1.

Counting Principles
Counting means establishment of an one-one and onto function µ : S → [n], where S is a non-empty
set [n] = {1, 2, . . . , n}. We denote |S| = n. We also have the convention that if S is an empty set, i.e. S
does not contain any element, then we denote |S| = 0.
Broadly, Combinatorics is the subject which studies “this” non empty set S.
Question. How many bones are there in a healthy adult homo sapiens?
Let S be the set of bones. After “studying” the set S, it is concluded that |S| = 206.
Our “study” consists of following two observations. The first one is there is an axis of symmetry, which
implies the number of bones in the “left side” equals the number of bones in the “right side”. The second
one is the axis of symmetry contains bones of head, chest (partly), spinal cord and pelvis (partly).
Definition. Let S be a non-empty set. A partition of S into k many parts is a collection {B1 , . . . , Bk },
k
where for each i, j ∈ [k], Bi ⊂ S; Bi ∩ Bj = ∅ if and only if i 6= j and ∪ Bi = S.
i=1

Definition. If a finite set S is partitioned into k many parts say {B1 , B2 , . . . , Bk }. Then the addition
principle of counting is
k
X
|S| = |B1 | + |B2 | + . . . + |Bk | = |Bi |.
i=1
Example. If the first box contains m many objects and the second box contains n many objects, then
the number of ways of choosing one object from either of the two boxes is m + n.
Definition. Let S ⊆ T be a finite set. We define S̄ := T r S, the complement of S in T . Then the
subtraction principle of counting is
|S̄| = |T | − |S|
Example. Let T is the set of students studying at NISER and S the set of students studying neither
Mathematics nor Physics. If we know |T | = 2000 and |S| = 1800, then we can compute the number |S̄|,
i.e the number of students studies either Mathematics or Physics is |S̄| = |T | − |S| = 2000 − 1800 = 200.
Definition. Let the finite set S = B1 × B2 × . . . × Bk = {(x1 , x2 , . . . , xk ) : xi ∈ Bi ∀ i ∈ [k]}, then the
multiplication principle of counting is
k
Y
|S| = |B1 | × |B2 | × . . . × |Bk | = |Bi |.
i=1
Example. On assuming, any registered vehicle from Odisha contains a number of the form OD02AC6234,
what is the maximum possible vehicle can be registered by Government of Odisha?
Definition. Let S and R be two sets. If there exist a bijective (i.e. one-one and onto) mapping f : S → R,
then the bijective principle of counting states that |S| = |R|.
Example. We consider the following three examples.
(a) S = {a, b, c} and R = {1, 2, 3} = [3], where a 7→ 3, b 7→ 1 and c 7→ 2.
(b) For positive integers m and n. Let R = {r ∈ N : 1 ≤ r ≤ mn, gcd (r, mn) = 1} and
S = {(a, b) ∈ N × N : 1 ≤ a ≤ m, 1 ≤ b ≤ n, gcd (a, m) = 1 = gcd (b, n)} There exists a bijective
mapping between R and S. This establishes |R| = φ(mn) = φ(m)φ(n) = |S|. Here φ(n) denotes the
cardinality (size) of the set {r ∈ N : 1 ≤ r ≤ n, gcd (r, n) = 1}.
(c) Let A = {x1 , x2 , . . . , xn } be a set of size n (in short an n−set) and 2A denotes the set of all subsets
of A and {0, 1}n = {(1 , 2 , . . . , n ) : i ∈ {0, 1}∀i ∈ [n]}. We take S = 2A and R = {0, 1}n . We map
{xi1 , xi2 , . . . , xik } to (1 , 2 , . . . , n ), where

1 if j = i1 , j = i2 , . . . , j = ik
j =
0 otherwise
We call such map as θ, i.e. θ : 2A → {0, 1}n . Note that θ is an one-one and onto function. (Exercise!)
This implies |2A | = |{0, 1}|n = 2n .
1
2

Definition. Let A and B be two finite non-empty sets and S = A × B. The double counting principle
is to count the size of S in two ways
P
(a) First way: We count |{b ∈ B : (a, b) ∈ S}|.
a∈AP
(b) Second way: We count |{a ∈ A : (a, b) ∈ S}|.
b∈B

Remark. We count S in two ways and such counting gives us the following identity
X X
|{a ∈ A : (a, b) ∈ S}| = |{b ∈ B : (a, b) ∈ S}|
b∈B a∈A

Example (Handshaking Lemma). Suppose there are n many people at a party and everyone will shake
hands with everyone else. How many handshakes will occur?
Here we count the set {(x, {x, y}) : x, y ∈ [n], x 6= y} = S (say). Note that [n] = {1, 2, . . . , n} represents
the set of one hand (right hand) of each person.
First way: There are n many choices for x and chosen x there are (n − 1) many choices for {x, y}. Hence
|S| = n(n − 1).
Second way: Suppose there are N number of handshakes, i.e. there are N choices for {x, y} and chosen
a handshake {x, y}, there are 2 choices namely (x, {x, y}) and (y, {x, y}). Hence
|S| = 2N
n(n−1)
Thus using the double counting principle we have 2N = |S| = n(n−1), i.e. N = 2 no of handshakes.
Remark. If we label the people with 1 to n. To avoid counting a hand twice, we count for the person i
only handshakes with the persons of upper numbers. Then the total number of handshakes is
n
X n−1
X
(n − i) = i
i=1 i=1
Thus we have the following identity
n n−1
X X n(n − 1)
(n − i) = i= .
i=1 i=1
2
Moral. One can deduct an identity using counting.
 1. Permutation and its application in Counting
Definition. An one-one and onto function λ : [n] → [n] is called permutation, where n is a positive
integer and [n] = {1, 2, . . . , n}. The set of all permutations is denoted by Sn .
Example. We consider the following examples.
(a) Take n = 3, λ : [3] → [3] given by λ(1) = 2, λ(2) = 1 and λ(3) = 3. Such function can be also
represented as 1 7→ 2, 2 7→ 1 and 3 7→ 3. Note that |S3 | = 6.
(b) Take n = k, λ : [k] → [k] given by λ(1) = 2, λ(2) = 3, . . ., λ(k − 1) = k, λ(k) = 1.
Model. We collect k many balls from an urn which contains n many different (or labelled) balls. Obvi-
ously k ≤ n. There are two types of collection:- first one is we collect it in one grip and the second one is
we collect one ball at time and repeat this for k many times. In this example we discuss the second type
of collection.
We consider the set
S = {(x1 , x2 , . . . , xk ) : ∀ i ∈ [k], xi is the ball drawn from the urn at the ith repetition}.
Our purpose is to count the size of the set S.
During the one-by-one at a time collection of k many balls we may replace or we may not replace the
ball into the urn. If we replace the ball into the urn at each step then |S| = nk (Exercise!). If we do
n!
not replace then |S| = n(n − 1) · · · (n − k + 1) = (n−k)! (Exercise! Hint use multiplication principle of
counting)
Exercise. Show that Sn contains exactly n! many elements. Hint : Establish an one-one and onto map
with Sn and S, where
S = {(x1 , x2 , . . . , xn ) : ∀ i, j ∈ [n] with i 6= j, xi ∈ [n] and xi 6= xj }
Also establish |S| = n!.
Theorem 1.1 (Stirling’s approximation
√ formula). For large positive integer n, the integer n! can be
1
approximated by the real number 2πnn+ 2 e−n .
Proof : Omitted. 
Remark. Let λ : [n] → [n] and µ : [n] → [n] be two mappings Then λoµ : [n] → [n] is defined by
λoµ(i) = λ(µ(i)) is called composition of two permutations λ and µ. In general λoµ and µoλ are not
same (Find examples).
Let for each i ∈ [t], there be a collection of ni identical copies of the object ai . Here the set of objects
is {a1 , a2 , . . . , at } and multi-set of objects is {n1 · a1 , n2 · a2 , . . . , nt · at }. The main difference is in a set
each element is distinguishable i.e for all i, j ∈ [t], ai 6= aj .
Definition. Let for each i ∈ [t], there be a collection of ni identical copies of the object ai and k be an
integer with 0 ≤ k ≤ n1 + n2 + . . . + nt .
• A k−tuple (x1 , x2 . . . , xk ), where for each r ∈ [k], xr = ai for some i ∈ [t], is said to be unordered
if for each permutation π : [k] → [k], (xπ(1) , xπ(2) . . . , xπ(k) ) is same as (x1 , x2 , . . . , xk ).
• A k−tuple (x1 , x2 . . . , xk ), where for each r ∈ [k], xr = ai for some i ∈ [t], is said to be
ordered if for each permutation π : [k] → [k], other than identity permutation and the permu-
tations µ : [k] → [k] with the property that xµ(i) = xµ(j) , whenever xi = xj , where i, j ∈ [k],
(xπ(1) , xπ(2) , . . . , xπ(k) ) is not same as (x1 , x2 , . . . , xk ).
Model. We consider the example of collecting k many balls from an urn, which contains n many different
(or labelled) balls. We draw one ball at a time with replacement and repeat this process k many times.
We consider the set
S = {(x1 , . . . , xk ) : ∀ i ∈ [k], xi is the ball drawn from the urn at the ith repetition}
We relate two k−tuples say (x1 , . . . , xk ) and (y1 , . . . , yk ) by the relation ∼ into the set S, if and only
if, there exists a permutation π : [k] → [k] such that xπ(i) = yi for each i ∈ [k]. This is an equivalence
1
2

relation on S. (Exercise!) With respect to this equivalence relation, our aim is to identify each equivalent
class.
Suppose the equivalent classes are S1 , S2 , . . . , SN . Since |S| = nk , we have N ≤ nk . Let the k−tuple
(x1 , . . . , xk ) be a member of S1 and π : [k] → [k] be a permutation. Then note that the k−tuple
(xπ(1) , . . . , xπ(k) ) is also a member of S1 . Hence we may represent this equivalent class S1 by the unique
unordered k−tuple (x1 , . . . , xk ) and each element of S1 by the ordered k−tuple (x1 , . . . , xk ). Some-
times the unordered k−tuple (x1 , . . . , xk ) also denoted as (multi-)set notation {x1 , . . . , xk } (written with
repetitions)
Definition. Let n and k be positive integers. A k−permutation of an n−set S (n−set means a set of
size n) is an ordered k−tuple of elements of S.
Note here, an n−permutation of an n−set S is simply a permutation (of S).
Definition. Let n and k be positive integers. A k−combination of an n−set S is an unordered k−tuple
of elements of S.
n!
Theorem 1.2. Let n and k be positive integers with n ≥ k. Then there are n(n−1) · · · (n−k+1) = (n−k)!
many k−permutations of an n−set.
Proof : Let X be an n−set and (x1 , x2 , . . . , xk ) be a k−permutation. The first element x1 can be chosen
in n many ways. Having chosen the first element x1 , the second element x2 can be chosen in (n − 1)
many ways (out of the remaining (n − 1) elements). Proceeding in this manner, we have the last, i.e. the
k−th element xk can be chosen in {n − (k − 1)} = n − k + 1 many ways. The result follows using the
multiplication principle of counting. 
Notation: Suppose P (n, k) counts the number of k−permutations of an n−set. Then by the above the-
n!
orem P (n, k) = n(n−1) · · · (n−k +1) = (n−k)! . Similarly, C(n, k) counts the number of k−combinations.
P (n,k)
Theorem 1.3. Let n and k be positive integers with n ≥ k. Then C(n, k) = P (k,k) .

Proof : We count the set of all k−permutations in two ways

First way: It is P (n, k).
Second way: Fix a k−combination (x1 , x2 , . . . , xk ) in C(n, k) many ways. If we order the chosen
k−combination (x1 , x2 , . . . , xk ), it yields a k−permutation. So we order the chosen k−combination in
P (k, k) many ways. Hence we have P (n, k) = C(n, k)P (k, k) and the result follows. 
Further explanation on the above proof
Proof : We consider the following set and count such set in two ways.
S = {(x1 , x2 , . . . , xk ) : (x1 , x2 , . . . , xk ) is a ordered k − tuple from the set {a1 , a2 , . . . , an }}
First way: |S| = P (n, k).
Second way: We say two elements of S say (x1 , . . . , xk ) and (y1 , . . . , yk ) are related (equivalent) if (and
only if) there exists a permutation π : [k] → [k] such that y1 = xπ(1) , y2 = xπ(2) , . . ., yk = xπ(k) .
Exercise. Show that the relation on S is an equivalence relation.
Note that an equivalence relation produces equivalence classes. Here one equivalence class can be
represented by a unique unordered k−tuple (x1 , . . . , xn ).
We count the number of equivalent classes and the size of a equivalent class. There are C(n, k) many
equivalent classes and each equivalent class contains exactly P (k, k) = k! many k−tuples.
Hence we have P (n, k) = C(n, k)P (k, k) and the result follows. 
Definition. Let S be a finite multi-set of the objects a1 , a2 , . . . , an and for each i ∈ [n], there are ki
identical copies of the object ai , i.e. S = {k1 · a1 , k2 · a2 , . . . , kn · an }.
A k−permutation (similarly, k−combination) of the multi-set S is an ordered (similarly, unordered)
k−tuple of the objects a1 , a2 , . . ., an , where 1 ≤ k ≤ k1 + k2 + . . . + kn is an integer.
 3

Theorem 1.4. Let N be a large positive integer and S be a finite multi-set of objects a1 , . . . , an and for
each i ∈ [n], there are N identical copies of the object ai , i.e. S = {N · a1 , N · a2 , . . . , N · an }. Then there
are at most nk many k−permutations of the multi-set S.
Note here N is large enough positive integer compare to n and k is small positive integer compare to
kN . (In notation, N >> n and k << nN .)
Proof : We choose a k−permutation of the multi-set S (say) (x1 , x2 , . . . , xk ). So for each j ∈ [k], we
can choose xj is one of a1 , a2 , . . ., an . Hence by multiplication principle, there are at most nk many
k−permutations. 
Theorem 1.5. Let S be a multi-set of objects a1 , . . . , an and for each i ∈ [n], there are ki many identical
copies of the object ai , i.e. S = {k1 · a1 , k2 · a2 , . . . , kn · an }. Let k = k1 + k2 + . . . + kn . Then there are
exactly k1 !k2k!!···kn ! many k−permutations of the multi-set S.
Proof : Recall that a k−permutation of the multi-set S is a k−tuple (x1 , x2 , . . . , xk ), where for each
i ∈ [k], xi ∈ {a1 , . . . , an }. Note that for each j ∈ [n], there exist ij1 , ij2 , . . . , ijkj ∈ [k] such that
xij1 = xij2 = . . . = xijkj = aj .
There are k many boxes and we want to put exactly one of the objects a1 , . . . , an in each of the boxes.
Since there are k1 many
identical copies of the object a1 , we choose k1 many boxes and fill it with a1 .
We can do this in kk1 many ways. 1 Having chosen k1 many boxes, we choose k2 many boxes and fill it
with a2 . We can do this in k−k


k2
1
many ways. In general, having chosen k1 + . . . + kj−1 boxes, we choose
kj boxes and fill it with aj , where j ∈ {2, 3, . . . , n}, we can do this in k+...+k j−1


kj
many ways. Hence by
multiplication principle, there are exactly
     
k k − k1 k − k1 − k2 k − k1 − . . . − kn−1 k!
··· =
k1 k2 k3 kn k1 ! · · · kn !.
many k−permutations. 
Proof : Revised one 
Theorem 1.6. Let N be a large positive integer and S be a finite multi-set of objects a1 , . . . , an and for
each i ∈ [n], there are

N identical copies of the object ai , i.e. S = {N · a1 , N · a2 , . . . , N · an }. Then there
are exactly n+k−1
k many k−combinations of the multi-set S, where k ≤ nN is a positive integer.
Proof : It will be done later. 

1Exercise! Hint: Identify here the set S as the “counting object” and associate each element of S uniquely with an unoerderd
k1 −tuple or k1 −combination. Also note here the symbol n n!


k
means the number k!(n−k)! , where n and k are positive integers
satisfies n ≥ k.
 1. Binomial and Multinomial Theorem
Definition. Let S be an n−set, a k−subset of S is a subset of S with size k and Sk denotes the collection

of all k−subsets of S. We define

   
n  S 
:=  .
k k 

 

Exercise. If X and Y are two sets satisfies |X| = |Y | < ∞, then show that  X
  Y 
k  =  k . (Hint: Use


bijective principle of counting technique.)

Lemma 1.1. Let S be an n−set and A, B ∈ Sk . Then there exist a bijective mapping π : S → S (a

permutation of S) such that π(A) = B.

Proof : Since |A| = |B| = k and |S r A| = |S r B| = n − k, there exist a bijective mapping f : A → B

and g : S r A → S r B. Consequently π : S → S is defined by

f (x) if x ∈ A
π(x) =
g(x) if x ∈ S r A

The result holds since π(A) = B. 

Question. Let A and B be a finite set with |A| = |B|. How you prove that there exists a bijective
function f : A → B? (Hint: Let A = {x1 , . . . , xn } and B = {y1 , . . . , yn }. Construct the map xi 7→ yi for
each i ∈ [n].)

Let X be an n−set

and Sn (X) denote the set of all bijective mappings from X to itself (i.e permutations
of X). Fix A ∈ Xk , now by the above lemma we have
 
X
{π(A) : π ∈ Sn (X)} =
k

Note that π 7→ (π|A , π|XrA ) is a bijective mapping from {λ ∈ Sn (X) : λ(A) = A} to

Sk (A) × Sn−k (X r A) (Verify it! Exercise!)
Therefore by bijective mapping principle of counting, we have

|{λ ∈ Sn (X) : λ(A) = A}| = |Sk (A) × Sn−k (X r A)|

= |Sk (A)| · |Sn−k (X r A)| = k!(n − k)!

Note that π ∼ µ (read as:- π is related to µ by the relation ∼), where π, µ ∈ Sn (X), if and only if
π(A) = µ(A) is an equivalence relation. (Exercise!)
An equivalence class with respect to the aforementioned equivalence relation into Sn (X) is
  
X
π ∈ Sn (X) : π(A) = B, for some B ∈ .(Prove it!)
k

Since there are exactly | X X

k | many k−subsets of X, using the above lemma, we have exactly | k |
number of equivalent classes. Again using the above lemma, we have (Prove that) two different equivalent
class have same size. This means, for two different k−subsets of X say C and D the equivalent classes
{π ∈ Sn (X) : π(A) = C} and {λ ∈ Sn (X) : π(A) = B} have same size, i.e

|{π ∈ Sn (X) : π(A) = C}| = |{π ∈ Sn (X) : π(A) = C}|.

1
2

This implies the set S
n (X) can be decomposed into the equivalent classes namely
{π ∈ Sn (X) : π(A) ∈ Xk }. Hence
Sn (X) = t {π ∈ Sn (X) : π(A) = B}
B∈(X
k)
X
|Sn (X)| = |{π ∈ Sn (X) : π(A) = B}|
X
B∈( ) k
 
 X 
n! =   |{π ∈ Sn (X) : π(A) = B}|
k 
 
 X 
n! =   k!(n − k)!
k 
Thus we have proved the following theorem
Theorem 1.2. Given an n−set X and X n



k be the collection of all k−subsets of X. We define k :=
| X


k |, then

if k ≥ n + 1
  
n 0
= n!
k k!(n−k)! if 0 ≤ k ≤ n

Remark. The above technique to decompose the set Sn (X) is known as orbit-stabilizer technique. Here
the set {π(A) : π ∈ Sn (X)} is called the orbit and the set {λ ∈ Sn (X) : λ(A) = A} is called the stabilizer.
Example. Recall the example of collecting k many
balls from an urn which contains n many balls. If
we
collect k many balls in one grip, then there are nk many grips are possible. In other words, there are
n
k many ways to collect k many balls in one grip from an urn which contains n many balls.

Exercise. Solve the following:-

(A) Let X be an n−set and 2X denote the set of all subsets of X. Show that there is a bijective function
between 2X and {0, 1}X , where {0, 1}X stands for all functions from X to {0, 1}. Conclude that
n n
n
t X



|2X | = 2n and deduce that n X
P
k = 2 , using 2 = k=0 k .
k=0
(B) Let X, Y be two mutually disjoint n−sets. Find bijection between XtY


n and
n X
n
Y 2n n


P
n
Y

Y


t k × n−k . Hence prove n = k n−k . But there is a bijection between k and n−k .
k=0 k=0
n
n n 2n n 2



P

(Find it.) Hence prove k = n−k . Therefore n = k .
k=0

Theorem 1.3 (Binomial identity). Let A be an a−set and B be an b−set with A ∩ B = ∅. Then
n  
n
X n i n−i
(a + b) = ab .
i=0
i

Proof : Let S = {(x1 , . . . , xn ) : ∀ i ∈ [n], xi ∈ A t B} and

 
n [n]
R= t {(x1 , . . . , xp ) : ∀ i ∈ [n], xi ∈ A} × {(y1 , . . . , yn−p ) : ∀ i ∈ [n − p], yi ∈ B}
p=0 p
Claim : There exist a bijection between S and R.
Proof of claim : Exercise!
n
n


Note that |S| = (a + b)n and |R| = ai bn−i .
P
i 
i=0

Let X be an n−set and p and q are integers satisfying 0 ≤ p + q ≤ n.

 
X
= {(A, B) : A ⊂ X with |A| = p, B ⊂ Y with |B| = q, A ∩ B = ∅}
p, q
 3

Exercise. Using orbit-stabilizer technique establish that

 
 X  n!
 p, q  = p!q!(n − p − q)! .
 

X


Hint: Firstly establish for each (A, B) and (C, D) ∈ p,q , there exist bijective mapping π : X → X such
that (π(A), π(B)) = (C, D)

X
Secondly, fix (A, B) ∈ p,q establish (the formation of the orbit) that
 
X
{(π(A), π(B)) : π ∈ Sn (X)} =
p, q

the number of permutations λ : X → X such that π(A) = A and π(B) = B, where
Thirdly, count
X
(A, B) ∈ p,q . This step is about the formation of stabilizer. Now establish that
|{λ ∈ Sn (X) : λ(A) = A, λ(B) = B}| = |Sp (A)| × |Sq (B)| × |Sn−p−q (X r (A t B))| = p!q!(n − p − q)!.
Finally, establish the equivalence relation on Sn (X). Count the number of equivalent class (size of the
orbit). Count the size of the equivalent classes (size of the stabilizer).
Theorem 1.4 (Multinomial identity). Let for each i ∈ [k], Ai be a set of size ai and for each i, j ∈ [n]
with i 6= j, Ai ∩ Aj = ∅. Then
X n!
(a1 + . . . + ak )n = at1 · · · atkk
t1 ! · · · tk ! 1
Here the summation extends over all non-negative integer solution of t1 , . . . , tk satisfies t1 + . . . + tk = n.
Proof : Omitted. 
Theorem 1.5 (Pascal’s Identity). For each positive integers m and k, with 1 ≤ k ≤ m
     
m m m+1
+ = ,
k k−1 k
Proof : Let A = {X ∈ [m+1] : m + 1 ∈ X} and B = {X ∈ [m+1]



k k :m+1∈ / X}. Hence
 
m+1
|A| + |B| = .
k
Note that for each X ∈ B, X ⊂ [m], i.e. B ⊂ [m] and for each X ∈ [m]



k k , we have X ∈ B. Hence
|B| = | [m] m [m]




k | = k−1 . Again, X →
7 X r {m + 1} is a bijective mapping from A to k−1 . Therefore,
   
 [m]  m
|A| =  =
k−1  k−1
and the identity follows. 
Theorem 1.6. Given two positive integers N and i, there is a unique way to expand N as a sum of
binomial coefficients as follows:-
     
ni ni−1 nj
N= + + ... + ,
i i−1 j
where j is a positive integer and ni ≥ ni−1 ≥ . . . ≥ nj ≥ j ≥ 1.
Proof : Set ni = max{n : ni ≤ N } and ni−1 = max{n : i−1 n
≤ N − nii }. In general, for k ∈

{0, 1, 2, . . .},       
n ni ni−k−1
ni−k = max n : ≤N− − ... −
i−k i i−k−1
The procedure of constructing the integers ni , ni−1 , . . . must terminate, since n1 = n for each n ∈ N

and k ∈ {0, 1, . . . , i − 1}. This implies

     
ni ni−1 nj
N= + + ... + ,
i i−1 j
4

where j is an integer satisfies 1 ≤ j ≤ i. Note that from the constructions of the integers ni , ni−1 , . . .,
nj we have ni , ni−1 , . . ., nj are uniquely determined by N .
Claim : ni−1 ≤ ni .
Proof of claim : Note that using Pascal’s identity nii ≤ N ≤ 1+n = nii + i−1
ni





i
i
. Hence we have
0 ≤ N − nii ≤ i−1 ni



. Note that since i ≥ 1, for each m ≥ 1 we have
   
m m+1
≤ .
i−1 i−1
m
≤ N − nii } ≤ ni . This establishes the claim.



Therefore max{m : i−1
Repeated use of similar argument as in the claim we have ni ≥ ni−1 ≥ . . . ≥ nj . (Exercise ! Hint: Use
backward induction on i.)
If N − nii = 0, then we have N = nii and the result holds with j = i. If N − nii 6= 0, i.e.

N − nii ≥ 1 = i−1



i−1 . This implies
    
n ni
i − 1 ≤ max n : ≤N− = ni−1 .
i−1 i
Similarly we have (Exercise!) i − 2 ≤ ni−2 , i − 3 ≤ ni−3 , . . ., j ≤ nj . 
 1. Pigeonhole Principle
Theorem 1.1 (Pigeonhole Principle- Simple Version). If (n + 1) many objects are distributed among n
many boxes, where n is a positive integer, then there exist at least one box which contains two or more
objects.
Proof : Suppose none of the n many boxes contains two or more objects. This means each of the n many
boxes contains at most one object. Hence there are at most n many objects, however we are distributing
(n + 1) many objects. A contradiction arises. 
Example. There are 30 faculties in SPS, NISER. At least 3 faculties share the same birth month.
Here 12 boxes are labelled with the months of the year and 30 birth months of 30 faculties are 30
objects.
Theorem 1.2 (Pigeonhole Principle- Strong and Uniform Version). If (mn + 1) many objects are dis-
tributed among n many boxes, where m and n are positive integers, then there exist at least one box which
contains at least (m + 1) objects.
Proof : Suppose each of the n boxes contains at most m many objects after distribution. Then there are
mn many objects, which are to be distributed. Hence our assumption that each of the n boxes contains
at most m many objects is wrong, i.e. a contradiction arises. Hence at least one of the n boxes contains
at least m + 1 many objects. 
Theorem 1.3 (Pigeonhole Principle- Strong Version). If m1 +. . .+mn −n+1 many objects are distributed
among n many labelled boxes with labelling 1, . . . , n, where m1 , . . ., mn , n are positive integers. Then
there exists at least one i ∈ [n] such that the box with label i contains mi many objects.
Proof : Suppose for each i ∈ [n], the box with label i contains less than mi many objects after distri-
n
P
bution. Then there are at most (mi − 1) = m1 + . . . + mn − n many objects are to be distributed. But
i=1
there are m1 + . . . + mn − n + 1 many objects. Hence our assumption for each i ∈ [n], the label i box
contains at most mi many objects. is wrong. i.e. a contradiction arises. Hence for some i, the box with
label i contains at least mi many objects. 
Remark. Objects are pigeons and boxes are pigeonholes.
Model. Let m and n be two positive integers with gcd (m, n) = 1. Let a and b be two integers with
0 ≤ a ≤ m − 1 and 0 ≤ b ≤ n − 1. Then there exists a positive integer x satisfies both the conditions
x ≡ a mod m and x ≡ b mod n.
The pigeonhole principle is applied to prove the existence of such x. The above existential statement
is referred as Chinese remainder theorem.
To prove the Chinese remainder theorem, we construct the n many numbers, a, a+m, . . . , a+(n−1)m.
Note that all the above numbers satisfies the modulo equation x ≡ a mod m. Now we claim the following
Claim : If for each i ∈ {0, 1, . . . , n − 1}, a + im ≡ ri mod n, then for each i, j ∈ {0, 1, . . . , n − 1}, with
i 6= j satisfies ri 6= rj .
Proof of claim : Suppose there exist i, j ∈ {0, 1, . . . , n − 1} with i < j and ri = rj . This means
n|(j − i)m Since gcd (m, n) = 1, we have n|(j − i). A contradiction arises, since 1 ≤ j − i ≤ n − 1, which
establishes the claim.
Here we use the pigeonhole principle, considering the remainders r0 , r1 , . . . , rn−1 and b as (n + 1) many
objects and the complete (Why?) list of remainders r0 , r1 , . . . , rn−1 as n many boxes. We conclude there
exist a remainder (say) rk , such that b = rk mod n, i.e. a + km ≡ b mod n. It shows that a + km is
the required x satisfies x ≡ a mod m and x ≡ b mod n.
Model. Let a1 , a2 , . . ., an be n integers, then there exist integers i and j with 0 ≤ i < j ≤ n such that
n|(ai+1 + . . . + aj ).
1
2

The pigeonhole principle is applied to prove the existence of such integers i and j. We construct the
following n many numbers
a1 , a1 + a2 , a1 + a2 + a3 , . . . , a1 + . . . + an
If one of the n many numbers are divisible by n, then the result follows. Otherwise we have none of them
is divisible by n. In this case, suppose
a1 + . . . + ak ≡ rk mod n,
then rk ∈ {1, 2 . . . , n − 1}, for each k ∈ {1, 2, . . . , n}. Here we use the pigeonhole principle. We consider
the n many numbers a1 , a1 + a2 , . . . , a1 + . . . + an are n many objects and the (n − 1) many remainders
r1 , r2 , . . . , rn−1 as (n − 1) many boxes. We conclude there exists i, j ∈ {1, . . . , n} with i < j and a
remainder r such that
a1 + . . . + ai ≡ a1 + . . . + aj ≡ r mod n
This means a1 + . . . + ai = pn + r and a1 + . . . + aj = qn + r for some integers p and q, i.e.
(a1 + . . . + aj ) − (a1 + . . . + ai ) = qn + r − pn − r
⇔ ai+1 + . . . + aj = (q − p)n
This means n divides the integer (ai+1 + . . . + aj ).
Theorem 1.4 (Erdős-Szekeres). Let m and n be positive integers and {xn }∞ n=1 be a sequence of real
numbers. Then x1 , x2 , . . ., xmn+1 contains a monotone increasing sub-sequence with m + 1 many real
numbers or a strictly monotone decreasing sub-sequence with n + 1 many real numbers.
Proof : Suppose x1 , . . . , xmn+1 does not have any monotonically increasing sub-sequence with m + 1
real numbers. Suppose for each i ∈ [mn + 1], the largest monotonically increasing sub-sequence of
xi , xi+1 . . . , xmn+1 contains ki many real numbers. Then 1 ≤ ki ≤ m.
We take k1 , k2 , . . . , kmn+1 as objects and m many labelled boxes with the labels 1, . . . , m. If kj = i,
where j ∈ [mn + 1] and i ∈ [m], we put kj into the box with label i. By using pigeonhole principle,
we conclude that there exists at least one box with some label say r contains at least n + 1 objects. It
means among the mn + 1 many objects k1 , k2 , . . . , kmn+1 there are at least n + 1 many many objects say
kl1 , kl2 , . . . , kln+1 have same value r.
Claim : xl1 > xl2 > . . . > xln+1 is the required strictly decreasing sub-sequence with n + 1 real numbers.
Proof of claim : Suppose for some i, j ∈ [n + 1] with i < j, xli ≤ xlj . This implies the sub-
sequence xli , xli +1 , . . . , xmn+1 and xlj , xlj +1 , . . . , xmn+1 have monotonically increasing sub-sequence with
kli = klj = r many real numbers. Therefore the sub-sequence xli , . . . , xlj , . . . xmn+1 have monotonically
increasing sub-sequence with more than r many real numbers. (Why?) A contradiction arises which
establishes the claim.
The result follows from the claim. 
Theorem 1.5 (Pigeonhole Principle-Twin version). If at most (n−1) many objects are distributed among
n many boxes, where n is a positive integer, then there exist at least one empty box.
Proof : Exercise. 
Theorem 1.6 (Dirichlet’s Approximation).
  For each irrational number α there exist infinitely many
rational numbers pq satisfies α − pq  < q12 .
 

Proof : We fix α and prove the following claim.  

Claim : For each positive integer n, there exist positive integers p and q satisfies α − pq  < 1
≤ 1
q2 .
 
nq

Proof of claim : For the positive integer m, note that mα − bmαc is an irrational number satisfies
0 < mα − bmαc < 1. We consider (0, n1 ), ( n1 , n2 ), . . ., ( n−1
n , 1) as n many boxes and n + 1 many rational
numbers α − bαc, 2α − b2αc, . . ., (n + 1)α − b(n + 1)αc as n + 1 many objects. These n + 1 many objects
 3

are to be distributed among those n many boxes. Therefore by pigeonhole principle one box must contain
at least two such irrational numbers (say) rα − brαc and sα − bsαc with r < s. This immediately implies,
1
|(sα − bsαc) − (rα − brαc)| <
n
1
⇔ |(s − r)α − (bsαc − brαc)| <
n
We set p = bsαc − brαc and q = s − r, then q ≤ n (Why?) and |qα − p| ≤ n1 . This establishes the claim,
since  
α − p  < 1 ≤ 1 .
 
 q  nq q2
 
Let pq be a rational number satisfying α − pq  < q12 . We consider the positive irrational number
 

β = |α − pq |. By using the Archimedean property, we have the positive integer N , with N > β1 . Using
 
the above claim there exist a rational number rs satisfies α − rs  < N1s ≤ s12 . This implies
 
 r  1 β  p 
α −  < < < β = α −  .
 
s Ns s q
p
 
r
Consequently s 6= q . This means we construct a different rational number rs satisfies α − rs  ≤ s12 . Using
 
the mathematical induction, we have infinitely many rational numbers pq satisfying α − pq  < q12 . This
 
completes the result. 
 1. Counting problems on labelled and unlabelled objects
Definition. A labelled collection of n many objects means n many distinguishable objects and unlabelled
collection collection of n many objects means n many indistinguishable objects.
Example. A collection of n many apples or a collection of n many identical objects are examples of
unlabelled collection.
1.1. n identical objects and k distinguishable boxes.
Problem. How many ways you can distribute n many identical objects into k many distinguishable
boxes? (The special case) How many ways you can distribute 10 (identical) biscuits among 3 students
Amit, Naren and Raju?
Theorem 1.1. One can distribute n many identical objects into k many labelled boxes into n+k−1


k−1 many
ways.
Proof : Let B be the set of labelled boxes. Note that |B| = k. For each distribution D of n many
identical objects into B, we associate a function rD : B → N, where for each b ∈ B, rD (b) counts the
number of identical objects received by the box b, with respect to the distribution D.
We note that for each distribution D ∈ D, where D denotes the set of all distributions of n many
identical objects into k many labelled boxes. Note that |D| = |{rD : D ∈ D}| (Exercise! Hint: Show that
D 7→ rD from D to {rD : D ∈ D} is a one-one and onto map.) Also note that |{rD : D ∈ D}| = |{f :
[k] → N : f (1) + f (2) + . . . + f (k) = n}| 

Claim : |{f : [k] → N : f (1) + f (2) + . . . + f (k) = n}| =  [n+k−1]

k−1 .

Proof of claim : For each B ∈ [n+k−1]

k−1 , we associate in a unique way a function fB : [k] → N with
fB (1) + fB (2) + . . . + fB (k) = n.
Let B ∈ [n+k−1]


k−1 and B = {i1 , . . . , ik−1 }, without loss of generality we assume i1 < i2 < . . . < ik−1 .
Set
fB (1) = |{i ∈ [n + k − 1] : i < i1 }|
fB (2) = |{i ∈ [n + k − 1] : i1 < i < i2 }|
..
.
fB (k − 1) = |{i ∈ [n + k − 1] : ik−2 < i < ik−1 }|
fB (k) = |{i ∈ [n + k − 1] : ik−1 < i}|
Conversely, for a function f : [k] → N with f (1) + f (2) + . . . + f (k) = n, we associate the (k − 1)−set
{i1 , i2 , . . . , ik−1 } in the “inverse way” where
i1 = f (1) + 1
i2 = f (1) + f (2) + 2
..
.
ik−1 = f (1) + . . . + f (k − 1) + k − 1
Hence the claim is established and the result follows.

Corollary 1.2. One can distribute n many identical
objects into k many labelled boxes such that each
box contains at least one identical object, into n−1
k−1 many ways.

Proof : We first distribute one object to each box. It results exactly n − k many objects are
to be
distributed among the labelled k boxes. By the above theorem, it can be done in n+k−1 n−1


k−1 = k−1 many
ways. 
1
2

k
P
Exercise. Consider the equation xi = n, where n and k are positive integers. Show that
i=1
(a) the number of non-negative integer solutions of the above equation is n+k−1


k−1 .
n−1


(b) the number of positive integer solutions of the above equation is k−1 .
1.2. n distinguishable objects and k identical boxes.
Definition. Let n and k be positive integers with n ≥ k. The Stirling’s number of second kind, denoted
as sII (n, k), counts the total number of partitions of the set [n] into k many non-empty parts.
The set [n] is chopped in k many (non-empty) parts. There are sII (n, k) many equivalence relations
on the set [n], which yields exactly k many (non-empty) equivalent classes.
Definition. The Bell number B(n), where n is a positive integer, counts the total number of partitions
of the set [n], i.e.
Xn
B(n) = sII (n, k).
k=0

Theorem 1.3. Let n and k be positive integers with n ≥ k ≥ 2. Then

sII (n, k) = sII (n − 1, k − 1) + ksII (n − 1, k).
Proof : Let S([n], k), A and B respectively denote the set of all partitions of [n] into k many non-empty
parts; the set of all partitions of [n] into k many non empty parts, where n belongs to an singleton set
(i.e. one chop of the set [n] is {n}.) and the set of all partitions of [n] into k many non empty parts,
where n belongs to non-singleton set.
Claim 1 : |A| = sII (n − 1, k − 1).
Proof of claim : Let X1 , . . . , Xk be a partition of the set [n] into k many non-empty parts and one
of X1 , . . . , Xk is {n}. Without loss of generality, let Xk = {n}, i.e. [n] = X1 t . . . t Xk , where for each
i ∈ [k], Xi 6= ∅ and Xk = {n}. This induces a partition of the set [n − 1] into k − 1 many non empty
parts namely X1 , . . . , Xk−1 .
Conversely, if Y1 , . . . , Yk−1 is a partition of the set [n − 1] into k − 1 many non-empty parts, then
Y1 , . . . , Yk−1 , {n} is a partition of the set [n] into k many non-empty parts. Hence a bijective correspon-
dence exists between A and S([n − 1], k − 1), and the claim is established.
Claim 2 : |B| = ksII (n − 1, k).
Proof of claim : We note that
X1 , . . . , Xk 7→ X1 r {n}, . . . , Xk r {n} (1)
where X1 , . . . , Xk is a partition of the set [n] into k many non-empty parts, where n belongs to non-
singleton set and note that X1 r {n}, . . . , Xk r {n} is a partition of the set [n − 1] into k many non-empty
parts. Again for each partition of the set [n − 1] into k many non-empty parts Y1 , Y2 . . . , Yk has exactly
k many pre-images with respect to the aforementioned mapping namely
Y1 t {n}, Y2 , . . . , Yk ;
Y1 , Y2 t {n}, . . . , Yk ;
..
.
Y1 , Y2 , . . . , Yk t {n}.
Hence such mapping is k − to − 1, onto mapping. Therefore the claim is established.
The result follows from the above two claims. 
Theorem 1.4. For each integer n ≥ 2, sII (n, 2) = 2n−1 − 1.
 3

Proof : Note that sII (2, 2) = 1. Hence by using the previous theorem
sII (n, 2) = 1 + 2sII (n − 1, 2) = 1 + 2 + 22 + . . . + 2n−2 sII (2, 2) = 2n−1 − 1.

Alternative proof : Let S([n], 2) denote the set of all partitions of [n] into 2 non-empty parts and
6 B}. Then |P| = 2n − 2 (Exercise!).
P = {(A, B) : A t B = [n], A 6= ∅ =
Note that if (A, B) ∈ P, then (B, A) ∈ P and (A, B), (B, A) represents exactly one partition of the set
[n] into two non-empty parts namely A, B. (Why! Hint: Boxes are not labelled.) Hence such mapping
P to S([n], 2) is a 2 − to − 1. onto mapping. Hence
2|S([n], 2)| = |P|
2n − 2
i.e. sII (n, 2) = = 2n−1 − 1
2

n


Theorem 1.5. sII (n, n − 1) = 2 .
Proof : Exercise! (Hint: Apply Pigeonhole principle, to conclude that partition of the set [n] into n − 1
many non-empty parts produces a unique part consist of two elements of [n].) 
Theorem 1.6. One can distribute n many distinguishable objects into k many identical boxes such that
each box contains at least one distinguishable object, into sII (n, k) many ways.
Proof : Let D denote the set of all distributions of n many distinguishable objects into k many identical
boxes such that each box contains at least one distinguishable object.
Note here [n] = {1, . . . , n} is the set of all distinguishable objects. We distribute these n many objects
into k many identical boxes such that each box is non-empty. For x, y ∈ [n], we say x ∼ y if and only
if x and y are in same box. This yields an equivalence relation on the set [n]. Such equivalence relation
produces equivalent classes, which is the required partition of [n].
We associate each D ∈ D, with unique equivalence relation on [n] with respect to D. (i.e. x ∼ y, where
x, y ∈ [n], if and only if x and y are in same box under D). Thus we have a bijective correspondence
(Exercise!) between D and the set of all equivalence relations on the set [n], which produces exactly k
many non-empty equivalent classes. Hence
|D| = |S([n], k)| = sII (n, k).

Theorem 1.7. One can distribute n many distinguishable objects into k many identical boxes into
k
X
sII (n, 1) + sII (n, 2) + . . . + sII (n, k) = sII (n, i)
i=1
many ways.
Proof : Exercise! (Hint: Establish an equivalence relation on some set S (Find such set S!) with exactly
k many equivalent classes and for each i ∈ [k] each equivalent class contains sII (n, i) many members of
S.) 
1.3. n distinguishable objects and k distinguishable boxes.
Theorem 1.8. One can distribute n many distinguishable objects into k many distinguishable boxes into
k n many ways.
Proof : Distribution of n many distinguishable objects into k many distinguishable boxes yields a func-
tion d : [n] → [k], where for each i ∈ [n], d(i) denotes the label of the box where object i is distributed.
Conversely, if f : [n] → [k] is a function then f −1 (1), . . ., f −1 (k) is a partition of the set [n] into at most
k many parts. So such distribution can be done in
|{d : d : [n] → [k] is a function}| = k n
4

many ways 
Theorem 1.9. One can distribute n many distinguishable objects into k many distinguishable boxes such
that each box contains at least one distinguishable object, into k!sII (n, k) many ways.
Proof : Distribution of n many distinguishable objects into k many distinguishable boxes such that
each box contains at least one distinguishable object, yields an onto function d : [n] → [k], where for each
i ∈ [n], d(i) denotes the label of the box where object i is distributed. Conversely, if f : [n] → [k] is an
onto function then f −1 (1), . . ., f −1 (k) is a partition of the set [n] into k many non-empty parts. So such
distribution can be done in
|{d : d : [n] → [k] is an onto function}|
many ways
Let O([n], [k]) denote the set of all onto functions from the set [n] to the set [k], i.e.
O([n], [k]) = {d : d : [n] → [k] is an onto function}.
We define an equivalence relation on O([n], [k]):- f ∼ g, where f, g ∈ O([n], [k]) if and only if there exists
a permutation π : [k] → [k] such that g = πof . (Exercise! Verify it.) Let [f ] denote the equivalent class
containing f ∈ O([n], [k]). Therefore |[f ]| = k!.
Claim : With respect to the aforementioned equivalence relation, there are exactly sII (n, k) many equiv-
alent classes.
Proof of claim : We first establish that if f, g ∈ O([n], [k]) such that the unordered k−tuples
(f −1 (1), . . . , f −1 (k)) and (g −1 (1), . . . , g −1 (k)) are same. Then there exists a permutation π : [k] → [k]
such that f = πog.
Let λ : [k] → [k] and φ : [k] → [k] be the permutations such that λof = φog. Hence π = λ−1 oφ is the
required permutation.
This implies two different partitions of [n] into k non-empty parts represent two different equivalent
classes. Since there are sII (n, k) many number of partitions of [n] into k non-empty parts. So such
distribution can be done in |O([n], [k])| = k!sII (n, k) many ways which equals the number of partitions
of [n] into k non-empty distinguishable parts. 
Example. There are six onto functions from the set [3] to [2] namely f1 , f2 , g1 , g2 and h1 , h2 , where
f1 (1) = 1, f1 (2) = 2, f1 (3) = 2; g1 (1) = 2, g1 (2) = 1, g1 (3) = 2; h1 (1) = 2, h1 (2) = 2, h1 (3) = 1 and
f2 = πof1 ; g2 = πog1 ; h2 = πoh1 . Here π : [2] → [2] is given by π(1) = 2 and π(2) = 1. Note
that the unordered pair (f1−1 (1), f1−1 (2)) = (f2−1 (1), f2−1 (2)) = ({1}, {2, 3}). Similarly, the unordered
pairs (g1−1 (1), g1−1 (2)) = (g2−1 (1), g2−1 (2)) = ({2}, {1, 3}) and (h−1 −1 −1 −1
1 (1), h1 (2)) = (h2 (1), h2 (2)) =
({3}, {1, 2})
Corollary 1.10. There are k!sII (n, k) many onto functions from the set [n] to the set [k].
Remark. As we consider onto functions from the set [n] to the set [k], here n and k are positive integers
satisfies n ≥ k.
1.4. n identical objects and k identical boxes.
Definition. Let n and k be positive integers with n ≥ k. An unordered k−tuple (a1 , . . . , ak ) is said
to be a partition of the integer n into k many positive integers if a1 , . . ., ak are positive integers and
n = a1 + . . . + ak . The integer Pk (n) counts the total number of ways in which the integer n can be
partitioned into k many positive integers.
Theorem 1.11. Let n and k be positive integers with n ≥ k, then
Pk (n) = Pk−1 (n − 1) + Pk (n − k)
Proof : Let P(n, k) denote the set of all partitions of the integer n into k many positive integers, i.e.
P(n, k) = {(a1 , . . . , ak ) : (a1 , . . . , ak ) is an unordered k − tuple of positive integers satisfy
a1 + . . . + ak = n}
 5

A = {(a1 , . . . , ak ) ∈ P(n, k) : ai = 1 for some i ∈ [k]}

B = {(a1 , . . . , ak ) ∈ P(n, k) : ∀ i ∈ [k], ai ≥ 2}
Claim : |A| = Pk−1 (n − 1).
Proof of claim : Let (a1 , . . . , ak ) ∈ A, without loss of generality let ak = 1. This induces a partition of
the integer (n − 1) into (k − 1) many positive integers namely a1 , . . . , ak−1 . Conversely, if (b1 , . . . , bk−1 ) ∈
P(n − 1, k − 1), then (b1 , . . . , bk−1 , 1) ∈ A. Hence a bijective correspondence exists between P(n − 1, k − 1)
and A. Hence the claim is established.
Claim : |B| = Pk (n − k).
Proof of claim : We note that
(a1 , . . . , ak ) 7→ (a1 − 1, . . . , ak − 1),
where (a1 , . . . , ak ) ∈ B and (a1 − 1, . . . , ak − 1) ∈ Pk (n − k), is an one-one and onto mapping, which
establishes the claim.
The result follows from the above two claims. 
Theorem 1.12. One can distribute n many identical objects into k many identical boxes such that each
box contains at least one identical object, into Pk (n) many ways.
Proof : Let D denote the set of all distributions of n many identical objects into k many identical boxes
such that each box contains at least one identical object.
There is a bijective correspondence between D and P(n, k). (Exercise!) Distribution of n many
identical objects into k many identical boxes such that each box is non-empty yields a partition of the
integer n into k many positive integers and vice-versa. (Establish it to complete the above exercise.) 
k
P
Theorem 1.13. One can distribute n many identical objects into k many identical boxes into Pi (n)
i=1
many ways.
Proof : Exercise 
Example. In Statistical Mechanics, one encounters the situation of putting n particles into k distinct
energy levels. The n particles can thus be considered as n objects and the k different energy levels as
k distinguishable boxes. Three different statistics (situations) are obtained by making three different
assumptions:
• Maxwell-Boltzman : Here n distinguishable objects (particles) are distributed into k distin-
guishable boxes (distinct energy levels). Hence the number of possibilities is k n .
• Bose-Einstein : Here n identical objects (particles) are distributed
into k distinguishable boxes
(distinct energy levels). Hence the number of possibilities is n+k−1
k−1 .
• Fermi-Dirac : Here n identical objects (particles) are distributed into k distinguishable boxes
(distinct energy levels)

but no box can hold more than one object (particle). Hence the number
of possibilities is nk .
 1. Principle of Inclusion and Exclusion
Theorem 1.1 (Inclusion and exclusion theorem: Version I). Let A be a finite set and for each integer i,
with 1 ≤ i ≤ n, Ai ⊂ A. Then
 
n
X X
|A1 ∪ . . . ∪ An | = |Ai | + (−1)2−1  |Ai ∩ Aj | + . . . +
i=1 1≤i<j≤n
 
X
+ (−1)k−1  |Ai1 ∩ . . . ∩ Aik | + . . . + (−1)n−1 |A1 ∩ . . . ∩ An |.
1≤i1 <...<ik ≤n

Proof : Exercise! Hint: Use induction on n. 

Remark. Note the following identity:-
         
n n n n n n
0 = {1 + (−1)} = (−1)0 + (−1)1 + (−1)2 + . . . + (−1)r + . . . + (−1)n
0 1 2 r n
X X X
= (−1)0 + (−1)1 + (−1)2 + . . . +
A⊂[n] A⊂[n] A⊂[n]
|A|=0 |A|=1 |A|=2
X X
+ (−1)r + . . . + (−1)n
A⊂[n] A⊂[n]
|A|=r |A|=n
n X
X
= (−1)|A|
r=0A⊂[n]
|A|=r
X
= (−1)|A|
A⊂[n]

[n]
Theorem 1.2 (Inclusion and exclusion theorem:
P Version II). Let Θ : 2 → R be a mapping and
[n]
Π : 2 → R be a mapping, defined by Π(X) = Θ(S), then
SjX
X
Θ(X) = (−1)|X|−|S| Π(S).
SjX

Here 2[n] denotes the set of all subsets of [n].

Proof : We begin with the right hand side of the required equality:
 
X X X
(−1)|X|−|S| Π(S) = (−1)|X|−|S|  Θ(T )
SjX SjX T jS
 
X X
= Θ(X) + Θ(Y )  (−1)|X|−|S| 
Y $X Y jS$X
 
X X
= Θ(X) + Θ(Y )  (−1)|P | 
Y $X P j(XrY )
X  
= Θ(X) + Θ(Y ) (−1 + 1)|X|−|Y |
Y $X

= Θ(X).
This completes the proof. 
1
2

Theorem 1.3 (Inclusion and exclusion theorem: Version III). Let X be a finite set and for each i ∈ [m],
Pi is a property. For each x ∈ X and i ∈ [m], either x satisfies property Pi or (exclusive “or”) x does not
satisfy property Pi . Let S j [m], N (S) := {x ∈ X : x satisfies property Pi for each i ∈ S}. The number
of elements of X that satisfy none of the properties P1 , . . . , Pm is given by
X
(−1)|S| |N (S)|.
Sj[m]

Proof : Let x ∈ X and Sx := {i ∈ [m] : x satisfies Pi }. We note that N (S) = {x ∈ X : S j Sx }.

X X
(−1)|S| |N (S)| = (−1)|S| |{x ∈ X : S j Sx }|
Sj[m] Sj[m]
X X
= (−1)|S| 1
Sj[m] x∈X
SjSx
X X
= (−1)|S|
Sj[m] x∈X
SjSx
X X
= (−1)|S|
SjSx j[m]x∈X
X X
= (−1)|S|
SjSx x∈X
X X
= (−1)|S|
x∈X SjSx
XX X X
= 1+ (−1)|S|
x∈X Sx =∅ x∈X Sx 6=∅
SjSx
X X
= |{x ∈ X : Sx = ∅}| + (−1)|S|
x∈X Sx 6=∅
SjSx
X
= |{x ∈ X : Sx = ∅}| + (−1 + 1)|Sx |
x∈X
= |{x ∈ X : Sx = ∅}|.
The result follows, since |{x ∈ X : S¯x = [m]}| = |{x ∈ X : Sx = ∅}| 
1.1. Applications of inclusion and exclusion theorem.
Definition. A permutation π : [n] → [n] is called derangement if for each i ∈ [n], π(i) 6= i. The set of
all derangements on [n] is denoted by Dn .
Model. There are n many letters and n many corresponding envelops. In how many ways the letters
can be placed in the envelop (one letter in each envelop), so that no letter is put into right envelop.
Theorem 1.4. For each positive integer n, |Dn | = (n − 1)(|Dn−1 | + |Dn−2 |).
Proof : Exercise! 
n
n


(−1)j
P
Theorem 1.5. For each positive integer n, |Dn | = j (n − j)!.
j=0

Proof : Let Sn denote the set of all permutations on [n] and for each i ∈ [n], Pi denote the property
that π(i) = i where π ∈ Sn . So the set of elements of Sn that satisfy none of the properties P1 , . . . , Pn is
the set of all derangements namely Dn . For S j [n], let
N (S) := {π ∈ Sn : π satisfies property Pi for each i ∈ S}.
 3

Claim : For S j [n], |N (S)| = (n − |S|)!.

Proof of claim : We note the following:

π ∈ N (S) ⇔ π satisfies property Pi for each i ∈ S
⇔ π(i) = i for each i ∈ S
⇔ π : ([n] r S) → ([n] r S) is an induced permutation on [n] r S.
Hence the claim is established.

Therefore using the Version III of Inclusion and exclusion theorem and the above claim, we have
X X
|Dn | = (−1)|S| |N (S)| = (−1)|S| (n − |S|)!
Sj[n] Sj[n]

Hence the result follows. 

Theorem 1.6 (Derangements Approximation Formula).

|Dn | 1
lim = .
n→∞ n! e
Proof : Exercise! 

Theorem 1.7. Let sII (n, k) denote the Stirling number of second kind, then
k  
II 1X i k
s (n, k) = (−1) (k − i)n .
k! i=0 i

Proof : Let A([n], [k]) denote the set of all functions from [n] to [k]. Recall that |A([n], [k])| = k n . We
define the properties P1 , P2 , . . . , Pk , where for each i ∈ [k], Pi denotes the property that i(∈ [k]) does not
belong to image of f ∈ A([n], [k]). Note that the set of elements of A([n], [k]) that satisfy none of the
properties P1 , P2 . . . , Pk is the set of all onto functions from [n] to [k], namely O([n], [k]).
For S j [k], let
N (S) := {f ∈ A([n], [k]) : f satisfies property Pi for each i ∈ S}.
Claim : For S j [k], |N (S)| = (k − |S|)n .

Proof of claim : We note the following:

f ∈ N (S) ⇔ f satisfies property Pi for each i ∈ S
⇔ i does not belong to the image of f for each i ∈ S
⇔ S is not the image of f
⇔ f is a function from [n] to [k] r S.
Hence the claim is established.

Therefore using the Version III of Inclusion and exclusion theorem and the above claim, we have
k  
X X X k
|O([n], [k])| = (−1)|S| |N (S)| = (−1)|S| (k − |S|)n = (−1)i (k − i)n .
i=0
i
Sj[k] Sj[k]

1
The result follows since sII (n, k) = k! |O([n], [k])|. 
n
n


Exercise. Show that sII (n, n) = 1 and establish the identity n! = (−1)i (n − i)n .
P
i
i=0
4

1.2. Applications of inclusion and exclusion theorem in number theoretic results.

Theorem 1.8. If n = pn1 1 pn2 2 . . . pnk k , where for each i ∈ [k], ni ≥ 1 are integers and pi is a prime
number. Then following formula holds.
    
1 1 1
φ(n) = n 1 − 1− ... 1 − .
p1 p2 pk
Here φ denotes the Euler’s phi function.
Proof : For each i ∈ [k], let Pi denote the property that x(∈ [n]) is divisible by the prime pi .
Claim : The set of elements of [n] that satisfy none of the properties P1 , . . . , Pk is the set
R := {x ∈ [n] : gcd(x, n) = 1}
Recall that φ(n) = |R|.
Proof of claim : We note that if x does not satisfy Pi for some i ∈ [k], then gcd(x,!pi ) = 1. This
i −1
nP
ni

n −j−1 n −j j j
implies there exists u, v ∈ Z such that xu + pi v = 1. Hence x j x
i
u i p v + pni v ni = 1.
j=0
ni
Consequently, if x does not satisfy Pi for some i ∈ [k], then gcd(x, Q p ) = n1.i So if x does not satisfy
P1 , . . . , Pk , then there exist ui , vi ∈ Z, with i ∈ [k], such that (xui + p vi ) = 1 ⇔ xU + nV = 1
i∈[k]
for some U, V ∈ Z. Therefore, gcd(x, n) = 1 and consequently x ∈ R. Conversely, let x ∈ R, i.e.
gcd(x, n) = 1. Then there exists U, V ∈ Z such that xU + nV = 1, i.e. xU + pj ( pnj V ) = 1 for each j ∈ [k].
Hence gcd(x, pj ) = 1 for each j ∈ [k] and the claim is established.
For S j [k], let
N (S) := {x ∈ [n] : x satisfies property Pi for each i ∈ S}.
Claim : For S j [k], |N (S)| = Qnpi .
i∈S

Proof of claim : We note the following:

x ∈ N (S) ⇔ x satisfies property Pi for each i ∈ S
⇔ pi |x for each i ∈ S
!
Y
⇔ pi |x since for i, j ∈ S with i 6= j, gcd(pi , pj ) = 1.
i∈S
Therefore we have the following:
 ! 

 Y n 
N (S) = x ∈ [n] : x = r pi where r ∈  Q 
 pi 
i∈S
i∈S
Hence the claim is established.
Therefore using the Version III of Inclusion and exclusion theorem and the above claim, we have
X
φ(n) = |R| = (−1)|S| |N (S)|
Sj[k]
X n
= (−1)|S| Q
pi
Sj[k] i∈S
X 1
=n (−1)|S| Q
pi
Sj[k] i∈S
k  
Y 1
=n 1− .
i=1
pi
 5

Q
The result follows because of the convention that pi = 1. 
i∈∅
P
Theorem 1.9. For each integer n ≥ 1, n = φ(d).
d|n

Proof : Let gcd(x, n) = d, then there exist u, v ∈ Z such that xu + nv = d ⇔ xd u + nd v = 1. So

x 7→ y := xd is a bijective mapping from {x ∈ [n] : gcd(x, n) = d} to {y ∈ [ nd ] : gcd(y, nd ) = 1}. Hence
|{x ∈ [n] : gcd(x, n) = d}| = φ( nd ). We note that {x ∈ [n] : gcd(x, n) = d} = ∅ if and only if d - n. Hence
n
n = | t {x ∈ [n] : gcd(x, n) = d}| = | t {x ∈ [n] : gcd(x, n) = d}|
d=1 d|n
X
= |{x ∈ [n] : gcd(x, n) = d}|
d|n
X n
= φ .
d
d|n

n
P
P
The result follows since φ d = φ(d). 
d|n d|n

1.3. The Mőbius Function.

Definition. The Möbius function µ : {n ∈ N : n ≥ 1} → {−1, 0, 1} is defined in the following way:
k
pni i , where i ∈ [k] and ni ≥ 1 are integers,
Q
µ(1) = 1; for n =
i=1

k
!
(−1)k

Y if n1 = · · · = nk = 1,
µ(n) = µ pni i =
0 if nj ≥ 2 for some j ∈ [k].
i=1

Theorem 1.10. For each positive integer n,


X 1 if n = 1,
µ(d) =
0 if n 6= 1.
d|n
P
Proof : If n = 1, then the only divisor of 1 is d = 1. Hence µ(d) = µ(1) = 1 So suppose n ≥ 2 and
d|n
k
pni i . Then we have :
Q
n=
i=1
X X X
µ(d) = µ(d) = (−1)|S| = 0.
d|n d|(p1 ···pk ) Sj[k]

This completes the proof. 

k
pni i ,
Q
Corollary 1.11. For each positive integer n =
i=1

φ(n) X µ(d)
= .
n d
d|n

Proof : We use Theorem 1.9 and and the arguments to prove Theorem 1.10 :
k   X (−1)|S|
Y 1 X µ(d) X µ(d)
φ(n) = n 1− =n Q =n =n .
i=1
pi pi d d
Sj[k] i∈S d|(p1 ···pk ) d|n

This completes the proof. 

6

P 1.12 (Mőbius Inversion Formula). Let f : N → R be a function and g : N → R is defined as

Theorem
g(n) = f (d), then
d|n
X n
f (n) = µ(d)g( ).
d
d|n

Proof : We begin with the right hand side of the required equality and use Theorem 1.10 :
X n X n
µ(d)g( ) = µ( )g(d)
d d
d|n d|n
X n X 0
= µ( ) f (d )
d 0
d|n d |d
 
X X 0
= µ(1)f (n) + f (d)  µ(d )
d|n d0 | n
d
d6=n
X
= µ(1)f (n) + f (d)(0)
d|n
d6=n

The result follows since µ(1) = 1. 

 1. Recurrence Relations
1.1. Stirling Numbers of first kind and their recurrence relations.
Definition. Let n and k be positive integers with n ≥ k. A cycle of length l is a permutation π : [n] → [n],
such that there exists l integers say 1 ≤ i1 < i2 < . . . < il ≤ n with the property that π(il ) = i1 , for
each j ∈ [l − 1], π(ij ) = ij+1 and for each i ∈ [n] r {i1 , . . . , il }. We denote such cycle of length l as
(i1 i2 . . . il ). We say two such cycles (i1 i2 . . . il ) and (j1 j2 . . . jm ) are disjoint if the sets {i1 , i2 , . . . , il )}
and {j1 j2 , . . . , jm } have empty intersection.
Theorem 1.1 (Cycle decomposition theorem). For each permutation π : [n] → [n] there exist an integer
k = k(π) such that π can be written as product of k many disjoint cycles.
Proof : Omitted. 
Definition. c(n, k) counts the number of permutations of [n], which can be written as a product of k
disjoint cycles. The Stirling’s number of first kind is defined and denoted as
sI (n, k) = (−1)n−k c(n, k).
Theorem 1.2. Let n and k be positive integers with n ≥ k, then
c(n, k) = c(n − 1, k − 1) + (n − 1)c(n − 1, k).
Proof : Let C([n], k), A and B denote the set of all permutations of [n], which can be written as a
product of k many disjoint cycles; the set of all permutations of [n], which can be written as a product
of k many disjoint cycles, where n belongs is a cycle of length 1 and the set of all permutations of [n],
which can be written as a product of k many disjoint cycles, where n belongs is a cycle of length at least
2.
Claim : |A| = c(n − 1, k − 1).
Proof of claim : Exercise!
Claim : |B| = (n − 1)c(n − 1, k).
Proof of claim : Exercise!

I
Theorem 1.3. Let s (n, k) denote the Stirling number of first kind, then
n
X n
X
n! = c(n, k) = (−1)n−k sI (n, k).
k=1 k=1

1.2. Fibonacci Numbers and their recurrence relations.

Definition. The Fibonacci sequence (numbers) {fn }∞
n=0 is defined by the recurrence relation

fn = fn−1 + fn−2
for each n ≥ 2, with f0 = 0, f1 = 1.
Theorem 1.4. Let fn denote the n−the Fibonacci number. Then
Xn Xn
1+ f2i = f2n+1 and 1+ f2i−1 = f2n .
i=1 i=1

Proof : Exercise! Hint: Use induction on the parameter n. 

Theorem 1.5 (Cassini). Let n be a positive integer and fn denote the n−th Fibonacci number, then
(fn )2 = fn+1 fn−1 + (−1)n+1 .
Proof : Exercise! Hint: Use induction on the parameter n. 
1
2

Definition. The continued fraction associates with the sequence {xn }∞

n=0 is the number
1
x0 +
x1 + x + 1 1
2 x3 +...
√
Lemma 1.6. The continued fraction associates with the {1n (= 1)}∞
n=0 is τ =
1+ 5
2 .

Proof : Note that τ > 1 and satisfies the property τ = 1 + τ1 . So τ is the positive root of the equation
√
x2 − x − 1 = 0, which is 1+2 5 . 
Theorem 1.7. Let fn denote the n−th Fibonacci number, then
fn+1
lim = τ.
n→∞ fn

1 2 f3 1 3 f4
Proof : We note that 1 + 1 = 1 = f2 , 1+ 1+ 11
= 2 = f3 and in using induction on the parameter k,
we have the continued fraction for the finite sequence up to k terms i.e. for {1n (= 1)}kn=1 is fk+1
fk := τk .
Note that τn+1 = 1 + τ1n , for each positive integer n. Therefore for each integer n ≥ 2, we have:
1 1
|τn − τ | = |1 + −1− |
τn−1 τ
1
= |τn−1 − τ |
τ τn−1
1
≤ |τn−1 − τ |
τ √
1 5 1
≤ n−1 |τ1 − τ | = .
τ 2 τ n−1
1
The result follows since 0 < τ < 1. 
Theorem 1.8. Let fn denote the n−th Fibonacci number, then
τn (1 − τ )n
fn = √ − √ .
5 5
Proof : It will be proved later. 
Definition. Let k be a positive integer. A real valued sequence {xn }∞
is said to satisfy the k−term
n=0
recurrence relation or simply recurrence relation if there exists a function λ : Rk → R such that x0 , . . . xk−1
is known (referred as initial conditions) and for each integer n ≥ 0,
xn+k = λ(xn+k−1 , . . . , xn ).
We say such function λ as a k−term recurrence function or simply recurrence function (if there is no
ambiguity on the number of initial conditions) with initial conditions x0 , . . . , xk−1 .
A function θ : Rk → R is said to be a linear function if θ(x̄ + αȳ) = θ(x̄) + αθ(ȳ), where x̄, ȳ ∈ Rk
and α ∈ R. A function ν : Rk → R is said to be a homogeneous function of degree n if for each α ∈ R,
ν(αx1 , . . . , αxk ) = αn ν(x1 , . . . , xk ).
A real valued sequence {xn }∞ n=0 is said to satisfy the k−term linear recurrence relation or simply
linear recurrence relation if there exists a k−term recurrence function λ : Rk → R with initial conditions
x0 , . . . , xk−1 and such λ is a linear function. If such λ is non linear, then {xn }∞ n=0 is said to satisfy the
k−term non linear recurrence relation or simply non linear recurrence relation.
A real valued sequence {xn }∞ n=0 is said to satisfy the k−term homogeneous recurrence relation of degree
n or simply homogeneous recurrence relation of degree n, if there exists a k−term recurrence function
ν : Rk → R with initial conditions x0 , . . . , xk−1 and such ν is a homogeneous function of degree n. If such
ν is non homogeneous, then {xn }∞ n=0 is said to satisfy the k−term non homogeneous recurrence relation
or simply non homogeneous recurrence relation.
Example. Suppose for each non-negative integer i, ci is a real number. Verify the following:-
 3

(a) (x, y) 7→ c0 x + c1 y is a linear and degree 1 homogeneous function from R2 to R.

(b) (x, y, z) 7→ c0 x + c1 y + c2 z is a linear and degree 1 homogeneous function from R3 to R.
(c) (x, y) →7 c0 x3 + c1 x2 y + c2 xy 2 + c3 y 3 is a non-linear and degree 3 homogeneous function from R2 to
R.
(d) (x, y, z) 7→ c0 x2 + c1 y 2 + c3 z 2 + c4 xy + c5 xz + c6 yx is a non-linear and degree 2 homogeneous function
from R3 to R.
1.3. Linear Recurrence Relation. The set of all complex valued sequences is denoted by S(C). It
forms an infinite dimensional vector space over the field C.
Definition. A linear operator is a mapping T : S(C) → S(C), which satisfies for each α ∈ C
T ({xn }∞ ∞ ∞ ∞
n=0 + {α · yn }n=0 ) = T ({xn }n=0 ) + αT ({yn }n=0 ).

The kernel of the operator T , is denoted as ker T , is the set

{{xn }∞ ∞ ∞
n=0 ∈ S(C) : T ({xn }n=0 ) = {0n (= 0)}n=0 } .

The set ker T form a vector subspace of the vector space S(C). An eigen vector of the operator T is a
non zero vector {xn }∞ ∞ ∞
n=0 such that there exist α ∈ C with, T ({xn }n=0 ) = {α · xn }n=0 . Such α ∈ C is
called the eigen value of the operator T .
Exercise. Suppose T : C2 → C2 is a operator.
(a) If T (x, y) = (x + y, x − y), then show that T is linear operator with ker T = {(0, 0)}.
(b) If T (x, y = (x + y, αx + αy)), for some α ∈ C, then show that T is linear operator with ker T =
{(x, −x) : x ∈ C}.
Definition. A linear operator R : S(C) → S(C) is said to be the right shift operator if {xn }∞
n=0 7→
{x1+n }∞
n=0 .

Example. Let α be a non zero complex number and {xn (= αn )}∞ n=0 . Then for each positive integer k,
{xn (= αn )}∞
n=0 is an eigen vector with eigen value α k
of the right shift operator Rk .
Remark. We make the the convention that R0 is the identity operator I.
Definition. A k−term recurrence relation is said to be linear homogeneous if it is of the form
k
X
xn+k = ci xn+k−i , (1)
i=1

where ci ∈ Z for each i ∈ [k] and x0 , . . . , xk−1 are known complex numbers (referred as initial conditions).
We associate the unique polynomial
Xk
xk = ci xk−i
i=1
with the above k−term recurrence relation, such polynomial is called the characteristic polynomial of
such k−term recurrence relation.
Example. Let β be a non zero complex number. Then for each positive integer k and for each integer
p, with 0 ≤ p ≤ k − 1,(using induction on the parameter p) we have,
(R − βI)k ({xn (= np β n )}∞ ∞
n=0 ) = {0n (= 0)}n=0 .

1.4. Solving homogeneous linear recurrence relations with constant coefficients.

Problem. Solve the following linear recurrence relation:
x0 = 2, x1 = 5, and for each integer n ≥ 0, xn+2 = 5xn+1 − 6xn .
Solution. We note that (R2 −5R+6I)({xn }∞ ∞
n=0 ) = {0n (= 0)}n=0 . Therefore we associate the polynomial
2
(called characteristic polynomial ) p(x) = x − 5x + 6 = (x − 2)(x − 3) with this recurrence relation. So
2 and 3 are roots of p(x) = 0. Hence the general solution is xn = c1 2n + c2 3n , where c1 and c2 are
4

complex numbers. In order to find the the complex numbers c1 and c2 , we use initial conditions x0 = 2
and x1 = 5.
Therefore we have, c1 + c2 = 2 and 2c1 + 3c2 = 5. Hence c1 = 1 = c2 . Hence the unique solution of
the above equation is xn = 2n + 3n for each integer n ≥ 0. 
Problem. Solve the following linear recurrence relation:
x0 = 0, x1 = 1, and for each integer n ≥ 0, xn+2 = xn+1 + xn .
Solution. We note that (R2 − R − I)({xn }∞ ∞
n=0 ) = {0n (= 0)}n=0 . Therefore the√characteristic polynomial
with this recurrence relation is p(x) = x −x−1 = (x−τ )(x+ τ1 ), where τ = 1+2 5 . So τ and − τ1 are roots
2

of p(x) = 0. Hence the general solution is xn = c1 τ n + c2 (− τ1 )n , where c1 and c2 are complex numbers.
In order to find the the complex numbers c1 and c2 , we use initial conditions x0 = 0 and x1 = 1.
Therefore we have, c1 + c2 = 0 and τ c1 − τ1 c2 = 1. Hence c1 = √15 , c2 = − √15 . Hence the unique
τn √1 (1
solution of the above equation is xn = √
5
− 5
− τ )n for each integer n ≥ 0. 

Problem. Solve the following linear recurrence relation:

x0 = 1, x1 = 1, x2 = 2 and for each integer n ≥ 0, xn+3 = 6xn+2 − 12xn+1 + 8xn .
Solution. We note that (R3 − 6R2 + 12R − 8I)({xn }∞ ∞
n=0 ) = {0n (= 0)}n=0 . Therefore the characteristic
3 2 3
polynomial is x −6x +12x−8 = (x−2) . Hence 2 is the only root with multiplicity 3 of the characteristic
polynomial. Hence the general solution is for each integer n ≥ 0, xn = (c0 + c1 n + c2 n2 )2n , where c0 , c1 , c2
are complex numbers. Applying the initial conditions x0 = 1, x1 = 1, x2 = 2 we get the unique solution
2
c0 = 1, c1 = − 34 and c2 = 14 . Therefore the unique solution is for each integer n ≥ 0, xn = 2n (1− 34 n+ 14 ).
Problem. Solve the following linear recurrence relation:
x0 = 0, x1 = 1, x2 = 3 and for each integer n ≥ 0, xn+3 = 5xn+2 − 8xn+1 + 4xn .
Solution. We note that (R3 − 5R2 + 8R − 4I)({xn }∞ ∞
n=0 ) = {0n (= 0)}n=0 . Therefore the characteristic
3 2 2
polynomial is x − 5x + 8x − 4 = (x − 2) (x − 1). Hence the roots of the characteristic polynomial
are 2 and 1 with respective multiplicity 2 and 1. Hence the general solution is for each integer n ≥ 0,
xn = (c0 + c1 n)2n + c2 1n , where c0 , c1 , c2 are complex numbers. Applying the initial conditions x0 =
1, x1 = 1, x2 = 3 we get the unique solution c0 = −2, c1 = 1 and c2 = 3. Therefore the unique solution
is for each integer n ≥ 0, xn = (n − 2)2n + 3.
Exercise. Solve the following homogeneous linear recurrence relations:
(i) x0 = 1, x1 = 1 and for each integer n ≥ 0, xn+2 + 2xn+1 + 2xn = 0.
(ii) x0 = 3, x1 = −3, and for each integer n ≥ 0, xn+2 = −6xn+1 − 9xn .
(iii) x0 = 6, x1 = 8, and for each integer n ≥ 0, xn+2 = −4xn+1 − 4xn .
(iv) x0 = 6, x1 = 8, and for each integer n ≥ 0, xn+2 = 4xn+1 − 4xn .
(v) x0 = 1, x1 = 2, x2 = 3 and for each integer n ≥ 0, xn+3 = 9xn+2 − 27xn+1 + 27xn .
1.5. Background of the intermediate steps in solving linear recurrence relations.
Theorem 1.9. Let α be a non zero complex number. ker (R − αI) is one dimensional vector subspace of
S(C). Moreover,
ker (R − αI) = h{αn }∞
n=0 i.

Proof : Note that {αn }∞ ∞

n=0 ∈ ker R − αI. Hence ker R − αI is at least one dimensional. Let {xn }n=0 ∈
ker R − αI. This implies
{xn+1 − αxn }∞ ∞
n=0 = {0n (= 0)}n=0 ⇔ xn+1 = αxn

So if x0 = 0, then {xn }∞ ∞ ∞ n ∞
n=0 = {0n (= 0)}n=0 otherwise {xn }n=0 = {x0 α }n=0 and the result follows. 
 5

k
P
Theorem 1.10. Let xn+k = ci xn+k−i , where for each i ∈ [k] ci ∈ Z, be a k−term linear homogeneous
i=1
recurrence relation with initial condition x0 , . . . , xk−1 . Then
k
X
p(R)({xn }∞ ∞ k
n=0 ) = {0n (= 0)}n=0 , where p(R) = R − ci Rk−i .
i=1

Moreover, ker p(R) is a k dimensional vector subspace of S(C).

Proof : We consider the mapping Λ : ker p(R) → Ck by Λ({xn }∞
n=0 ) = (x0 , . . . , xk−1 ).
Claim : The Λ is a linear one-to-one and onto mapping.

Proof of claim : We note that for each α ∈ C and {xn }∞ ∞

n=0 , {yn }n=0 ∈ ker p(R),

Λ({xn }∞ ∞ ∞
n=0 + α{yn }n=0 ) = Λ({xn + αyn }n=0 )
= (x0 + αy0 , . . . , xk−1 + αyk−1 )
= (x0 , . . . , xk−1 ) + α(y0 , . . . , yk−1 ).
which proves Λ is a linear mapping. Now note that
Λ({xn }∞ ∞
n=0 ) = Λ({yn }n=0 ) ⇔ (x0 , . . . , xk−1 ) = (y0 , . . . , yk−1 )
⇔ x0 = y0 , . . . , xk−1 = yk−1

This implies x0 = y0 , . . . , xk−1 = yk−1 and xn+k = yn+k for each n ≥ 0. Hence {xn }∞ ∞
n=0 = {yn }n=0 , which
k
proves Λ is one-to-one. Suppose (a0 , . . . , ak−1 ) ∈ Ck and for each integer n ≥ 0, an+k =
P
ci an+k−i ,
i=1
then {an }∞
n=0 ∈ ker p(R), which proves Λ is onto and this establishes the claim.

The above mapping establishes vector space isomorphism between ker p(R) and Ck . The result follows
since Ck is a k dimensional vector space. 

Lemma 1.11 (Vandermonde Determinant). Let α0 , . . . , αn be a non zero distinct complex numbers.
Then  
1 1 1 ... 1 

 α0 α1 α2 . . . αn  Y
..  = (αi − αj )

 .. .. .. ..
 .
 . . . .  0≤i<j≤n
n n n
α
0 α 1 α 2 . . . αn 
n

Proof : Exercise! Hint first show that using backward induction on the parameter n and the fact that
αip − α0p − α0 (αip−1 − α0p−1 ) = (αi − α0 )αip−1
   
1 1 1 ... 1   1 1 ... 1 
 n

 α0 α1 α2 . . . αn  Y  α1 α2 α3 ... αn 
= (αi − α0 )  .
  
 .. .. .. . . . . . .. 
.. ..   .. .. .. ..

 .
 . .  i=1  n−1 . 
αn αn αn . . . αn  α n−1 n−1 n−1 
0 1 2 n 1 α2 α 3 . . . αn

Theorem 1.12. Let α0 . . . αk−1 be a non zero distinct complex numbers. ker (R − α0 I) ◦ · · · ◦ (R − αk−1 I)
is k dimensional vector subspace of S(C). Moreover,
ker (R − α0 I) ◦ · · · ◦ (R − αk−1 I) = h{α0n }∞ n ∞
n=0 , . . . , {αk−1 }n=0 i.

Proof : We first prove the following claim:

Claim : The vectors {α0n }∞ n ∞
n=0 , . . . , {αk−1 }n=0 are linearly independent vectors in S(C).
6

Proof of claim : Let c0 , . . . , ck−1 ∈ C such that

c0 ({α0n }∞ n ∞ n ∞ ∞
n=0 ) + c1 ({α1 }n=0 ) + . . . + ck−1 ({αk−1 }n=0 ) = {0n (= 0)}n=0
k−1
ci αij = 0. In matrix equation form we have:
P
This implies that for each integer j, with 0 ≤ j ≤ k − 1,
i=0
    
1 1 1 ... 1 c0 0
 α0 α1 α2 ... αk−1   c1  0
.. .. .. ..    =  .. 
    
 .. ..
 . . . . .  .  .
α0k−1 α1k−1 α2k−1 ... k−1
αk−1 ck−1 0
Since α0 , . . . , αk−1 are non zero distinct complex numbers, we have the Vandermonde determinant is non
zero, i.e  
 1 1 1 ... 1 

 α0 α1 α2 . . . αk−1  Y
 .. .. .. . = (αi − αj ) 6= 0

 . . .. .. 
 k−1 . .  0≤i<j≤k−1
α
0 α1k−1 α2k−1 . . . αk−1
k−1 

Therefore c0 = 0 = c1 = . . . = ck−1 and the claim establishes.

We note that for each integers i, j with 0 ≤ i < j ≤ k − 1, we have
(R − αi I) ◦ (R − αj I) = (R − αj I) ◦ (R − αi I).
Therefore we have :
(R − α0 I) ◦ · · · ◦ (R − αi I) ◦ · · · ◦ (R − αk−1 I)({αin }∞
n=0 )
= (R − α0 I) ◦ · · · ◦ (R − αi−1 I) ◦ (R − αi+1 I) ◦ · · · ◦ (R − αk−1 I) ◦ (R − αi I)({αin }∞
n=0 )
= (R − α0 I) ◦ · · · ◦ (R − αi−1 I) ◦ (R − αi+1 I) ◦ · · · ◦ (R − αk−1 I)({0n (= 0)}∞
n=0 )
= {0n (= 0)}∞
n=0

for each integer i with 0 ≤ i ≤ k − 1. Consequently, for each integer i with 0 ≤ i ≤ k − 1 we have
{αin }∞
n=0 ∈ ker (R − α0 I) ◦ · · · ◦ (R − αk−1 I).

Since ker (R − α0 I) ◦ · · · ◦ (R − αk−1 I) is a vector subspace of S(C), so h{α0n }∞ n ∞

n=0 , . . . , {αk−1 }n=0 i is a
subspace of ker (R − α0 I) ◦ · · · ◦ (R − αk−1 I). The result follows by compiling Theorem 1.10 and the
above claim. 
Theorem 1.13. Let α be a non zero complex number. ker (R − αI)k is k dimensional vector subspace of
S(C). Moreover,
ker (R − αI)k = h{αn }∞ n ∞
n=0 , {nα }n=0 , . . . , {n
k−1 n ∞
α }n=0 i.
Proof : We first prove the following claim:
Claim : The vectors {αn }∞ n ∞
n=0 , {nα }n=0 , . . . , {n
k−1 n ∞
α }n=0 are linearly independent vectors in S(C).
Proof of claim : Let c0 , . . . , ck−1 ∈ C such that
c0 ({αn }∞ n ∞
n=0 ) + c1 ({nα }n=0 ) + . . . + ck−1 ({n
k−1 n ∞
α }n=0 ) = {0n (= 0)}∞
n=0

Since α is a non zero complex number, this implies that c0 = 0 and for each integer p, with 1 ≤ p ≤ k − 1,
k−1
P i
ci p = 0. In matrix equation form we have:
i=1
    
1 1 1 ... 1 c1 0
 2 22 23 ... 2k−1  c2  0
 =  .. 
    
 .. .. .. .. ..  ..
 . . . . .  .  .
(k − 1) (k − 1)2 (k − 1)3 ... (k − 1)k−1 ck−1 0
 7

Since α is non zero distinct complex numbers, we have the Vandermonde determinant is non zero, i.e
 
 1 1 1 ... 1 
2 3
2k−1 
 
 2 2 2 ...
 6= 0

 .. .. .. .. ..
 .
 . . . . 

(k − 1) (k − 1)2 (k − 1)3 . . . (k − 1) k−1 

Therefore c0 = 0 = c1 = . . . = ck−1 and the claim establishes.

We note that for each integer p with 0 ≤ p ≤ k − 1,
(R − αI)k ({np αn }∞ ∞
n=0 ) = {0n (= 0)}n=0

for each i ∈ [k]. Consequently, for each integer p with 0 ≤ p ≤ k − 1 we have

{np αn }∞ k
n=0 ∈ ker (R − αI) .

Since ker (R − αI)k is a vector subspace of S(C), so h{αn }∞ n ∞

n=0 , {nα }n=0 , . . . , {n
k−1 n ∞
α }n=0 i is a subspace
k
of ker (R − αI) . The result follows by compiling Theorem 1.10 and the above claim. 
Theorem 1.14. Let α1 , . . . , αk and m1 , . . . , mk be non zero distinct complex numbers and positive inte-
k
gers. ker (R − α1 I)m1 ◦ · · · ◦ (R − αk I)mk is
P
mi dimensional vector subspace of S(C). Moreover,
i=0
m1
ker (R − α1 I) ◦ · · · ◦ (R − αk I)mk = hS1 , . . . , Sk i.
where for each i ∈ [k],
Si = {αin }∞ n ∞ mi −1 n ∞

n=0 , {nαi }n=0 , . . . , {n αi }n=0 .
Proof : Exercise! Hint: Use Theorem 1.12 and Theorem 1.13. 
1.6. Non Homogeneous Recurrence Relation.
Example (Tower of Hanoi). Consider the recurrence relation x0 = 0 and for each positive integer n,
xn = 2xn−1 + 1. Then
xn = 2(2xn−2 + 1) = 22 xn−2 + 2 + 1
= 22 (2xn−3 + 1) + 2 + 1 = 23 xn−3 + 22 + 2 + 1
..
.
= 2k xn−k + 2k−1 + 2k−2 + . . . + 2 + 1
..
.
2n − 1
= 2n−1 + 2n−2 + . . . + 2 + 1 = = 2n − 1
2−1
The sequence {xn (= 2n − 1)}∞
n=0 is a solution of Tower of Hanoi non homogeneous recurrence relation.
Definition. A k−term recurrence relation is said to be linear non homogeneous if it is of the form
k
X
xn+k = αn + ci xn+k−i , (2)
i=1

where {αn }∞
n=0 ∈ S(C), ci ∈ Z for each i ∈ [k] and x0 , . . . , xk−1 are known complex numbers.
k
Check that (2) can be expressed as p(R)({xn }∞ ∞ k
ci R i .
P
n=0 ) = {αn }n=0 , where p(R) = R −
i=1

Problem. Solve the following recurrence relation:

(R − 2I)2 ({xn }∞ n ∞
n=0 ) = {αn (= 3 + 2n)}n=0 . (3)
with initial condition x0 and x1 .
8

Solution. Firstly, We need to know the general solution,

(R − 2I)2 ({xn }∞
n=0 ) = {0n (= 0)}.
The general solution is (c0 + c1 n)2n , where c0 , c1 ∈ C.
Secondly, we prepare particular solution and trial solution. Let {yn }∞ n=0 be a particular solution, i.e.
we have
(R − 2I)2 ({yn }∞ n ∞
n=0 ) = {3 + 2n}n=0 .
Suppose yn = d1 3n + d2 n + d3 for each integer n ≥ 0 (such assumption is called trial solution). Then we
have for each integer n ≥ 0,
3n + 2n = yn+2 − 4yn+1 + 4yn
= d1 3n + d2 n + (d3 − 2d2 ).
Thirdly, we form the particular solution equation coefficients with the trial solution. Here it is d1 = 1
and d2 = 2 and d3 = 4, i.e for each integer n ≥ 0, yn = 3n + 2n + 4.
Fourthly, we formulate the general solution for (3) and it is for each integer n ≥ 0,
xn = (c0 + c1 n)2n + 3n + 2n + 4.
Finally one solution (may not unique) of (3) can be determined by computing c0 and c1 from the initial
condition x0 and x1 .
Remark. Here we list down some trial solutions.

Given for each integer n ≥ 0,

αn , a non-negative integer and c ∈ C Trial Solution
αn = cnk rn (a0 + a1 n + . . . + ak nk )rn , where a0 , . . . , ak ∈ C
e.g. (a) nrn (a) (a0 + a1 n)rn
(b) n2 rn (b) (a0 + an n + a2 n2 )rn
k n
αn = cn r , here r is one root (a0 + a1 n + . . . + ak nk )nm rn ,
of characteristic polynomial with multiplicity m where a0 , . . . , ak ∈ C
of the given recurrence relation
e.g. (a) nrn (a) (a0 + a1 n)nm rn
(b) n2 rn (b) (a0 + an n + a2 n2 )nm rn
n
αn = cr cos (βn), β ∈ R a1 rn cos (βn) + a2 rn sin (βn)
αn = crn sin (βn), β ∈ R a1 rn cos (βn) + a2 rn sin (βn)
Exercise. Solve the homogeneous linear recurrence relation:- x0 = 1, x1 = 1 and for each integer n ≥ 0,
xn+2 + 6xn+1 + 9xn = cos (αn)
 1. Generating Functions
Definition. A generating function of the complex valued sequence {an }∞
n=0 is the (formal) power series

∞
X
a0 + a1 x + a2 x2 + . . . = an xn .
n=0

denoted as G{an }∞
n=0
(x) or more simply as Ga (x).

Remark. By the word “formal”, we mean that we shall not bother about the radius of convergence of
the above power series Ga (x).

Example.
∞
(a) For {1n (= 1)}∞ xn = 1
P
n=0 , G1 (x) = 1−x .
n=0
∞
1 ∞ xn
= ex .
P
(b) For {en (= n! )}n=0 , Ge (x) = n!
n=0
(c) For {an }∞ ∞
n=0 , {bn }n=0 ∈ S(C), we have

Ga+b (x) = Ga (x) + Gb (x) = (Ga + Gb )(x).

n
(d) For {an }∞ ∞
P
n=0 , {bn }n=0 ∈ S(C), we define c0 = a0 b0 and for n ≥ 1, cn = ai bn−i . The convolution
i=0
of {an }∞ ∞ ∞
n=0 and {bn }n=0 is {cn }n=0 . We have,

Gc (x) = (Ga ∗ Gb )(x) = (Gb ∗ Ga )(x).

(e) A function δk : N → C is called Dirac delta sequence if


1 if n = k
δk (n) =
0 6 k
if n =

∞
We note that (δk ∗ Ga )(x) = xk Ga (x), where Ga (x) = an xn .
P
n=0
∞
an xn we define
P
(f) Let Ga (x) =
n=0

∞ ∞
0 X X
Ga (x) = nan xn−1 = (n + 1)an+1 xn .
n=1 n=0

1.1. Solution of recurrence relation using Generating Function.

Example. We solve the recurrence relation an+1 = 2an for each integer n ≥ 0 with initial condition
a0 = 1.
STEP 1: Formation of generating function related to {an }∞
n=0 . Here it is

Ga (x) = a0 + a1 x + a2 x2 + . . .
∞
X
= an xn
n=0

1
2

STEP 2: Use the recurrence relation to obtain the formula. Here note that
∞
X
Ga (x) = a0 + an+1 xn+1
n=0
X∞
= a0 + 2an xn+1
n=0
X∞
= a0 + 2x a0 xn
n=0
= a0 + 2xGa (x)
(1 − 2x)Ga (x) = a0
∞
a0 1 X
Ga (x) = = = 2n xn
1 − 2x 1 − 2x n=0

FINAL: Conclude the solution. Here the required the solution is {an (= 2n )}∞
n=0 .

Example. We solve the recurrence relation an+1 = 2an + 4n for each integer n ≥ 0 with initial condition
a0 = 1.
STEP 1: Formation of generating function related to {an }∞n=0 . Here it is

Ga (x) = a0 + a1 x + a2 x2 + . . .
∞
X
= an xn
n=0

STEP 2: Use the recurrence relation to obtain the formula. Here note that
∞
X
Ga (x) = a0 + an+1 xn+1
n=0
X∞
= a0 + (2an + 4)xn+1
n=0
X∞ ∞
X
= a0 + 2x 2an xn + x 4 n xn
n=0 n=0
x
= a0 + 2xGa (x) +
1 − 4x
x 1
(1 − 2x)Ga (x) = a0 + =1+
1 − 4x 1 − 4x
1 x 1 1 (1 − 2x) − (1 − 4x)
Ga (x) = + = +
(1 − 2x) (1 − 2x)(1 − 4x) 1 − 2x 2 (1 − 2x)(1 − 4x)
1 1 1 1 1
= + −
(1 − 2x) 2 (1 − 4x) 2 (1 − 2x)
1 1 1 1
= +
2 (1 − 2x) 2 (1 − 4x)
∞   X ∞  
X 1 1 1 n 1 n
= (2x)n + (4x)n = 2 + 4 xn
n=0
2 2 n=0
2 2

FINAL: Conclude the solution. Here the required the solution is {an (= 12 2n + 12 4n )}∞
n=0 .

Example. We solve the recurrence relation an+2 + an+1 − 6an = 0 for each integer n ≥ 0 with initial
condition a0 = 1 and a1 = 3.
 3

STEP 1: Formation of generating function related to {an }∞ n=0 . Here it is

X∞
Ga (x) = a0 + a1 x + a2 x2 + . . . = an xn
n=0
STEP 2: Use the recurrence relation to obtain the formula. Here note that
(1 + x − 6x2 )Ga (x) = 1 + 4x
∞  
X 6 n 1
⇔ Ga (x) = 2 − (−3) xn
n

n=0
5 5
FINAL: Conclude the solution. Here the required the solution is {an (= 56 2n − 15 (−3)n )}∞
n=0 .

1.2. Application in distribution of balls in boxes problems.

Example. There are 3 boxes. The first box contains 2 balls, the second box contains 3 balls and the
third box contains 4 balls.
• The number of ways of selecting k balls with all or none balls from three boxes is the coefficient
of xk in the expression (1 + x2 )(1 + x3 )(1 + xk ).
• The number of ways of selecting k balls with one or none balls from three boxes is the coefficient
of xk in the expression (1 + x)(1 + x)(1 + x).
Remark. Suppose all the boxes contains large number of balls. Our job is to select k many balls from
all the boxes.

Conditions on the selection Coefficient of xk in the expression

3
(1 + x + x2 + . . . + xni )
Q
From Box i choose at most ni many balls, where i ∈ [3]
i=1
From Box i choose at least mi many balls and at most ni
3
(xmi + . . . + xni )
Q
many balls, where i ∈ [3]
i=1
From Box i choose exactly mi1 or mi2 or . . . or miji
3
mij
(xmi1 + xmi2 . . . + x
Q
many balls, where i ∈ [3] i )
i=1

Theorem 1.1. Let S be a multi-set of objects b1 , b2 , . . . , bk with repetitions n1 , n2 , . . . , nk , where for each
i ∈ [k], ni is a positive integer i.e.
S = {n1 · b1 , n2 · b2 , . . . , nk · bk }.
Let an denote the number of ways to express the integer n as a sum of k many non-negative integers,
whenever 0 ≤ n ≤ n1 + . . . + nk , otherwise an = 0, then
Y
1 + x + x2 + . . . + xni .


Ga (x) =
i∈[k]

Proof : Note that

Ga (x) = a0 + a1 x + a2 x2 + . . . ,
where an = 0 for each integer n ≥ 1 + n1 + . . . + nk . Also if we expand the expression
Y
1 + x + x 2 + . . . + x ni

i∈[k]
m1 xm2
as a sum, a typical summand is x . . . xmk , where m1 , . . . , mk are integers satisfies for each i ∈ [k],
0 ≤ mi ≤ ni .

Therefore the coefficient of xn in the summand expression of 1 + x + x2 + . . . + xni is
Q
i∈[k]
X
1 = an .
0≤mi ≤ni ∀i∈[k],
m1 +...+mk =n
4


Definition. An exponential generating function of the complex valued sequence {an }∞
n=0 is the (formal)
power series
∞
a1 x a2 x2 X an n
a0 + + + ... = x .
1! 2! n=0
n!
denoted as eG{an }∞
n=0
(x) or more simply as eGa (x)
Theorem 1.2. Let S be a multi-set of objects b1 , b2 , . . . , bk with repetitions n1 , n2 , . . . , nk , where for each
i ∈ [k], ni is a positive integer i.e.
S = {n1 · b1 , n2 · b2 , . . . , nk · bk }.
Let an denote the number of n permutations of the multi-set S, whenever 0 ≤ n ≤ n1 + . . . + nk , otherwise
an = 0, then
Y x2 xni

x
eGa (x) = 1+ + + ... + .
1! 2! ni !
i∈[k]

Proof :
a1 a2
eGa (x) = a0 + x + x2 + . . . ,
1! 2!
where an = 0 for each integer n ≥ 1 + n1 + . . . + nk . Also if we expand the expression
Y x2 x ni

x
1+ + + ... +
1! 2! ni !
i∈[k]

as a sum, a typical summand is

xm1 xm2 xm k xm1 +...+mk
· ··· = ,
m1 ! m2 ! mk ! m1 ! · · · mk !
where m1 , . . . , mk are integers satisfies for each i ∈ [k], 0 ≤ mi ≤ ni . This implies
Y x2 x ni X xm1 +...+mk

x
1+ + + ... + =
1! 2! ni ! m1 ! · · · mk !
i∈[k] 0≤mi ≤ni
i∈[k]
n1 +...+n
X k X 1
= xm1 +...+mk
n=0
m1 ! · · · mk !
0≤mi ≤n∀i∈[k]
m1 +...+mk =n
n1 +...+n
X k X n! xn
= .
n=0
m1 ! · · · mk ! n!
0≤mi ≤n∀i∈[k]
m1 +...+mk =n
xn
Therefore the coefficient of n! in the summand expression is
X n!
.
m1 ! · · · mk !
0≤mi ≤n∀i∈[k]
m1 +...+mk =n
n!
Now note that m1 !···m k!
, where for each i ∈ [k], mi is an integer with 0 ≤ mi ≤ ni and m1 + . . . + mk = n,
counts the number of n permutations of the multi-set {m1 · b1 . . . , mk · bk }. Hence for each integer
n ≤ n1 + . . . + nk ,
X n!
an = .
m1 ! · · · mk !
0≤mi ≤n∀i∈[k]
m1 +...+mk =n