions to linear equations

Question 1: How many solutions (x1 , x2 , · · · , xn ) to the equation
x1 + x2 + · · · + xn = k
are there if
xi ∈ {0, 1}?
Answer: Clearly, k of the unknowns have to be equal to 1 and the remaining n − k of the
unknowns have to equal to 0. If we choose the k variables which will be equal to 1 then we
automatically choose the remaining n−k variables that will

be set equal to 0. Hence it suffices
to choose k of then n variables. This can be done in nk ways.
Question 2: How many solutions (x1 , x2 , · · · , xn ) to the equation
x1 + x2 + · · · + xn = k
are there if
xi ∈ {0, 1, 2, 3, ...., k}?
Answer: We have k units to distribute among the n unknowns. However, in contrast to the
above example, in this case we can distribute more than one unit to each variable. In other
words, we can choose each variable more than once. In fact, each variable can be chosen either
• 0 times (in which case is given the value 0) or
• 1 times (in which case is given the value 1) or
• 2 times (in which case is given the value 2) or
• ···
• k times (in which case is given the value k).
This means that we need to choose k of the given n variables but where we allow repetition,
that is we allow

each variable to be chosen more than once (and up to k times). Theren+k−1are in

total n+k−1
k combinations with k elements with repetition. Hence, there are in total k
different solutions to the above equation.

35
6.2 The Path Problem
The path problem
Consider all points (x, y) on the plane with integer coordinates. How many different paths
are there from (0, 0) to (n, n), if we can only move towards the right, that is

(x, y) → (x + 1, y)

and towards the top of the plane, that is

(x, y) → (x, y + 1)?

2n


The answer is n . Indeed, any such path is completely determined by the sequence of the
2n individual moves, n of which are towards the right (each such move is denoted by R) and
the other n moves are towards the top of the plane (each of which is denoted by T ). How
many different lists of length 2n are there if n of the elements of the lists are R and the other
n elements are T ? We clearly only need to specify the n locations where have the symbol R
in our lists. There are
2n positions and we need to specify (that is, to choose) n of them.
This can be done in 2n n ways.
Alternatively, we can say thatthe total number of paths is equal to all permutations with
repetition of n + m objects where n of them are identical (horizontal moves) and m of them
are also identical (vertical moves). According to the formula above, this is equal to
 
(n + n)! 2n
= .
n! · n! n

Restricted paths
Now suppose we want to consider all paths from (0, 0) to (n, n) which however satisfy one
additional restriction, namely that they do not intersect the line y = x + 1. This line will be
called the forbidden line. The paths that intersect the forbidden line are called bad paths,
and the paths that do not intersect the forbidden line are called good paths.
We will compute the number of bad paths first. Then the number of good paths is simply
equal to the number of all paths (given by the formula above) minus the number of all bad
paths.
Consider the point (n − 1, n + 1). This is the reflection of (n, n) across the forbidden line
y = x + 1. We will show that there is a bijection between all bad paths from (0, 0) to (n, n)
and all regular paths from (0, 0) to (n − 1, n + 1). Indeed, any path from (0, 0) to (n − 1, n + 1)
must intersect the forbidden line (why?). By refleting across the forbidden line the portion
of the path after its intersection with the forbidden line we obtain a bad path from (0, 0) to
(n, n). This process yields

a bijection (why?).
2n 2n


Since there are n−1 paths from (0, 0) to (n − 1, n + 1), there must be n−1 bad paths
from (0, 0) to (n, n).
Hence, there are in total
     
2n 2n 1 2n
cn = − =
n n−1 n+1 n

good paths. The natural numbers cn are known as the Catalan numbers.

36
6.3 Practice Problems
1. Compute how many (non-empty) subsets of the set {1, 2, 3, · · · , 10} satisfy the property
that the sum of their elements is even.

2. We are given a bin with 5 red balls, 5 green balls and 5 yellow balls. In how many ways
can we choose 5 balls from this bin?

3. How many permutations of the set T10 = {1, 2, ..., 10} are there which begin with an
odd number?

4. Consider the set Tn = {1, 2, · · · , n}. If we can choose, with repetition but without order,
2 numbers from the set Tn in 15 ways, then compute n.

5. Using the theory of linear equations, compute in how many ways can we choose four
non-consecutive numbers from the set T10 = {1, 2, · · · , 10}? For example, the following
four numbers {1, 2, 5, 7} cannot be selected because 1,2 are consecutive.

6. Using the theory of linear equations, compute in how many ways can we choose four
non-consecutive numbers from the set Tn = {1, 2, · · · , n}?

7. How many paths are there from (0, 0) to (n, n) such that we can only move up and right,
and such that the path lies entirely in the region of the plane above the diagonal y = x
(that is, the path lies in the region y ≥ x).

8. Imagine that there are 2n people in front of a box office. Each of n of these people holds
a 10 dollar bill and each of the other n people holds a 20 dollar bill. The ticket costs 10
dollars and at the beginning the box office has no cash. How many ways are there to
line up the 2n people if we want to make sure the box office never runs out of change?

9. A committee of 10 officials is to be chosen from 8 mathematicians and 7 physicists. If

the committee is to contain at least 4 mathematicians and at least 3 physicists, in how
many ways can it be chosen?

10. Find the number of all permutations (p1 , p2 , ..., p9 ) of (1, 2, ..., 9) such that

p1 > p2 > p3 > p4 < p5 < p6 < p7 < p8 < p9 .

11. How many combinations with repetitions for 4 numbers are there from the set {1, 2, 3, 4, 5, 6}?

12. In how many different ways can we roll a dice four times so that each time the roll is at
least as large as the roll preceding it?

7 Inclusion–Exclusion principle
7.1 Theory
Let |A| denote the cardinality of a set A. The inclusion-exclusion principle says that

|A ∪ B| = |A| + |B| − |A ∩ B|.

37
That is the number of the elements of the unionA∪B is equal to the number of the elements in
A plus the number of the elements in B minus the number of the elements that were already
counted twice (there are elements in the intersection of the sets). Similarly, we obtain

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

Hence, the inclusion-exclusion principle allows us to compute the cardinality of the union of
sets in terms of the cardinality of the intersection of sets. This is very convenient, since in
general it is easier to compute the cardinality of the intersection.

7.2 Solved Problems

Problem 1. How many natural numbers between 1 and 600 are there which are multiples
of 2 or 3?
Answer:
There are 600/2 = 300 multiples of 2. There are 600/3 = 200 multiplies of 3. And
there are 600/6 = 100 multiples of 2 and 3. Hence, by the inclusion-exclusion principle there
are 300 + 200 − 100 = 400 numbers which are multiples of 2 or 3. And clearly, there are
600 − 400 = 200 numbers which are neither a multiple of 2 nor of 3.
The above example is important because it is related to a fundamental function in number
theory, namely Euler’s φ function. For any natural number, φ(n) is equal to the number of
natural numbers m ≤ n which have no common prime divisor with n. We can produce a
formula for φ(n) using the inclusion-exclusion principle.

Problem 2. Compute φ(n) if we assume that

n = pk11 · pk22

Answer:
Since p1 , p2 are the prime divisors of n then we immediately obtain that there are n/p1
multiples of p1 , there are n/p1 p2 multiples of p2 and there are n/p1 p2 multiples of both p1
and p2 . Hence there are
n n n
+ −
p1 p2 p1 p2
numbers which are divisible by either p1 or p2 . Hence,
   
n n n 1 1
φ(n) = n − − + =n· 1− · 1− .
p1 p2 p1 p2 p1 p2

In a similar fashion we can produce a formula for general natural numbers n = pk11 pk22 · · · pkl l .

Problem 3. Compute the number of all possible rearrangements of the following sequence
of letters
AAABBBCCC
such that no three identical consecutive letters exist (that is no strings of the form AAA or
BBB or CCC exist in the rearrangment).
Answer:
9!
First of all, the number of all possible re-arrangements is 3!·3!·3! = 1680. Indeed, there are
in total 9 letters, but 3 of them are A, 3 are B and 3 are C (hence we need to use the formula

38
of permutations with repetition). Consider next the set RAAA of all re-arrangements that do
contain AAA. Define similarly RB and RC . We have
 
7 6!
|RAAA | = ·
1 3! · 3!
since there are 7 different locations for the position of AAA (why 7 and not 9?) and for each
6!
of these 7 locations of AAA we have 3!·3! re-arrangements of BBBCCC. Similarly,
 
7 6!
|RBBB | = |RCCC | = ·
1 3! · 3!
Furthermore,
|RAAA ∩ RBBB | = 20
since any re-arrangement that contains both AAA and BBB is in fact determined by uniquely
choosing the position of the first A and the B of these two strings. Indeed, having chosen the
first A then the other two A’s are automatically located and similarly for the B’s. Having
chosen the location for the strings AAA and the BBB, the location for the 3 C’s is uniquely
determined. In how many ways can we choose the first A and the first B of the strings AAA
and BBB? To answer this question consider 5 boxes. It is easy to show that there is bijection
between any ordered pair of these 5 boxes and a re-arrangement in RAAA ∩ RBBB (exercise).
Similarly,
|RBBB ∩ RCCC | = |RAAA ∩ RCCC | = 20.
Finally,
|RAAA ∩ RBBB ∩ RCCC | = 3!
since we simply have to commute the three strings AAA, BBB and CCC.
By the inclusion-exclusion principle we have
|RAAA ∪ RBBB ∪ RCCC | = 420 − 60 + 6 = 366.
Hence, the final answer is 1680 − 366 = 1314.

7.3 Practice Problems

1. Among 100 students, 70 study mathematics, 50 study physics, and 20 study both. How
many of these students study neither mathematics nor physics?
2. How many natural numbers are there between 1 and 2018 which are not multiples of 2
or 4 or 8 or of 5?
3. How many arrangements of the digits 0,1,...,9 are there that do not end with a 1 and
do not begin with a 9?
4. A derangement of (1, 2, 3, 4) is a permutation that moves every number away from its
correct position. For example (2, 4, 1, 3) is a derangement, but (2, 4, 3, 1) is not. How
many derangements of (1, 2, 3, 4) are there?
5. How many different re-arrangments of the word MATHEMATICS are there?
6. Compute the number of all possible re-arrangements of the letters TORONTO so that
there are no three consecutive identical letters.
7. Compute the number of all possible re-arrangements of the letters TORONTO so that
there are no two consecutive identical letters.

39
8 Recurrence Relations
8.1 Theory
Recurrence relations is a very powerful method in combinatorics to compute all the states of
a system that depends on n variables.
Let an denote the number of all different states that a system with n variables can take.
Instead of computing an directly, we assume that an−1 (and possibly an−2 ) are known and we
compute an in terms of an−1 (and an−2 if needed).
This technique is somehow the analogue of the inductive principle in number theory where
we assuming that a statement holds for k = n − 1 (and needed for k = n − 2) we show that
it holds for k = n.
Hence, we want to obtain a relation of the following form:

an = f (an−1 ) (3)

or
an = f (an−1 , an−2 ). (4)
Solving (3) requires knowing a1 and solving (4) requires knowing a1 , a2 .
There is a huge theory developed for solving recurrence equations of the form (3) and
(4). In this course, we will only consider basic equations that can be solved with elementary
methods. For example, it is important to recall the formulas for the geometric progressions
and the arithmetic progressions:
Arithmetic progression

an = an−1 + d ⇒ an = (n − 1)d + a1 .

The method of telescopic sums can be used to derive the above formula. Indeed, if we use
the equation
ak − ak−1 = d
for k = 2, 3, ..., n and we sum them then all but the first and last terms on the LHS will cancel
(telescopic sum) yielding
an − a1 = (n − 1)d.

Geometric progression
an = r · an−1 ⇒ an = rn−1 · a1 .
The method of telescopic product can be used to derive the above formula. Indeed, if we
use the equation
ak
=r
ak−1
for k = 2, 3, ..., n and we multiply them then all but the first and last terms on the LHS will
cancel (telescopic product) yielding
an
= rn−1 .
a1

40
8.2 Solved Problems
Problem 1. Compute the number of all subsets of the set

In = {1, 2, 3, ..., n} .

Answer.
Let an be the number of all subsets of In . Then clearly a1 = 2 since we have two subsets
in I1 , namely the empty set and the subset {1}. Similarly, we see that a2 = 4 and a3 = 8. We
will show that an = 2n for all n using the method of recurrent relations.
All subsets in In can be split in two categories:

• Category I consists of all subsets of In which contain n.

• Category II consists of all subsets of In which do not contain n.

How many subsets are there in Category I and how many in Category II?
Clearly, any subset in Category II is also a subset of In−1 (why?). Hence, there are an−1
subsets in Category II. Also, any subset in Category I is in fact the union of {n} and a subset
in Category II. This yields a bijection between the subsets in Category I and the subsets in
Category II. Hence, by the bijection principle, there are an−1 subsets in Category I. Hence,
by the additive principle we finally get

an = an−1

which with a1 = 2 yields that

an = 2n .

Problem 2. Consider an equilateral triangle OAB and add n points on each of the sides
OA, OB, AB so in total we have 3n new points on the three sides. Specifically, on OA we
consider the points P1 , P2 , ..., Pn such that

OP1 = P1 P2 = ... = Pn A.

Similarly, on OB we consider the points Q1 , Q2 , ..., Qn such that

OQ1 = Q1 Q2 = ... = Qn B

and on AB we consider the points T1 , T2 , ..., Tn such that

AT1 = T1 T2 = ... = Tn B.

We consider all the segments that connect the points Pi , Qi , Ti and we consider all the inter-
section points of these segments. Consecutive intersection points are connected by what we
call quantum segments.
How many paths (which cannot revisit the same vertex more than once) are there from O
to A if we can only move along quantum segments such that we can only move in the following
directions:
~
• parallel to the vector AB,

41
 ~
• parallel to the vector BA,
~
• parallel to the vector OA,
~
• parallel to the vector OB.
and if n = 6?
Answer.
Let an be the number we want to compute.
First observe that once a path has reached a point on the line AB then there is only one
way to continue to the path so it terminates at A. This implies that the number of paths
from O to A is the same as the number of paths from O to any other point on the line AB
(why?). This is a very crucial observation for this problem.
Next consider the line that connects the points Pn Qn . This is the last line L parallel to
AB. Any path from O to A must intersect the line L at one of its n + 1 points. Once a path
has reached its last point on the line L then there are only two ways to continue the path to
the point A (why?). Hence, it suffices to compute the number of all paths from O to any of
the n + 1 points of the line Pn Qn . Clearly there are
(n + 1) · an−1
such paths (why??). Hence,
an = 2(n + 1)an−1 .
There desired number a6 can be computed using the method of telescopic products (exercise).
Problem 3. How many different sets of n pairs can be formed from 2n people?
Answer.
Let an be the number we want to compute.
Consider any person. Then there are 2n − 1 pairs that can be formed with this one person.
After forming any of these pairs, we need to form n−1 pairs from the remaining 2n−2 people.
This can be done in an−1 ways. Hence,
an = (2n − 1)an−1 .
Clearly a1 = 1. Hence, by the method of telescopic products we obtain that an is equal to
the product of all odd numbers less than 2n. Indeed, by multiplying the following equations
an
=(2n − 1)
an−1
an−1
=(2n − 3)
an−2
an−2
=(2n − 5)
an−3
...
a3
=5
a2
a2
=3
a1
we obtain
an an−1 an−2 a3 a2
· · ··· · = (2n − 1) · (2n − 3) · (2n − 5) · · · 5 · 3 · 1
an−1 an−2 an−3 a2 a1

42
and hence (since a1 = 1) we obtain
n
Y
an = (2k − 1).
k=1

Now, an can be rewritten as follows

an =(2n − 1) · (2n − 3) · (2n − 5) · · · 5 · 3 · 1

(2n) · (2n − 1) · (2n − 2) · (2n − 3) · (2n − 4) · (2n − 5) · · · 5 · 4 · 3 · 2 · 1
=
(2n) · (2n − 2) · (2n − 4) · · · 4 · 2
(2n) · (2n − 1) · (2n − 2) · (2n − 3) · (2n − 4) · (2n − 5) · · · 5 · 4 · 3 · 2 · 1
=
2n · n · (n − 1) · (n − 2) · · · 2 · 1
(2n)!
= n .
2 · n!
That is,
(2n)!
an = .
n! · 2n

Problem 4. (Hanoi Towers) We are given three pegs A, B, C and n disks of graduated
size with holes in their centers. Initially the discs are at peg A such that any disc is always
on top of a bigger disk. We want to transfer all discs to another peg by moving one disk at a
time and without placing a larger disk on top of a smaller disk. What is the minimal number
of moves required to transfer the n disks?
Answer.
Let an be the number we want to compute. It takes an−1 moves to transfer the first (n−1)
disks from peg A to peg B. Then it takes one move to transfer the biggest disk from peg A
to peg C. And it takes again an−1 moves to transfer the n − 1 disks from peg B to peg C.
This is really the minimal number of moves (why?). Hence,

an = 2an−1 + 1.

Clearly, a1 = 1. To solve this recurrence relation, we add 1 to both sides and we obtain

(an + 1) = 2(an−1 + 1).

Hence, if we denote
xn = an + 1
then
xn = 2xn−1
with x1 = a1 + 1 = 2. Hence xn = 2n and hence

an = 2n − 1.

Problem 5. Consider an equilateral triangle OAB and add n points on each of the sides
OA, OB, AB so in total we have 3n new points on the three sides. Specifically, on OA we
consider the points P1 , P2 , ..., Pn such that

OP1 = P1 P2 = ... = Pn A.

43
Similarly, on OB we consider the points Q1 , Q2 , ..., Qn such that

OQ1 = Q1 Q2 = ... = Qn B

and on AB we consider the points T1 , T2 , ..., Tn such that

AT1 = T1 T2 = ... = Tn B.

How many parallelograms are formed by all segments that connect the points O, A, B, Pi , Qi , Ti ?
Answer.
We call nodes the intersections of the segments. Clearly there are
1
1 + 2 + · · · + (n + 1) + (n + 2) = (n + 2)(n + 3)
2
nodes.
Let an be the number we want to compute. All parallelograms can be divided in two
classes:
• Type 1 consists of all parallelograms which do not have any vectices on AB,

• Type 2 consists of all parallelograms which have at least one vertex on AB.
Clearly, there are an−1 parallelograms of type 1. We will next compute the number of all
parallelograms of type 2. For we will establish a bijection between all parallelograms of type
2 and specific pairs (u, v) of vertices such that v is one of the n + 2 nodes on the line AB.
The bijection is the following: Any parallelogram has two opposite angles equal to 60 degrees
and two opposite angles equal to 120 degrees. We associate any parallelogram to the pair of
vertices of the angles which are equal to 60 degrees. This association is a bijection (why??).
Any pair of vertices that comes from a parallelogram in the above way is called admissible.
Clearly, if (u, v) is an admissible pair of vertices then u lies in one of the three planar sections
(angles) that are created by the three lines that pass through v and each of which has angle
60 degrees. This immediately implies that u cannot line on any of the lines that pass through
v.
We will next compute all admissible pairs (u, v) such that v ∈ AB. We have n + 2 nodes
on AB. For each of these n + 2 nodes (call it v), we need to compute the total number of
admissible nodes, that is the total number of nodes which do not lie on any of the three lines
that pass through the node v ∈ AB. Clearly, one of these three lines is the line AB which in
total contains n + 2 nodes. The number of the nodes on the other two lines passing through
v, and excluding v since it has already been counted in the nodes of AB, is equal to n + 1
(why??).
Therefore, there are
1 n(n + 1)
(n + 2)(n + 3) − (n + 2) − (n + 1) = .
2 2
admissible pairs for the node v. There are n + 2 modes v on AB hence in total we have
 
n(n + 1)(n + 2) n+2
=3
2 3
admissible pairs, and hence parallelograms of type 2.

44
 We have  
n+2
an = an−1 + 3 .
3
Also, clearly a0 = 0, a1 = 3. By the telescopic sums method we obtain
n  
X k+2
an − a0 =
3
k=0

and hence, by the bonus problem of assignment 3, we obtain

 
n+3
an = 3
4
.
Remark: For an alternative, more direct, proof, please see:
http://www.laurentlessard.com/bookproofs/counting-parallelograms/

8.3 Practice Problems

1. Solve the recurrence relation
a0 = 1,
an+1 = an + n.

2. Using recurrence relations, compute in how many ways we can choose one (yes one!)
number from the set Tn = {1, 2, · · · , n}?

3. Using recurrence relations, compute in how many ways we can choose two non-consecutive
numbers from the set Tn = {1, 2, · · · , n}?

4. Using recurrence relations, compute in how many ways we can choose three non-consecutive
numbers from the set Tn = {1, 2, · · · , n}?

5. Using recurrence relations, compute in how many ways we can choose four non-consecutive
numbers from the set Tn = {1, 2, · · · , n}?

6. We call an ordered n-tuple (a1 , a2 , · · · , an ) “good” if

1) for all i = 1, 2, · · · , n we have ai ∈ {0, 1, 2}, and
2) the sum a1 + a2 + · · · + an is even.

Let An denote the number of all possible “good” n-tuples. Compute An as a function
of n.

7. Solve the recurrence relation

a0 = 1, a1 = 1
an+1 = 2an − an−1 .

8. Solve the recurrence relation

a0 = 1, a1 = 2
an+1 = 2an − an−1 .

45
 9. Solve the recurrence relation
a0 = 1, a1 = 6
an+1 = 6an − 9an−1 .

10. Find a recurrence relation for the number of strings with 1s and 0s of length n that do
not have two consecutive 0s.

11. (*) Obtain a recurrence relation for the Catalan numbers Cn . Recall that Cn is the
number of paths from (0, 0) to (n, n) which do not cross or intersect the line y = x + 1.

9 Generating Functions
9.1 Theory
The method of generating functions is a very powerful tool in combinatorics developed by Euler
in 1748. The main idea is to turn combinatorial considerations into algebraic manipulations.
The latter algebrain operations can in principle be done by a computer system.
In most problems in combinatorics, given any natural number n, we want to compute the
number an of all possible ways to perform a given process that depends on n. The method
of generating functions provides a new way to compute an by first computing the so-called
associated generating function of an :
∞
X
gan (x) = a0 + a1 x + a2 x2 + · · · an xn + ... = ai xi .
i=0

Clearly we have
gan (x) = gbn (x) if and only if an = bn for all n ∈ N.
We will not consider issues of convergence here. We will assume that x takes values such that
the powerseries (i.e. the generating functions) are finite without worrying about the exact
convergence interval.
The main idea of the method is to compute an by first computing the associated generating
function gan (x). Once the generating function has been found, then the sequence is trivially
computed since it is simply given by the coefficients of the monomials xn for all n.
Algebraic operations of generating functions
1. Sum of generating functions:

gan (x) + gbn (x) = gcn (x),

where
cn = an + bn .
2. Product of generating functions:

gan (x) · gbn (x) = gcn (x),

where
cn = a0 · bn + a1 · bn−1 + a2 · bn−2 + · · · + an · b0 .

46
The sequence cn is known as the Cauchy product or the convolution of the sequences an , bn .
Special examples of generating functions
1. The most important example of a generating function is the one associated to the
constant sequence
an = 1
that is
1
gan (x) = 1 + x + x2 + x3 + · · · = . (5)
1−x
Note that this formula follows from the following identity
1 − xn+1
1 + x + x2 + · · · xn =
1−x
and using that xn → 0 as x → ∞. Furthermore, if

an = b

then
b
gan (x) = .
1−x
On the other hand, if
an = bn
then
1
gan (x) = .
1 − bx
2. If ak = 1 for k = n and ak = 0 for all k 6= n (for some fixed n) then

gan (x) = xn .

3. If  
n
ak = for k ≤ n, and ak = 0 for k > n
k
then
gan (x) = (1 + x)n .
The above follows immediately by the binomial identity.
Differentiation of generating functions
Recall from calculus the following identity
 
d 1 1
=
dx 1 − x (1 − x)2
On the other hand, by differentiating (9) we obtain
 
1 d 1
= = 1 + 2x + 3x2 + · · · + (n + 1)xn + · · · . (6)
(1 − x)2 dx 1 − x
1
Hence, (1−x)2
is the generating function for the sequence an = n + 1. Moreover,
 
d 1 2
2
= .
dx (1 − x) (1 − x)3

47
Moreover, by differentiating (10) we obtain
 
1 1 d 1 (n + 1)(n + 2) n
3
= 2
= 1 + 3x + · · · + x + ···
(1 − x) 2 dx (1 − x) 2

1 (n+1)(n+2)
Hence, (1−x)3 is the generating function for the sequence an = 2 .
Exercise: Compute the generating functions of the sequences an = n and bn = n2 .

Combinatorics and generating functions

In this section we will consider problems where we need to select a certain number of
elements (usually we want to select n elements) from a certain set of elements such that each
of the elements satisfies given restrictions.
A brief general description of the method is as follows (note that the method will become
more clear by reading the examples that follow):
Step 1.
We introduce a variable x which indicates if an element has been chosen. Specifically,

• x0 indicates that an element has not been selected,

• x1 indicates that an element has been selected 1 time,

• x2 indicates that an element has been selected 2 times,

• ....

• xn indicates that an element has been selected n times.

Step 2.
For each of the elements of the set we construct an associated powerseries, that is, sums
of those powers of x meet the restrictions of the element.
Step 3.
We consider the product of all the powerseries of Step 2. This product gives the generating
function associated to the problem. The coefficient of xn in the generating function gives the
desired sequence an .
The above will become more clear in the examples that follow:

Example 1.
Consider the set
S = {a, b, c}
such that

• a can be selected only an even number of times,

• b can be selected only an odd number of times,

• c can be selected at most 10 times.

48
Find the number of ways that we can select n elements by repeatedly selecting elements from
S so that the above restrictions are met.
Answer: Let an be the number of ways of the desired selection. We will first compute the
generating function gan (x).
Step 1:
We introduce the variable x which indicates if an elements has been selected.
Step 2:
For the element a we have the following considerations:

• x0 indicates that a has not been selected (i.e. selected 0 times, and 0 is even),

• x2 indicates that a has been selected 2 times,

• x4 indicates that a has been selected 4 times,

• ....

• x2n indicates that a has been selected 2n times,

• ....

Hence, the associated powerseries for the element a is

1 + x2 + x4 + · · · + x2n + · · ·

For the element b we have the following considerations:

• x1 indicates that b has been selected 1 time,

• x3 indicates that b has been selected 3 times,

• x5 indicates that b has been selected 5 times,

• ....

• x2n+1 indicates that b has been selected 2n + 1 times,

• ....

Hence, the associated powerseries for the element b is

1 + x3 + x5 + · · · + x2n+1 + · · ·

For the element c we have the following considerations:

• x0 indicates that c has not been selected,

• x1 indicates that c has been selected 1 time,

• x2 indicates that c has been selected 2 times,

• ....

• x10 indicates that c has been selected 10 times.

49
Hence, the associated powerseries for the element c is

1 + x + x2 + x3 + · · · + x10 .

Step 3.
Finally we obtain the generating function by multiplying the powerseries associated to
each of the variables a, b, c:

gan (x) = (1+x2 +x4 +· · ·+x2n +· · · )·(1+x3 +x5 +· · ·+x2n+1 +· · · )·(1+x+x2 +x3 +· · ·+x10 ).

The coefficient of xn in the above expression gives the desired number an . Note that if n is
large, say n = 1000 then it is difficult to find the value an by hand, however it is always easy
for a computer to quickly determine the value of an .

Example 2.
In how many ways can we select 4 elements from the set

S = {a, b, c, d}

such that

• a can be selected at most 2 times,

• b can be selected at most once,

• c can be selected at most 2 times,

• d can be selected at most once.

Answer: We will first compute the generating function gan (x). Then the answer will be
equal to the coefficient of x4 .
Step 1:
We introduce the variable x which indicates if an elements has been selected.
Step 2:
For the element a we have the following considerations:

• x0 indicates that a has not been selected (i.e. selected 0 times, and 0 is even),

• x1 indicates that a has been selected once,

• x2 indicates that a has been selected 2 times,

Hence, the associated powerseries for the element a is

1 + x + x2

For the element b we have the following considerations:

• 1 = x0 indicates that b has been selected 0 times,

• x1 indicates that b has been selected 1 time.

50
Hence, the associated powerseries for the element b is
1+x
For the element c we have the following considerations:
• x0 indicates that c has not been selected,
• x1 indicates that c has been selected 1 time,
• x2 indicates that c has been selected 2 times,
Hence, the associated powerseries for the element c is
1 + x + x2
For the element d we have the following considerations:
• 1 = x0 indicates that d has been selected 0 times,
• x1 indicates that d has been selected 1 time.
Hence, the associated powerseries for the element d is
1+x
Step 3.
Finally we obtain the generating function by multiplying the powerseries associated to
each of the variables a, b, c, d:
gan (x) = (1 + x + x2 ) · (1 + x) · (1 + x + x2 ) · (1 + x)
The coefficient of x4 in the above expression gives the desired number.

9.2 Solved Problems

Problem 1. (Solutions to linear equations with restrictions) Find the number of all
natural solutions (x1 , x2 , x3 , x4 ) to the linear equation
x1 + x2 + 2x3 + 3x4 = n
if
• x1 ≥ 2,
• x2 takes only even values,
• x3 ≥ 0,
• x4 ≥ 3.

Answer: We will first compute the generating function gan (x). Then the answer will be
equal to the coefficient of xn .
Step 1:
We introduce the variable x which represent a unit. In total we need to select (i.e. dis-
tribute) n units.
Step 2:
For the element x1 we have the following considerations:

51
 • x2 indicates that x1 has been given 2 units (that is x1 = 2),

• x3 indicates that x1 = 3,

• ...

Hence, the associated powerseries for the element x1 is

x2 + x3 + x4 + · · · =x2 · (1 + x + x2 + · · · )
1
=x2 · .
1−x
For the element x2 we have the following considerations:

• 1 = x0 indicates that x2 = 0,

• x2 indicates that x2 = 2,

• ...

• x2n indicates that x2 = 2n (even),

• ...

Hence, the associated powerseries for the element x2 is

1 + x2 + x4 + · · · + x2n + · · · =1 + (x2 ) + (x2 )2 + · · · + (x2 )n + · · ·

1
= .
1 − x2
For the element x3 we have the following considerations: It is the value 2x3 that we need to
consider. Since x3 ≥ 0 we have that 2x3 takes any even value greater or equal to 0. Hence,

• 1 = x0 indicates that 2x3 = 0,

• x2 indicates that 2x3 = 2,

• ...

• x2n indicates that 2x3 = 2n (even),

• ...

Hence, the associated powerseries for the element 2x3 is

1 + x2 + x4 + · · · + x2n + · · · =1 + (x2 ) + (x2 )2 + · · · + (x2 )n + · · ·

1
= .
1 − x2
For the element x4 we have the following considerations:It is the value 3x4 that we need to
consider. Since x4 ≥ 3 we have that 3x4 takes any value greater or equal to 9 which is a
multiple of 3. Hence,

• x9 indicates that 3x4 = 9,

• x12 indicates that 3x4 = 12,

52
 • ...

• x3n indicates that 3x4 = 3n (multiple of 3),

• ...

Hence, the associated powerseries for the element d is

x9 + x12 + · · · + x3n + · · · =x9 · (1 + x3 + x6 + · · · + x3n + · · ·

1
=x9 · .
1 − x3
Step 3.
Finally we obtain the generating function by multiplying the powerseries associated to
each of the variables x1 , x2 , 2x3 , 3x4 :
1 1 1 1 1 1 1
gan (x) = x2 · · 2
· 2
· x9 · 3
= x11 · · · .
1−x 1−x 1−x 1−x 1 − x (1 − x ) 1 − x3
2 2

The coefficient of xn in the above expression gives the desired number.

Problem 2. In how many ways can we toss a dice three times and get a total of 14?

Answer: We toss the dice three times and hence we have three results to consider (the
sum of which has to be 14). Each individual result is a natural number between 1 and 6.
Hence if x indicates a unit, then xk indicates that the result of a toss if exactly k. In total we
need to have 14 units so we need to compute the coefficient of x14 in the associated generating
function.
The powerseries corresponding to each die toss is

x + x2 + x3 + x4 + x5 + x6

and since we we toss the dice three time the final generating function is

(x + x2 + x3 + x4 + x5 + x6 )3 .

The coefficient of x14 in the above function is the desired number (it can be easily computed
by a computer).

Problem 3. Compute the generating function of an if an is the number of ways to select

n elements from the set
S = {e1 , e2 , · · · , em }
such that the element ei can be selected at most ki times.
Answer: For any i = 1, 2, ..., m, the element ei can be selected at most ki times and hence
its associated powerseries is
1 + x + x2 + · · · + xki .
The final generating function is simply the product of all the powerseries of all elements ei :
m
Y
gan (x) = (1 + x + x2 + · · · + xki ).
i=1

53
The coefficient of xn gives the desired answer.

Problem 4. Find the number of ways to select 2n balls from n identical red balls, n
identical blue balls and n identical white balls.
Answer: We will use the method of generating functions. The associated powerseries of
each of the three types of balls (ie. red, blue and white balls) is

1 − xn+1
1 + x + x2 + · · · + xn = .
1−x
Hence, the generating function is

(1 − xn )3
gan (x) = .
(1 − x)3

Note that

(1 − xn+1 )3 = 1 − 3xn+1 + 3(xn+1 )2 − (xn+1 )3 = 1 − 3xn+1 + 3x2n+2 − x3n+3

and recall from Lecture notes 20–21 that

1 (n + 1)(n + 2) n
3
= 1 + 3x + · · · + x + ···
(1 − x) 2

Hence, we want to compute the coefficient of x2n in the product

 
n+1 2n+2 3n+3 (n + 1)(n + 2) n
g = (1 − 3x + 3x −x ) · 1 + 3x + · · · + x + ···
2

The answer is given by

(2n + 1)(2n + 2) n(n + 1) (n + 1)(n + 2)

−3 = .
2 2 2
Remark: The problem is equivalent to finding the number of all natural solutions (x1 , x2 , x3 )
of the linear equation
x1 + x2 + x3 = 2n
such that 0 ≤ xi ≤ n for i = 1, 2, 3.

Problem 5. Consider 5 boxes in a row.

• Let an denote the number of ways that we can distribute n identical objects into these
5 boxes such that the first, the third and the fifth boxes are always non-emptry (that
is, they all contain at each one objects).

• Let bn denote the number of ways that we can distribute n identical objects into these
5 boxes such that the second and the fourth boxes contain at least two objects.

Prove that an = bn+1 for all n.

54
Proof. We will use the method of generating functions (it is left as an exercise to the students
to find alternative solutions which do not make use of generating functions).
We will compute the generating functions gan and gbn .
Let xi denote the number of objects we put in the ith box. We need to find the number
of solutions (x1 , x2 , x3 , x4 , x5 ) of the linear equation

x1 + x2 + x3 + x4 + x5 = n

with additional restrictions on xi .

We start with gan . By assumption, we need to have

x1 ≥ 1, x3 ≥ 1, x5 ≥ 1.

Hence the powerseries associated to each of the variables x1 , x3 , x5 is

x
x + x2 + · · · + xn + · · · = x · (1 + x + x2 + · · · xn + · · · ) = .
1−x
The only restrictions for the variables x2 , x4 are simply

x2 ≥ 0, x4 ≥ 0.

Hence the powerseries associated to each of the variables x2 , x4 is

1
1 + x + x2 + · · · xn + · · · = .
1−x
Hence, the generating function gan of an is

x 1 x 1 x x3
gan (x) = · · · · = . (7)
1−x 1−x 1−x 1−x 1−x (1 − x)5

We next compute gbn . By assumption, we need to have

x2 ≥ 2, x4 ≥ 2.

Hence the powerseries associated to each of the variables x2 , x4 is

x2
x2 + x3 + · · · + xn + · · · = x2 · (1 + x + x2 + · · · xn + · · · ) = .
1−x
The only restrictions for the variables x1 , x3 , x5 are simply

x1 ≥ 0, x3 ≥ 3, x5 ≥ 0.

Hence the powerseries associated to each of the variables x1 , x3 , x5 is

1
1 + x + x2 + · · · xn + · · · = .
1−x
Hence, the generating function gbn of bn is

1 x2 1 x2 1 x4
gbn (x) = · · · · = . (8)
1−x 1−x 1−x 1−x 1−x (1 − x)5

55
Hence, from (7) and (8) it immediately follows that

gbn (x) = x · gan (x)

We will show that the above relation of the two generating functions implies the desired
identity an = bn+1 . We have

gbn (x) =x · gan (x) = x · (a0 + a1 x + a2 x2 + · · · an xn · · · )

(9)
=a0 x + a1 x2 + a2 x3 + · · · + an xn+1 · · ·

On the other hand, we also have (by the definition of generating functions)

gbn (x) = b0 x + b1 x + b2 x2 + · · · + bn xn + bn+1 xn+1 + · · · (10)

Therefore, from (9) we have the coefficient of xn+1 in gbn (x) is an and from (10) we have the
coefficient of xn+1 in gbn (x) is bn+1 . Hence, we must have

an = bn+1

for all n.

Note that an is the left shift of bn . Indeed, if we shift every entry of bn towards the left
then the new nth entry is the old (n + 1)th entry which was shifted to the left. Hence, indeed,
an = bn+1 is the left shift of bn .
The proof above shows that shifting the sequence to left amounts to multiplying the gen-
erating function by x1 2 and similarly shifting the sequence to the right amounts to multiplying
the generating function by x. These observations are very important since they can be used
to solve recurrence relations which provide a relation between a sequence an and its shifted
sequences an+1 or an−1 .

9.3 Applications in recurrence relations

Let’s first show how to use generating functions to solve the following general recurrent rela-
tion:
an = r · an−1 + f (n) (11)
for all n ≥ 1, where r ∈ R and f (n) is a given function of n (that is a given sequence). We
multiply (11) by xn and we add over all n = 1, 2, 3, ... to obtain
X X X
an xn = r · an−1 xn + f (n)xn
n≥1 n≥1 n≥1

Therefore, X X X
an xn − a0 = rx · an−1 xn−1 + f (n)xn
n≥0 n−1≥0 n≥1

and so X X X
an xn − a0 = rx · an xn + f (n)xn
n≥0 n≥0 n≥1

2
Note that we assume that a−1 = b0 = 0.

56
Hence X
gan (x) − a0 = rx · gan (x) + f (n)xn (12)
n≥1

We can easily solve (12) with respect to gan to obtain

a0 + n≥1 f (n)xn
P
gan (x) = . (13)
1 − rx
Hence, (13) provides the generating function of an . Therefore, the sequence an can be com-
puted. Note that we need to know a priori the value of a0 . The following result is very useful
in computing the actual an from (13):
1
= 1 + rx + r2 x2 + · · · rn xn + · · · = gbn =rn (x)
1 − rx
Concluding, an is the convolution of the sequences bn = rn and the sequence whose
generating function is x · gf (n) (x) + a0 that is of the sequence

(a0 , f (1), f (2), f (3), ....)

Example 1.
Solve
an = 2an−1 + 2n
for all n ≥ 1 with a0 = 1.
Answer: We multiply both sides with xn and we add over all n ≥ 1 to obtain
X X X
an xn = 2 an−1 xn + 2n xn
n≥1 n≥1 n≥1

Hence X X X
an xn − a0 = 2x an−1 xn−1 + 2n xn
n≥0 n≥1 n≥1

Therefore, X
gan − a0 = 2xgan + 2n xn
n≥1

Hence we reached to
n n n n
P P
a0 + n≥1 2 x 1+ n≥1 2 x 1
gan (x) = = =
1 − 2x 1 − 2x (1 − 2x)2

Hence, we obtain that (why?)

an = (n + 1) · 2n .

Example 2. (Fibonacci sequence)

Solve the Fibonacci relation
Fn = Fn−1 + Fn−2
for all n ≥ 2 with F0 = 0, F1 = 1.
Answer:

57
 We multiply the Fibonacci relation with xn and add over all n ≥ 2 to obtain:
X X X
Fn xn = Fn−1 xn + Fn−2 xn
n≥2 n≥2 n≥2

Hence, X X X
Fn xn − F0 − F1 x = x Fn−1 xn−1 + x2 Fn−2 xn−2
n≥0 n≥2 n≥2
and so X X X
Fn xn − F0 − F1 x = x Fn−1 xn−1 + x2 Fn−2 xn−2
n≥0 n−1≥1 n−2≥0
which yields X X X
Fn xn − F0 − F1 x = x Fk xk + x2 Fk xk
n≥0 k≥1 k≥0
Thus  
gFn − F0 − F1 x = x · gFn − F0 + x2 gFn
Therefore,
F0 + F1 x − F0 x x
gFn = 2
= .
1−x−x 1 − x − x2
It is easily verified that
A B
gFn =
+
1 − ax 1 − bx
where √ √
1 1 1+ 5 1− 5
A = √ , B = −√ , a = ,b = .
5 5 2 2
This implies that (why?)
√ !n √ !n
1 1+ 5 1 1− 5
Fn = √ · −√ · .
5 2 5 2

9.4 Practice Problems

1. Compute the generating function of an , where an denotes the number of solutions to
the equation
x1 + x2 + · · · + x7 = n
with xi ∈ N such that xi ≥ 1. Show that the coefficient of x18 in
(x + x2 + x3 + · · · )7
is  
17
.
11
2. Compute the generating functions of the number kn of partitions of n which contain
only perfect squares. For example, 25 = 3 + 16 is such a partition of 25.
3. Explain how the generating function of an , n ≥ 0 is related with the generating function
of the sequences
• bn = an+2 , n ≥ 0,
• cn = an−2 , n ≥ 2 and c0 = c1 = 0.

58
10 Partitions of Natural Numbers
10.1 Theory
We next present a very nice application of generating functions in Number Theory. We will
consider the problem of counting all partitions of natural numbers.
A partition of a positive integer n is a collection of positive integers with sum equal to n.
For example,
3=2+1
is a partition of 3. The precise order of the positive integers does not matter. Hence,

3=2+1

and
3=1+2
represent the same partition of 3. Hence, since order does not matter, it is useful to write the
summands in increasing order. For example, we have the following partitions:

1=1
2=1+1
2=2
3=1+1+1
3=1+2
3=3
4=1+1+1+1
4=1+1+2
4=1+3
4=2+2
4=4

Hence, we see that 1 has 1 partition, 2 has 2 partitions, 3 has three partitions, 4 has 5
partitions.
In general we denote by pn the number of different partitions of a natural number n. Any
partition is completely determined by the number of 1’s present, the of 2’s present, the number
of 3’s present, ..., the number of n’s present. For example for the following partition of 35

35 = 1 + 2 + 3 + 3 + 5 + 7 + 7 + 7

we have that

• 1 is present 1 time

• 2 is present 1 time

• 3 is present 2 times,

• 4 is present 0 times,

• 5 is present 1 time,

59
 • 6 is present 0 times,

• 7 is present 3 times,

• 8 is present 0 times,

• all natural numbers greater of equal to 9 are present 0 times.

Therefore, we can rewrite this partition of 35 as follows

35 = 1 · 1 + 2 · 1 + 3 · 2 + 4 · 0 + 5 · 1 + 6 · 0 + 7 · 3 + 8 · 0 + .... + k · 0 + ...

More generally, any partition of n corresponds to a solution (x1 , x2 , ..., xn ) of the equation

n = x1 + 2x2 + 3x3 + 4x4 + 5x5 + 6x6 + 7x7 + 8x8 + · · · + ixi + · · ·

where xi denotes the number of i’s present in the partition, and hence xi ≥ 0.
Problem 1.
Compute the generating function gpn (x) of the number pn of all partitions of n.
Answer: It suffices to compute the generating function of the number of all solutions
(x1 , x2 , ..., xn ) of the equation

n = x1 + 2x2 + 3x3 + 4x4 + 5x5 + 6x6 + 7x7 + 8x8 + · · · + ixi + · · ·

Let x correspond to a unit. The variable xi contributes ixi units in the partition of n since xi
denotes the number of i’s present in the partition of n. Hence, xi contributes either 0 units,
or i units, or 2i units, etc. Therefore, the associated powerseries to the variable xi is

1 + xi + (xi )2 + (xi )3 + ...

Hence the generating function of pn is equal to the product of all the associated powerseries
and hence equal to

gpn (x) = 1+x+x2 +· · · · 1+x2 +(x2 )2 +· · · · 1+x3 +(x3 )2 +· · · ·...· 1+xi +(xi )2 +· · · · · ·

The ith factor provides all the possibilities for the contribution of the number i in the partitions
of n. Clearly we have
1 1 1 1
gpn (x) = · 2
· 3
··· · ···
1−x 1−x 1−x 1 − xi
therefore
∞
Y 1
gpn (x) = .
1 − xk
k=1

60
10.2 Partitions with restrictions
The theory of generating functions provides neat solutions to partition problems with addi-
tional restrictions.
Problem 1.
Derive the generating function of the number of partitions of n which consist of the number
of 1,2 and 3 only.
Answer: We need to find the number of all solutions (x1 , x2 , x3 ) to the equation

x1 + 2x2 + 3x3 = n

with xi ≥ 0. The powerseries associated to x1 is

1
1 + x + x2 + · · · = .
1−x
The powerseries associated to x2 is
1
1 + x2 + (x2 )2 + · · · = .
1 − x2
The powerseries associated to x3 is
1
1 + x3 + (x3 )2 + · · · = .
1 − x3
Hence, the desired generating function is
1 1 1
· · .
1 − x 1 − x 1 − x3
2

Problem 2.
Derive the generating function of the number of partitions of n which consist of distinct
numbers (that it, each number cannot appear more than once).
Answer: We need to find the number of all solutions (x1 , x2 , x3 , ..., xi , ...) to the equation

x1 + 2x2 + 3x3 + ...ixi + ... = n

with 0 ≤ xi ≤ 1 since each number can appear at most once. The powerseries associated to
x1 is
1 + x.
The powerseries associated to x2 is
1 + x2 .
More generally, the powerseries associated to xi is

1 + xi .

Hence, the desired generating function is

∞
Y
2 3
(1 + x) · (1 + x ) · (1 + x ) · · · = (1 + xk ).
k=1

61
 Problem 3.
Derive the generating function of the number of partitions of n which consist of odd
numbers.
Answer: We need to find the number of all solutions (x1 , x3 , x5 , ..., x2k+1 , ...) to the equa-
tion
x1 + 3x3 + ... + (2k + 1)x2k+1 + ... = n
with x2k+1 ≥ 0. The powerseries associated to x1 is
1
1 + x + x2 + · · · =
1−x
The powerseries associated to x3 is
1
1 + x3 + (x3 )2 · · · =
1 − x3
More generally, the powerseries associated to x2k+1 is
1
1 + x2k+1 + (x2k+1 )2 + · · · =
1 − x2k+1
Hence, the desired generating function is
∞
1 1 1 Y 1
· 3
· 5
· · · = .
1−x 1−x 1−x (1 − x2k+1 )
k=0

Let’s consider the partitions of 6 in problems 2 and 3.

The partitions of 6 with distinct numbers are the following

6=1+5
6=2+4
6=1+2+3
6=6

Hence, there are 4 partitions of 6 with distinct numbers.

Let’s consider now the partitions of 6 with odd numbers. We have the following partitions

6=1+1+1+1+1+1
6=1+1+1+3
6=3+3
6=1+5

Hence, there are 4 partitions of 6 with odd numbers.

The fact that we have 4 partitions of 6 with distinct number and 4 partitions of 6 with
odd number is not a coincidence. See the following subsection.

62
10.3 Euler’s Theorem
One of the most important results in partition theory is due to Euler

Theorem 1 (Euler). The number of partitions of any natural number n with distinct numbers
is equal to the number of partitions of n with odd numbers.

Proof. Let dn denote the number of partitions of n with distinct numbers and let on partitions
of n with odd numbers.
It suffices to prove that the generating functions gdn , gon of the corresponding sequences
are equal. By problems 2 and 3 above we have
∞
Y
2 3
gdn (x) = (1 + x) · (1 + x ) · (1 + x ) · · · = (1 + xk )
k=1

and
∞
1 1 1 Y 1
gon (x) = · 3
· 5
· · · = .
1−x 1−x 1−x (1 − x2k+1 )
k=0

Using the identity

1 − a2
1+a=
1−a
we obtain
gdn (x) =(1 + x) · (1 + x2 ) · (1 + x3 ) · (1 + x4 ) · (1 + x5 ) · · ·
1 − x2 1 − x4 1 − x6 1 − x8 1 − x10
= · · · · ···
1 − x 1 − x2 1 − x3 1 − x4 1 − x5
1 1 1
= · · ···
1 − x 1 − x3 1 − x5
=gon (x),

where we observed that the numerator and the denumerator contain all terms of the form
(1 − x2k ) and hence all these terms cancel out.
Since the generating functions are equal we obtain dn = on for all n.

63