Lecture Notes on Laplace

Transform, Fourier Series,

Fourier Integral, Fourier
Transform, Partial Differential
Equations, Linear Algebra,
Statistics & Probability Theory
(Engineering Mathematics-II)

by

Prof Shiv Datt Kumar

Department of Mathematics

Motilal Nehru National Institute of

Technology Allahabad, Prayagraj (Uttar

Pradesh), India, Pin-211004

E-mail: sdt@[Link]
2
Contents

1 Laplace Transform 7

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.2 Laplace transforms and their inverses . . . . . . . . . . . . . . . 8

1.2.1 Piecewise continuous function . . . . . . . . . . . . . . . 8

1.2.2 Functions of exponential order . . . . . . . . . . . . . . . 9

1.2.3 Basic Theorems . . . . . . . . . . . . . . . . . . . . . . . 10

1.2.4 Finding the Laplace Transform . . . . . . . . . . . . . . 13

1.2.5 Finding an Inverse Laplace Transform . . . . . . . . . . 15

1.2.6 Heaviside function and t-shifting (Multiplication by e−as ) 16

1.2.7 Solving initial value problems . . . . . . . . . . . . . . . 19

1.2.8 How to handle initial conditions that are not at t = 0 . . 21

1.3 Impulses and the Dirac Delta function . . . . . . . . . . . . . . 22

1.4 Convolutions, and Transfer Theorems . . . . . . . . . . . . . . . 24

1.4.1 Convolutions . . . . . . . . . . . . . . . . . . . . . . . . 24

1.4.2 Transfer Theorems . . . . . . . . . . . . . . . . . . . . . 25

2 Fourier Series 49

2.0.1 Dirichlet Conditions . . . . . . . . . . . . . . . . . . . . 50

3
 2.0.2 Orthogonal functions . . . . . . . . . . . . . . . . . . . . 50

2.0.3 Gibbs phenomenon . . . . . . . . . . . . . . . . . . . . . 55

2.0.4 Parseval Formula . . . . . . . . . . . . . . . . . . . . . . 56

2.0.5 Complex form of Fourier Series . . . . . . . . . . . . . . 57

3 Fourier integral 61

3.1 Fourier integral . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.2 Fourier cosine and sine integral . . . . . . . . . . . . . . . . . . 64

4 Fourier Transform 69

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.2 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.2.1 Fourier cosine transform and Fourier sine transform . . . 71

4.2.2 Relation between Fourier transform and Laplace transform 72

4.3 Properties of Fourier transform . . . . . . . . . . . . . . . . . . 73

4.3.1 Fourier transform of Dirac delta function(Unit impulse

function) . . . . . . . . . . . . . . . . . . . . . . . . . . 77

4.3.2 Amplitude spectrum . . . . . . . . . . . . . . . . . . . . 78

4.3.3 Parseval Identity . . . . . . . . . . . . . . . . . . . . . . 79

5 Partial Differential Equations (PDE) 81

5.1 Linear partial differential equation: . . . . . . . . . . . . . . . . 81

5.1.1 Formation of PDE by eliminating arbitrary constants . . 83

5.1.2 Formation of PDE by eliminating arbitrary functions . . 84

5.1.3 Lagrange linear differential equation . . . . . . . . . . . 86

5.1.4 Surface orthogonal to a given family of surfaces . . . . . 93

4
 5.2 Non-linear partial differential equations . . . . . . . . . . . . . 94

5.2.1 Charpit’s Method . . . . . . . . . . . . . . . . . . . . . . 94

5.3 Higher order linear partial differential equations with constant

coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

5.3.1 Rules for finding CF for homogeneous PDE . . . . . . . 101

5.3.2 Rules for finding CF for non-homogeneous PDE . . . . . 102

5.3.3 General Method for finding CF . . . . . . . . . . . . . . 103

5.3.4 Rules for finding particular integral (PI) . . . . . . . . . 103

6 Applications of PDE 109

6.1 Classification of second order linear PDE’s . . . . . . . . . . . . 109

6.2 Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

6.3 Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

6.3.1 One dimensional heat flow . . . . . . . . . . . . . . . . . 115

6.3.2 Two dimensional heat flow . . . . . . . . . . . . . . . . . 118

6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

7 Matrix Theory and Linear Algebra 123

7.1 Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

7.2 Group Homomorphism . . . . . . . . . . . . . . . . . . . . . . . 126

7.3 Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

7.4 Linear Transformation . . . . . . . . . . . . . . . . . . . . . . . 134

7.5 Matrix representation of linear transformation . . . . . . . . . . 135

7.5.1 Null Space and Range space . . . . . . . . . . . . . . . . 136

7.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

7.6.1 Eigen value (Latent root) . . . . . . . . . . . . . . . . . 138

5
Shiv Datt Kumar Laplace Transform

7.6.2 Eigen vector (Characteristic vector) . . . . . . . . . . . . 139

7.7 Similar matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.8 Matrix Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.9 Solution of general linear system of equations . . . . . . . . . . 155

8 Statistics and Probability 161

8.1 Measures of central tendency . . . . . . . . . . . . . . . . . . . . 162
8.2 Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
8.2.1 Sampling Table . . . . . . . . . . . . . . . . . . . . . . . 166
8.2.2 Definition of Probability . . . . . . . . . . . . . . . . . . 166
8.2.3 Independent events . . . . . . . . . . . . . . . . . . . . . 166
8.2.4 Law of Total Probability (LOTP) . . . . . . . . . . . . . 168
8.2.5 Expected Value . . . . . . . . . . . . . . . . . . . . . . . 168
8.3 Probability distributions: . . . . . . . . . . . . . . . . . . . . . . 170
8.3.1 Binomial distribution . . . . . . . . . . . . . . . . . . . . 172
8.3.2 Poisson distribution . . . . . . . . . . . . . . . . . . . . . 173
8.3.3 Normal (Gaussian) distribution . . . . . . . . . . . . . . 175
8.4 Markov and Chebyshev inequalities . . . . . . . . . . . . . . . . 177

6
Chapter 1

Laplace Transform

1.1 Introduction

Laplace transform is an important tool to solve differential equations. Laplace

transform has ability to convert a differential equation into an algebraic equa-
tion which can be solved purely with algebra. The Laplace transform can be
applied to functions which are not continuous, which makes it extremely use-
ful in real world applications. For example, if a hammer hits a mass-spring
system, or an electrical switch is turned on or off, the Laplace transform can
handle the complications which these simple changes bring to an ODE. The
methods we have learned before require that our driving forces are continuously
differentiable functions.

Laplace transforms replace ODEs with equations which can be solved alge-
braically. Laplace transforms are more advantageous than the classical meth-
ods. Initial values are automatically taken care of when using Laplace trans-
forms. In addition, Laplace transforms increase the types of functions, which

7
Shiv Datt Kumar Laplace Transform

can include as inputs to an ODE, so that non continuous functions and impulses
can be included.

The idea is a three step process.

1. Convert and ODE into an algebraic equation using the Laplace transform.

2. Solve the corresponding algebraic equation (which often means we have

to use a partial fraction decomposition and completing the square).

3. Use an inverse Laplace transform to find the solution of the ODE.

1.2 Laplace transforms and their inverses

1.2.1 Piecewise continuous function

A function is called piecewise continuous function on a given interval if the

interval can be broken into a finite number of subintervals on which the function
is continuous on each open subinterval and it has a finite limit at the endpoints
of each subinterval.

For example, consider a function,






−1 if x < 0,


f (x) = 0 if x = 0





1
 if x > 0.

The function is continuous for all x ∈ R except x = 0. .

8
Shiv Datt Kumar Laplace Transform

1.2.2 Functions of exponential order

A function f (t) is said to be of the exponential order α if there exists α and

M > 0 such that

f (t)
|f (t)| ≤ M eαt , t > 0 or lim = f inite number.
t→∞ eαt

This means geometrically that graph of f (t) does not grow faster than the
graph of the function g(t) = M eαt , α > 0. We write f (t) = O(eαt ).

f (t)
Example 1.2.1. 1. f (t) = tn is of exponential order, for limt→∞ =0
eαt

2. f (t) = e−3t is of exponential order, for |e−3t | ≤ et .

3. f (t) = cos t or sin t, t ≥ 0 are of exponential order,

2
4. f (t) = e2t is not of exponential order, for it is not possible to find real
2
numbers α and M such that e2t ≤ M eαt .

Definition 1.2.2. If f (t) is a function of t defined on [0, ∞) such that

1. f (t) is piecewise continuous in any subinterval (t1 , t2 ) in which [0, ∞) can

be divided

2. f (t) is of exponential order i.e. f (t) = O(eα t ), where α = exponential

order

3. s > α.

R∞
Then the integral 0
e−st f (t)dt exists and is absolutely convergent and is called
Laplace transform of f (t). The Laplace transform of a function f (t) defined

9
Shiv Datt Kumar Laplace Transform

for t ≥ 0 is
Z ∞
F (s) = L(f (t)) = e−st f (t)dt.
0

The function f (t) is called the inverse Laplace transform of F (s), and we
write f (t) = L−1 (F (s)).

Remark 1.2.1. 1. If |f (t)| ≤ M eαt , and s > α. then

R∞ R∞ R∞ M
| e−st f (t)dt| ≤ | 0 e−st M eαt dt| = |M 0 e(α−s)t dt| =
0
.. This
s−α
R∞
means that 0 e−st f (t) dt is absolutely convergent.

2. Conditions in the definition of Laplace transform are sufficient but not

1
necessary. For example f (t) = √ is not continuous on any interval [0, T ],
R t R
∞ 1 ∞
2
as limt→0 f (t) = ∞ but e−st √ dt (put t = u2 ) =, 2 e−su du =
R 0 t 0
2 ∞ √
2
p
√ e−z dz (put z = u s) = π/s, s > 0.
s 0

1.2.3 Basic Theorems

1. Laplace transform is a linear operator i.e.

L(a f (t) + b g(t)) = a L(f (t)) + b L(g(t)) for functions f, g and constants
a, b

2. Transform of a derivative:

Z ∞ Z ∞
0 −st 0 −st ∞
L(f ) = e f (t)dt = (e f (t)) 0
+s e−st f (t)dt = sL(f )−f (0).
0 0

We can use this formula to get

L(f 00 ) = sL(f 0 )−f 0 (0) = s[sL(f )−f (0)]−f 0 (0) = s2 L(f )−sf (0)−f 0 (0).

10
Shiv Datt Kumar Laplace Transform

In general
Pn
Lf (n) (t) = sn F (s) − k=1 sn−k f (k−1) (0), where F (s) = L(f (t)).

3. Laplace transform of integral of f (t)

R t
!
1 1 0
R
L f (t) dt = F (s) + f (t) dt
α s s α

Rt
Proof. Let φ(t) = 0
f (u) du be piecewise continuous on [0, ∞) and of
exponential order and φ0 (t) = f (t) except for points for which f (t) is
discontinuous. L(f (t)) = L [φ0 (t)] = sLφ(t)−φ(0) = sLφ(t), since φ(0) =
 
Rt 1 −1 F (s) Rt
0. Therefore L 0 f (u) du = L(f (t) implies L = 0 f (u) du.
s s

4. Laplace transform of f (t) multiplied by power of t (say tn )

dn F (s)
L[tn f (t)] = (−1)n .
dsn

5. Laplace transform of f (t) divided by t


f (t)
 R∞
L = F (s) ds
t s

6. Shifting theorem

L [e−at f (t)] = L(f (t) s→s+a

= F (s + a).

s-shifting: Multiplication by eat

The s-shifting theorem states that F (s − a) = L(eat f (t))

or alternatively L−1 (F (s − a)) = eat f (t).

1 1
7. L [f (kt)] = L(f (t) s→s/k
= F (s/k)
k k
11
Shiv Datt Kumar Laplace Transform

8. Laplace transform of periodic function

If f (t + T ) = f (t), where T is the period of the function, then

1
R T
Lf (t) = e−st f (t) dt
1 − e−sT 0

R∞
Proof. L(f (t)) = 0
e−st f (t)dt

Break up the integration interval in pieces of size T

RT R 2T R 3T
L(f (t)) = 0
e−st f (t)dt + T
e−st f (t)dt + 2T
e−st f (t)dt + · · ·

Put u = t − T for first, u = t − 2T for second integration and so on, then

RT RT RT RT
Lf (t) = 0
e−su f (u)dt+ 0 e−s(u+T ) f (u)du+ 0 e−s(u+2T ) f (u)du+ 0 e−s(u+3T ) f (u)du+
···

RT
= 1 + e−sT + e−2sT + e−3sT · · · 0
e−su f (u)du

on the RHS, geometric series converges if |e−sT | < 1, for Re(s) > 0. Then
1
R T
L(f (t)) = e−st f (t) dt
1 − e−sT 0

9. Convolution of functions
Rt
Convolution of f (t) and g(t)= f ∗ g = 0
f (t − λ)g(λ) dλ

L(f ∗ g) = L(f (t).L(g(t)),

where f (t) ∗ g(t) is called the convulsion or falting of f (t) and g(t).

Convolution Theorem:

If L−1 [(F (s)] = f (t), and L−1 [(G(s)] = g(t), then L−1 [(F (s) G(s)] =
Rt
f (t) ∗ g(t) = 0 f (u)g(t − u)du,

Rt
Proof. Let φ(t) = 0
f (u)g(u − t)du. Then
R∞ Rt R∞Rt
L(φ(t)) = 0
e−st [ 0
f (u)g(u − t)du]dt = 0 0
e−st f (u)g(u − t)du dt.

12
Shiv Datt Kumar Laplace Transform

Now change the order for this double integration between the lines u = 0
and u = t
R∞R∞
L(φ(t)) = 0 u
te−st f (u)g(u − t)dt du
R∞ R ∞
e−st f (u)

= 0 u
g(u − t)dt du

Put t − u = v. Then
R ∞  R ∞
e−su f (u) du 0 e−sv g(v) dv .

= 0

= F (s)G(s)

1.2.4 Finding the Laplace Transform

∞
R∞ e−st
1. If f (t) = 1, then F (s) = 0
e−st 1dt = = 1s , provided s > 0.
−s 0

n!
2. Integration by parts shows that L(tn ) = for integers n.
sn+1
13
Shiv Datt Kumar Laplace Transform

∞
at
R∞ −st at
R∞ −(s−a)t e−(s−a)t
3. If f (t) = e , then F (s) = 0
e e dt = 0
e dt = =
−(s − a) 0
1
, provided s > a. The improper integral computed with the Laplace
s−a
transform will normally exist only for certain values of s.

1 s + iω
4. If f (t) = eiωt , then F (s) = = 2 , provided s > ω. Since
s − iω s + ω2)
s ω
eiωt = cos ωt + i sin ωt, then L(cos ωt) = 2 2
and L(sin ωt) = 2
s +ω s + ω2
e + e−at
 at   
1 at −at 1 1 1
5. L(cosh at) = L = (L(e ) + L(e )) = + =
2 2 2 s−a s+a
s a
. Similarly L(sinh at) = 2 .
s −a
2 2 s − a2

Example 1.2.3. Find the Laplace transform of f (t) = t cos 3t. We have
s
L(cos 3t) = .
s2 +9
 
d s
Hence L(f (t)) = L(t cos 3t) = − .
dx s2 + 9

Example 1.2.4. Error function is defined by

Z t
2 2
err(t) = √ e−x dx
π 0

√ 1
Show that L(err( t)) = p .
s (s + 1)

Solution:

√ Z √t Z √t 
x4 x6

2 −x 2 2 2
err( t)) = √ e dx = √ 1−x + − + · · · dx
π 0 π 0 2! 3!

x3 x5 x7 t3/2 t5/2 t7/2

   
2 2 1/2
= √ x− + − + ··· = √ t − + − + ···
π 3 2!.5 3!7 π 3 2!.5 3!7
Γ(n + 1)
Using the Laplace transform formula L(tn ) =
sn+1
14
Shiv Datt Kumar Laplace Transform

√
 
2 Γ(3/2) Γ(5/2) Γ(7/2)
L(err( t)) = √ − + − ···
π  s3/2 3s5/2 5(2!)s7/2
√

1 1 1 1.3 1 1.3.5 1 1
L(err( t)) = 3/2 1 − . + 2
− 3
− +··· = p
s 2 s 2.4 s 2.4.6 s s (s + 1)

1.2.5 Finding an Inverse Laplace Transform

If the Laplace transforms of two functions are the same, then the two functions
must be the same as well. This allows us to invert the Laplace transform and
obtain the only function with a given Laplace transform. Inverting a Laplace
transform often involves matching the transformed function up with a function
from a table, and then using the table to invert the transform.

7
Example 1.2.5. 1. Find the inverse Laplace transform of F (s) = 3 , first
s
2 2 7 7 2!
note that 3 is the transform of t . We rewrite F (s) = 3 = , and
s s 2 s3
7
then find the inverse transform as L−1 (F (s)) = t2 .
2
3s + 4
2. Find the inverse Laplace transform of F (s) = , note that the
s2 + 25
transforms of cos(5t) and sin(5t) are s/(s2 + 52 ) and 5/(s2 + 52 ). We
 
3s + 4 3s 4 5
rewrite F (s) = 2 = 2 + , and then the inverse
s + 25 s + 25 5 s2 + 25
4
transform is 3 cos(5t) + sin(5t).
5
3s + 1
3. Find the inverse Laplace transform of F (s) =
s2 + 3s + 2
3s + 1 5 2
Since = − , then inverse Laplace is f (t) = 5e−2t −
s2 + 3s + 2 s+2 s+1
2e−t .

s−3
Example 1.2.6. 1. L(e3t cos(πt)) = . In other words, multi-
(s − 3)2 + π 2
plication of the function f (t) by eat means that you first find the Laplace
transform of f and the replace each s with s − a.

15
Shiv Datt Kumar Laplace Transform

2! 2!
2. Compute L(t2 e−5t ) = 3
(start by finding L(t2 ) = 3 and the re-
(s + 5) s
place the s with s − (−5)).

3
3. Inverse Laplace transform of is related to the inverse transform
(s − 4)3
3 2 3 32
of 3 . The transform of t2 is 3 , so the inverse transform of 3 =
s s s 2 s3
3 2
is t . Because there was an s − 4 in the original denominator, we have
t
3
to multiply by e4t to obtain the inverse Laplace transform of as
(s − 4)3
3 2 4t
te .
t
s−2
4. Inverse Laplace transform of is found by first obtaining an
(s − 4)2 + 1
s − 4 in the numerator and then breaking the fraction into two parts,
s−4+4−2 s−4 1
= +2 . We then compute the in-
(s − 4) + 1
2 (s − 4) + 1
2 (s − 4)2 + 1
verse transform as e4t cos(t) + 2e4t sin(t).

5. To use the s shifting theorem, we have to complete the square in the de-
s+2 s+2
nominator, as 2 = has Laplace inverse e−2t cos 3t.
s + 4s + 13 (s + 2)2 + 9
The reason the shifting theorem is true is because (here’s the 3 step short
proof )

Z ∞ Z ∞
−st
at
L(e f (t)) = e at
(e f (t))dt = e−(s−a)t f (t)dt = F (s − a).
0 0

1.2.6 Heaviside function and t-shifting (Multiplication

by e−as )


0, t < 0

The Heaviside function u(t) =

1, t ≥ 0

can be used as a function which turns other functions on and off.

16
Shiv Datt Kumar Laplace Transform

For example, the function t2 u(t) is zero to the left of 0, and then the function
t2 is turned on at t = 0. The function t2 (u(t) − u(t − 3)) turns on the function
t2 at t = 0 and then turns it off at t = 3. The Heaviside function can be used to
piece together piecewise continuous functions by using this “on-off” approach.
For example,
 the function

0, t<0






t2 ,

 0≤t≤1
f (t) =




 2 − t, 1 < t ≤ π




cos t, π < t

can be written using the Heaviside function as
f (t) = t2 [u(t) − u(t − 1)] + (2 − t)[u(t − 1) − u(t − π)] + cos t[u(t − π)].

This section describes how to compute Laplace transforms and inverse

transforms of such piecewise defined functions. Such functions are needed to
model turning on a switch in an electrical network, modifying the driving force
in a mechanical spring system, and many other practical applications.

The t-shifting theorem states

L(f (t − a)u(t − a)) = e−as F (s)

or L−1 (e−as F (s)) = f (t − a)u(t − a), where F (s) = L(f (t)). This theorem is
easiest to use when computing inverse transforms, as it says that multiplication
of F (s) by e−as means that you multiply the inverse of F (s) by u(t) and then
replace all the t’s with t − a. For example we can find the inverse transform
2 2
of F (s) = 2 e−3s by first noting that the inverse transform of 2 is
s +4 s +4
sin(2t), and so we multiply by u(t) and replace each t with t − 3 to obtain

17
Shiv Datt Kumar Laplace Transform

L−1 (F (s)) = sin(2(t − 3))u(t − 3). This is just the curve sin(2t) which has been
shifted 3 units to the right and turned on at t = 3.

Notice that the t-shifting theorem has an e−as , while the s shifting theorem
has an eat . Pay attention to this sign difference. The reason the t-shifting theo-
R∞ R∞
rem is true is because 0 f (t−a)u(t−a)e−st dt = a f (t−a)e−st dt. By the sub-
R∞ R∞
stitution w = (t−a), this becomes 0 f (w)e−sw+a dw = e−as 0 f (w)e−sw dw =
e−as F (s). Again it is a fairly short proof.

To use the t-shifting theorem to compute forward Laplace transforms, you

have to rewrite f (t) in terms of t−a. For example, the function t2 for t > 1 and
0 otherwise is written f (t) = t2 u(t − 1). To use the second shifting theorem, we
write f (t) = t2 u(t−1) = ((t−1)+1)2 u(t−1) = ((t−1)2 +2(t−1)+1)2u(t−1).
Notice how I purposefully inserted a −1 + 1 next to the t. By shifting theorem
 
2 2 1 −s
F (s) = 3
+ 2+ e . Our function f (t − 1) = (t − 1)2 + 2(t − 1) + 1 is
s s s
2 2 2 1
simply f (t) = t +2t+1 whose transform is 3 + 2 + , which we then multiply
s s s
−s
by e to complete the transform. An alternate version of this transformation
uses the formula
L(f (t)u(t − a)) = e−as L(f (t + a))

(meaning replace each t with t + a and then find the transform and multiply
by e−as ). This formula shows that L(t2 u(t − 1)) = e−s L((t + 1)2 ) = e−s L(t2 +
 
−s 2 2 1
2t + 1)) = e + + as before.
s3 s2 s
Both shifting theorems can be applied simultaneously.
(s − 3)e−2s
If F (s) = , then without the e−2s the inverse transform would
(s − 3)2 + ω 2
be e3t cos(ωt) by the s-shifting theorem. The t-shifting theorem says that the
inverse transform is e3(t−2) cos(ω(t − 2))u(t − 2) (multiply by u(t) and then

18
Shiv Datt Kumar Laplace Transform

replace each t with t − 2).

1.2.7 Solving initial value problems

Given an ODE, take the Laplace transform of each side, giving what is called
the subsidiary equation. This equation can be solved for L(y) = Y (s) using
only algebra. Then compute L−1 (Y (s)) to find the solution to the IVP y(t).
This may involve finding a partial fraction decomposition. Laplace transforms
reduce many IVPs to a 3 step process (1) convert to the subsidiary equation,
(2) use algebra to solve for Y , performing a partial fraction decomposition if
needed, (3) find inverse Laplace transform.

Example 1.2.7. 1. Solve the homogeneous IVP y 0 + 2y = 0, y(0) = 1.

Taking the Laplace transform both side. Then sL(y) − y(0) + 2L(y) = 0,
and using the notation L(y) = Y , we have sY − 1 + 2Y = 0. Solving for
1
Y gives the equation Y = . The inverse Laplace transform of both
s+2
sides gives y(t) = e−2t .

2. Solve the nonhomogeneous IVP y 0 + 2y = 3, y(0) = 1

Take the Laplace transform of both side, then we have sY − 1 + 2Y =

3 s+3
. Solving for Y gives the equation Y = . Partial fraction
s s(s + 2)
s+3 3 1
decomposition = − . Inverse Laplace transform of
s(s + 2) 2s 2(s + 2)
3 1
both sides gives y(t) = − e−2t .
2 2

3. Now for a problem with a Heaviside function. Consider the IVP

y 00 + 4y = u(t − 5), y(0) = 0, y 0 (0) = 1, which represents a mass-spring
system with no friction which has a constant driving force of magnitude 3

19
Shiv Datt Kumar Laplace Transform

applied at time t = 5. Take the Laplace transform of each side to obtain.

e−4s e−5s
s2 Y −sy(0)−y 0 (0)+4Y = or (s2 +4)Y = 1+ . Solving for Y gives
s s
1 1
the equation Y = 2 + e−5s . The partial fraction decomposi-
s + 4 s(s2 + 4)
1 A Bs + C
tion 2
= + 2 becomes 0s2 +0s+1 = A(s2 +4)+Bs2 +Cs =
s(s + 4) s s +4
(A+B)s2 +(C)s+(4A). We obtain the 3 equations 0 = A+B, 0 = C, 1 =
4A whose solution is A = 1/4, B = −1/4, C = 0. Our subsidiary equation
 
2 1 s
is now Y = + − e−5s . The inverse Laplace trans-
2(s2 + 4) 4s 4(s2 + 4)  
1 1 1
form of both sides gives y(t) = sin(2t) + − cos(2(t − 5)) u(t − 5),
2 4 4
using the t-shifting theorem.

Example 1.2.8. Solve the IVP y 00 + 4y 0 + 3y = 30e2t , y(0) = 1, y 0 (0) = 3,

take the Laplace transform of each side. This gives the subsidiary equation
30
s2 Y − sy(0) − y 0 (0) + 4(sY − y(0)) + 3Y = .
s−2
Solving for Y gives us
30
(s2 + 4s + 3)Y = s + 7 + or
s−2
s+7 30
Y (s) = 2 + 2
s + 4s + 3 (s + 4s + 3)(s − 2)
(s + 7)(s − 2) + 30 s2 + 5s + 16
= = .
(s + 1)(s + 3)(s − 2) (s + 1)(s + 3)(s − 2)
−2 1 2
Y (s) = + + .
s+1 s+3 s−2
The inverse Laplace transform of Y (s) is

L−1 (Y (s)) = y(t) = −2e−t + e−3t + 2e2t .

Consider an example where a factor appears more than once in the denom-
inator of a partial fraction decomposition.

Example 1.2.9. Solve the IVP y 00 + 4y 0 + 4y = 0, y(0) = 1, y 0 (0) = 3.

20
Shiv Datt Kumar Laplace Transform

Take Laplace transforms of both sides to obtain the subsidiary equation

s2 Y − sy(0) − y 0 (0) + 4(sY − y(0)) + 4Y = 0 or
s2 Y − s − 3 + 4(sY − 1) + 4Y = 0. Solving for Y gives
s+7 s+7
Y = 2 = . Since we have a double factor in the denomi-
s + 4s + 4 (s + 2)2
nator, partial fraction decomposition is
s+7 A B 1 5
2
= + 2
= + . The t-shifting theorem
(s + 2) s+2 (s + 2) s+2 (s + 2)2
gives us the inverse transform y = e−2t + 5te−2t . This is yet another reason
why you multiply by t when you get a repeated factor.

Example 1.2.10. Solve y 00 + 4y 0 + 5y = e−2t cos t, y(0) = 0, y 0 (0) = 1.

s+2
Subsidiary equation is s2 Y − 1 + 4sY + 5Y = .
(s + 2)2 + 1
1(s2 + 4s + 5) + (s + 2) s2 + 5s + 7
Solving for Y gives Y = = .
((s + 2)2 + 1)2 ((s + 2)2 + 1)2
A partial fraction decomposition would be
s2 + 5s + 7 As + B Cs + D
2 2
= + , or
((s + 2) + 1) (s + 2) + 1 ((s + 2)2 + 1)2
2

s2 + 5s + 7 = (As + B)(s2 + 4s + 5) + (Cs + D) = (A)s3 + (4A + B)s2 +

(5A + 4B + C)s + (5B + D).
This gives the system of equations 0 = A, 1 = 4A + B, 5 = 5A + 4B +
C, 5B + D = 7 which has solutions A = 0, B = 1, C = 1, D = 2. The
1 s+2
subsidiary equation is Y = + which has Laplace
2
(s+ 2) + 1 {(s + 2)2 + 1)}2
t
inverse y = e−2t sin(t) + e−2t sin(t) .
2

1.2.8 How to handle initial conditions that are not at

t=0

If the initial conditions are not stated in terms of t0 = 0, then change variables
t̂ = t − t0 so that the initial conditions are at t̂0 = 0.

21
Shiv Datt Kumar Laplace Transform

Example 1.2.11. Solve the IVP y 00 + y = 3, y(2) = 0, y 0 (2) = 1,

Let t̂ = t−2, y(t̂) = ŷ so that the IVP becomes ŷ 00 + ŷ = 1, ŷ(0) = 0, ŷ 0 (0) = 3

where derivatives in this latter equation are with respect to t̂. The subsidiary
3 1
equation is (taking Laplace transforms) s2 Ŷ − 0s − 1 + Ŷ = , or Ŷ = 2 +
s s +1
3 3
2
. A partial fraction decomposition of the latter term 2 =
(s + 1)s (s + 1)s
A Bs + C
+ 2 gives the equation 3 = A(s2 +1)+(Bs+C)s. Equating coefficients
s s +1
1 3 −3s
gives 0 = A + B, 0 = C, 3 = A, so Ŷ = 2 + + 2 . The inverse
s +1 s s +1
Laplace transform is ŷ = sin t̂ + 3 − 3 cos t̂. Change variables back to obtain
y(t) = sin(t − 2) + 3 − 3 cos(t − 2).

1.3 Impulses and the Dirac Delta function

Dirac delta function is defined as



0
 t 6= a
δ(t − a) = ,

∞ t = a


R∞
which satisfies 0 δ(t − a) dt = 1, and has the sifting property
R∞
0
g(t)δ(t − a) dt = g(a). Dirac delta is really a distribution, and is
technically studied using limits. Consider the function fk (t − a) which has
1 R∞
value k for a < t < t + , yet is zero elsewhere. The integral 0 fk (t) dt =
k
R a+1/k
a
k dt = 1 for all k, so the function fk (t − a) is essentially a short impulse
1
of magnitude k over a time interval of length (so that the total area under
k
the curve is 1). The Dirac delta distribution is studied by considering limits of
fk as k → ∞. The limit limk→∞ fk (t − a) is point-wise the zero function, so it
should have no area underneath. The Dirac delta function is defined so that it

22
Shiv Datt Kumar Laplace Transform

behaves like the limit of the fk functions, but has a positive area under it when
your integral bounds include a. The reason we study this function is because it
is used to describe events which happen instantaneously as in a hammer blow,
flickering a light switch, when lightning strikes, or if a driving force is applied
instantaneously at t = a. In other words, if the force is turned on and then
off essentially instantaneously, then the Dirac delta function is used instead of
the Heaviside function. The Laplace transform of the Dirac delta distribution
R∞
comes easily from the sifting property, as 0 e−st δ(t − a) dt = e−as .

Dirac delta distribution represents derivative of the unit impulse function.

The IVP y 0 = δ(t − 3), y(0) = 0 has subsidiary equation sY = e−3s or Y =
1 −3s
e . The Laplace inverse is y(t) = u(t − 3). In other words, at time 3 a
s
hammer hits the constant solution y(t) = 0 and causes a jump upwards of 1
unit. The derivative of such a jump is undefined in terms of what we learned
in first semester calculus, but the Dirac delta distribution allows us to define
the derivative of a jump discontinuity. The IVP y 0 = 4δ(t − 3), y(0) = 0 would
have solution y = 4u(t − 3) which would result in a jump upwards of 4 units.

We notice that whenever an input involves a Dirac delta distribution, the

solution y will involve a Heaviside function (the change in the solution will be
turned on when the hammer blow occurs).

Example 1.3.1. In the IVP y 00 + 2y 0 + 2y = (1 − u(t − 3))et + 30δ(t − 6),

y(0) = 0, y 0 (0) = 0 has an input force equal to et for 0 < t < 3, and then at
t = 6 it receives an impulse of 30 units. Corresponding subsidiary equation is
1
s2 Y − 0s − 0 + 2sY − 0 + 2Y = + e−3s L(et+3 ) + 30e−6s ,
s−1
−3s+3
1 e
or (s2 + 2s + 2)Y = − + 30e−6s .
s−1 s−1

23
Shiv Datt Kumar Laplace Transform

Since the zeros of s2 + 2s + 2 are imaginary, we cannot factor it any further

and so we complete the square to obtain s2 + 2s + 2 = (s + 1)2 + 1. We have

1 e−3s+3 1
Y (s) = − + 30e−6s .
(s − 1)((s + 1) + 1) (s − 1)((s + 1) + 1)
2 2 (s + 1)2 + 1

1 A B(s + 1) + C
Partial fractions gives = + , where
(s − 1)((s + 1) + 1)
2 s−1 (s + 1)2 + 1
I purposefully wrote B(s + 1) + C as we will have to get a multiple of s + 1
in the numerator anyways when we compute the inverse transform which will
involve the first shifting theorem and a cosine. Multiplying both sides by the
denominator on the left gives 1 = A(s2 + 2s + 2) + (B(s + 1) + C)(s − 1) =
s2 (A + B) + s(2A + B + C) + (2A − B − C), which means 0 = A + B, 0 =
2A + C, 1 = 2A − B − C. So B = −A, C = −2A, and the solution is A =
1/5, B = −1/5, C = −2/5. This means we can write
 
1 (s + 1) 2
Y (s) = − −
5(s − 1) 5((s + 1) + 1) 5((s + 1) + 1) 
2 2

1 (s + 1) 2 1
−e−3s e3 − − +30e−6s
 5(s − 1) 5((s + 1) + 1) 5((s + 1) + 
2 2 1) (s + 1)2 + 1 
1 t 1 −t 2 1 t−3 1 −(t−3)
y(t) = e − e cos(t) − e−t sin(t) −e3 u(t−3) e − e cos(t − 3) −
5 5 5 5 5
2 −(t−3)
e sin(t − 3) + 30 e−(t−6) sin(t − 6).
5

1.4 Convolutions, and Transfer Theorems

1.4.1 Convolutions

The Laplace transform of the product f · g of two functions is not the product
of the Laplace transforms of each (L(f g) 6= L(f )L(g)). Instead, the Laplace
Rt
inverse of H(s) = L(f )L(g) is equal to a quantity h(t) = f ∗g(t) = 0 f (p)g(t−

24
Shiv Datt Kumar Laplace Transform

p)dp called the convolution of f and g (the proof involves interchanging the
order of integration on a double integral). The variable p is a dummy variable
of integration, and could be called anything else. The convolution satisfies
various properties (commutative, distributive, associative), however f ∗ 1 6= f ,
and f ∗ f may be negative.
1 1 1
If H(s) = = 2 = F (s)G(s), where f (t) = t and g(t) =
s2 (s− 1) s (s − 1)
et , then the inverse Laplace transform of H(s) is the convolution of f and
Rt Rt t
g. We compute h(t) = 0 pet−p dp = et 0 pe−p dp = et (−pe−p − e−p ) 0 =
et (−te−t − e−t + 1) = −t−1+et . The convolution is an alternate approach (in-
stead of a partial fraction decomposition) to finding inverse Laplace transforms.
A mass-spring system with ODE my 00 + cy 0 + ky = f (t), y(0) = 0, y 0 (0) = 0 has
1
subsidiary equation s2 mY + scY + kY = F (s) or Y = F (s) =
ms2 + cs + k
W (s)F (s), where W (s) is called the transfer function of the system. Engi-
neers often study mass-spring systems by letting w(t) = L−1 (W (s)) (called the
weight function) and then writing the solution using the convolution y(t) =
w(t) ∗ f (t). This gives an extremely easy way to represent the solution of a
mass-spring system, with a single integral (though the integral may be rather
complex). This formula is called Duhamel’s principle for the system.

1.4.2 Transfer Theorems

Transfer theorems give us rules for computing transforms and inverse trans-
forms of derivatives and integrals. The key transfer theorem we have already
been using is L(y 0 ) = sY − y(0). Also L( ydt), L−1 (F 0 (s)), and L−1 ( F (s)ds)
R R

are important.

25
Shiv Datt Kumar Laplace Transform

Rt F (s)
Laplace transform of f ∗ 1 = f (p)1dp is
0
, which is immediate by
s
1
the convolution theorem since L(1) = . This can be used to find transforms
s
Rt F
of integrals, namely L( 0 f (t)dt) = (remember that the transform of a
s
derivative resulted in multiplying by s, so it seems reasonable the transform of
an integral should involve division by s). Remember that the voltage drop due
Q Rt
to a capacitor is V = , where Q(t) = 0 I(t)dt for the current I(t). Hence
C
the charge is Q(t) = I(t) ∗ 1, the convolution of I and 1, which means that
Q(s) I Q
= . Hence the subsidiary equation of the ODE LI 0 + RI + =
C sC C
E(t) for a power source supplying E(t) volts with I(0) = 0 and Q(0) = 0 is
L(I)
L(sI − I(0)) + RL(I) + = L(E).
sC
Derivative of a transform satisfies the rule L(−tf (t)) = F 0 (s). This can be
1 − s2
 
d s
used to show that L(−t cos(t)) = = , or that L(−t sin(wt)) =
ds s2 + 1 (s2 + 1)2
d w 2ws
= .
ds s2 + w2 (s2 + w2 )2 R
∞
 
f (t)
Integral of a transform satisfies the rule L = F (p)dp, provided
t s
f (t)
limt→0+ exists and is finite.
t

Example 1.4.1. 1. Using the linearity of the Laplace transform, calculate

the Laplace transform of

eat − e−at
f (t) = sinh(at) =
2

Solution:

eat − e−at
 
1 1
L(sinh(at)) = L = L(eat ) − L(e−at )
2 2 2

26
Shiv Datt Kumar Laplace Transform

1 1
= −
2(s − a) 2(s + a)
s + a − (s − a) a
= = (1.1)
2(s2 − a2 ) s 2 − a2

2. Using the shift theorem find the Laplace transform of

f (t) = e2t t2

Solution: By first shift theorem

L e−at f (t) = F (s − a)


(1.2)

where L(f ) = F (s). Now, we know that

2! 2
L t2 = 3 = 3 (1.3)
s s

so, by the shift theorem

2
L e2t t2 =


(1.4)
(s − 2)3

The next two questions are about the Laplace transform of f 0 , recall the
formula
L(f 0 ) = sL(f ) − f (0)

3. Find the Laplace transform of both side of the identity

d
cosh 3t = 3 sinh 3t
dt
27
Shiv Datt Kumar Laplace Transform

and verify that you get the same answer on each side. The idea is that
you do the right hand side using the table entry for sinh(3t) and the left
hand side using the formula for f 0 with f = cosh(3t). cosh(0) = 1 by the
way.

Solution: We know that

a s
L(sinh at) = , L(cosh at) = (1.5)
s 2 − a2 s 2 − a2

and cosh 0 = 1 so

 
d
L cosh 3t = L (3 sinh 3t)
dt
3
sL (cosh 3t) − 1 = 3 2
s −9
s 3
s 2 −1 = 3 2
s −9 s −9
2
s s −9 9
s 2 − 2 = 2
s −9 s −9 s −9
9 9
= 2 (1.6)
s −9
2 s −9

4. Find the Laplace transform of both sides of the differential equation

df
2 =1
dt

with initial conditions f (0) = 4. By solving the resulting equations find

F (s).

Solution: Using linearity of L, plus the property of Laplace transforms of

28
Shiv Datt Kumar Laplace Transform

derivatives, we get

 
dx
L 2 = L(1)
dt
 
df 1
2L =
dt s
1
2sF (s) − 8 = (1.7)
s
(1.8)

This means that

4 1
F (s) = + 2 (1.9)
s 2s

and, since, L(tn ) = n!/sn+1

1
f =4+ t (1.10)
2

To verify that this solves the equation note that f (0) = 4 as required and
f 0 = 1/2.

5. Using the Laplace transform solve the differential equation

f 00 − 4f 0 + 3f = 1 (1.11)

with boundary conditions f (0) = f 0 (0) = 0.

Solution: First, take the Laplace transform of the equation. Since f 0 (0) =
f (0) = 0, if L(f ) = F (s) then L(f 0 ) = sF (s) and L(f 00 ) = s2 F (s). Thus,

29
Shiv Datt Kumar Laplace Transform

the subsidiary equation is

1
s2 F − 4sF + 3F = (1.12)
s

and so

1
(s2 − 4s + 3)F =
s
1 1
F = (1.13)
s s − 4s + 3
2

and, since s2 − 4s + 3 = (s − 3)(s − 1), this gives

1
F = (1.14)
s(s − 3)(s − 1)

Before we can invert this, we need to do a partial fraction expansion.

1 A B C
= + +
s(s − 3)(s − 1) s s−3 s−1
1 = A(s − 3)(s − 1) + Bs(s − 1) + Cs(s − 3)(1.15)

So substituting in s = 0 we get A = 1/3, s = 3 gives B = 1/6 and s = 1

gives C = −1/2. Hence

1 1 1
F = + − (1.16)
3s 6(s − 3) 2(s − 1)

and so
1 1 3t 1 t
f (t) = + e − e (1.17)
3 6 2

30
Shiv Datt Kumar Laplace Transform

6. Using the Laplace transform solve the differential equation

f 00 − 4f 0 + 3f = 2et (1.18)

with boundary conditions f (0) = f 0 (0) = 0.

Solution: We have L(2et ) = 2/(s − 1) on the right hand side. This means
that the subsidiary equation is

2
(s2 − 4s + 3)F = (1.19)
s−1

so
2
F = (1.20)
(s − 1)2 (s − 3)

We need to do partial fractions again, but this is one of those cases with
a repeated root:

1 A B C
= + + (1.21)
(s − 1)2 (s − 3) s − 1 (s − 1)2 s − 3

and multiplying across

1 = A(s − 1)(s − 3) + B(s − 3) + C(s − 1)2 (1.22)

so s = 1 gives B = −1/2 and s = 3 gives C = 1/4. No value of s gives

A on its own, so wee try s = 2:

1 1
1 = −A + + (1.23)
2 4
31
Shiv Datt Kumar Laplace Transform

which means that A = −1/4. Hence

1 1 1
F =− − + (1.24)
2(s − 1) (s − 1) 2 2(s − 3)

and
1 1
f = − et − tet + e3t (1.25)
2 2

7. Using the Laplace transform solve the differential equation

f 00 − 4f 0 + 3f = 0 (1.26)

with boundary conditions f (0) = 1 and f 0 (0) = 1.

Solution: Since

L(f 0 ) = sF − f (0) (1.27)

L(f 00 ) = s2 F − sf (0) − f 0 (0) (1.28)

the subsidiary equation in this case is

s2 F − s − 1 − 4sF + 4 + 3F = 0 (1.29)

so
(s2 − 4s + 3)F = s − 3. (1.30)

Hence
1
F = (1.31)
s−1

32
Shiv Datt Kumar Laplace Transform

and
f (t) = et (1.32)

8. Using the Laplace transform solve the differential equation

y 00 − 2ay 0 + a2 y = 0 (1.33)

with boundary conditions y 0 (0) = 1 and y(0) = 0. a is some real constant.

Solution: Taking the Laplace transform we get

s2 Y − 1 − 2aY + a2 Y = 0 (1.34)

and hence
1
Y = (1.35)
(s − a)2

which means that

y = teat (1.36)

9. Using the Laplace transform solve the differential equation

f 00 + f 0 − 6f = e−3t (1.37)

with boundary conditions f (0) = f 0 (0) = 0.

Solution: The subsidiary equation is

1
s2 F + sF − 6F = (1.38)
s+3

33
Shiv Datt Kumar Laplace Transform

or
1
F = (1.39)
(s + 3)2 (s − 2)

we do partial fractions

1 A B C
= + +
(s + 3)2 (s − 2) s + 3 (s + 3)2 s − 2
1 = A(s + 3)(s − 2) + B(s − 2) + C(s + 3)2 (1.40)

s = −3 gives B = −1/5 and s = 2 gives C = 1/25. Putting in s = 1 we

find
1 16
1 = −4A + + (1.41)
5 25

and so A = −1/25. Putting all this together says that

1 −3t t −3t 1
f =− e − e + e2t (1.42)
25 5 25

10. Using the Laplace transform solve the differential equation

f 00 + 6f 0 + 13f = 0 (1.43)

with boundary conditions f (0) = 0 and f 0 (0) = 1.

Solution: Taking the Laplace transform of the equaiton we get,

s2 F − 1 + 6sF + 13F = 0 (1.44)

34
Shiv Datt Kumar Laplace Transform

and, hence,
1
F = . (1.45)
s2 + 6s + 13

Now, using minus b plus or minus the square root of b squared minus
four a c all over two a, we get

s2 + 6s + 13 = 0 (1.46)

if
√
−6 ± 36 − 52
s= = −3 ± 2i (1.47)
2

which means

s2 + 6s + 13 = (s + 3 − 2i)(s + 3 + 2i) (1.48)

Next, we do the partial fraction expansion,

1 A B
= + (1.49)
s2 + 6s + 13 s + 3 − 2i s + 3 + 2i

and multiplying across we get

1 = A(s + 3 + 2i) + B(s + 3 − 2i) (1.50)

therefore we choose s = −3 + 2i to get

1 i
A= =− (1.51)
4i 4

35
Shiv Datt Kumar Laplace Transform

and s = −3 − 2i to get
1 i
B=− = (1.52)
4i 4

and so
i 1 i 1
F =− + . (1.53)
4 s + 3 − 2i 4 s + 3 + 2i

If we take the inverse transform

i i
f = − e−(3−2i)t + e−(3+2i)t
4 4
i −3t −2it
= e (e − e2it )
4
1 −3t
= e sin 2t (1.54)
2

11. Using the Laplace transform solve the differential equation

f 00 + 6f 0 + 13f = et (1.55)

with boundary conditions f (0) = 0 and f 0 (0) = 0.

Solution: Taking the Laplace transform of the equation gives

1
s2 F + 6sF + 13F = (1.56)
s−1

so that
1
F = . (1.57)
(s − 1)(s + 3 + 2i)(s + 3 − 2i)

We write

1 A B C
= + + (1.58)
(s − 1)(s + 3 + 2i)(s + 3 − 2i) s + 3 − 2i s + 3 + 2i s − 1

36
Shiv Datt Kumar Laplace Transform

giving

1 = A(s−1)(s+3+2i)+B(s−1)(s+3−2i)+C(s+3−2i)(s+3+2i). (1.59)

s = −3 + 2i gives

1 = A(−4 + 2i)(4i) = A(−8 − 16i) (1.60)

so
1 1 8 − 16i 1 + 2i
A=− =− =− (1.61)
8 + 16i 8 + 16i 8 − 16i 40

In the same way, s = −3 − 2i leads to

1 − 2i
B=− (1.62)
40

and, finally, s = 1 gives

1
C= . (1.63)
20

Putting all this together we get

1 + 2i 1 1 − 2i 1 1 1
F =− − + (1.64)
40 s + 3 − 2i 40 s + 3 + 2i 20 s − 1

and so

1 + 2i −(3−2i)t 1 − 2i −(3+2i)t 1
f = − e − e + et
40 40 20
1 −3t  1
(1 + 2i)e2it + (1 − 2i)e−2it + et

= − e (1.65)
40 20
37
Shiv Datt Kumar Laplace Transform

We then substitute in

e2it = cos 2t + i sin 2t

e−2it = cos 2t − i sin 2t (1.66)

to end up with

1 −3t 1
f= e [2 sin 2t − cos 2t] + et (1.67)
20 20

12. Use Laplace transform methods to solve the differential equation


 1, 0 ≤ t < c

00 0
f + 2f − 3f = (1.68)
 0, t ≥ c


subject to the initial conditions f (0) = f 0 (0) = 0.

Solution: Taking Laplace transforms of both sides

1 − e−cs
(s2 + 2s − 3)F =
s
1 − e−cs
F =
s(s2 + 2s − 3)
1
= (1 − e−cs )
s(s − 1)(s + 3)
 
−cs A B C
= (1 − e ) + + (1.69)
s s−1 s+3

38
Shiv Datt Kumar Laplace Transform

Concentrating on the partial fractions part, we have

1 A B C
= + +
s(s − 1)(s + 3) s s−1 s+3
1 = A(s − 1)(s + 3) + Bs(s + 3) + Cs(s − 1)

s=0:

1 = −3A
1
A = −
3
s=1:

1 = 0 + 4B + 0
1
B =
4
s = −3 :

1 = 0 + 012C
1
C =
12

Hence we have

 
−cs 11 1 1 1 1
F = (1 − e ) − + + (1.70)
3 s 4 s − 1 12 s + 3

From the tables, we know that

 
1 1 t 1 −3t 11 1 1 1 1
L − + e − e =− + + (1.71)
3 4 12 3 s 4 s − 1 12 s + 3

39
Shiv Datt Kumar Laplace Transform

and then using the second shift theorem

 
1 1 t 1 −3t 1 1 (t−c) 1 −3(t−c)
f (t) = − + e + e − Hc (t) − + e + e (1.72)
3 4 12 3 4 12

13. Use Laplace transform methods to solve the differential equation


0, 0 ≤ t < 1





f 00 + 2f 0 − 3f = 1, 1 ≤ t < 2 (1.73)



 0, t ≥ 2


subject to the initial conditions f (0) = 0 and f 0 (0) = 0.

Solution: Heaviside function:


 0 t<a

Ha (t) = (1.74)
 1 t≥a


so the Heaviside function is zero until a and then it is one. The right
hand side is zero until t = 1 and then it is one until t = 2 and then it
is zero again. Consider H1 (t) − H2 (t), this is zero until you reach t = 1,
then the first Heaviside function switches on, the other one remains zero.
Things stay like this until you reach t = 2, then the second Heaviside
function switches on as well and you get 1 − 1 = 0. Thus

0, 0 ≤ t < 1





H1 (t) − H2 (t) = 1, 1 ≤ t < 2 (1.75)



0, t ≥ 2



40
Shiv Datt Kumar Laplace Transform

Now, using
e−as
L(Ha (t)) = (1.76)
s

we take the Laplace transform of the differential equation:

e−s e−2s
s2 F + 2sF − 3F = − (1.77)
s s

This gives

1 −s
(s2 + 2s − 3)F = e − e−2s


s
1
e−s − e−2s


F = (1.78)
s(s − 1)(s + 3)

Now
1 1 1 1
=− + + (1.79)
s(s − 1)(s + 3) 3s 4(s − 1) 12(s + 3)

and we know that

 
1 1 t 1 −3t 1 1 1
L − + e + e =− + + (1.80)
3 4 12 3 4(s − 1) 12(s + 3)

However, we know from the second shifting theorem that the affect of the
exponential e−as is to change t to t − a and to introduce an overall factor
of Ha (t). Thus

   
1 1 t−1 1 −3t+3 1 1 t−2 1 −3t+6
f = H1 (t) − + e + e − H2 (t) − + e + e
3 4 12 3 4 12
(1.81)

41
Shiv Datt Kumar Laplace Transform

14. Use Laplace transform methods to solve the differential equation

f 00 + 2f 0 − 3f = δ(t − 1) (1.82)

subject to the initial conditions f (0) = 0 and f 0 (0) = 1.

Solution: We take the Laplace transform using

L(δ(t − a)) = e−as (1.83)

hence
(s2 + 2s − 3)F − 1 = e−s (1.84)

Now, if we do partial fractions on 1/(s2 + 2s − 3) we get

1 1 1
=− + (1.85)
s2 + 2s − 3 4(s + 3) 4(s − 1)

Hence
 
1 1
1 + e−s


F = − + (1.86)
4(s + 3) 4(s − 1)

Since
 
1 −3t 1 t 1 1
L − e + e =− + (1.87)
4 4 4(s + 3) 4(s − 1)

then, by the second shift theorem we have

   
1 −3t 1 t 1 −3t+3 1 t−1
f= − e + e + H1 (t) − e + e (1.88)
4 4 4 4

15. Find the convolution (f ∗ g)(t) when f (t) = t, g(t) = e2t (t ≥ 0).

42
Shiv Datt Kumar Laplace Transform

Solution: From the definition of convolutions

Z t Z t
(f ∗ g)(t) = f (τ )g(t − τ ) dτ = τ e2(t−τ ) dτ
Z0 t Z t0
= τ e2t e−2τ dτ = e2t τ e−2τ dτ
0 0
−2τ
Use integration by parts with u = τ, dv = e dτ
1
du = dτ, v = − e−2τ
Z t 2 Z t 
2t 2t t
= e u dv = e [uv]0 − v du
0 0
h Z t 
2t τ −2τ it 1 −2τ
= e − e − − e dτ
2 0 0 2
1 t −2τ
 Z 
2t t −2t
= e − e +0+ e dτ
2 2 0
t
t e2t

1 −2τ
= − + − e
2 2 2 0
2t
 
t e 1 −2t 1
= − + − e +
2 2 2 2
t 1 1
= − − + e2t
2 4 4

16. Use the convolution theorem to find the function f (t) with

1
L(f ) = .
s2 (s − 4)

1 4t 1
Solution: We know L(t)) = and L(e ) = . From the convolution
s2 s−4
theorem, we see

1
L(f ) = = L(t)L(e4t ) = L(t ∗ e4t )
s2 (s− 4)

43
Shiv Datt Kumar Laplace Transform

so that f (t) is the convolution t ∗ e4t .

Z t
f (t) = τ e4(t−τ ) dτ
Z0 t Z t
4t −4τ
= τe e dτ = e 4t
τ e−4τ dτ
0 0
−4τ
Use integration by parts with U = τ, dV = e dτ
1
dU = dτ, V = − e−4τ
Z t 4 Z t 
4t 4t t
= e U dV = e [U V ]0 − V dU
0 0
h Z t 
4t τ −4τ it 1 −4τ
= e − e − − e dτ
4 0 0 4
1 t −4τ
 Z 
4t t −4t
= e − e +0+ e dτ
4 4 0
t
t e4t

1 −4τ
= − + − e
4 4 2 0
4t
 
t e 1 −4t 1
= − + − e +
4 4 4 4
t 1 1
= − − + e4t
4 16 16

17. Use the formula for the Laplace transform of a periodic function with
period c:
Z c
1
L(f ) = f (t)e−st dt (1.89)
1 − e−cs 0

to find the Laplace transform of a half-rectified wave


 sin t sin t > 0

f (t) =
 0
 sin t ≤ 0

44
Shiv Datt Kumar Laplace Transform

We substitute this into the formula

Z 2π Z π
1 −st 1
L(f ) = f (t)e dt = sin t e−st dt
1 − e−2πs 0 1 − e−2πs 0

We need to do the integral. There are two obvious ways, the first is to
split the sine into exponentials

Z π Z π Z π 
−st 1 (i−s)t −(i+s)t
sin te dt = e dt − e dt
0 2i 0 0
 
1 1 (i−s)π

1 −(i+s)π


= e −1 + e −1
2i i − s i+s

Now, we use
eiπ = e−iπ = −1

and

1 1 −i − s s+i
= =− 2
i−s i − s −i − s s +1
1 1 −i + s s−i
= = 2
i+s i + s −i + s s +1

to get
π
1 + e−sπ
Z
sin te−st dt =
0 1 + s2

or
1 1 + e−sπ 1 1
L(f ) = =
s2 + 1 1 − e−2sπ s2 + 1 1 − e−sπ

where the final equality uses

1 − e−2sπ = 1 − e−sπ 1 + e−sπ

45
Shiv Datt Kumar Laplace Transform

The other way to do the integral is to integrate by parts. Write

π
1 π
Z Z
−st
I= sin te dt = − cos te−st dt
0 s 0
 
1 1 −πs
1
= − − e +1 + I
s s s

and solve for I to get the answer.

Exercise 1.4.1. 1. Apply definition to find directly Laplace transform of

f (t) = sin2 t

2. Use the table of transforms to find Laplace transform of

f (t) = sin 2t + cos 2t

3. Use the table of transforms to find the Inverse Laplace transform of

3s + 1
F (s) = 2
s +4

4. Use the integral

eax
Z
eax cos bx dx = (a cos bx + b sin bx) + C
a2 + b 2

to obtain Laplace cos kt directly from the definition of the Laplace trans-
form.

5. Apply the translation theorem to find the Laplace transform of

f (t) = t5 e−4t

6. Use partial fractions to find the inverse Laplace transforms

5s − 6
(a) F (s) =
s2 − 3s

46
Shiv Datt Kumar Fourier Series

1
(b) F (s) =
s4 − 16

7. Use Laplace transforms to solve the given initial value problem

(a) x00 + 9x = 0 x(0) = 3 x0 (0) = 4

(b) x00 + 4x = cos t x(0) = 0 x0 (0) = 0


x00 + 2x + 4y = 0


(c)
y 00 + x + 2y = 0



x(0) = y(0) = 0 x0 (0) = y 0 (0) = −1.

8. Use Laplace transforms to solve the initial value problem

x00 − 4x = 3t x(0) = x0 (0) = 0

47
Shiv Datt Kumar Fourier Series

48
Chapter 2

Fourier Series

Periodic functions occur frequently in many physical investigations and engi-

neering problems and their representation in simple periodic function such as
sine, cosine is a matter of great importance which leads to Fourier series. Sine
wave (or sinusoidal) is a mathematical curve that describes a smooth repetitive
oscillations. It occurs in signal processing and many other fields.

There are many functions which are periodic functions but are not sinu-
soidal. For example voltage input to circuit or the force on a spring mass
system may be periodic but possesses discontinuities. These functions can be
expressed as a trigonometric series in terms of sine or cosine function within
desired range. This series is called Fourier series i.e. a series expansion of
a function in terms of trigonometric functions cos mx and sin mx is called
Fourier series. Many functions including some discontinuous periodic func-
tions can be expressed in Fourier series if it satisfies the following Dirichlet
conditions.

49
Shiv Datt Kumar Fourier Series

2.0.1 Dirichlet Conditions

For the expansion of a function f (x) in Fourier series the following conditions
are to be satisfied:

1. f (x) is single valued and bounded

2. f (x) is periodic i.e. f (x + 2T ) = f (x), ∀ x ∈

/ [−T, T ],

3. f (x) has atmost finite number of discontinuities in the given interval

4. f (x) has atmost finite number of maxima and minima

Example 2.0.1. 1. f (x) = x sin x, 0 < x < 2π satisfies Dirichlet condi-

tions and has Fourier series expansion.

1
2. f (x) = sin , x 6= 0, can not be expanded as Fourier series in [−π, π].
x
2
Maximum of f (x) occurs at infinite number of points x = . In
(2n − 1)π
the neighbourhood of x = 0, there are infinite number of extreme values.

1
3. f (x) = x2 sin , x 6= 0, can not be expanded as Fourier series in [−π, π].
x
The function f (x) has infinite number of extreme values.

2.0.2 Orthogonal functions

1. A set of continuous functions {f1 (x), f2 (x), . . . , fn (x)}, which do not van-
ish identically in [a, b] is said to be orthogonal if
Rb Rb
a
fm (x) fn (x) dx = 0, if m 6= n and a
fm (x) fn (x) dx 6= 0, if m = n.

2. A set of continuous functions {f1 (x), f2 (x), . . . , fn (x)}, is said to be or-

thogonal on [a, b] with respect to weight function w(x), w(x) > 0, if

50
Shiv Datt Kumar Fourier Series

Rb Rb
a
w(x) fm (x) fn (x) dx = 0, if m 6= n and a
w(x)fm (x) fn (x) dx 6= 0, if
m = n.

Example 2.0.2. 1. Set of functions {1, cos x, sin x, cos 2x, sin 2x, . . . , } is
orthogonal on [−π, π].

πx πx 2πx 2πx
2. Set of functions {1, cos , sin , cos , sin . . . , } is orthogonal on
c c c c
[−c, c].

Useful Integrals
R α+2π R α+2π
α
cos nx dx = α sin nx dx = 0.
R α+2π R α+2π
α
cos mx cos nx dx = α sin mx cos nx dx = 0,
R α+2π R α+2π
α
cos2 nx dx = α sin2 nx dx = π.
Rc mπx nπx Rc mπx nπx
−c
cos cos dx = −c cos sin dx = 0.
c c c c
Rc nπx Rc nπx
−c
cos2 dx = −c sin2 dx = c.
c c
Remark 2.0.1. 1. Sum of periodic functions is also periodic function. Let
1 1 1
f (x) = sin x + sin 2x + sin 3x + sin 4x. Then
2 3 4
period of f (x) = LCM of periods of all functions.

2. Discontinuous function can be represented by Fourier series. This is not

true for Taylor series, for derivative of discontinuous functions do not
exists.

3. It is useful in expanding the periodic functions since outside the closed

interval, there exists a periodic extension of function.

4. Given all modes of oscillation, expansion of an oscillating function by

Fourier series is extremely useful in Physics.

51
Shiv Datt Kumar Fourier Series

5. Fourier series of discontinuous function is not uniformly convergent at all

points.

6. Term by term integration of a convergent Fourier series is always valid

and it may valid even when series is not convergent. However a term by
term differentiation is not valid in most cases.

Definition 2.0.3. Suppose f (x) is a periodic function defined on [−l, l] i.e.

f (x + 2l) = f (x) and satisfies Dirichlet conditions . Then f (x) can be written
as
∞ ∞
a0 X nπx X nπx
f (x) = + an cos + bn sin
2 n=1
l n=1
l

where

Z l Z l Z l
1 1 nπx 1 nπx
a0 = f (x) dx, an = f (x)cos dx, bn = f (x)sin dx
l −l l −l l l −l l

Remark 2.0.2. If period of the function f (x) is 2π i.e. f (x) is defined on [−π, π],
then Fourier series is

∞ ∞
a0 X X
f (x) = + an cos nx + bn sin nx (2.1)
2 n=1 n=1

where

Z π Z π Z π
1 1 1
a0 = f (x) dx, an = f (x) cos nx dx, bn = f (x) sin nx dx−−(∗)
π −π π −π π −π

Remark 2.0.3. 1. From the definition of definite integral, if f (x) is contin-

uous or piecewise continuous (i.e. continuous except finite number of
jump discontinuities), then integrals (∗) exist and f (x) can be expressed

52
Shiv Datt Kumar Fourier Series

in Fourier series.

2. Value of a0 is obtained by integrating (2.1), value of an is obtained by

integrating (2.1) after multiplication of cos nx both sides, value of bn is
obtained by integrating (2.1) after multiplication of sin nx both sides,
using useful integrals given above.

Theorem 2.0.1. If f (x) and f 0 (x) are piecewise continuous on interval [−l, l],
then Fourier series of f (x) at the point of continuity converges to f (x) and at
f (x + 0) + f (x − 0)
the point of discontinuity, it converges to , where f (x + 0),
2
f (x − 0) denote RHL and LHL respectively.

Remark 2.0.4. At the end points of the interval [−l, l], fourier series converges
f (−l + 0) + f (l − 0)
to .
2

Example 2.0.4. Find the Fourier series of



π + x,
 if − π < x < 0
f (x) = , and f (x + 2π) = f (x)

0,
 if 0 ≤ x < π

1 1 1 π2
Hence show that 2 + 2 + 2 + . . . =
1 3 5 8

Solution:
∞ ∞
a0 X X
f (x) = + an cos nx + bn sin nx
2 n=1 n=1

where

π 0 0
x2
Z Z 
1 1 1 π
a0 = f (x) dx = (π + x) dx = πx + =
π −π π −π π 2 −π 2

53
Shiv Datt Kumar Fourier Series

π
1 0
Z Z
1
an = f (x) cos nx dx = (π + x) cos nx dx,
π −π π −π

1 0 1 0
Z Z
= π cos nx dx + x cos nx dx,
π −π π −π
 0
1 πsin nx xsin nx cos nx
= + +
π n n n2 −π
 
1 1 1
= 2
(1 − cos nπ) = 2
[1 − (−1)n ]
π n πn


0,
 if n is even
= ,
 2

 , if n is odd
πn2
1 π 1 0
Z Z
bn = f (x) sin nx dx = (π + x) sin nx dx,
π −π π −π
0
1 −(π + x)cos nx sin nx

= +
π n n2 −π

1 −π
   
1 1 1
= (1 − cos nπ) = = −
π n2 π n n

Therefore Fourier series is

   
π 2 cos x cos 3x sin x sin 2x
f (x) = + + + ... − + + ...
4 π 12 32 1 2

Since x = 0 is a point of discontinuity, then at x = 0, Fourier series converges

 
1 π π 2 1 1
to [f (0+) + f (0−)] = = + + + ... .
2 2 4 π 12 32
1 1 1 π2
Hence 2 + 2 + 2 + . . . = .
1 3 5 8

Theorem 2.0.2. (Term by term differentiation of Fourier series) Let f (x) be

continuous and f 0 (x) be piecewise continuous on interval [−l, l]. Let f (−l) =

54
Shiv Datt Kumar Fourier Series

f (l). Then at every x in (−l, l), where f 0 0(x) exists,

∞
0 πX h  nπx   nπx i
f (x) = n bn cos − an sin .
l n=1 l l

Theorem 2.0.3. (Term by term integration of Fourier series) Let f (x) be piece-
wise continuous on interval [−l, l] and its Fourier series is given by

∞ ∞
a0 X nπx X nπx
f (x) = + an cos + bn sin
2 n=1
l n=1
l

Then for every x in (−l, l), we have

Z l ∞
a0 l X1h  nπx  n  nπx oi
f (x) dx = (x+l)+ an sin + bn cos nπ − cos
−l 2 π n=1 n l l

2.0.3 Gibbs phenomenon

Consider the
 Fourier series expansion of the function

1
 if − π < x < 0
f (x) =

−1
 if 0 ≤ x < π
Function is odd. Then Fourier series is

f (x) = ∞
P
n=1 bn sin nπx, where
2 Rπ 2
bn = 0
− sin nx dx = [cos nπ − 1]
π  nπ

2 0
 if n is even
n
= [(−1) − 1] =
nπ
− 4

 if n is odd
nπ
 
4 sin 3x sin 5x
Then f (x)) = sin x + + +·
nπ 3  5 
4sin x 4 sin 3x
Partial sums S1 = , S2 = sin x + ,
nπ nπ 3

55
Shiv Datt Kumar Fourier Series
 
4 sin 3x sin 5x
S3 = sin x + + , etc
nπ 3 5

2.0.4 Parseval Formula

Let f (x) be periodic with period 2π and is piecewise continuous on (−π, π).
Then " ∞
#
Z π 2
1 a 0
X
|f (x)|2 dx = π + (a2n + b2n ) ,
π −π 2 n=1

provided the Fourier series converges uniformly in (−π, π).

Proof. Consider the Fourier series

∞ ∞
a0 X X
f (x) = + an cos nx + bn sin nx − − − − − − − − − (1)
2 n=1 n=1

1 Rπ 1 Rπ 1 Rπ
where a0 = f (x) dx, a n = f (x) cos nx dx, b n = f (x) sin nx dx.
π −π π −π π −π
Multiply equation (1) by f (x) and integrate term by term from −π to π, then
1 Rπ 2 a0 R π P∞ Rπ P∞ Rπ
|f (x)| dx = f (x) dx+ n=1 a n f (x) cos nx dx+ n=1 b n f (x) sin nx dx
π −π 2 −π −π −π
a0
= (πa0 ) + ∞
P
n=1 [an (πan ) + bn (πbn )],
2
2

a0 P∞ 2 2
=π + n=1 (an + bn ) ,
2

Example 2.0.5. Using Fourier series expansion of f (x) = |x| in (−π, π), show
π2 1 1 1 π4
= 1 + 2 + 2 + · · · (ii) ∞
P
that (i) n=1 =
8 3 5 (2n − 1)4 96

Proof. Since f (x) = |x| is even function in (−π, π), then bn = 0 and Fourier
series is
∞
a0 X
f (x) = + an cos nx
2 n=1

56
Shiv Datt Kumar Fourier Series

2 Rπ
where a0 = |x| dx = π,
π 0  π
2 Rπ 2 xsin x cos nx 2
an = 0
x cos nx dx = + 2
= 2
[(−1)n − 1].
π π x n 0 nπ
Therefore Fourier series is
π 4 P∞ cos (2n − 1)x
   
π 4 cos 3x cos 5x
|x| = − cos x + + + ··· = − n=1 .
2 π 32 52 2 π (2n − 1)2
π2 1 1
Put x = 0 to get (i) = 1 + 2 + 2 + ···.
8 3 5
Now apply Parseval’s formula

" ∞
#
Z π 2
π X 16
|x|2 dx = π + 2 (2n − 1)4
,
−π 2 n=1
π
" ∞
#
2π 3 π 2 16 X 1
=π + 2 .
3 2 π n=1 (2n − 1)4

Hence
∞
X 1 π4
= .
n=1
(2n − 1)4 96

2.0.5 Complex form of Fourier Series

We know that eix = cos x + i sin x and e−ix = cos x − i sin x. Thus
eix + e−ix eix − e−ix
cos x = and sin x = .
2 2i
 inπx 
 

A set of exponential functions e l , n = 0, ±1, ±2, . . . , is orthogonal

 
in the interval (c, c + 2l). Therefore it is possible to represent any arbitrary
function f (x) by linear combination of exponential functions.

57
Shiv Datt Kumar Fourier Series

Then Fourier Series

∞ ∞
a0 X inπx X inπx
f (x) = + an cos + bn sin
2 n=1
l n=1
l

 inπx inπx   inπx inπx 

∞ − ∞ −
a0 X e l + e l  X  e l −e l 
f (x) = + an  + bn  
2 n=1
 2 
n=1
 2i 

∞  inπx X∞   inπx
an − ibn

a0 X an + ibn −
f (x) = + e l + e l
2 n=1
2 n=1
2

a0 an − ibn an + ibn
Let c0 = , cn = , c−n = . Since f (x) is real, cn and c−n are
2 2 2
conjugate of each other.

∞ inπx ∞ inπx
X X −
f (x) = c0 + cn e l + c−n e l ,
n=1 n=1

∞ inπx inπx
1 l
Z
X −
f (x) = cn e l , where cn = f (x) e l dx, n = 0, ±1, ±2, . . . , .
n=−∞
2l −l

Example 2.0.6. Find the complex Fourier series of f (x) = e−x , −π < x < π.
Solution: Then complex form of Fourier series is given by

∞
X
f (x) = cn einx , f or n = 0, ±1, ±2, . . . , where
n=−∞

e−(1+in)π − e(1+in)π
 
1 R π −x −inx 1 R π −(1+in)x
cn = e e dx = e dx = − ,
2π −π 2π −π 2π(1 + in)
e−π (cos nπ − isin nπ) − eπ (cos nπ + isin nπ)
=−
2π(1 + in)
cos nπ(eπ − e−π ) cos nπ sinh π (1 − in)cos nπ sinh π (−1)n (1 − in) sinh π
= = = =
2π(1 + in) π(1 + in) π(1 + n2 ) π(1 + n2 )

58
Shiv Datt Kumar Fourier Series

Hence
∞
!
sinh π X (−1)n (1 − in) inx
f (x) = 2)
e
π n=−∞
(1 + n

Example 2.0.7. Find the frequency spectrum of the periodic pulse defined by


−1, if − 1 < x < 0

f (x) = , and f (x + 2) = f (x)

1,
 if 0 ≤ x < 1

Solution: Period T = 2l = 2, Fundamental angular frequency (ω) = π. Then

complex form of Fourier series is given by

∞ inπx inπx
1 l
Z
X −
f (x) = cn e l , where cn = f (x) e l dx, n = 0, ±1, ±2, . . . , .
n=−∞
2l −l

Z 1 Z 0 Z 1 
1 1
c0 = f (x) dx = − dx + dx = 0,
2 −1 2 −1 0
Z 1
1 i
cn = f (x)e−inx dx = [1 − cos nπ] , n 6= 0
2 −1 nπ

59
Shiv Datt Kumar Fourier Series


1 0,
 if n is even
|cn | = [1 − (−1)n ] =
|nπ|  2

 , if n is odd
|nπ|

60
Chapter 3

Fourier integral

3.1 Fourier integral

If a function f (x) is not periodic, then it can not be expanded by Fourier series
over the entire real line. However we may be able to represent f (x) in an
integral form.

Theorem 3.1.1. Let the function f (x) have the following properties:

1. f (x) is piecewise continuous on every interval [−l, l].

R∞
2. f (x) is absolutely integrable on the x-axis i.e. −∞
|f (x)| dx converges.

3. At every x on the real line, f (x) has left and right hand derivatives.

Then f (x) can be written in integral form as

1 ∞
Z
f (x) = [A(ω) cos ωx + B(ω) sin ωx ] dω,
π 0
R∞ R∞
where A(ω) = −∞ f (t)cosωt dt and B(ω) = −∞ f (t)sin ωt dt.

61
Shiv Datt Kumar Fourier integral

Proof. Consider the Fourier series representation of f (x) on the interval [−l, l].

∞ ∞
a0 X nπx X nπx
f (x) = + an cos + bn sin (3.1)
2 n=1
l n=1
l

where

Z l Z l Z l
1 1 nπx 1 nπx
a0 = f (x) dx, an = f (x) cos dx, bn = f (x) sin dx
l −l l −l l l −l l

nπ nπ (n − 1)π π 1 δω
Set ωn = , ∆ω = ωn − ωn−1 = − = . Thus = Then
l l l l l π
expression (3.1) can be written as

Z l ∞ Z l  
∆ω 1X
f (x) = f (t)dt + f (t)cosωn t dt cos(ωn x) ∆ω
2π −l π n=1 −l

∞ Z l  
1X
+ f (t) sinωn tdt sinωn x ∆ω
π n=1 −l

∆ω R l R∞
As l → ∞, ∆ω → 0 and −l
f (t) dt → 0, since −∞ |f (x)| dx converges,
2π
the summation resembles with Riemann sum of definite integral, we have

Z ∞ Z ∞  Z ∞  
1
f (x) = f (t) cosωt dt cos ωx + f (t) sin ωt dt sinωx dω
π 0 −∞ −∞
(3.2)
R∞ R∞
Denote A(ω) = −∞
f (t)cosωt dt and B(ω) = −∞
f (t)sin ωt dt. Then
equation (3.2) can be written as

Z ∞
1
f (x) = [A(ω) cos ωx + B(ω) sin ωx ] dω
π 0

62
Shiv Datt Kumar Fourier integral

This equation is called Fourier integral representation of f (x).

Theorem 3.1.2. If f (x) satisfies the following properties,

1. f (x) is piecewise continuous on every interval [−l, l].

R∞
2. f (x) is absolutely integrable on the x-axis i.e. −∞
|f (x)| dx converges.

3. At every x on the real line, f (x) has left and right hand derivatives,

then Fourier integral representation of f (x) converges to f (x) at the point of

f (x + 0) + f (x − 0)
continuity and converges to at the point of discontinuity.
2
Example 3.1.1.
 Find the Fourier integral representation of the function




 0, if x < 0


f (x) = 1 , if 0 ≤ x ≤ 1





0 ,
 if x > 1
R ∞ sin(x/2) π
Hence show that 0 =
x 2
R∞ R1 sin ω
Solution: A(ω) = −∞ f (t)cos ωt dt = 0 cos ωt dt = and
ω
R∞ R1 1 − cos ω
B(ω) = −∞ f (t)sin ωt dt = 0 cos ωt dt = .
ω
Then Fourier integral representation of the function

Z ∞
1
f (x) = [A(ω) cos ωx + B(ω) sin ωx ] dω
π 0

∞
1 − cos ω
Z     
1 sin ω
= cos ωx + sin ωx dω
π 0 ω ω
Z ∞
2 1 1
= sin (ω/2) cos ω (x − ) dω
π 0 ω 2
2 R ∞ sin(ω/2) R ∞ sin(x/2) π
Now put x = 1/2. Then 1 = 0
dω. Hence 0 dx = .
π ω x 2
63
Shiv Datt Kumar Fourier integral

Example 3.1.2.
 Find the Fourier integral representation of the function

cos x , if |x| < π/2

f (x) =

0 ,
 otherwise

Solution:
Then Fourier integral representation of the function is

Z ∞
1
f (x) = [A(ω) cos ωx + B(ω) sin ωx ] dω,
π 0

R∞ R π/2
where A(ω) = −∞ f (t)cos ωt dt = −π/2 cos t cos ωt dt
R π/2
= 0 [cos (ω + 1)t + cos (ω − 1) t]dt
π/2  π/2
sin(ω − 1)t

sin(ω + 1)t
= +
1+ω o ω−1 o
1
= 2 [ω (sin (1 + ω)π/2 + sin(ω − 1)π/2 + sin(ω − 1)π/2 − sin(ω + 1)π/2)]
(ω − 1)
2 cos (ωπ/2)
= .
(1 − ω 2 )
R∞ R π/2
B(ω) = −∞ f (t)sin ωt dt = −π/2 cos t. sin ωt dt = 0, for it is an odd function.
2 R ∞ cos (ωπ/2)
Hence f (x) = cos ωx dω, ω 6= 1
π 0 (1 − ω 2 )

3.2 Fourier cosine and sine integral

Let f (x) be defined on half interval. Assume f (x) is defined [0, ∞) and
R∞
0
|f (x)| dx converges. To obtain Fourier cosine integral, we use even ex-
tension of f (x) to 
whole real line.

f (x) ,
 x≥0
Define g(x) =

f (−x) , x < 0

R∞
Since g(x) is even function. A(ω) = 2 0
g(t)cos ωt dt and
R∞
B(ω) = −∞ f (t)sin ωt dt = 0.

64
Shiv Datt Kumar Fourier integral

Therefore Fourier cosine integral representation of the function on [∞, ∞)

1 R∞ R∞
is f (x) = 0
A(ω) cos ωx dω, where A(ω) = 2 0 f (t)cos ωt dt.
π
Let f (x) be defined on half interval [0, ∞). To obtain Fourier sine integral,
we use odd extension
 of f (x) to whole real line.

f (x) ,
 x≥0
Define g(x) =

−f (−x) , x < 0

R∞
Since g(x) is odd function, A(ω) = −∞ f (t)cos ωt dt = 0 and
R∞
B(ω) = 2 0 f (t)sin ωt dt. Therefore Fourier sine integral representation of
1 R∞
the function on [−∞, ∞) is f (x) = B(ω) sin ωx dω, where B(ω) =
π 0
R∞
2 0 f (t) sin ωt dt.

Theorem 3.2.1. If f (x) satisfies the following properties:

1. f (x) is piecewise continuous on every interval [0, l].

R∞
2. f (x) is absolutely integrable on the real axis i.e. −∞
|f (x)| dx converges.

3. At every x ∈ (0, ∞), f (x) has left and right hand derivatives,

then Fourier cosine and sine integral representation converges to f (x) at the
f (x + 0) + f (x − 0)
point of continuity and converges to at the point of dis-
2
continuity.

Example 3.2.1. Let f (x) = e−kx , x ≥ 0, where k is positive constant. Then

find the Fourier cosine integral representation of the function.
 −kx b
R∞ R ∞ −kx e 1
Since 0 |f (x)| dx = 0 e dx = limb→∞ = , then the im-
−k 0 k
proper integral exists. To obtain Fourier cosine integral representation, we
compute
R∞ 2k
A(ω) = 2 0
e−kt cos ωt dt = and
k2 + ω2

65
Shiv Datt Kumar Fourier integral
 
1 R∞ 1 R∞ 2k
f (x) = A(ω) cos ωx dω = cos ωx dω.
π 0 π 0 k2 + ω2

Example 3.2.2. Find solution of the integral equation

R∞
0
f (x) cos ax dx = e−a .

Solution: Given integral is Fourier cosine integral representation. Let g(x) =

1 R∞ R∞
A(ω) cos ωx dω, where A(ω) = 2 g(t)cos ωt dt. Comparing with
π 0 0

given equation x = a, and π g(a) = e−a , f (x) = A(x). Hence A(ω) =

R ∞ e−t
 
2 1
2 0 cos ω t dt = , with k = 1. Therefore f (x) = A(x) =
π π 1 + ω2
2
, x > 0.
π(1 + x2 )

Example 3.2.3. Find the Fourier integral of the function



a , |x| < c

f (x) = .

0 , |x| > c


R ∞ sin cω cos ωx
Hence evaluate 0
dω
ω
1 R∞ a Rc 2a
Solution: A(ω) = −∞
f (t)cos ω t dt = −c
cos ωt dt = sin cω.
π π πω
1 R∞ 1 Rc
B(ω) = −∞
f (t)sin ωt dt = a sin ωt dt = 0.
π π −c
R ∞ 2a 2a R ∞ sin cω cos ωx
f (x) = 0 sin cω cos ωx dω = dω.
πω π 0 ω
Since x = c is the point of discontinuity,
1 a
f (c) = [limx→c−0 f (x) + limx→c+0 f (x)] = .
2 2
Hence





 π/2 , −c < x < c
∞
Z 
sin cω cos ωx π 
dω = f (x) = π/4 , x = ±c
0 ω 2a 




0 ,
 |x| > c

66
Shiv Datt Kumar Fourier integral


1 , |x| < 1

Example 3.2.4. Express the function f (x) =

0 , |x| > 1

R∞ sin ω cos ωx
as Fourier integral and hence evaluate 0
dω.
ω

Solution: By Fourier integral formula

Z ∞ Z ∞ Z ∞ Z 1
1 1
f (x) = f (t)cos ω(t − x) dω dt = cos ω(t − x) dω dt
π −∞ −∞ π ω=0 t=−1

∞ 1 ∞
sin ω(t − x) sin ω(1 − x) − sin ω(−1 − x)
Z  Z  
1 1
= dω = dω
π ω=0 ω −1 π ω=0 ω

1 R ∞ 2 sin ω cos ωx
= dω.
π ω=0 ω
R ∞ sin ω cos ωx π
Hence 0 dω = f (x).
ω 2
R ∞ sin ω π
Put x = 0, then 0 dω = .
ω 2

67
Shiv Datt Kumar Fourier integral

68
Chapter 4

Fourier Transform

4.1 Introduction

Fourier transform is the generalization of complex form of Fourier series. Fourier

transform helps to to extend the Fourier series to non-periodic functions.
Fourier transform is an important concept used in image processing, which
helps to decompose an image into its sine and cosine components. The output
of the transform is a complex valued function and represents the image in the
frequency domain, while the input image is the spatial domain equivalent. In
the frequency domain, image at each point represents a particular frequency
contained in the spatial domain. The Fourier transform has a wide range of
applications, such as image analysis, image filtering, image reconstruction and
image compression.

The Fourier transform is a major cornerstone in the analysis and repre-

sentation of signals and linear, time-invariant systems. Fourier Transform has
important interpretations and meaning in the context of signals and signal

69
Shiv Datt Kumar Fourier Transform

processing. Duality between the time and frequency domains is another im-
portant property of Fourier transforms. This property relates to the fact that
the analysis equation and synthesis equation look almost identical except for
a factor of 1/2π and the difference of a minus sign in the exponential in the
integral.

Two major properties that form the basis for a wide array of signal process-
ing systems are the convolution and modulation properties. According to the
convolution property, the Fourier transform maps convolution to multiplica-
tion; that is, the Fourier transform of the convolution of two time functions is
the product of their corresponding Fourier transforms. For the analysis of lin-
ear, time-invariant systems, this is particularly useful because through the use
of the Fourier transform we can map the sometimes difficult problem of eval-
uating a convolution to a simpler algebraic operation, namely multiplication.
Furthermore, the convolution property highlights the fact that by decomposing
a signal into a linear combination of complex exponentials, which the Fourier
transform does, we can interpret the effect of a linear, time invariant system
as simply scaling the (complex) amplitudes of each of these exponentials by
a scale factor that is characteristic of the system. This ”spectrum” of scale
factors which the system applies is in fact the Fourier transform of the system
impulse response. This is the underlying basis for the concept and implemen-
tation of filtering.

70
Shiv Datt Kumar Fourier Transform

4.2 Definition

Definition 4.2.1. Let f (x) be a function defined for all x ∈ R with values in
C. Then Fourier transform of f (x) denoted as F(f (x)) is defined by

Z ∞
F (s) = F(f (x)) = f (x)e−isx dx. (4.1)
−∞

In other words, Fourier Transform is a mapping F : R → C defined by

R∞
F (s) = F(f (x)) = −∞ f (x)e−isx dx.
Inverse Fourier transform is defined by

Z ∞
−1 1
f (x) = F (f (s)) = F (s)eisx ds. (4.2)
2π −∞

p 1
Remark 4.2.1. Some authors take the coefficient 2/π, in place of 1 and
2π
in (4.1) and (4.2).

4.2.1 Fourier cosine transform and Fourier sine trans-

form

Let f (x) be function defined on 0 < x < ∞. Then Fourier cosine transform of
f (x) denoted as Fc (f (x)) is defined by

Z ∞
Fc (s) = Fc (f (x)) = f (x) cos sx dx. (4.3)
0

Inverse Fourier cosine transform is defined by

Z ∞
2
f (x) = Fc−1 (f (s)) = Fc (s) cos sx ds. (4.4)
π 0

71
Shiv Datt Kumar Fourier Transform

Fourier sine transform of f (x) denoted as Fs (f (x)) is defined by

Z ∞
Fs (s) = Fs (f (x)) = f (x) sin sx dx. (4.5)
0

Inverse Fourier sine transform is defined by

Z ∞
2
f (x) = Fs−1 (f (s)) = Fs (s) sin sx ds. (4.6)
π 0

p
Remark 4.2.2. 1. Some authors take the coefficient 2/π, in place of 1 and
1
.
2π
2. Some authors define Fourier transform

Z ∞
F (s) = F(f (x)) = f (x)eisx dx. (4.7)
−∞

and inverse Fourier transform by

Z ∞
−1 1
f (x) = F (f (s)) = F (s)e−isx ds. (4.8)
2π −∞

Both definition lead to the same conclusion.

4.2.2 Relation between Fourier transform and Laplace

transform

Define 
e−xt φ(t) ,

 t>0
f (t) =

0 ,
 t<0

72
Shiv Datt Kumar Fourier Transform
Z ∞ Z 0 Z ∞
−iyt −iyt
F(f (t)) = e f (t) dt = e f (t) dt + e−iyt f (t) dt
−∞ −∞ 0
Z 0 Z ∞ Z ∞
−iyt −iyt −xt
= e .0 dt + e e φ(t) dt = e−(x+iy)t φ(t) dt
−∞ 0 0
R∞
= 0
e−st φ(t) dt = L(φ(t)), where s = x + iy.

4.3 Properties of Fourier transform

1. Fourier transform is a linear operator i.e.

F(a f (t)+b g(t)) = a F(f (t))+b F(g(t)), for functions f, g and constants
a, b.

1
2. Change of scale property: If F[f (t)] = F (s), then F[f (at)] = F (s/a),
a
a > 0.

3. Shifting property: If F[f (x)] = F (s) and a is a real constant, then

F[f (x − a)] = eisa F (s).

4. Convolution property: Let f1 (t), f2 (t) be piecewise continuous on every

R∞ R∞
interval [−l, l] and −∞ |f1 (t)| dt < ∞, −∞ |f2 (t)| dt < ∞.

Convolution of two functions f1 (t) and f2 (t) is defined by

R∞
f1 ∗ f2 = −∞
f1 (t).f2 (t − λ) dλ.

Let F[f1 (t)] = F1 (s), F[f2 (t)] = F2 (s). Then

F(f1 (t) ∗ f2 (t)) = F(f1 (t).F(f2 (t)) = F1 (s).F2 (s) (convolution with re-
spect to time t). Then
1
F(f1 (t)f2 (t)) = [F1 ∗ F2 ](s) (convolution with respect to frequency s).
2π
73
Shiv Datt Kumar Fourier Transform

Then inverse Fourier transform is given by

F −1 (F1 (s)F2 (s)) = (f1 ∗ f2 )(t).

F −1 [F1 ∗ F2 ](s) = 2π f1 (t)f2 (t).

5. Modulation property: Fourier transform of the product of two time func-

tions is proportional to the convolution of their Fourier transforms. This
simple property provides the basis for the understanding and interpre-
tation of amplitude modulation which is widely used in communication
systems. Amplitude modulation also provides the basis for sampling,
which is the major bridge between continuous-time and discrete-time
signal processing and the foundation for many modern signal processing
systems using digital and other discrete-time technologies. Modulation
property is the dual of the convolution property.
1
If F[f (t)] = F (s), then F[f (t) cos ωt] = [F (s − ω) + F (s + ω)]
2
1
F[f (t) sin ωt] = [F (s + ω) − F (s − ω)].
2

6. Fourier transform of a derivative: Let f (t) be continuous and

f (k) (t), k = 1, 2, . . . , n, be piecewise continuous on every interval [−l, l]
R∞
and −∞ |f (k−1) (t)| dt < ∞ and f (k) (t) → 0, k = 0, 1, 2, . . . , n − 1. Then
F(f (n) (t)) = (is)n F (s).
R∞ ∞ R∞
F(f 0 (t)) = −∞
e−ist f 0 (t) dt = (e−ist f (t)) −∞
+ is −∞
e−ist f (t) dt =
(is) F (s).

We can use this formula to get F(f 00 (t)) = −s2 F (s). and

F(f (n) (t)) = (is)n F (s).

74
Shiv Datt Kumar Fourier Transform

7. Fourier transform of f (t) multiplied by power of t:

R∞
Let f (t) be piecewise continuous on every interval [−l, l] and −∞
|tn f (t)| dt <
∞. Then F(tn f (t)) = in F (n) (s).

Proof:

Z ∞ Z ∞
0 d −ist d
F (s) = f (t)e dt = [f (t)e−ist ] dt
ds −∞ −∞ ds

Z ∞
= −i tf (t) e−ist dt = −iF(tf (t)).
−∞

8. Fourier transform of an integral of f (t)

Let f (t) be piecewise continuous on every interval [−l, l] and

R∞
−∞
|f (k−1) (t)| dt < ∞ and F(f (t)) = F (s). Then
R ∞
!
1
F f (t) dt = F (s).
−∞ is

Example 4.3.1. Find Fourier transform of



t ,
 |t| ≤ a
f (t) =

0 ,
 |t| > 0

Z ∞ Z −a Z a Z ∞
−ist −ist −ist
F(f (t)) = e f (t) dt = e f (t) dt+ e f (t) dt+ e−ist f (t) dt
−∞ −∞ −a a

Z a Z a Z ∞
−ist
= isy
e f (−y) (−dy) + e t dt + e−ist . 0 dt
∞ −a a

75
Shiv Datt Kumar Fourier Transform

.
∞ a
e−ist t e−ist
Z 
isy
= e f (−y) dy + + 2
a −is s −a
Z ∞  
isy a −isa isa 1 −isa isa
= e .0 dy + − (e + e ) + 2 (e −e )
a is s
ia 1 2i
= .2cos sa + 2 (−2i sin sa) = 2 (as cos sa − sin sa)
s s s

Example 4.3.2. Find Fourier transform of



A ,
 |t| ≤ T
f (t) =

0 ,
 |t| > T

Picture in time domain

y
A

t [s]
−T T

Solution: Fourier transfom of f (t) is given by

Z ∞ Z −T Z T Z ∞
−iωt −iωt −iωt
F(f (t)) = e f (t) dt = e f (t) dt+ e f (t) dt+ e−iωt f (t) dt
−∞ −∞ −T T

T
eiωT − e−iωT
Z  
−iωt 2A sin(T ω)
= e A dt = = 2A
−T ω 2i ω

76
Shiv Datt Kumar Fourier Transform

F
sin(T ω)
2T A 2A
ω

ω [rad/s]
3π 2π π π 2π 3π
− − −
T T T T T T

Picture in Frequency Domain

4.3.1 Fourier transform of Dirac delta function(Unit im-

pulse function)

Impulse is a force of very high magnitude for just an instant. A function

representing theimpulse is called Dirac delta function and is defined by
1 , a ≤ t ≤ a + 


δ (t − a) = 

0 , otherwise


u(t) − u(t − a)
or δ(t) = lima→0 ,
a
where u(t) is the unit
 step function and




 0, t<0


u(t) − u(t − a) = 1 , 0 < t < a





0 , t ≥ a


77
Shiv Datt Kumar Fourier Transform

Fourier transform of δ(t)

1
= F(δ (t)) = F(u(t) − u(t − a))
a
R a −iωt eiωa − 1
= lima→0 0 e dt = lima→0 = 1.
iωa

Example 4.3.3. Solve the

 integral equation

R∞ −1 , 0 < ω < 1

0
f (t) cos ωt dt =

1 ,
 1<ω<2
using Fourier transform.

Solution:
R∞
Fourier cosine transform
 is given by Fc (ω) = 0
f (t) cos ωt dt.

−1 , 0 < ω < 1

Then Fc (ω) =

1 ,
 1<ω<2
Taking inverse Fourier cosine transform

Z ∞
2
f (t) = Fc−1 (f (ω) = Fc (ω) cos ωt dω
π 0

1 2
2(sin 2t − 2 sin t)
Z Z 
2
= (−1). cos ωt dω + 1. cos ωt dω = .
π 0 1 πt

4.3.2 Amplitude spectrum

If F(f (t) = F (ω). Then graph (ω, |F (ω)|) is called the amplitude spectrum of
function f (t) and ω is called frequency of the spectrum.

Example  4.3.4. Find the amplitude spectrum of the function


3 , −2 < t < 2

f (t) =

0 , otherwise


78
Shiv Datt Kumar Fourier Transform

R∞ R2
Solution: F (ω) = F(f (t) = f (t) e−iωt dt = −2 3 e−iωt dt
−∞
2 6
= −3 [e−iωt ]−2 = sin 2ω.
ω
Graph of ω verses |F (ω)| is the amplitude spectrum of f (t).

4.3.3 Parseval Identity

If F(f (t) = F (ω), then

Z ∞ Z ∞
2 1
|f (t)| dt = |F (ω|2 dω
−∞ 2π −∞

i.e. power of a signal represented by a function f (t) is the same whether

computed in signal space or frequency(spectrum) space.

Proof. Let F(g(t)) = G(ω) and ḡ(t) = F −1 (Ḡ(ω)), where bar denotes complex
conjugate. Then by definition of inverse Fourier transform

Z ∞ Z ∞  Z ∞ 
1 iωt
f (t)ḡ(t) dt = f (t) Ḡ(ω)e dω dt
−∞ −∞ 2π −∞

By changing the order of integration

Z ∞ Z ∞ 
1 iωt
= Ḡ(ω) f (t)e dt dω
2π −∞ −∞

Z ∞
1
= Ḡ(ω)F (ω) dω.
2π −∞

Now taking f (t) = g(t)

Z ∞ Z ∞
2 1
|f (t)| dt = |F (ω|2 dω.
−∞ 2π −∞

79
Shiv Datt Kumar Fourier Transform

Example 4.3.5. Using Parseval indenty

R∞ 1 π R∞ 1 π
(i) 0 2 2
dt = (ii) 0 2 2
dt =
(t + 1) 4 (t + 1)(t + 4) 12
1
Solution: (i) Let f (t) = g(t) = e−t , t > 0. Then Fc (ω) = Gc (ω) = .
ω2 +1
Now using Parseval identity

Z ∞ Z ∞
2
Fc (ω)Gc (ω) dω = f (t)g(t) dt
π 0 0

Z ∞ Z ∞
2 1 1
dω = e−t e−t dt =
π 0 (ω + 1)2
2
0 2

This implies
Z ∞
1 π
dt =
0 (t2 + 1) 2 4
1 2
(ii) Let f (t) = e−t , g(t) = e−2t , t > 0. Then Fc (ω) = , Gc (ω) = .
ω2 + 1 ω2 + 4
Now using Parseval indenty we have

Z ∞
1 π
dt =
0 (a2 + t2 )(b2 + t2 ) 2ab(a + b)

This implies
Z ∞
1 π
dt = .
0 (t2 2
+ 1)(t + 4) 12

80
Chapter 5

Partial Differential Equations

(PDE)

A partial differential equation is an equation that involves both a function and

its partial derivatives. An equation containing one or more partial derivatives
of dependent variables, independent variable and dependent variables is called
PDE. Dependent variable is denoted by z and independent variables are x and
y.
∂z ∂z ∂ 2z ∂ 2z ∂ 2z
Notations: p = ,q= ,r= , s = , t = .
∂x ∂y ∂x2 ∂x∂y ∂xy2
Order of partial differential equation: Order of the highest derivative
occurring in equation is called order of the partial differential equation.

5.1 Linear partial differential equation:

An equation in which unknown function and its partial derivatives appear lin-
early is called linear PDE. An equation containing x, y, z, p, q i.e. f (x, y, z, p, q) =

81
Shiv Datt Kumar Partial Differential Equations

0 defines a first order P DE. Most general second order linear PDE in two in-
∂ 2z ∂ 2z ∂ 2z ∂z ∂z
dependent variables is a 2 + b +c 2 +d +e + f z = g, where
∂x ∂x∂y ∂y ∂x ∂y
a, b, c, d, e, f, g are functions of x and y. In other words F (x, y, z, p, q, r, s, t) = 0
is a second order PDE.

Example 5.1.1. 1. One dimensional heat equation: ut = c uxx .

2. Laplace equation: uxx + uyy = 0 (homogeneous).

3. One dimensional wave equation : utt = c2 uxx (homogeneous).

4. Poisson equation: uxx + uyy = f (x, y) (non-homogeneous).

Tangent to a curve: Let C : x = x(t), y = y(t), z = z(t) be a curve, where

0 0 0
a ≤ t ≤ b. Then x (t0 ), y (t0 ), z (t0 ) are the direction ratios of the tangent
0
line to the curve at t = t0 . If arc length is parameter of the curve, then x (s0 ),
0 0
y (s0 ), z (s0 ) are the direction cosines of the tangent line to the curve at s = s0 .

Now let C be a curve of intersection of two surfaces S1 : f (x, y, z) = c1 ,

S2 : g(x, y, z) = c2 . Let P (a, b, c) be a point on the curve of intersection. Then
equation of tangent plane to the surface S1 is
∂f ∂f ∂f
(x − a) (P ) + (y − b) (P ) + (z − c) (P ) = 0 − − − − − −(1)
∂x ∂y ∂z

Equation of tangent plane to the surface S2 is

∂g ∂g ∂g
(x − a) (P ) + (y − b) (P ) + (z − c) (P ) = 0 − − − − − −(2).
∂x ∂y ∂z

Two planes intersect along tangent line to the curve. Solving equations

(1) and (2) we get

82
Shiv Datt Kumar Partial Differential Equations

x−a y−b z−c

= = ,
∂(f, g) ∂(f, g) ∂(f, g)
∂(y, z) ∂(z, x) ∂(x, y)

5.1.1 Formation of PDE by eliminating arbitrary con-

stants

Let f (x, y, z, a, b) = 0, where a and b are arbitrary constants. (5.1)

be an equation. Differentiating partially with respect to x and y we get

∂f ∂f ∂z
+p = 0, where p = . (5.2)
∂x ∂z ∂x

∂f ∂f ∂z
+q = 0, where q = . (5.3)
∂y ∂z ∂y

Eliminating a and b between 1.1, 1.2, and 1.3 we obtain first order PDE
g(x, y, z, p, q) = 0.

Example 5.1.2. Form a partial differential equation from

(i) z = axy + b (ii) x2 + y 2 + (z − c)2 = a2 (iii) z = cx + dy + cd
(iv) z = ceωt sin ωx.

Solution: (i) Differentiating partially with respect to x and y we have p = ay

and q = ax from which we obtain PDE px − qy = 0.
(ii) Differentiating partially with respect to x and y we have
x + (z − c) p = 0 ——— (1) and
y + (z − c) q = 0 ——— (2).
−x
Putting (z − c) = from (1) in (2) we get PDE yp − xq = 0.
p
(iii) Differentiating partially with respect to x and y we have p = c, q = d.

83
Shiv Datt Kumar Partial Differential Equations

Then PDE z = px + qy + pq.

(iv) z = ceωt sin ωx ⇒ zxz + ztt = 0.

5.1.2 Formation of PDE by eliminating arbitrary func-

tions

PDE can also be formed by eliminating arbitrary functions from a given equa-
tion. First order PDE can be obtained by eliminating one arbitrary constant.
Second order PDE can be obtained by eliminating two arbitrary constants.
However it may not be always possible to obtain n order PDE by eliminating
n arbitrary constants. We may require higher order PDE and such PDE may
be not be unique.

Example 5.1.3. Derive PDE by eliminating arbitrary function from

(i) f (x2 + y 2 , x2 − z 2 ) = 0. (ii) φ(u, v) = 0, u = xyz, v = x + y + z.

Solution:

(i) Let u(x, y, z) = x2 + y 2 and v(x, y, z) = x2 − z 2 . Then given relation

becomes f (u, v) = 0.

Differentiating partially with respect to x and y we have

   
∂f ∂u ∂u ∂f ∂v ∂v
+p + +p = 0 − − − − − − − − − (1) and
∂u ∂x ∂z ∂v ∂x ∂z

   
∂f ∂u ∂u ∂f ∂v ∂v
+q + +q = 0 − − − − − − − − − (2)
∂u ∂y ∂z ∂v ∂y ∂z

84
Shiv Datt Kumar Partial Differential Equations

∂u ∂u ∂u ∂v
= 2x, = 2y, = 0, = −2z
∂x ∂y ∂z ∂z

∂f ∂f
2x + (2x − 2zp) = 0 − − − − − − − − − (3)
∂u ∂v

.
∂f ∂f
2y − 2zq = 0 − − − − − − − − − − − −(4)
∂u ∂v

.
∂f ∂f
Eliminating and
∂u ∂v

2x 2x − 2pz
= 0 or xqz + xy = pyz.
2y −2qz

(ii) Let φ(u, v) = 0, u(x, y, z) = xyz and v(x, y, z) = x + y + z.

Differentiating φ(u, v) = 0 partially with respect to x and y we have

   
∂φ ∂u ∂u ∂φ ∂v ∂v
+p + +p = 0 − − − − − − − − − (1) and
∂u ∂x ∂z ∂v ∂x ∂z

   
∂φ ∂u ∂u ∂φ ∂v ∂v
+q + +q = 0 − − − − − − − − − (2)
∂u ∂y ∂z ∂v ∂y ∂z

85
Shiv Datt Kumar Partial Differential Equations

∂u ∂u ∂u ∂v ∂v ∂v
= yz, = xz, = xy, = = = 1.
∂x ∂y ∂z ∂z ∂y ∂z

∂φ ∂φ
[yz + xyp] + [1 + p] = 0 − − − − − − − − − (3)
∂u ∂v

.
∂φ ∂φ
[xz + xyq] + [1 + q] = 0 − − − − − − − − − − − −(4)
∂u ∂v

.
∂φ ∂φ
Eliminating and
∂u ∂v

yz + xyp 1 + p
= 0 or px(y − z) + qy(z − x) = z(x − y).
xz + xyq 1 + q

5.1.3 Lagrange linear differential equation

Derive Lagrange linear equation P p + Qq = R − − − − − − − − − (1)

∂z ∂z
where P, Q, R are the function of x, y, z and p = ,q= . Show that its
∂x ∂y
general solution is f (u, v) = 0, where f is arbitrary function and u(x, y, z) = c1 ,
v(x, y, z) = c2 are linearly independent solutions of the subsidiary equations

dx dy dz
= =
P Q R

Solution: Differentiate f (u, v) = 0 with respect to x and y we get

   
∂f ∂u ∂u ∂f ∂v ∂v
+p + +p = 0 − − − − − − − − − (1) and
∂u ∂x ∂z ∂v ∂x ∂z

86
Shiv Datt Kumar Partial Differential Equations

   
∂f ∂u ∂u ∂f ∂v ∂v
+q + +q = 0 − − − − − − − − − (2)
∂u ∂y ∂z ∂v ∂y ∂z

∂f ∂f
Eliminating and
∂u ∂v

∂u ∂u ∂v ∂v
+p +p
∂x ∂z ∂x ∂z =0
∂u ∂u ∂v ∂v
+q +q
∂y ∂z ∂y ∂z
i.e.

   
∂u ∂v ∂u ∂v ∂u ∂v ∂u ∂v ∂u ∂v ∂u ∂v
p − +q − = −
∂x ∂y ∂y ∂z ∂x ∂z ∂z ∂x ∂y ∂x ∂x ∂y

. Or
p P +q Q=R

Now suppose u(x, y, z) = c1 and v(x, y, z) = c2 . Differentiating these equa-

tion we get

∂u ∂u ∂u
dx + dy + dz = 0
∂x ∂y ∂z

.
∂v ∂v ∂v
dx + dy + dz = 0.
∂x ∂y ∂z

Solving these equations we get Lagrange auxiliary or subsidiary equations

dx dy dz
= = ,
P Q R

87
Shiv Datt Kumar Partial Differential Equations

∂(u, v) ∂(u, v) ∂(u, v)

where P = ,Q= ,R= .
∂(y, z) ∂(z, x) ∂(x, y)
Geometric Interpretation of Lagrange linear PDE

Solution of PDE are surfaces of the form f (x, y, z) = c. Normal vector to

∂f ∂f ∂f ∂f ∂f ∂f
the surface is given by ∇f = î+ ĵ + k̂. Thus , , are direction
∂x ∂y ∂z ∂x ∂y ∂z
ratios of the normal to the surface. Direction ratios of the normal at a point
∂z ∂z
on the surface z = f (x, y) are , , −1 or p, q, −1. Lagrange equation is
∂x ∂y
∂(u, v) ∂(u, v) ∂(u, v)
p P + q Q = R, where P = ,Q= ,R= are direction
∂(y, z) ∂(z, x) ∂(x, y)
ratios of the tangent line to the integral curve given by u(x, y, z) = c1 and
v(x, y, z) = c2 . Therefore p P + q Q + (−1)R = 0 shows that the normal N to
the surface is perpendicular to the tangent line to the integral curve.

Solution of first order PDE:

Complete Integral or solution: Any relation of the form f (x, y, z, a, b) = 0,

which contains two arbitrary constants and satisfies the differential equation
F (x, y, z, p, q) = 0 is called complete integral.

Particular Integral : The solution obtained by determining the arbitrary

constants in the complete integral by using some specified condition is called
particular integral.
Particular solution passing through a given curve: Consider Lagrange
equation P p+Qq = R and let φ(u, v) = 0 be the solution, where u(x, y, z) = C1
and v(x, y, z) = C2 are linearly independent solutions. Let the integral surface
pass through a curve C : x = x(t), y = y(t), z = z(t). Since curve lies on the
integral surface, we have u(x(t), y(t), z(t)) = C1 , (v(x(t), y(t), z(t)) = C2 . If
we eliminate t from these two equations, we obtain a relation between C1 and
C2 . Substituting C1 = u(x, y, z) and C2 = v(x, y, z) in this relation, we obtain

88
Shiv Datt Kumar Partial Differential Equations

particular solution.
Singular solution: The equation of the envelope of two parameter family
of surfaces representing the complete integral of given PDE is called singular
solution. If f (x, y, z, a, b) = 0 is complete solution, then by eliminating a and
∂f ∂f
b from f = 0, = 0, = 0 we get singular solution, which is constant free
∂a ∂b
and is also called envelope.

Note: Singular solution may not exist for every PDE.

Example 5.1.4. Solve Lagrange partial differential equations:

1. (x2 − y 2 − z 2 ) p + 2xy q − 2zx = 0

2. (x2 − yz)p + (y 2 − zx)q = z 2 − xy

3. (x + y 2 )p + yq = z + x2

4. px(z − 2y 2 ) = (z − qy)(z − y 2 − 2x3 )

5. p cos (x + y) + q sin (x + y) = z

Solution:

(1)
dx dy dz
= = ,
x2 −y −z
2 2 2xy 2zx
dy dz dy dz
= ⇒ = or log y = log z + log C1 ⇒ y/z = C1 .
2xy 2zx y z

x dx + y dy + z dz dz
N ow =
x(x2 + y 2 + z 2 ) 2xz

2x dx + 2y dy + 2z dz dz x2 + y 2 + z 2
= ⇒ = C2
x2 + y 2 + z 2 z z

89
Shiv Datt Kumar Partial Differential Equations

Answer: x2 + y 2 + z 2 = zf (y/z).

(2) Lagrange auxiliary equations

dx dy dz
= 2 = 2 ,
x2 − yz y − zx z − xy

d(x − y) d(y − z) d(x − y) d(y − z)

= ⇒ = ,
(x − y)(x + y + z) (y − z)(x + y + z) (x − y) (y − z)
x−y
log (x − y) = log (y − z) + log C ⇒ = C1 − − − − − (i)
y−z

.
x dx + y dy + z dz dx + dy + dz
N ow = 2
x + y + z − 3xyz
3 3 3 x + y + z 2 − yz − zx − xy
2

x dx + y dy + z dz
= dx + dy + dz
x+y+z
Z Z
x dx + y dy + z dz = (x + y + z)dx + dy + dz ⇒ xy + yz + zx = C2

x−y
Answer : = f (xy + yz + zx)
y−z

(3)
dx dy dz
2
= =
x+y y z + x2

dx x 1 dx x
From first two − = y or − 2 = 1 or x/y = y + a
dy y y dy y
dy dz
Using this relation we have =
y z + y (y + a)2
2

1 dz z y 1
− 2 = (y + a)2 or = (y + a)3 + b.
y dy y z 3
x − y 2 3zy 2 − x3
 
General solution is φ (a, b) = 0 or φ , = 0.
y 3y 3
(4) px(z − 2y 2 ) + qy(z − y 2 − 2x3 ) = z(z − y 2 − 2x3 )

90
Shiv Datt Kumar Partial Differential Equations

dx dy dz
= =
x(z − 2y )
2 y(z − y − 2x )
2 3 z(z − y 2 − 2x3 )

Using second and third

dy dz
= ⇒ y = az.
y z

Now from first and third,

dx dz
= , U sing y = az
x(z − 2a z )
2 2 z(z − a z 2 − 2x3 )
2

z 2 dx − a2 z 3 dx − 2x3 dx = xz dz − 2a2 xz 2 dz
2
xdz − zdx 2 (2xz dz − z dx) z a2 z 2
Or − a + 2x dx or − + x2 + b.
x2 x2 x x

Answer: y/z = f (z/x − a2 z 2 /x + x2 ).

(5)
dx dy dz
= =
cos (x + y) sin (x + y) z

dx + dy dz 1 1
= or √ cosec (π/4 + x + y)(dx + dy) = dz
sin (x + y) + cos (x + y) z 2 z

√ √
2
log tan(π/8 + (x + y)/2) + log C1 = 2 log z or C1 = z cot(π/8 + (x + y)/2).

91
Shiv Datt Kumar Partial Differential Equations

Now

dx + dy dx − dy
=
sin (x + y) + cos (x + y) cos (x + y) − sin (x + y)

cos (x + y) − sin (x + y)
 
(dx + dy) = dx − dy
sin (x + y) + cos (x + y)

log [sin (x + y) + cos (x + y)] − log a = x − y

Or sin (x + y) + cos (x + y) = a ex−y

1 √ 1
√ sin(π/4 + x + y) = C2 ex−y or 2e(x−y) cosec (π/4 + x + y) = = C2 .
2 a
Hence general solution is
√ √
φ(z 2 cot(π/8 + (x + y)/2), 2e(x−y) cosec (π/4 + x + y) = 0

Example 5.1.5. Find general solution of PDE (3 − 2yz) p + x(2z − 1) q =

2x(y − 3) and hence find particular solution which passes through the curve
z = 0, x2 + y 2 = 9.

Solution: Lagrange auxiliary equations

dx dy dz xdx + zdz
= = =
3 − 2yz x(2z − 1) 2x(y − 3) −3x(2z − 1)

From second and third, we have (2y − 6) dy = (2z − 1) dz ⇒ y 2 − 6y =

z 2 − z + C1 .
From second and fourth, we have −3 dy = xdx + z dz ⇒ x2 + z 2 + 6y = C2 .
Therefore general solution is φ(C1 , C2 ) = 0 or φ( y 2 −6y −z 2 +z, x2 +z 2 +6y) =

92
Shiv Datt Kumar Partial Differential Equations

0. Integral surface passes through the curve z = 0, x2 + y 2 = 9, then we have

y 2 − 6y = C1 and x2 + 6y = C2 . This gives x2 + y 2 = C1 + C2 ⇒ C1 + C2 = 9.
Hence particular solution is x2 + y 2 + z = 9.

5.1.4 Surface orthogonal to a given family of surfaces

Let surface S1 : f (x, y, z) = C cuts the family of the surfaces S2 : z = g(x, y)

∂f ∂f ∂f
orthogonally. Normal N1 to S1 = ∇f = î + ĵ + k̂.
∂x ∂y ∂z
∂z ∂z
Normal N2 to S2 = î + ĵ − k̂ = p î + q ĵ − k̂.
∂x ∂y

∂f ∂f ∂f
N1 ⊥ N2 ⇒ N1 .N2 = 0 ⇒ p +q − = 0. Hence surface orthogonal
∂x ∂y ∂z
to surface f (x, y, z) = c are the surfaces generated by solution of curves of the
equation

dx dy dz
= =
∂f ∂f ∂f
∂x ∂y ∂z
.

Example 5.1.6. Find the equation of surface which cuts orthogonally the fam-
ily of surfaces z 2 = c(x2 + y 2 ). Obtain particular surface which pass through
the circle z = 3, x2 + y 2 = 9.

Solution: Here f = c(x2 + y 2 ) − z 2 = 0.

∂f ∂f ∂f
= 2cx, = 2cy, = −2z
∂x ∂y ∂z

93
Shiv Datt Kumar Partial Differential Equations

Surface orthogonal to surface f (x, y, z) = c are the surfaces generated by

solution of curves given by

dx dy dx
= =
2cx 2cy −2z

dx dy x
From first two we have = or = C1 . Also
x y y

dx xdx + ydy + zdz

= ⇒ xdx + ydy + zdz = 0
2cx 2[c(x2 + y 2 ) − z 2 ]

Or x2 + y 2 + z 2 = C2 . This pass through z = 3, x2 + y 2 = 9. Thus C2 = 18

and we get required the surface x2 + y 2 + z 2 = 18.

5.2 Non-linear partial differential equations

5.2.1 Charpit’s Method

This is general method for complete integral of non-linear PDE. Consider the
partial differential equation f (x, y, z, p, q) = 0 − − − − − − − − − − − − − (1)
∂z ∂z
Since z depends on x and y, dz = dx + dy = p dx + q dy − − − − − −(2)
∂x ∂y
Now if we can find another relation involving x, y, z, p, q such that

φ(x, y, z, p, q) = 0 − − − − − − − − − − − − − −(3)

then we can solve (1) and (3) for p and q and substitute in (2). This will give
a solution provided (2) is integrable. To determine φ, we differentiate (1) and

94
Shiv Datt Kumar Partial Differential Equations

(3) with respect to x and y,

∂f ∂f ∂f ∂p ∂f ∂q
+ p+ + . = 0 − − − − − − − − − (4)
∂x ∂z ∂p ∂x ∂q ∂x

∂φ ∂φ ∂φ ∂p ∂φ ∂q
+ p+ + . = 0 − − − − − − − − − (5)
∂x ∂z ∂p ∂x ∂q ∂x
∂f ∂f ∂f ∂p ∂f ∂q
+ q+ + . = 0 − − − − − − − − − (6)
∂y ∂z ∂p ∂y ∂q ∂y
∂φ ∂φ ∂φ ∂p ∂φ ∂q
+ q+ + . = 0 − − − − − − − − − (7)
∂y ∂z ∂p ∂y ∂q ∂y
∂p
Eliminating between (4) and (5)
∂x
     
∂f ∂φ ∂φ ∂f ∂f ∂φ ∂φ ∂f ∂f ∂φ ∂φ ∂f ∂q
− + − p+ − = 0−−−−−−(8)
∂x ∂p ∂x ∂p ∂z ∂p ∂z ∂p ∂q ∂p ∂q ∂p ∂x

∂q
Eliminating between (6) and (7)
∂y
     
∂f ∂φ ∂φ ∂f ∂f ∂φ ∂φ ∂f ∂f ∂φ ∂φ ∂f ∂p
− + − q+ − = 0−−−−−−(9)
∂y ∂q ∂y ∂q ∂z ∂q ∂z ∂q ∂p ∂q ∂p ∂q ∂y

∂p ∂ 2z ∂q
Adding (8) and (9) and using = = we find that last term in both
∂y ∂x∂y ∂x
cancel and we get

         
∂f ∂f ∂φ ∂f ∂f ∂φ ∂f ∂f ∂φ ∂f ∂φ ∂f ∂φ
+p + +q + −p −q + − + − =0
∂x ∂z ∂p ∂y ∂z ∂q ∂p ∂q ∂z ∂p ∂x ∂q ∂y

         
∂f ∂φ ∂f ∂φ ∂f ∂f ∂φ ∂f ∂f ∂φ ∂f ∂f ∂φ
− + − + −p −q + +p + +q =0
∂p ∂x ∂q ∂y ∂p ∂q ∂z ∂x ∂z ∂p ∂y ∂z ∂q

95
Shiv Datt Kumar Partial Differential Equations

This is a Lagrange differential equation with x, y, z, p, q as independent variable

and φ as dependent variable. Subsidiary equations are

dx dy dz dp dq
= = = =
∂f ∂f ∂f ∂f ∂f ∂f ∂f ∂f
− − −p −q +p +q
∂p ∂q ∂p ∂q ∂x ∂z ∂y ∂z
dx dy dz dp dq
Or = = = =
fp fq fp + qfq −(fx + pfz ) −(fy + qfz )

These are called Charpit’s equations. Solving these equations taking in pair
gives required solution.

When PDE is of special form, then Charpit’s equations are given by

Case 1: Equations which are independent of x, y, z, Charpits equation are

dx dy dz dp dq
= = = =
fp fq −fp − qfq 0 0

p = a, q = φ(a) (depends on the problem). dz = a dx + φ(a) dy

⇒ z = ax + ψ(a, y).

Example 5.2.1. Find complete integral of partial differential equation p2 −

7q 2 = 3.
√ √
Here q = a, p = 3 + 7a2 . Hence dz = ( 3 + 7a2 ) dx + a dy ⇒ z =
√
( 3 + 7a2 ) x + ay + b.

Case 2: Equations which are independent of x, y. Then Charpits equation are

dx dy dz dp dq
= = = =
fp fq −fp − qfq −pfz −qfz

From last two equations log p = a = log q + log a or p = aq. Now solve for

96
Shiv Datt Kumar Partial Differential Equations

p and q from the equations p = aq and f (z, p, q) = 0. Then complete integral

is given by dz = p dx + q dy.

Example 5.2.2. Find complete integral of p (5 + q) = 2qz.

Solution: Given equation is free from x and y. Then p = aq, and aq (5 + q) =
2z − 5a dz
2qz ⇒ q = and p = 2z − 5a. Hence dz = p dx + q dy ⇒ =
a 2z − 5a
1
dx + 1/a dy ⇒ log (2z − 5a) = x + y/a + b∗ ⇒ 2z − 5a = c e2z+(2/a)y , where
2
∗
b = 2b and c = eb .

Case 3: Separable equation of the form f (x, p) = g(y, q).

F = f (x, p) − g(y, q) = 0. Equations which are independent of x, y. Then
Charpit’s equation are

dx dy dz dp dq
= = = =
fp −gq pfp − qgq −fx gy

dx dp
= or fx dx + fp dp = 0 Or d(f (x, p)) = 0 ⇒ f (x, p) = a (constant).
fp −fx

Similarly g(y, q) = b. Now solve for p and q. Then complete integral is given
by dz = p dx + q dy.

Example 5.2.3. Find complete integral of p2 q 2 = 9p2 y 2 (x2 + y 2 ) − 9x2 y 2 .

9x2 1
Solution: Divide by p2 y 2 , then 2 (p2 − 1) = 2 (q 2 − 9y 4 ) = a2 (say). We
p y
get
3x p 3x
p= √ , q = y 9y 2 + a2 and dz = p dx + q dy = √ dx +
9x2 − a2 9x2 − a2
p 1√ 2 1
y 9y 2 + a2 dy ⇒ z = 9x − a2 + (9y 2 + a2 ) + b.
3 27
97
Shiv Datt Kumar Partial Differential Equations

Example 5.2.4. Solve by Charpit’s method yq + 3xp = 2(z − y 2 p2 ) − − − −(1)

Solution: Let f = yq + 3xp − 2 (z − y 2 p2 ) = 0. Then fx = 3p, fy = q + 4yp2 ,

fz = −2, fp = 3x + 4y 2 p, fq = y. Charpits auxiliary equations are

dx dy dz dp dq
= = = =
fp fq fp + qfq −(fx + pfz ) −(fy + qfz )

dx dy dz dp dq
= = = =
2
3x + 4y p y 2 2
3px + 4y p + qy −p q − 4y 2 p2
dy dp
From second and fourth = ⇒ log y+log p = log a or py = a or p = a/y.
y −p
2z − y 2 p2 − 3xp 2z − 2a2 − 3xa/y 2zy − 2a2 y − 3ax
From (1) q = = = .
y y y2

2zy − 2a2 y − 3ax

 
a
dz = dx + dy
y y2

2a2
 
2zy a 3ax
dz − dy = dx − dy − dy
y2 y y2 y

y 2 dz − 2zy dy ay dx − 3ax dy 2a2 dy

= −
y2 y2 y
y 2 dz − 2zy dy ay 3 dx − 3ay 2 x dy 2a2 dy
= −
y4 y6 y3
     2
z ax a
d = d + d
y2 y3 y2

z ax a2 ax
Integrating we get 2 = 3 + 2 + b ⇒ z = + a2 + by 2 , where a, b are
y y y y
constants.

Example 5.2.5. Find the singular solution of the following differential equa-
tions

(i) 6yz − 6pxy − 3qy 2 + pq = 0 (ii) px + qy + z = x q 2

98
Shiv Datt Kumar Partial Differential Equations

Solution: (i) Let f = 6yz−6pxy−3qy 2 +pq = 0, fp = −6xy+q, fq = −3y 2 +p,

fz = 6y, fx = −6py, fy = 6z − 6px − 6qy. Charpit’s auxiliary equations are

dx dy dz dp dq
= = = =
fp fq pfp + qfq −(fx + pfz ) −(fy + qfz )

dx dy dz dp dq
= = = =
(−6xy + q) 2
(−3y + p) −[p(−6xy + q) + q(−3y + p)]
2 0 −(6z − 6px)

From fourth and any other term, we get p = a and from given differential
6y(z − ax) 6y(z − ax)
equation q = . Therefore dz = p dx + q dy = adx + dy.
3y 2 − a 3y 2 − a
dz − a dx z − ax z − ax
 
6y dy
Then + 2 =0⇒ d =0⇒ 2 = b(constant).
z − ax 3y − a 3y − a
2 3y − a

Now we find the envelope of two parameters family of surfaces, if it exists.

φ(a, b) = z − ax − b(3y 2 − a) − − − − − − − − − −(2)

. Differentiating with respect to a and b,

∂φ
= −x + b = 0 ⇒ x = b.
∂a

,
∂φ
= −(3y 2 − a) = 0 ⇒ y 2 = a/3.
∂b

Then from (2), z − ab − b(3a/3 − a) = 0 ⇒ z = ab. Hence singular solution is

z = 3xy 2 .
(ii) F = px + qy + z − xq 2 , Fp = x, Fz = 1, Fy = q.

99
Shiv Datt Kumar Partial Differential Equations

dx dq dx dq
= ⇒ = ⇒ qx2 = a.
fp −(fy + qfz ) x −(q + q)

From given differential equation

 2
ay a a2 − axy − x3 z
px + 2 + z − x = 0 ⇒ p = .
x x4 x4
a2 − axy − x3 z a z a2 a
dz = pdx + qdy = 4
dx + 2
= − dx + 4
dx + 3 (xdy − ydx)
x x x x x

a2 a −a2 ay
xdz + zdx = dx + (xdy − ydx) ⇒ xz = + +b
x3 x2 2x2 x

Thus φ(a, b) = 2bx2 + 2axy − a2 − 2x3 z = 0.

Differentiating with respect to a and b,

∂φ ∂φ
= 2xy − 2a = 0, = 2x2 = 0
∂a ∂b

Thus a = xy and x = 0 ⇒ a = 0.
∂φ ∂φ
Therefore no singular solution exists, for = 2xy−2a = 0, = 2x2 = 0.
∂a ∂b
Thus a relation can not be obtained by eliminating a and b. Hence no singular
solution exists.

5.3 Higher order linear partial differential equa-

tions with constant coefficients

An equation of the type

∂ nz ∂ nz ∂ nz
a0 + a1 + . . . + a n = g(x, y), (5.4)
∂xn ∂xn−1 ∂y ∂y n

100
Shiv Datt Kumar Partial Differential Equations

where a0 , a1 , . . . , an are constants, is called homogeneous partial differential

equation if g(x, y) = 0 otherwise it is a non-homogeneous.
∂ ∂ 0 ∂2 0 ∂2 2 ∂
2
0
Put = D, =D, = DD , 2
= D , 2
= (D )2 etc.
∂x ∂y ∂x∂y ∂x ∂y
Then equation (1.4) can be written as
0 0
F (D, D )z = ni=0 ai Dn−i (D )i z = g(x, y)
P

General solution of this differential equation is

z = CF (complementary function) + PI (particular integral).

5.3.1 Rules for finding CF for homogeneous PDE

Consider the equation

∂ 2z ∂ 2z ∂ 2z
a0 + a 1 + a 2 =0
∂x2 ∂x∂y ∂y 2

0 02 0 0
(a0 D2 + a1 DD + a2 D )z = (D − m1 D )(D − m2 D )z = 0 (5.5)

Case (i): If m1 6= m2 .
0
Equation (1.5) will be satisfied by the solution of (D − m2 D )z = 0 or

p − m2 q = 0, (5.6)

which is a Lagrange type differential equation. Then auxiliary equation is

dx dy dz
= =
1 −m2 0

101
Shiv Datt Kumar Partial Differential Equations

We get solution y+m2 x = C1 , z = C2 i.e. z = f2 (y+m2 x). Similarly solution of

0
(D−m1 D )z = 0 is z = f1 (y+m1 x). Then CF is z = f1 (y+m1 x)+f2 (y+m2 x).
0 0
Case (ii): If m1 = m2 = m (say). Then we have (D−mD )(D−mD )z = 0. Put
0 0 0
(D −mD )z = u. Then we get (D −mD )u = 0. Solution of (D −mD )u = 0 is
0
u = f (y +mx). Thus PDE is (D −mD )z = f (y +mx) i.e. p−mq = f (y +mx)
is a Lagrange PDE. Then auxiliary equation is

dx dy dz
= =
1 −m f (y + mx)

First and second gives y + mx = C1 and first and third gives dz = f (C1 ) dx i.e.
z = xf (C1 ) + C2 = xf (y + mx) + f (C1 ). Thus z = f (y + mx) + xf (y + mx).

0
Remark 5.3.1. For homogeneous linear PDE, put D = m, D = 1. Then
auxiliary equation is a0 m2 + a1 m + a2 = (m − m1 )(m − m2 ) = 0. Then CF is
z = f1 (y + m1 x) + f2 (y + m2 x).

5.3.2 Rules for finding CF for non-homogeneous PDE

0
Consider the equation (D − mD − a)z = 0. Then auxiliary equation is

dx dy dz
= =
1 −m az

First and second give y + mx = C1 and first and third give z = eax C2 .
Combining these two we have solution z = eax f (y + mx).

102
Shiv Datt Kumar Partial Differential Equations

5.3.3 General Method for finding CF

Consider homogeneous linear partial differential equation

0 0 0n
F (D, D )z = (a0 Dn + a1 Dn−1 D + . . . + an D )z = 0. (5.7)

0
We have observed that z = φ(y + mx) is a solution of (D − mD )z = 0.
0
Suppose z = φ(y + mx) is a solution of (1.7). Then F (D, D )z = (a0 mn +
a1 mn−1 + . . . + an )φ(y + mx) = 0. If m1 , m2 , . . . , mn are distinct roots of
the equation a0 mn + a1 mn−1 + . . . + an = 0, then φi (y + mi x) are solutions
0 0 0 0
of F (D, D )z = (D + m1 D )(D + m2 D ) . . . (D + mn D )z = 0. The general
solution is

z = φ1 (y + m1 x) + φ2 (y + m2 x) + . . . + φn (y + mn x).
0 0 0 0
If PDE is F (D, D )z = (D + m1 D + a1 )(D + m2 D + a2 ) . . . (D + mn D + an )z.

Then general solution is

z = ea1 x φ1 (y + m1 x) + ea2 x φ2 (y + m2 x) + . . . + ean x φn (y + mn x).

5.3.4 Rules for finding particular integral (PI)

0
Let f (D, D ) = g(x, y) be a partial differential equation. Then

eax+by eax+by
1. If g(x, y) = eax+by , then P I = = , if f (a, b) 6= 0.
f (D, D0 ) f (a, b)
 
eax+by eax+by 
If f (a, b) = 0, then = x ,

f (D, D0 ) d

0
f (D, D )
dD
sin(ax + by) sin(ax + by)
2. if g(x, y) = sin(ax + by), P I = 02
= if
2 0
f (D , DD , D ) f (−a2 , −ab, −b2 )

103
Shiv Datt Kumar Partial Differential Equations

f (−a2 , −ab, −b2 ) 6= 0.

cos(ax + by) cos(ax + by)

3. If g(x, y) = cos(ax + by), P I = 02
= , if
2 0
f (D , DD , D ) f (−a2 , −ab, −b2 )
f (−a2 , −ab, −b2 ) 6= 0.

xm y n 0
4. If g(x, y) = xm y n , then P I = = f (D, D )−1 (xm y n ). Expand
f (D, D0 )
0 0
f (D, D )−1 in ascending powers of D, D and operate on xm y n term by
term.

5. If g(x, y) = eax+by u(x, y), then

u(x, y)eax+by u(x, y)eax+by
 
u(x, y)
PI = = = eax+by .
f (D, D0 ) f (D, D0 ) f (D + a, D0 + b)

6. (General method) If g(x, y) is any function of x and y, then

g(x, y) R
PI = 0 = g(x, c − mx) dx − − − − − −(∗)
(D − mD )
and replace c by y + mx after integration.
0 0 0 0
If f (D, D ) = (D − m1 D )(D − m2 D ) . . . (D − mn D ), then
g(x, y)
PI = .
(D − m1 D )(D − m2 D0 ) . . . (D − mn D0 )
0

Operations are performed in succession similar to ∗.

7. If g(x, y) is of the form φ(ax + by),

0 0
Case (i) f (a, b) 6= 0 and f (D, D ) is a homogeneous function of D &, D
of degree n, then
φ(ax + by) 1
RR R
0 = ... φ(V ) dV n . After integration V is
f (D, D ) f (a, b)
replaced by ax + by, then
0 0
case (ii) If f (a, b) = 0, then (bD − aD ) must be factor of F (D, D ).
0 0 0 0
Suppose f (D, D ) = (bD − aD )m G(D, D ) and G(D, D ) 6= 0. Then

104
Shiv Datt Kumar Partial Differential Equations

0
solution for G(D, D ) is obtained by using method of case (i) and also
φ(ax + by) xm φ(ax + by)
making use of the following formula = .
(bD − aD0 )m bm m!
0 0
Example 5.3.1. (D2 − DD + D − 1) z = cos(x + 2y) + ey

0
Solution: Given PDE is (D − 1)(D − D + 1) z = cos(x + 2y) + ey

Consider the equation (D − 1)z = 0. Then auxiliary equation is

dx dy dz
= =
1 0 z

First and second imply y = C1 . First and third imply

log z = x + C ⇒ z = ex C2 = ex φ1 (y).
0
Now consider (D − D + 1) z = 0. Then auxiliary equation is

dx dy dz
= =
1 −1 −z

First and second imply y + x = A1 .

First and third imply log z = −x + A ⇒ z = e−x A2 = e−x φ2 (y + x).

Therefore CF = ex φ1 (y) + e−x φ2 (y + x).
cos(x + 2y) ey
PI = +
(D2 − DD + D − 1) (D2 − DD0 + D0 − 1)
0 0

cos(x + 2y) ey
= +
(−1 + 2 + D0 − 1) (D2 − D(D0 + 1) + (D0 + 1) − 1)
cos(x + 2y) 1
= 0 + ey 2
D (D − DD − D + D0 )
0

0
ey D 0
− D − D)−1 (1)
R
= cos(x + 2y)dy + (1 −
−D D
1
= sin(x + 2y) − xey
2
1
Complete solution z = ex φ1 (y) + e−x φ2 (y + x) + sin(x + 2y) − xey .
2
0 02
Example 5.3.2. (D2 − 4DD + D ) z = ex+2y

105
Shiv Datt Kumar Partial Differential Equations

0
Solution: Put D = m and D = 1. Auxiliary equation is m2 − 4m + 1 = 0.
1 1
Then m = , .
2 2
CF = f1 (2y + x) + xf2 (2y + x)
ex+2y
PI =
4D2 − 4DD0 + D0 2
ex+2y
=x
d
(4D2 − 4DD0 + D0 2 )
dD
ex+2y ex+2y x2 ex+2y
=x 0 = x2 = .
(8D − 4D ) d 0 8
(8D − 4D )
dD
ex+2y
Complete solution z = f1 (2y + x) + xf2 (2y + x) + x2 .
8

106
Shiv Datt Kumar Partial Differential Equations

Exercise 5.3.1. 1. Eliminate function to find PDE.

(a) z = f (z + ix) + g(x − iy), Ans: Zxx + zyy = 0

(b) f (x + yz, x2 + y 2 − z 2 ) = 0. Ans: p (y 2 + z 2 ) − q (z + xy) = xz − y.

(c) x2 = y 2 + z 2 = f (xy). Ans.: z(px − qy) = y 2 − x2

(d) (mz − xy)p + (nx − lz)q = ly − mx

2. Solve the following PDE by Lagrange method

x2 − yz
 
(a) p + zxq = y 2
x
∂z ∂z
(b) x(z 2 − y 2 ) + y(x2 − z 2 ) = z(y 2 − x2 ).
∂x ∂y
(c) (x2 − y 2 − z 2 )p + 2xyq = 2xz. Ans: x2 + y 2 + z 2 = zf (y/z)
x−y
(d) x2 − yz)p + (y 2 − zx)q = z 2 − xy. Ans: = f (xy + yz + zx)
y−z
(e) (2y + z)p + (y + 2x)q = 4xy − z. Also find particular solution which
pass through z = 1, x = y.

3. Solve the following PDE by Charpit’s method

(a) z 2 = pqxy. Hint: Use y dx − x dy = q dp − p dq.

(b) (p2 + q 2 )y = qz. Ans z 2 = (a + cx)2 + c2 y 2

(c) 2xz − px2 − 2qxy + pq = 0. Ans z = ay + b(x2 − a).

(d) 2z + p2 + qy + 2y 2 = 0. Ans y 2 (x − a)2 + 2z + y 2 = b.

4. Solve the following PDE by Charpit’s method

(a) r + s − 6t = y cosx. Ans z = f1 (y + 2x) + f2 (y − 3x) + sin x − y cos x

107
Shiv Datt Kumar Applications of PDE

(b) 4r − 4s + t = 16 log (x + 2y). Ans z = f1 (2y + x) + xf2 (2y + x) +

2x2 log (x + 2y).
0 02
(c) (D3 − 4D2 D + 4DD )z = 4 sin (2x + y).

Ans z = f1 (y) + f2 (y + 2x) − x2 cos(2x + y).

02 0
(d) (D2 − D + D + 3D − 2)z = x + y.
1
Ans z = e−2x f1 (y + x) + ex f2 (y − x) − [x2 y + 3x2 /2 + xy + 3y/2 +
2
3x + 21/4].
0 02
(e) (D2 + 3DD + 2D )z = x2 y.
(x + y)3
Ans z = f1 (y − x) + f2 (y − 2x) + .
36

108
Chapter 6

Applications of PDE

Most of the partial differential equations of interest arise in physics and en-
gineering applications. Laplacian operator is the most physically important
∂2 ∂2 ∂2
differential operator, given by ∇2 = + + .
∂x2 ∂y 2 ∂z 2

6.1 Classification of second order linear PDE’s

Let the various physical quantities be measured by function u = u(x, y, z, t)

which depends on all three spatial variable x, y, z and time t. A second order
linear PDE is an equation of the form

Auxx + Buxy + Cuyy + Dux + Euy + F u + G = 0,

where the coefficients A, B, C, D, E, F, G are constants, or specified functions

of the variables x, y. The equation is classified into one of three types, based
on the coefficients A, B, C, as

109
Shiv Datt Kumar Applications of PDE

1. Elliptic: if B 2 − 4AC < 0;

2. Parabolic: if B 2 − 4AC = 0;

3. Hyperbolic: if B 2 − 4AC > 0.

∂ 2u ∂ 2u
For e.g, Laplace’s equation + 2 = 0 is elliptic, one dim heat equation
 2  ∂x2 ∂y
∂u ∂ u ∂ 2u 2
2∂ u
= c2 is parabolic, and wave equation = c is hyperbolic.
∂t ∂x2 ∂t2 ∂x2
It is useful to classify equations because solution techniques, and properties
of the solutions are different, depending on whether the equation is elliptic,
parabolic, or hyperbolic. Also, the physical nature of the corresponding prob-
lems are different. For instance, elliptic equations often arise in steady-state
and equilibrium problems; parabolic equations arise in diffusion problems; hy-
perbolic problems arise in wave motion and vibrational problems.

6.2 Wave Equation

Wave equation describes the propagation of oscillations at a fixed speed in

some quantity u. Wave equation is a second order linear partial differential
equation describing waves, including travelling and standing waves. The wave
equation in one dimension is written as:

∂ 2u 2
2∂ u
= c ,
∂t2 ∂x2

∂ 2u
where u represents a displacement from rest situation, is a term for how
∂t2
∂ 2u
the displacement accelerates, is a term for how the displacement is varying
∂x2
at the point x in one of the dimensions.

110
Shiv Datt Kumar Applications of PDE

The equation states that at any given instance, at any given point, the
way the displacement accelerates is proportional to the way the displacement’s
changes are squashed up in the surrounding area.

Proof. Consider the vibrations of a uniform elastic string of length l stretched

between two fixed points O and A. Let P Q = δs, m = mass per unit length of
string , T1 = Tension at point P and T2 = Tension at point Q. Let u represents
displacement from rest situation at any point. Suppose y = u denotes the
displacement of any points from rest position.

Then using Newton Second Law of motion, the equation of motion in ver-
tical direction is given by

∂ 2u
m δs = T2 sin β − T1 sin α,
∂t2

Since there is no motion in horizontal direction,

T1 cos α = T2 cos β = T (constant)

mδs ∂ 2 u
 
T2 sin β T1 sin α
= − .
T ∂t2 T2 cos β T1 cos α

111
Shiv Datt Kumar Applications of PDE

mδs ∂ 2 u
   
∂u ∂u
= − .
T ∂t2 ∂x x+δx ∂x x

If δx is very small, then δs = δx.

    
∂u ∂u
−
∂ 2u T  ∂x

x+δx ∂x x

= .
∂t2 m δx



Taking limit as δx → 0, we get

∂ 2u 2
2 ∂ u T
2
= c 2
, where c2 = .
∂t ∂x m
Remark 6.2.1. In two dimensions, the wave equation is

utt = c2 (uxx + uyy ) .

Using the notations of Newtonian mechanics and vector calculus, the wave
equation can be written ü = c2 ∇2 u.

Method of Separation of Variables

Step 1: We find all solutions of (1) that are of the special form u(x, t) =
X(x)T (t) for some function X(x) that depends on x but and some function
T (t) that depends on t. Once again, if we find a bunch of solutions Xi (x)Ti (t)
P
of this form, then since (1) is a linear equation, ai Xi (x)Ti (t) is also a solution
for any choice of the constants ai .

Step 2: We impose the boundary conditions.

Step 3 We impose the initial condition.

Solution of the wave equation by separation of variables

∂u dT
Let u = XT , where X = X(x), and T = T (t). Then =X ,
∂t dt
∂u dX ∂ 2 u d2 T ∂ 2u d2 X
=T , = X 2 and =T 2.
∂x dx ∂t2 dt ∂x2 dx
112
Shiv Datt Kumar Applications of PDE

Substituting these values in the given equation (1), we get

1 d2 T 1 d2 X
= = k(say), since variables t and x are independent.
c2 T dt2 X dx2
d2 T 2 d2 X
− kc T = 0 and − kx = 0
dt2 dx2
Case I : if k = p2 > 0, then
X = c1 epx + c2 e−px , T = c3 ept + c4 e−pt
Case II: if k = −p2 < 0, then
X = c1 cos px + c2 sin px, T = c3 cos cpt + c4 sin cpt.
Case III: if k = 0, then X = c1 t + c2 , T = c3 X + c4 .
Due to periodic nature of wave equation, suitable solution of wave equation
equation (1) is

u(x, t) = (c1 cos px + c2 sin px)(c3 cos cpt + c4 sin cpt).

Example 6.2.1. A string is stretched and fastened to two points l apart. Mo-
tion is started by displacing the string into the form y = k(lx − x2 ) from which
it is released at time t = 0. Find the displacement of any point on the string
at a distance of x from one end at time t.

Solution 6.2.2. Vibrations of the string is given by wave equation

∂ 2u 2
2 ∂ u
= c − − − − − − − − − − − − − − − − − −(1).
∂t2 ∂x2

Solution of wave equation (1) is

y(x, t) = (c1 cos px + c2 sin px)(c3 cos cpt + c4 sin cpt)———-(2)

Since the end points of the string are fixed for all time,
y(0, t) = 0 for all t > 0————————- (3)

113
Shiv Datt Kumar Applications of PDE

y(l, t) = 0 for all t > 0————————- (4)

Note that initial transverse velocity of any point of the string is zero, therefore
 
∂y
= 0——————————————(5)
∂t t=0
and y(x, 0) = k(lx − x2 )————————(6)
By equation (3), y(0, t) = 0 gives 0 = c1 (c3 cos cpt + c4 sin cpt). Thus c1 = 0.
Then equation (2) becomes

y = c2 sin px(c3 cos cpt + c4 sin cpt)——–(7).

∂y
= c2 sin px(−c3 cp sin cpt + c4 cp cos cpt)——–(7).
∂t
 
∂y
By equation (5), = 0 we get c2 c4 cp sin px = 0, which gives c4 = 0,
∂t t=0
as c2 6= 0.

Then equation (7) becomes

y = c2 c3 sin px sin cpt——–(8).

Using condition (4), y(l, t) = 0 = c2 c3 sin pl cos cpt gives sin pl = 0

nπ
or pl = nπ or p = , n = 1, 2, 3 . . ..
l
nπ
On putting bn = c2 c3 and p = in equation (8), we have
l
nπx nπct
y = bn sin cos .
l l
Since we get solutions for all integral values of n, complete solution is
nπx nπct
y(x, l) = ∞
P
n=1 bn sin cos ————–(9).
l l
nπx
Now y(x, 0) = ∞ = lx − x2 ——-(10).
P
n=1 bn sin
l
Since equation (10) represents Fourier sine series of function f (x) = lx−x2 ,
R
2 l
2 nπx 8l2
bn = (lx − x )sin dx = 3 3 , if n is odd and bn = 0 if n is even.
l 0 l nπ
Thus

114
Shiv Datt Kumar Applications of PDE

∞
X 8l2 nπx nπct
y(x, t) = 3 3
sin cos .
n=1
n π l l

6.3 Heat Equation

Heat equation is a parabolic partial differential equation, describing the distri-

bution of heat in a given space over time.

6.3.1 One dimensional heat flow

Consider the heat flow in a uniform thin bar (rod) of length l. Let u(x, t)
denote the temperature at position x and time t in a long, from x = 0 to x = l.
Assume that the sides of the rod are insulated so that heat energy neither
enters nor leaves the rod through its sides. Also assume that heat energy is
neither created nor destroyed in the interior of the rod. Then u(x, t) obeys the

heat equation.

Let S = specific heat, A = cross sectional area, k = thermal conductivity

of the material. Then

rate of increase of heat in the slab = amount of heat retained by the slab

    
∂u ∂u ∂u
ρSAδx = kA −
∂t ∂x x+δx ∂x x

115
Shiv Datt Kumar Applications of PDE
    
∂u ∂u
−
∂u k  ∂x

x+δx ∂x x

= .
∂t ρS  δx



Taking limit as δx → 0, we get one dim heat equation

∂ 2u
 
∂u k
= c2 , where c2 = ., − − − − − − −(1)
∂t ∂x2 ρS

Solution of the Heat Equation by Separation of Variables

The problem is to determine u(x, t) for all x and t. Suppose function

u(x, t) = X(x)T (t) is a solution to the heat equation (1). Then
00 0
0 2 00 X (x) 1 T (t)
X(x)T (t) = c X (x)T (t) iff = 2 .
X(x) c T (t)
The left hand side is independent of t. The right hand side is independent
of x. The two sides are equal. So both sides must be independent of both x
and t and hence equal to some constant, say k. So we have
00 0
X (x) = kX(x) and T (t) = kc2 T (t).
2 p2 t
We get X = C1 cos px + C2 sin px and T = e−c , where k = p2 .
00 0
If k = 0, the equations (5) simplify to X (x) = 0, T (t) = 0 and the general
solution is X(x) = d1 + d2 x, T (t) = d3 for arbitrary constants d1 , d2 and d3 .

Due to nature of heat, general solution of heat equation is

2 p2 t
u(x, t) = (C1 cos px + C2 sin px) e−c .

Now we can impose boundary conditions to get values of constants c1 , c2 ,

and c. The boundary conditions are

u(0, t) = 0 for all t > 0————————- (2)

u(l, t) = 0 for all t > 0————————- (3)

Also there is some given function f (x) such that the initial condition

116
Shiv Datt Kumar Applications of PDE

u(x, 0) = f (x), for all 0 < x < l—————— (4) is satisfied.

Example 6.3.1. The ends A and B of a 30 cm long rod have the temperature
at 20o and at 80o until steady state reaches. The temperature of the ends of rod
are changed to 30o and 60o respectively. Find the temperature distribution in
the at time t.

Solution 6.3.2. Let u1 (x, t) be the transient temperature distribution and u2 (x)
be the steady state temperature distribution in the rod. To get u(x, t) in the
intermediate period, reckoning time from the instant, when end temperature
were changed, we assume

u = u1 (x, t) + u2 (x)

Now u1 (x, t) satisfies

∂ 2u
 
∂u
= c2 − − − − − − − (1)
∂t ∂x2

2 2
Its solution is given by u(x, t) = (ak cos kx + bk sin kx) e−c k t . The initial
(80 − 20)x
temperature distribution in the rod is u = 20 + = 20 + 2x and the
30
final distribution in steady state is
(60 − 30)x
u = 30 + = 30 + x.
30
2 2
Hence u is of the form u = 30 + x + (ak cos kx + bk sin kx) e−c k t
P

nπ
When x = 0, u = 30, and when x = 30, u = 80, ak = 0, k = .
30
nπ !2
P∞ n  nπx o −c2 t
Hence u = 30 + x + n=1 bn sin e 30 .
30
Using the initial condition u = 20 + 2x at t = 0,
P  nπx 
20 + 2x = 30 + x + bn sin
30
117
Shiv Datt Kumar Applications of PDE

P  nπx 
x − 10 = bn sin − − − − − − − − − (2),
R 30
2 30  nπx  20
where bn = (x − 10) sin dx = [(−1)n+1 2 − 1] .
30 0 30 nπ
Putting the value of bn in (2) we get the required solution.
 −c2 t nπ 2
!

20  nπx 
Hence u = 30 + x + ∞ [(−1)n+1 2 − 1] sin 30 .
P
n=1 e
nπ 30

6.3.2 Two dimensional heat flow

When heat flow is along all the curves lying in parallel planes, then the flow is
called two dimensional. Consider heat flow in a metal plane in the xoy plane.
Let h = thickness of metal plate, ρ = density, k = thermal conductivity, S=
specific heat. Consider a rectangular element ABCD of the plate with sides
δx, δy, edges being parallel to the coordinate axes.

Then the quantity of heat entering the element ABCD per second through
   
∂u ∂u
the surface AB and AD is −k δx.h, and −k δy.h, respectively.
∂y y ∂x x
∂u
The rate of gain heat by element ABCD is also given by ρδx.δ[Link]. .
∂t
The total gain of heat by rectangular element ABCD per sec = inflow-
outflow

118
Shiv Datt Kumar Applications of PDE
        
∂u ∂u ∂u ∂u
− −
∂u  ∂x

x+δx ∂x x
∂x y+δy ∂x 
y
ρδx.δ[Link]. = khδxδy  + .
∂t  δx δy 
        
∂u ∂u ∂u ∂u
− −
∂u k  ∂x

x+δx ∂x x
∂x y+δy ∂x 
y
= + .
∂t ρS  δx δy



As δx → 0, δy → 0, two dim heat equation in transit state is given by:

 2
∂ u ∂ 2u

∂u 2 k
=c 2
+ 2 , where c2 = .
∂t ∂x ∂y ρS
∂u
In steady state = 0, so two dim heat equation is given by the Laplace
∂t
equation:

∂ 2u ∂ 2u
+ = 0.
∂x2 ∂y 2

Solution of this heat equation using separation of variables u = X(x)Y (y)

is given by u(x, y) = (c1 cos kx + c2 sin kx)(c3 eky + c4 e−ky ), where c1 , c2 , c3 , c4
are constants to be determined using boundary conditions.

Example 6.3.3. A rectangular plate with insulated surfaces is 10 cm wide and

so long compared to its width that it may be considered infinite in length without
introducing an appreciable error. If the temperature along the short edge y = 0
is given by

u(x, 0) = 20x, 0<x≤5

= 20(10 − x), 5 < x < 10

while the two long edges x = 0 and x = 10 as well as other short edges
are kept at 0o C. Find the steady state temperature at any point P (x, y) of the
plate.

119
Shiv Datt Kumar Applications of PDE

Answer: In steady state, two dim heat equation is given by :

∂ 2u ∂ 2u
+ = 0.
∂x2 ∂y 2

Solution of this heat equation is given by

u(x, y) = (c1 cos kx + c2 sin kx)(c3 eky + c4 e−ky ),

Boundary conditions are

u(0, y) = 0, for all y,

u(10, y) = 0, for all y,
u(x, ∞) = 0, for all x,
u(x, 0) = 20x, 0<x≤5
= 20(10 − x), 5 < x < 10

nπ
Using these conditions we get c1 = c3 = 0, k = .
10
u(0, y) = 0 implies c1 (c3 eky + c4 e−ky = 0. Then c1 = 0. Thus

u(x, y) = c2 sin kx (c3 eky + c4 e−ky ),

nπ
u(10, y) = 0 implies sin10k = 0. This implies k = .
10
u(x, ∞) = 0 implies c3 = 0. Let bn = c2 c4 . Then

∞
X
u(x, y) = bn sin kx e−ky ,
i=0

Z 5 Z 10
2 2
bn = 20x sinkx dx + (10 − x) sinkx dx
10 0 10 5

120
Shiv Datt Kumar Applications of PDE

800 nπ 400
= 2 2
sin − 2 2 sinnπ
nπ 2 nπ

800
= (−1)n+1 , if n = odd , bn = 0 if n = even
n2 π 2

Temperature at any point P (x, y) is given by

∞ (2n − 1)πy
800 X (−1)n+1 (2n − 1)πx
 
−
u(x, y) = 2 sin e 10 .
π n=1 (2n − 1)2 10

6.4 Exercises

1. A tightly stretched string with fixed ends points x = 0 and x = l is

 πx 
initially in a position given by y = y0 sin3 . If it is released from
l
rest from this position, find the displacement y(x, t).
 
y0 πx cπt 3πx 3cπt
Ans: y = 3sin cos − sin cos .
4 l l l l

2. Determine the solution of one dimensional heat equation

∂u ∂ 2u
= c2 .
∂t ∂x2

under the boundary conditions u(0, t) = 0, u(l, t) = 0 (t > 0), where l is

the length of the bar and the initial condition u(x, 0) = x.
 nπx  n2 π 2 c2 t
P∞ −
Ans: y = n=1 (−1)
n+1
sin e l2 .
l

3. A rectangular plate with insulated surfaces is 8 cm wide and so long

compared to its width that it may be considered infinite in length without

121
Shiv Datt Kumar Applications of PDE

introducing an appreciable error. If the temperature along one short edge

πx
y = 0 is given by u(x, 0) = 100 sin , 0 < x < 8, while the two edges
8
x = 0 and x = 8 as well as other short edge are kept at 0o C. Show
that the steady state temperature at any point of the plate is given by
πy πx
u(x, y) = 100 e− 8 sin .
8
 πx  πy
Ans: u(x, y) = 100 sin e− 8 .
8

122
Chapter 7

Matrix Theory and Linear

Algebra

Definition 7.0.1. Binary Operation: Let S be any non empty set. Then a
mapping ? : S × S → S is a binary operation on the set S iff a ? b ∈ S for all
a, b ∈ S and a ? b is the unique element of S.

Definition 7.0.2. Algebraic Structure: A non empty set S equipped with

one or more binary operations and satisfying certain axioms is called an Alge-
braic Structure.

Groups, rings, fields, vector Spaces are examples of algebraic structures.

7.1 Group

Definition 7.1.1. Let G be a non empty set and o : G × G → G, sending

(x, y) → xoy is a binary operation. Then (G, o) is called a group if it satisfies
the following conditions:

123
Shiv Datt Kumar Matrix Theory and Linear Algebra

1. Closure property : xoy ∈ G, ∀ x, y ∈ G.

2. Associative property: (xoy)oz = xo(yoz), ∀ x, y, z ∈ G.

3. Existence of identity: There exists an element e ∈ G, such that xoe =

eox = x, ∀ x ∈ G. Then such e is called the identity element.

4. Existence of inverse: For all x ∈ G, there exists y ∈ G such that xoy =

yox = e. Such y is called the inverse of x, and is denoted by x−1 .

Definition 7.1.2. Group (G, o) is called abelian if it also satisfies commutative

property: xoy = yox, ∀ x, y ∈ G.

Example 7.1.3. 1. (Z, +), (Q, +), (R, +), (C, +) are abelian groups.

2. (Q∗ , ·), (R∗ , ·), (C∗ , ·) are abelian groups, where Q∗ = Q − {0} etc.

3. Zm = {0̄, 1̄, 2̄, . . . , m − 1}, is an abelian group with respect to +.

4. Kliens four group V4 = {e, a, b, c}, where aob = c, boc = a, aoc = b,

a2 = b2 = c2 = e.

5. Circle S 1 = {z = x + iy ∈ C||x2 + y 2 = 1} is a group with respect to

multiplication.

6. Set of n × n invertible matrices with entries in R is a non-abelian group

with respect to multiplication.

7. Un = {ā ∈ Zn |(a, n) = 1} is the group of units of Zn with respect to

multiplication.

Definition 7.1.4. (Subgroup) Let (G, o) be a group. Then a subset H of G is

called subgroup if xoy −1 ∈ H, for all x, y ∈ H.

124
Shiv Datt Kumar Matrix Theory and Linear Algebra

Example 7.1.5. 1. If G is a group, then {e} and G are always subgroups

of G and are called improper subgroups.

2. H1 = {e, a}, H2 = {e, b}, H3 = {e, c} are all proper subgroups of Kliens
four group V4 = {e, a, b, c}.

3. (Z, +) is a subgroup of (Q, +).

4. (R, +) is a subgroup of (C, +).

5. (Q∗ , ·) is a subgroup of (R∗ , ·).

Definition 7.1.6. (Normal Subgroup) Let (G, o) be a group. Then a subgroup

H of G is called normal subgroup if gH = Hg, for all g ∈ G, where gH =
{goh|h ∈ H} and Hg = {hog|h ∈ H} are called left and right cosets of H in
G respectively.

Example 7.1.7. 1. (Z, +) is a normal subgroup of (Q, +).

2. (nZ, +) is a normal subgroup of (Z, +).

3. H = {e, a} is a normal subgroup of Kliens four group V4 = {e, a, b, c}.

4. (Q∗ , ·) is a normal subgroup of (R∗ , ·).

Definition 7.1.8. (Cyclic group) A group G is called cyclic if there exists an

element a in G which generates the group i.e. G = [a] = {an |n ∈ Z}.

Theorem 7.1.1. Let Z = [a] be a cyclic group and o(G) = n. Then G = [ar ],
where (r, n) = 1, r < n. Therefore number of generators of G = φ(n).

Example 7.1.9. 1. nZ = [n] is a cyclic group.

125
Shiv Datt Kumar Matrix Theory and Linear Algebra

2. Group of nth roots of unity = {e2πir/n |0 ≤ r < n} is a cyclic group.

3. Kliens four group V4 = {e, a, b, c} is not cyclic.

4. (Q∗ , ) is not cyclic group of (R∗ , ·).

Definition 7.1.10. (Quotient group) Let (G, o) be a group and H be a normal

group of G. Then G/H = {aH|a ∈ G} is a group with respect to binary
operation defined by aH ∗ bH = aobH and is called called Quotient group.

Example 7.1.11. 1. Let H = {e, a}be a normal subgroup of Kliens four

group V4 = {e, a, b, c}. Then quotient group G/H = {H, bH}, where
H = eH = {e, a} and bH = {boe, boa} = {b, c}.

2. Since mZ = {0, ±m, ±2m, · · · } is a normal subgroup of (nZ, +). Then

Z/mZ is a quotient group.

Theorem 7.1.2. ( Lagrange Theorem)

Let G be a finite group and H is any subgroup of G. Then order of H

divides order of G, where order of G means number elements in G.

7.2 Group Homomorphism

Let (G1 , o1 ) and (G2 , o2 ) be two groups. Then a map f : G1 → G2 is called

group homomorphism if f (xo1 y) = f (x)o2 f (y), ∀ x, y ∈ G1 .

A bijective homomorphism f : G1 → G2 is called group isomorphism. Two

isomorphic groups are denoted as G1 ∼
= G2 .

126
Shiv Datt Kumar Matrix Theory and Linear Algebra

Kernel of a group homomorphism: Let f : G1 → G2 be group homo-

morphism. Then Ker f = {x ∈ G1 |f (x) = e2 }, where e2 is identity element of
G2 , is a normal subgroup of G1 (verify).

Fundamental theorem of group homomorphism: Let (G1 , o1 ) and (G2 , o2 )

be two groups and f : G1 → G2 is a surjective group homomorphism, then
G1 /Ker f ∼
= G2 ,

Example 7.2.1. 1. Map f : Z → Q defined by f (x) = x is a group homo-

morphism.

2. Map f : Z → Zm defined by f (x) = x̄ is a group homomorphism,

Kerf = {x ∈ Z|f (x) = 0} = mZ and Z/mZ ∼
= Zm .

3. Map f : R → S 1 defined by f (x) = eix is a group homomorphism and

R/Z ∼
= S 1.

Definition 7.2.2. Ring: An algebraic structure (R, +, .) together two binary

operations “+” and “ . ” is called a ring if the following properties hold:

1. (R, +) is an abelian group.

2. (R, .) is a semi group i.e. (a.b).c = a.(b.c), for all a, b, c ∈ R.

3. Distributive laws holds:

a.(b + c) = a.b + a.c and (a + b).c = a.c + b.c, ∀a, b, c ∈ R.

Example 7.2.3. Set of Real numbers (R, +, .), set of rational numbers (Q, +, .),
and set of Complex numbers (C, +, .), (Zp , +, 0), where Zp = {0̄, 1̄, . . . , (p − 1)}
are rings with multiplicative identity 1.

127
Shiv Datt Kumar Matrix Theory and Linear Algebra

(Zm , +, .), where Zm = {0̄, 1̄, . . . , (m − 1)} is a ring with multiplicative iden-
tity 1̄.

Definition 7.2.4. Field: An algebraic structure (F, +, .) together two binary

operations “+” and “ . ” is called a field if the following properties hold:

1. (F, +) is an abelian group.

2. (F ∗ , .) is an abelian group.

3. Distributive laws hold-

a.(b + c) = a.b + a.c and (a + b).c = a.c + b.c, ∀ a, b, c ∈ F .

Note: Here F ∗ = F − {0}, where 0 is the additive identity. Any commutative

ring with identity, in which every non zero element has multiplicative inverse
is a field.

Example 7.2.5. 1. Field real numbers (R, +, .).

2. Field of rational numbers (Q, +, .).

3. Field of complex numbers (C, +, .).

4. Field (Zp , +, .), where p is any prime and Zp = {0̄, 1̄, . . . , (p − 1)} is a
set of residue classes modulo p.

7.3 Vector Space

Definition 7.3.1. Let V be a non empty set and F be a Field. Then V is

called a vector space over F , if (V, +) is an abelian group together with a map
· : F × V → V satisfying the following conditions:

128
Shiv Datt Kumar Matrix Theory and Linear Algebra

1. α.(x + y) = α.x + α.y,

2. (α + β).x = α.x + β.x,

3. α.(β.x) = (αβ).x,

4. 1.x = x, ∀α, β ∈ F , ∀x, y ∈ V and 1 ∈ F .

Example 7.3.2. 1. V = Rn = {(x1 , x2 , . . . , xn ) | xi ∈ R} is vector space

over R with basis {e1 , e2 , . . . , en }.

2. C is a vector space over R.

3. Set of polynomials F [X] over a field F is a vector space.

Proposition 7.3.1. Let V be a vector space over a field F . Then:

1. a.0 = 0, ∀ a ∈ F

2. 0.x = 0

3. a.(−x) = (−a).x = −a.x for all x ∈ V and a ∈ F .

4. a.x = 0 ⇒ a = 0 or x = 0, where x ∈ V and a ∈ F , 0 ∈ V.

Proof. 1. a.0 = a.(0 + 0), ∀ a ∈ F

a.0 + 0 = a.0 + a.0

0 = a.0, by cancellation property of group (V, +).

2. 0.x = (0 + 0).x, ∀ x ∈ V

0.x + 0 = 0.x + 0.x

0 = 0.x, by cancellation property of group (V, +)

129
Shiv Datt Kumar Matrix Theory and Linear Algebra

3. Since 0 = 0.x,

0 = (a + (−a)).x, x ∈ V and for all a ∈ F .

0 = a.x + (−a).x,

−a.x + 0 = −a.x + a.x + (−a).x

−a.x = 0 + (−a).x

−a.x = (−a).x, ∀ x ∈ V and ∀ a ∈ F .

4. Suppose a.x = 0, and x 6= 0, then to prove that a = 0. Suppose contrary

that a 6= 0. Since a ∈ F , then a−1 exists and

a−1 (a.x) = a−1 0

(a−1 .a)x = 0

1.x = 0

x = 0, a contradiction. Hence a = 0

Definition 7.3.3. (Vector subspace) A subset W is called subspace of a vector

space V over a field F if αx + βy ∈ W for all x, y ∈ W and for all α, β ∈ F .

Exercises

1. If W1 and W2 are two subspaces of a vector space V , then W1 ∩ W2 is

also a subspace of V but W1 ∪ W2 need not be subspace of V .

2. If W1 and W2 are two subspaces of a vector space, then W1 ∪ W2 is a

subspace of V iff either W1 ⊂ W2 or W2 ⊂ W1 .

130
Shiv Datt Kumar Matrix Theory and Linear Algebra

Definition 7.3.4. (Linearly dependence and independence) A finite non-empty

set {x1 , x2 , . . . , xn } of vectors of a vector space V is said to be linearly dependent
if there exists scalars a1 , a2 , . . . , an , not all zero such that a1 x1 + a2 x2 + . . . +
an xn = 0.
If a1 x1 + a2 x2 + . . . + an xn = 0 implies ai = 0, for all i, then the set
{x1 , x2 , . . . , xn } is called linearly independent.
An infinite set of vectors S = {x1 , x2 , x3 , . . .} is called linearly independent if
every finite subset S is linearly independent, otherwise S is linearly dependent.

Properties

1. Every set containing 0 is linearly dependent.

2. Every set containing linearly dependent subset is linearly dependent.

3. Every subset of linearly independent set is linearly independent.

4. Null set φ is linearly independent.

5. If x is a linear combination of the vectors {x1 , x2 , . . . , xn }, then {x, x1 , x2 , . . . , xn }

in linearly dependent.

Basis and dimension: A nonempty subset B of a vector space V is called a

basis of V over a field F if
(i) B is linearly independent
(ii) B generates V i.e. V = [B]
i.e. each x ∈ V can be written as linear combination of elements of B i.e.
x = a1 x1 + a2 x2 + . . . + an xn , where ai ∈ F and xi ∈ B.
The number of elements in a basis B is called the dimension of V .

131
Shiv Datt Kumar Matrix Theory and Linear Algebra

Example 7.3.5. 1. Set {e1 , e2 , . . . , en } is a basis of Rn over R, where e1 =

(1, 0, . . . , 0), e2 = (0, 1, 0, . . . , 0), . . . , en = (0, 0, . . . , 1). Then dimR (Rn ) =
n.

2. Set {1, i} is a basis of vector space C over R. Then dimR (C) = 2.

3. Set {1, x, x2 , . . . , xn . . .} is a basis of R[x] over R. Then dimR (R[x]) = ∞

Rb
Exercise 7.3.2. 1. Show that the set V = {f : [a, b] −→ R| a
f (x) dx = 0}
is a vector space over R.

Rb
2. Show that the set V = {f : [a, b] −→ R| a
f (x) dx = 2} is not a vector
space over R.

3. Show that V = solution set of homogeneous linear differential equation

0
y ” − 3y + 2y = 0 is a vector space over R.

4. The set {(1, 0, 0), (1, 1, 0), (1, 1, 1)} is linearly independent.

5. The set {(1, 1, 1), (−1, 0, 1), (0, −2, 1)} is linearly independent.

6. The set {(1, 0, 0), (0, 1, 0), (1, 1, 1), (−1, 1 − 1)} is linearly dependent.

7. Determine whether the set {u, v, w}, where u = (2, 2, 0), v = (3, 0, 2), w =
(2, −2, 2) forms a basis of R3 over R? Ans: The set {u, v, w} is linearly
independent and if (x, y, z) = a(2, 2, 0) + b(3, 0, 2) + c(2, −2, 2) implies
2a = 2c + y, −b = x − y − 2z, 2c = 2x − 2y − 3z. Verify?

8. Find the dimension of subspace of R4 spanned by the set e1 = (1, 0, 0, 0), e2 =

(0, 1, 0, 0), e3 = (1, 2, 0, 1), e4 = (0, 0, 0, 1). Ans: Dim = 3, Note that
e4 = −e1 − 2e2 + e3

132
Shiv Datt Kumar Matrix Theory and Linear Algebra

Sum and direct sum of subspaces

If W1 and W2 are two subspaces of a vector space V over a field F , then

W1 + W2 = {x1 + x2 |x1 ∈ W1 , x2 ∈ W2 } is also a subspace of V .

If every x ∈ V can be expressed uniquely as x = x1 + x2 , where x1 ∈ W1

and x2 ∈ W2 , then V is called direct sum of W1 and W2 and is written as
V = W1 ⊕ W2 .

Proposition 7.3.3. Vector space V is called direct sum of W1 and W2 over a

field F if

(i) V = W1 + W2

(ii) W1 ∩ W2 = {0}.

Example 7.3.6. Let V = R3 , W1 = {(x, y, 0)|x, y ∈ R} and W2 = {(0, 0, z)|z ∈

R}. Then R3 = W1 + W2 and R3 = W1 ⊕ W2 .
     
 a b
 
  a b
 

Let V =  |a, b, c, d ∈ , W = |a, b ∈ ,
   
 R 1   R
 c d
 
  0 0
 

  
 c 0
 

W2 =   |c, d ∈ R .
 
 d 0
 

  
 a+c b
 

Then W1 + W2 =   |a, b, c, d ∈ R
 

 d 0 

  
 x y
 

=   |x, y, z ∈ R .
 
 z 0
 


Hence V 6= W1 + W2 .

133
Shiv Datt Kumar Matrix Theory and Linear Algebra

7.4 Linear Transformation

Let U and V be two vector spaces over a field F . Then a map T : U → V is

called linear transformation if

T (αx + βy) = αT (x) + βT (y), ∀x, y ∈ U &, α, β ∈ F .

1. Let U = C[0, 1] = {f : [0, 1] −→ R|f is continuous} and V = R. Then

R1
define T : C[0, 1] → R by T (f ) = 0 f dx. Then T is a linear transfor-
mation. Take any α, β ∈ R & f, g ∈ C[0, 1]. Then
R1
T (αf + βg) = 0
(αf + βg)dx
R1 R1
= 0
αf dx + 0
βg dx
R1 R1
= 0
f dx + β 0
g dx

= αT (f ) + βT (g)

2. Let U = F [x]. Then T : F [x] → F [x] defined by

d
T (f ) = f (x) is a linear transformation.
dx
 
 1 0 
3. Identity Linear Transformation I =  .
0 1

4. Double stretching along x-direction. Define T : R2 −→ R2 by

  
 2 0  x 
T (x, y) =     = (2x, y).
0 1 y
 
 1 0 
5. Double stretching along y-direction: A1 =  .
0 2

134
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 −1 0 
6. Reflection with respect to y-axis: A2 =  .
0 1

7. Double
stretching
 along y-direction and reflection with respect to y-axis:
 −1 0 
A3 =  
0 2
 
 cos θ −sin θ 
8. Rotation by an angle θ: A4 =  .
sin θ cos θ
 
π  0 −1 
9. Rotation by an angle θ = : A5 =  .
2 1 0
 
 1 1 
10. Shear: A6 =  .
0 1
 
 1 −1 
11. Shear and rotation : A7 =  .
0 1

7.5 Matrix representation of linear transfor-

mation

Let T : V −→ W be linear transformation and B1 = {e1 , e2 , . . . , en } and

B2 = {f1 , f2 , . . . , fm } be bases of V and W respectively. Then

T (e1 ) = α11 f1 + . . . + α1m fm

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

135
Shiv Datt Kumar Matrix Theory and Linear Algebra

T (en ) = αn1 f1 + . . . + αnm fm .

 
 α11 α12 . . . αn1 
 . .. .. .. 
M atrix of T =  .
 . . . . 

 
α1m α12 . . . αnm

1. Define T : R2 → R3 by T (x, y) = (x + y, x, x − y),

where T (1, 0) = (1, 1, 1) = 1(1, 0, 0) + 1(0, 1, 0) + 1(0, 0, 1)

T (0, 1) = (1, 0, −1) = 1(1, 0, 0) + 0(0, 1, 0) − 1(0, 0, 1)

 
 1 1 
 
Matrix (T ) =   1 0 .

 
1 −1

2. Define T : R2 → R2 by T (x, y) = (x + 2y, 3x + 4y)

 
 1 2 
Matrix of T =  
3 4

7.5.1 Null Space and Range space

Let T : U → V be linear transformation. Then Ker(T ) = {x ∈ U |T (x) = 0}

and Range(T ) = {y ∈ V |∃ x ∈ U such that T (x) = y} are subspaces of U and
V and are called null space and range space respectively. Dimension of null
space is called nullity and dimension of range space is called rank.

Theorem 7.5.1. (Rank-Nullity Theorem) Let T : U −→ V be linear transfor-

mation and dim(U ) = n. Then Rank(T ) + N ullilty(T ) = n.

Example 7.5.1. 1. Find the matrix, rank and nullity of linear transforma-
tion T : R3 −→ R2 defined by T (x, y, z) = (x + y, −x + z).

136
Shiv Datt Kumar Matrix Theory and Linear Algebra

Answer:
 
 1 1 0 
M atrix (T ) =  
−1 0 1

Ker(T ) = {(x, y, z)| T (x, y, z) = (0, 0)} = {(x, y, z)| x + y = 0, −x + z =

0} = {a, −a, a)| a ∈ F }. Nullity= dim Ker(T ) = 1.
 
  x  
 1 1 0    x+y 
 
Range(T) ={T (v)|v ∈ V } = {Av =   y =
 
−1 0 1   −x + z
z
           
 1   1   0   1   1   0 
= x +y  +z  }. Since   =   −  .
−1 0 1 −1 0 1
This implies dim (Range(T )) = Rank(T ) = 2. Also by Rank-Nullity
Theorem Rank(T ) + N ullilty(T ) = dim(R3 ) = 3. Then Rank(T ) =
3−1=2

d
2. Linear transformation T : P4 −→ P3 be defined by T (f (x)) = (f (x)).
dx
Since T (constant polynomial) = 0. This implies null space N (T ) = F
(Field) and nullity = 1.

137
Shiv Datt Kumar Matrix Theory and Linear Algebra

7.6 Exercises

1. If W1 and W2 are two finite dimensional subspaces of a vector space V

over a field F , then W1 + W2 is also finite dimensional and

dim(W1 + W2 ) = dim(W1 ) + dim(W2 ) − dim(W1 ∩ W2 ).

2. Let linear transformation T : R2 −→ R3 be defined by T (x, y) = (x −

y, x + y, x + 3y). Find matrix of T with respect to standard basis.

3. Let linear transformation T : P1 −→ P2 be defined by T (a + bx) =

R
(a + bx) dx. Find matrix of T with respect to standard basis, where
P1 = {a + bx|a, b ∈ R} and P2 = {a + bx + cx2 | a, b, c ∈ R}.

4. Determine matrix of the linear transformation T : R3 −→ R2 defined by

T (x, y, z) = (y + z, y − z) with respect to (i) standard basis (ii) basis of
R3 = {(0, 1, 1), (1, 0, 1), (1, 1, 0)} and basis of R2 = {(1, 1), (1, −1)}.

5. Find rank and nullity of linear transformation T : R2 −→ R3 defined by

T (x, y, z) = (2x + y, y − x, 3x − 4y).

6. Find matrix, rank and nullity of linear transformation T : R3 −→ R3

defined by T (x, y, z) = (x + y, z, x − y).

7.6.1 Eigen value (Latent root)

A number λ ∈ F is called an eigen value of a linear transformation T : V −→ V

if ∃u ∈ V , u 6= 0 such that T (u) = λu.

Characteristic polynomial and characteristic equation: If λ is an eigen-

value of the matrix A, then |A − λI| is called the characteristic polynomial and

138
Shiv Datt Kumar Matrix Theory and Linear Algebra

|A − λI| = 0 is the characteristic equation.

Remark 7.6.1. If the homogeneous system of equation AX − λX = 0 has a

non-trivial solution, then Rank(A − λI) < n = no. of variables. Then the
coefficient matrix (A − λI) is singular i.e. |A − λI| = 0 i.e.
λn − c1 λn−1 + . . . + (−1)n−1 cn−1 λ + (−1)n cn = 0.
If λ1 , λ2 , . . . , λn are roots of characteristic equation, then
λ1 + λ2 + . . . + λn = T race(A) = −c1 .
λ1 .λ2 . . . . λn = det(A) = (−1)n cn

7.6.2 Eigen vector (Characteristic vector)

Let V be a finite dimensional vector space over a field F and T : V → V be

a linear operator on V . Then x ∈ V , x 6= 0, is called a characteristic vector
corresponding to eigen value λ if T (x) = λx.
Characteristic space corresponding to λ = {x ∈ V |T (x) = λx}.

Theorem 7.6.1. Let V be a finite dimensional vector space over a field F and
T : V → V is a linear transformation on V . Then the following are equivalent:
(i) λ is a eigen value of T .
(ii) The operator (T − λI) is singular.
(iii) |T − λI| = 0.

Proof. If λ is a characteristic value of T , then ∃ x 6= 0 such that T x = λx.

Therefore (T − λI)x = 0. Homogeneous system of equations have non trivial
solution only if (T − λI) is singular i.e. |T − λI| = 0.

Eigen Value Problem: Consider a homogeneous system of equation in ma-

trix form AX = λX, where λ is a scalar. This system of equation always has

139
Shiv Datt Kumar Matrix Theory and Linear Algebra

a trival solution. We need to find values of λ for which non trivial solution of
this homogeneous system exists. The values of λ for which nontrival solution
of AX = λX exists are eigen values and corresponding eigen vector are solu-
tions. If X is a solution, then αX is also a solution, where α is scalar. Hence
eigen vector is unique up to constant multiple. Thus problem of determining
the eigen values and corresponding eigen vectors of a square matrix A is called
eigen value problem.

Theorem 7.6.2. (Cayley-Hamilton Theorem) Every square matrix A of order n

satisfies its own characteristic equation i.e.

An − c1 An−1 + . . . + (−1)n−1 cn−1 A + (−1)n cn I = 0.

Proof. Cofactors of the elements of the det(A − λI) are polynomial in λ of

degree n − 1 or less. Therefore the elements of adjoint matrix (transpose of
cofactor matrix) are also polynomial in λ of degree n − 1 or less. Hence

adj(A − λI) = B1 λn−1 + B2 λn−2 + . . . + Bn−1 λ + Bn .

Also (A − λI) adj(A − λI) = |(A − λI)| I.

(A − λI)(B1 λn−1 + B2 λn−2 + . . . + Bn−1 λ + Bn )

= λn I − c1 λn−1 I + c2 λn−2 I + . . . + (−1)n−1 cn−1 A + (−1)n cn I.

Comparing the coefficients of powers of λ

−B1 = I

AB1 − B2 = −c1 I

AB2 − B3 = c2 I
..
.

ABn−1 − Bn = (−1)n cn−1 I

ABn = (−1)n cn I

140
Shiv Datt Kumar Matrix Theory and Linear Algebra

Pre-multiplying these equations by An , An−1 , . . . , A, I and adding we have

An − c1 An−1 + . . . + (−1)n−1 cn−1 A + (−1)n cn I = 0.

Example 7.6.1. Find eigen values and eigen vectors of the matrix
 
 1 1 2 
 
A=
 −1 2 1 .

 
0 1 3

Answer: |A − λI| = 6 − 11λ + 6λ2 − λ3 = (λ − 1)(λ − 2)(λ − 3) = 0. λ =

1, 2, 3 and Eigen vectors X1 = a(−1, −2, 1), X2 = a(−1, −1, 1), X3 = a(1, 0, 1)
respectively.

Example 7.6.2. Verify Cayley-Hamilton theorem for the matrix

 
 1 2 0 
 
A=
 −1 1 2  .

 
1 2 1

Hence (i) find A3 and A−1 (ii) Verify that eigen values of A2 are squares of
eigen values of A. (iii) Find spectral radius.

Proof. |A − λI| = −λ3 + 3λ2 − λ + 3 = (λ − 3)(λ2 + 1) = 0. Then λ = 3, ±i.

   
 −1 4 4   −1 10 12 
   
A2 =   3 2
 0 3 4  and A = A .A =  1 11 10
 .

   
0 6 5 −1 16 17

141
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 −3 −2 4 
1 1
T hen A−1

= [A2 − 3A + I] =  3 1 −2 .
3 3



−3 0 3

Eigen values of A2 are 9, −1, −1. Spectral radius = Max|λi | = 3

Example 7.6.3. If
 
 1 0 0 
 
A=
 1 0 1 .

 
0 1 0

Then show that An = An−2 + A2 − I, for all n ≥ 3.

 
 1−λ 0 0 
 
Proof. |A−λI| =  −λ 1  = λ3 −λ2 −λ+1 = (1−λ)(λ2 −1) = 0.
 1 
 
0 1 −λ

A3 − A2 − A + I = 0

A3 − A2 = A − I

A4 − A3 = A2 − A
..
.

An−1 − An−2 = An−3 − An−4

An − An−1 = An−2 − I.

Adding these equations we have An − A2 = An−2 − I. Using these equations

n 1
recursively An = An−4 + A2 − I = An−4 + 2(A2 − I) = A2 − (n − 2)I.
2 2
142
Shiv Datt Kumar Matrix Theory and Linear Algebra

 
 1 0 0 
 
A2 = 
 1 1 0 .

 
1 0 1

     
 1 0 0   1 0 0   1 0 0 
     
Hence A50  1 1 0  − 24  0 1 0
= 25     =  25 1 0 
  
     
1 0 1 0 0 1 25 0 1

Definition 7.6.4. A square matrix A = [aij ] is called

−1
1. Unitary if AT = (A) .

2. Orthogonal if AT = A−1 . i.e. AAT = I.

3. Hermitian if A = AT or A = (A)T .

4. Skew Hermitian if A = −AT .

T T
5. Positive definite if X AX > 0, for X 6= 0 and X AX = 0, iff X = 0.

Properties: If λ is an eigenvalue of the matrix A and x is eigen vector corre-

sponding to λ, then

1. Matrix αA has eigen value αλ, where α ∈ F (field), for Ax = λx implies

αAx = αλx.

2. Matrix Am has eigen value λm .

143
Shiv Datt Kumar Matrix Theory and Linear Algebra

3. Matrix (A − kI) has eigen value λ − k, for Ax = λx implies (A − k I)x =

Ax − kIx = λx − kx = (λ − k)x.

4. Matrix A−1 has eigen value λ−1 , for Ax = λx implies A−1 Ax = λA−1 x
implies A−1 x = λ−1 x.

1
5. Matrix (A − k I)−1 has eigen value .
λ−k

6. Eigen values of matrix A and AT are same.

7. For a real matrix A, if α + iβ is an eigen value, then α − iβ is also an

eigen value.

8. Eigenvalues of A = roots of |A − λI| = 0 in the field.

9. Trace of A = sum of eigen values of A.

10. Det A = Product of eigen values of A.

11. All the eigen values of a Hermitian matrix (A = AT ) are real. A matrix
T is Hermitian if < T (x), y >=< y, T (x) >, ∀ x, y ∈ V . Then λ <
x, x >=< λx, x >=< T (x), x >=< x, T (x) >=< x, λx >= λ̄ < x, x >.
Hence λ = λ̄.

12. All the eigen values of a real symmetric matrix are real, for real symmetric
matrix can be taken as complex Hermitian matrix.

13. All nonzero eigen values of a skew Hermitian matrix are pure imaginary.

14. Eigen values of a positive definite matrix are real and positive.

15. All leading minors of a positive definite matrix are positive.

144
Shiv Datt Kumar Matrix Theory and Linear Algebra

Definition 7.6.5. A matrix A is said to be diagonalizable if there exists non-

singular matrix P such that P −1 AP = D.

Theorem 7.6.3. A square matrix A of order n is diagonalizable if it has n

linearly independent eigen vectors.

Proof. Let X1 , X2 , . . . , Xn be linearly independent vectors corresponding to

eigen values λ1 , λ2 , . . . , λn (not necessarily distinct) i.e. AXi = λi Xi . Con-
sider matrices P = [X1 X2 . . . Xn ] and D = diag(λ1 , λ2 , . . . , λn ). Then
AP = A[X1 X2 . . . Xn ] = [AX1 AX2 . . . AXn ] = [λ1 X1 λ2 X2 . . . λn Xn ] =
[X1 X2 . . . Xn ]D = P D. Since Rank P = n, matrix P is invertible, then
P −1 AP = P −1 P D = D. Hence A is diagonalizable.

Exercise 7.6.4. The eigen vectors of 3 × 3 matrix A corresponding to eigen

values 1, 1, 3 are (1, 0, −1)T , (0, 1, −1)T , (1, 1, 0)T . Find matrix A.
 
 1 −1 2 
 
Example 7.6.6. Change the matrix   0 2 3  to the diagonal form.

 
0 0 3

1−λ −1 2
Solution: Characteristic equation |A − λI| = 0 2−λ 3 = 0.

0 0 3−λ
(1 − λ)(2 − λ)(3 − λ) = 0 i.e. λ = 1, 2, 3.

For
 λ = 1, wehave (A  − λ
1 I)X= 0 i.e.

 0 −1 2   x   0 
    
 0 1 3   y  =  0  or −y + 2z = 0, y + 3z = 0, 2z = 0 i.e.
    
    
0 0 2 z 0
y = 0, z = 0, x = arbitrary. Then eigen vector is X1 = (1, 0, 0). Similarly for

145
Shiv Datt Kumar Matrix Theory and Linear Algebra

X2 = (1, −1,
λ = 2, eigen vector   X3 = (3, −2,
0), for λ = 3, eigen vector  2).
 1 1 3   2 2 −1 
, |P | = −2, P −1 = 1  0 −2 −2 ,
   
Modal matrix P =  1 −1 0



 2



0 0 3 0 0 1
 
 1 0 0 
 
P −1 AP = D =   0 2 0 .

 
0 0 3

 
 6 −2 2 
 
Example 7.6.7. Let A =  −2 3 −1 . Find matrix P such that P −1 AP
 
 
2 −1 3
is a diagonal matrix.
6−λ −2 2
Solution: Characteristic equation |A − λI| = −2 3−λ −1 = 0.

2 −1 3−λ
or (λ − 2)2 (λ − 8) = 0 i.e. λ = 2, 2, 8.
For λ = 2, eigen vector is given by (A − 2I)X = 0 i.e.
         
 4 −2 2   x   0   2 −1 1   x   0 
         
 −2 1 −1   y  =  0  or  0 0 0  
= 0 
y   
      
         
2 −1 1 z 0 0 0 0 z 0
or 2x − y + z = 0. This equation is satisfied by x = 0, y = 1, z = 1 and x = 1,
y = 3, z = 1. Then eigen vectors are X1 = (0, 1, 1) and X2 = (1, 3, 1),.
For λ = 8, eigen vector
 −1, 1).
X3 = (2,  
 0 1 2   4 1 −7 
  −1 1 
Modal matrix P =  , |P | = −6, P = − 6  −2 −2 2
3 −1  ,
 1 
   
1 1 1 −2 1 −1

146
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 2 0 0 
 
P −1 AP = D = 
 0 2 0 .

 
0 0 8
 
 1 2 3 
 
Example 7.6.8. Check whether the matrix 
 0 1 2 

 
0 0 1
is diagonalizable or not ?
1−λ 2 3
Solution: Characteristic equation |A − λI| = 0 1−λ 2 = 0.

0 0 1−λ
3
(1 − λ) = 0 i.e. λ = 1, 1, 1.
For λ = 1, we have
  (A − λ1
I)X = 0 i.e.
 0 2 3  x   0 
    
 0 0 2   y  =  0  Implies 2y +3z = 0, 2z = 0. Therefore x =
    
    
0 0 0 z 0
arbitrary, y = 0, z = 0. Hence eigen vector can be taken as (1, 0, 0), (2, 0, 0),
(3, 0, 0). These vectors are linearly dependent. Hence diagonalization is not
possible.

Definition 7.6.9. (Quadratic form)

Let X = (x1 , x2 , . . . , xn )T and A = [aij ] be a symmetric matrix. Then a real

quadratic form is an expression of the form Q = X T AX = ni=1 ni=1 aij xi xj .
P P

Note that if Rank(A) = r, then X T AX will contain only r terms.

  
 a h  x 
 
Example 7.6.10. 1. ax2 +2hxy+by 2 = x y     = X T AX.
h b y

147
Shiv Datt Kumar Matrix Theory and Linear Algebra
  
  a h g  x 
  
2. ax2 +by 2 +cz 2 +2hxy+2f yz+2gzx+2hxy = x y z 
 h b f   y .
 
  
g f c z

Exercise 7.6.5. 1. Obtain the symmetric matrix A for the quadratic form
(i) 4x21 + 5x1 x2 + x22 (ii) x21 + 2x1 x2 − 6x1 x3 + 8x2 x3 − 5x22 + 4x23 .

2. Examine which ofthe following

 matrix are positive definite (i) A =
 1 0 i 
 
4 1  
 (ii) A = 
 
 0 1 0 
 
1 4  
−i 0 3

Definition 7.6.11. (Canonical form) By a non-singular linear transformation

X = P Y , if the quadratic form X T AX is reduced to another quadratic form of
the type Y T BY = ni=1 λi yi2 , here B = dig(λ1 , . . . , λn ) and λi are eigen values
P

of A, then Y T BY is called canonical form of X T AX.

Index = p = number of positive terms in quadratic form,

q = r − p = number of negative terms in quadratic form,

Signature of quadratic form = p − q = 2p − r

Example 7.6.12. Reduce the quadratic form

3x21 + 3x22 + 3x23 + 2x1 x2 + 2x1 x3 − 2x2 x3

into canonical form and give the matrix of transformation.

   
 3 1 1   x1 
   
Solution: Here A =   1 3 −1  and X =  x .
  2 
   
1 −1 3 x3

148
Shiv Datt Kumar Matrix Theory and Linear Algebra

3−λ 1 1
Characteristic equation |A − λI| = 1 3−λ −1 = 0.

1 −1 3−λ
λ3 − 9λ2 + 24λ − 16 = 0 or λ = 1, 4, 4.

Canonical form = y12 + 4y22 + 4y32 .

For λ = 1, eigen vector is given by (A − I)X = 0 i.e.

    
 2 1 1   x1   0 
    
 1 2 −1   x  =  0  or
  2   
    
1 −1 2 x3 0
2x1 + x2 + x3 = 0

x1 + 2x2 − x3 = 0

x1 − x2 + 2x3 = 0

Solving we get x1 = −1, x2 = 1, x3 = 1. Eigen vector X1 = (−1, 1, 1), Nor-

−1 1 1
malized eigen vector =X 1 = ( √ , √ , √ ).
3 3 3

For λ = 4, eigen vector is given by (A − 4I)X = 0 i.e.

    
 −1 1 1   x1   0 
    
 1 −1 −1   x  =  0  or x1 − x2 − x3 = 0
  2   
    
1 −1 −1 x3 0

Let x1 = 0. Then x2 = −x3 and eigen vector X2 = (0, 1, −1). Normalized

1 −1
eigen vector X 2 = (0, √ , √ ). Now let x2 = x3 = k,. Then x1 = 2k and
2 2
2 1 1
eigen vector X3 = (2, 1, 1). Normalized eigen vector X 3 = ( √ , √ , √ ).
6 6 6

149
Shiv Datt Kumar Matrix Theory and Linear Algebra

 −1 2 
√ 0 √
 3 6 
 1 1 1 
Matrix of transformation is P =  √ √ √ .
 
 3 2 6 
 1 −1 1 
√ √ √
3 2 6
Rank = 3, Index = p = 3, Signature = 2r − p = 3.
X = P Y implies Y = P −1 X
   −1 1 1  
√ √ √
 y1   3 3 3 
x1 
   1 −1  
 y = 0 √ √  .

 2   x2 
  2 2  
  2 1 1 

y3 √ √ √ x3
6 6 6
Example 7.6.13. Solve by matrix method

d2 y dy 0

2
− 5 + 6y = 0, y(0) = 1, y (0) = 2
dx dx

.
dy1
Solution: Let y = y1 and = y2 − − − − − − − − − − − − − (1).
dx
dy2
Then = −6y1 + 5y2 − − − − − − − − − − − − − − − −(2)
dt
Differential equations (1) and (2) can be written in matrix form as

    
d  y1   0 1   y1 
=
dx y
  
2 −6 5 y2

−λ 1
Characteristic equation |A − λI| = = 0.
−6 5 − λ
2
 λ− 5λ 
+6=  0 or λ = 2, 3. Eigen vector
 for
 λ = 2 and λ= 3 are 
 1   1   1 1   3 −1 
  and  . Modal matrix P =   and P −1 =  
2 3 2 3 −2 1

150
Shiv Datt Kumar Matrix Theory and Linear Algebra
   
2x
 1 1  e 0   3 −1 
P eλx P −1 =    
3x
2 3 0 e −2 1
 
2x 3x 2x 3x
 3e − 2e −e + e
=


6e2x − 6e3x −2e2x + 3e3x
      
2x 3x
 y1   3e − 2e −e2x + e3x   1   e2x 
 =
    =  
2x 3x 2x 3x
y2 6e − 6e −2e + 3e 2 2e2x
.
dy
y1 = y = e2x and y2 = = 2e2x .
dx
Exercise 7.6.6. Solve by matrix method

d2 x 0
+ 4x = 0, x(0) = 1, x (0) = 0
dt2

7.7 Similar matrices

Two matrices A and B are said to be similar if ∃ an invertible matrix P such

that B = P AP −1 . Two linear operators T and S on a finite dimensional vector
space V are said to be similar if there exists an invertible linear transformation
P such that S = P T P −1 .

Proposition 7.7.1. Similar matrices have same eigen values.

Proof. Let A and B be two similar matrices. Then ∃P such that B = P AP −1 .

Now |B − λI| = |P AP −1 − λP P −1 | = |P AP −1 − λP IP −1 | = |P AP −1 −
P λIP −1 | = |P (A − λI)P −1 | = |P ||A − λI| |P −1 | = |A − λI|.

151
Shiv Datt Kumar Matrix Theory and Linear Algebra

Remark 7.7.1. If A and B are similar matrices, then

1. Trace A = Trace B

2. Characters polynomial |A − λI| = |B − λI|

3. Eigenvalues of A = Eigenvalues of B.

4. Det A = Det B.

Theorem 7.7.2. Let T be a linear operator on n dimensional vector space V [or

let A be n × n matrix]. Then the characteristic polynomial and the minimal
polynomial for T [or A] have same roots, except for multiplicity.

Proof. Let p be the minimal polynomial for T . Let c be a scalar. We want

show that p(c) = 0 iff c is a eigen value of T . Suppose p(c) = 0. Then
p = (x − c)q, where q is some polynomial. Since degree q < degree p, the
definition of minimal polynomial p tells that q(T ) 6= 0. Choose a vector n
such that q(T )v 6= 0 and p(T )(v) = 0. Let u = q(T )v. Then 0 = p(T )v =
(T − cI)q(T )v = (T − cI)u ⇒ T (u) = cu. This implies that c is an eigen value
of T . Conversely, suppose c is an eigen value of T say T u = cu, for u 6= 0. We
know that p(T )u = p(c)u. Since p(T ) = 0 & u 6= 0, we have p(c) = 0.

Theorem 7.7.3. Let T be a linear operator on a finite dimensional vector space

V . If f is the characteristic polynomial for T , then f (T ) = 0, In other words,
the minimal polynomial divides the characteristic polynomial for T .

152
Shiv Datt Kumar Matrix Theory and Linear Algebra

7.8 Matrix Theory

Elementary transformation in a matrix

1. Interchange of any two rows (Ri ↔ Rj ) or two columns (Ci ↔ Cj ).

2. Multiplication of a row or column by a non-zero number K

i.e. (Ri → KRi ) or (Ci → KCi ).

3. Adding a row or column by multiplying a non-zero number K to a row

or column (Ri → Ri + KRj ) or (Ci → Ci + KCj ).

Elementary matrix
The matrix obtained from unit matrix I after one or more row elementary
transformations.

Canonical matrix
The matrix obtained by applying a series of elementary row or column op-
erations such that there are some non-zero rows in the top and remaining rows
consists of all zeros.

Echelon form of a matrix

The matrix obtained by applying a series of elementary row or column
operations such that
(i) All the rows with zero elements will be below the non-zero rows.
(ii) Number of zeros before the first non-zero in a row is less than the number
of such zero in the next row.

153
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 2 0 3 5 
 
 0 1 2 0 
A=  Row echelon form
 
 0 0 5 0 
 
 
0 0 0 0

Rank of a matrix: A matrix M has rank r if

1. there is atleast one minor of M of order r, which does not vanish.

2. every minor of M of order r + 1 vanishes.

Remark 7.8.1. (Facts)

1. Elementary transformation do not change rank of matrix.

2. Rank(AB) ≤ Min {Rank(A), Rank(B)}.

3. If a matrix A is equivalent to a canonical matrix having exactly r non-zero

rows, then Rank(A) = r.

4. If M is non-singular matrix of order n, then Rank(M A) = Rank(A) =

Rank(AM ), where A is matrix of order n.

5. Rank of a canonical matrix = number of non-zero rows.

 
 0 1 −3 −1 
 
 1 0 1 1 
Example 7.8.1. Determine Rank of A = 
 

 3 1 0 2 
 
 
1 1 −2 0

154
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 0 1 −3 −1 
 
 1 0 1 1 
∼  Using R3 − R1 , R4 − R1
 
 3 0 3 3 
 
 
1 0 1 1
 
 0 1 −3 −1 
 
 1 0 1 1 
∼  Using R3 − 3R2 , R4 − R2
 
 0 0 0 0 
 
 
0 0 0 0
Hence Rank(A) = 2.

7.9 Solution of general linear system of equa-

tions

Consider the system of equation

a11 x1 + a12 x2 + . . . + a1n xn = b1

.. .. .. ..
. . . .
am1 x1 + am2 x2 + . . . + amn xn = bm
 
 b1 
 .. 
or in matrix form AX = b, where A = Coefficient matrix = [aij ], b = 
 . ,

 
bn
 
 x1 
 .. 
X=
 .  . Here m = number of rows = number of equations, n = number

 
xn
of columns. Crammer rule for solving a system of m equations in n unknown
AX = b requires |A| =
6 0.

155
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 a11 . . . a1n . b1 
 . .. .. .. .. 
Augmented matrix à = (A|b) =  ..
 . . . . .
 
am1 . . . amn . bm
A system of equations is said to be consistent if solution exists otherwise
non-consistent. There are three possibilities given in the following theorem.

Theorem 7.9.1. The non-homogeneous system of equation AX = b has

1. Unique solution if Rank A = Rank à = n (number of variables).

2. No solution if Rank A 6= Rank Ã

3. Infinite solution if Rank A = Rank à = m < n

Corollary 7.9.1. The homogeneous system of equation AX = 0 has

1. Trivial solution if Rank A = n (number of variables).

2. Infinite solution if Rank A = m < n.

Example 7.9.2. Test for consistency and solve

x + 2y + z = 3

2x + 3y + 2z = 5

3x − 5y − 5z = 2

3x + 9y − z = 4
 
 1 2 1 : 3 
 
 2 3 2 : 5 
Proof. Ã = 
 

 3 −5 5 : 2 
 
 
3 9 −1 : 4

156
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 1 2 1 : 3 
 
 0 −1 0 : −1 
∼  Using R2 − 2R1 , R3 − 3R1 , R4 − 3R1
 
 0 −11 2 : −7 
 
 
0 3 4 : −5
 
 1 2 1 : 3 
 
 0 −1 0 : −1 
∼  Using R3 − 11R2 , R4 + 3R2
 
 0 0 2 : 4 
 
 
0 0 −4 : −8
 
 1 2 1 : 3 
 
 0 −1 0 : −1 
∼  Using R4 + 2R2
 
 0 0 2 : 4 
 
 
0 0 0 : 0
Rank A = Rank à = 3 = no. of variables. Thus unique solution exists.

2z = 4 implies z = 2, −y− = −1 or y = 1, x = 3 − 3y − z = 3 − 2 − 2 = −1.

Hence unique solution x = −1, y = 1, z = 2.

Example 7.9.3. Test for consistency and solve

−2x + y + z = 2

x − 2y + z = 2

x + y − 2z = 2
 
 −2 1 1 : 2 
 
Proof. Ã = 
 1 −2 1 : 2 

 
1 1 −2 : 2

157
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 1 1 −2 : 2 
 
∼
 1 −2 1  Using R13
: 2 
 
−2 1 1 : 2
 
 1 1 −2 : 2 
 
∼
 0 −3 3 :  Using R2 − R1 , R3 + 2R1
0 
 
0 3 −3 : 6
 
 1 1 −2 : 2 
 
∼
 0 −3 3 :  Using R3 + R2
0 
 
0 0 0 : 6
Rank A = 2, Rank à = 3 = no. of variables. Thus system is inconsistent.

Example 7.9.4. For what values of λ, µ do the following system of equations

x+y+z =6

x + 2y + 3z = 10

x + 2y + λz = µ

have (i) No solution (ii) Unique solution (iii) Infinite solutions

 
 1 1 1 : 6 
 
Proof. Ã = 
 1 2 3 : 10 

 
1 2 λ : µ
 
 1 1 1 : 6 
 
∼  0 1 2 : 4   Using R2 − R1 , R3 − R1
 
0 1 λ−1 : µ−6

158
Shiv Datt Kumar Matrix Theory and Linear Algebra
 
 1 0 −1 : 2 
 
∼
 0 −1 2 : 4  Using R1 − R2 , R3 − R2

 
0 0 λ − 3 : µ − 10

1. If λ = 3, µ 6= 10, then Rank A = 2, Rank à = 3 = no. of variables.

Inconsistent. No solution.

2. If λ 6= 3, µ has any value, then Rank A = Rank à = 3 = Unique solution.

3. If λ = 3, µ = 10, then Rank A = Rank(Ã) = 2 < 3= no. of variables.

Infinite solutions.

159
Shiv Datt Kumar Matrix Theory and Linear Algebra

160
Chapter 8

Statistics and Probability

Frequency distribution

The frequency (f ) is the number of times the value occurs in the data.
The distribution of a variable is the pattern of frequencies associated with these
values. Frequency distributions are portrayed as frequency tables or charts.

Frequency distributions can be shown either by the actual number of ob-

servations falling in each range or by the percentage of observations. If the
distribution is the percentage of observations, then it is called a relative fre-
quency distribution. Frequency distribution can be better way visualized
by histogram. By joining the middle points of the upper sides of rectangles by
dotted straight lines, we obtain a frequency curve. The graph representing
the cumulative frequency is called ogive.

161
Shiv Datt Kumar Statistics and Probability

8.1 Measures of central tendency

Mean
P
xi
If x1 , x2 , . . . , xn are n values of a variable, then mean x̄ = . For a
n
grouped data, if x1 , x2 , . . . , xn are mid values of class intervals having frequen-
P
f i xi
cies f1 , f2 , . . . , fn , then x̄ = P is called the weighted mean.
fi

Remark 8.1.1. If x̄1 and x̄2 are the means of two samples of size n1 and n2 ,
n1 x̄1 + n2 x̄2
then mean is given by x̄ = .
n1 + n2

Median

It is a number such that half of the values of the data have values less than
and half values are greater than it.

For ungrouped data, if n values of a variable are arranged in the ascending

order of magnitude,
 then

middle term, if n is odd

Median =

mean of two middle terms, if n is even.

 
N
− C .h
2
For grouped data, median = L + , where
f
L = lower limit of the median class

f = frequency of the median class

h = width of the median class

C = cumulative frequency upto the class preceding the median class

P
N= fi = total frequency

162
Shiv Datt Kumar Statistics and Probability

Mode

The value or number in a data set, which has the highest frequency is called
mode or modal value.

The mode value depends on the given dataset. Mode for grouped data is
found using the following mode formula.
h(fm − f1 )
Mode formula =L +
(fm − f1 ) + fm − f2 )
where,

L = the lower limit of the modal class.

h = size of the class interval.

fm = frequency of the modal class.

f1 = frequency of the class preceding the modal class.

f2 = frequency of the class succeeding the modal class.

Remark 8.1.2. 1. Relation: M ean–M ode = 3(M ean–M edian) Or

M ode = 3 M edian–2 M ean.

2. For a symmetrical distribution mean, median, mode coincide.

Example 8.1.1. For the following data calculate (i) Mean (ii) Median (iii)
Mode

Class : 0-10 10-20 20-30 30-40 40-50 50-60

Frequency: 14 17 22 26 23 18

(i) Consider the mid value (x) of each class, for finding mean:

x: 5 15 25 35 45 55

f 14 17 22 26 23 18

163
Shiv Datt Kumar Statistics and Probability
P
f i xi 3810
Mean x̄ = P = = 31.75
fi 120

Class : 0-10 10-20 20-30 30-40 40-50 50-60

Frequency: 14 17 22 26 23 18

Cumulative Frequency: 14 31 53 79 102 120

P N
(ii) N = f = 120, h = 10. Now = 60. Cumulative frequency just greater
2
than 60 is 79 and corresponding class is 30 − 40. Then median class = 30 − 40,
L = 30, f = 26, C = 53. 
N
− C .h
2 (60 − 53)10
Median = L + = 30 + = 32.69.
f 26
(iii) For the given data maximum frequency = 26. Then modal class =30 −
40, L = 30, fm = 26, f1 = 22, f2 = 23.
h(fm − f1 ) (26 − 22)10
Mode formula =L + = 30 + =
(fm − f1 ) + fm − f2 ) (26 − 22) + (26 − 23)
35.7143.

8.2 Probability

Measure of chance or likelihood for the statement to be true is called the

probability of the statement. A branch of mathematics, where we investigate
and discuss rules for random experiments is called the theory of probability.

Random experiment: Experiment happening under uncertain situations is

called a random trial or random experiment. Random means haphazard. It is
performed by a set of rules and result depends on chance.

Outcome: Result of a single trial (performance of experiment).

Event: A subset of sample space,

164
Shiv Datt Kumar Statistics and Probability

Elementary events: Consider a random experiment when all possible out-

comes are known. Then each outcome of this experiment is called an elemen-
tary event.

Elementary events: Two or more events are called mutually exclusive if

occurrence of any one of them excludes the occurrence of other. In this case If
A and B two mutually exclusive events, then P (A ∩ B = 0.

Sample space: The set of all possible outcomes of a random experiment is

called sample space.

Example 8.2.1. 1. Let two coins be thrown simultaneously. Then sample

space S = {HH, HT, T H, T T }.

2. Sample space of a random experiment of throwing a dice S = {1, 2, 3, 4, 5, 6}.

Axioms of Probability:

Let S be a sample space and A, B are two mutually exclusive events. Then

1. 0 ≤ P (A) ≤ 1, 0 ≤ P (B) ≤ 1.

2. P (S) = 1

3. P (A ∪ B) = P (A) + P (B).

Consider an experiment with multiple components. If the 1st compo-

nent has n1 possible outcomes, the 2nd component has n2 possible outcomes,
. . . , and the rth component has nr possible outcomes, then overall there are
n1 n2 . . . nr possibilities for the whole experiment.

165
Shiv Datt Kumar Statistics and Probability

8.2.1 Sampling Table

The number of possible samples of size k out of a population of size n, under

various assumptions is given by:

Order Matters Not Matter

n+k−1
 
k
With Replacement n
k 
n! n
Without Replacement
(n − k)! k

8.2.2 Definition of Probability

If all outcomes are equally likely, the probability of an event A happening is:

number of outcomes favorable to A

Pnaive (A) =
number of outcomes

8.2.3 Independent events

Independent Events Two events A and B are independent if knowing whether

A occurred gives no information about whether B occurred. Events A
and B (which have nonzero probability) are independent if and only if
one of the following equivalent statements holds:

P (A ∩ B) = P (A)P (B)

P (A|B) = P (A)

P (B|A) = P (B)

Conditional Independence Two events A and B are conditionally indepen-

dent given C if P (A∩B|C) = P (A|C)P (B|C). Conditional independence

166
Shiv Datt Kumar Statistics and Probability

does not imply independence, and independence does not imply condi-
tional independence.

Joint Probability: P (A ∩ B) or P (A, B) – Probability of A and B.

Conditional Probability P (A|B) = P (A, B)/P (B) – Probability of A, given

that B occurred.

Conditional Probability is Probability: P (A|B) is a probability function

for any fixed B. Any theorem that holds for probability also holds for
conditional probability.

Bayes’ Rule: Bayes’ Rule unites marginal, joint, and conditional probabili-
ties. We use this as the definition of conditional probability.

P (A ∩ B) P (B|A)P (A)
P (A|B) = =
P (B) P (B)

Law of addition of probabilities: If A and B are any two events associated

with a random experiment, then

P (A ∪ B) = P (A) + P (B) − P (A ∩ B)

P (A ∪ B ∪ C) = P (A) + P (B) + P (C) − P (A ∩ B) − P (A ∩ C) − P (B ∩ C)

+ P (A ∩ B ∩ C).

167
Shiv Datt Kumar Statistics and Probability

8.2.4 Law of Total Probability (LOTP)

Let B1 , B2 , B3 , ...Bn be a partition of the sample space (i.e., they are disjoint
and their union is the entire sample space). Then

P (A) = P (A|B1 )P (B1 ) + P (A|B2 )P (B2 ) + · · · + P (A|Bn )P (Bn )

P (A) = P (A ∩ B1 ) + P (A ∩ B2 ) + · · · + P (A ∩ Bn )

For LOTP with extra conditioning, just add in another event C!

P (A|C) = P (A|B1 , C)P (B1 |C) + · · · + P (A|Bn , C)P (Bn |C)

P (A|C) = P (A ∩ B1 |C) + P (A ∩ B2 |C) + · · · + P (A ∩ Bn |C)

Special case of LOTP with B and B c as partition:

P (A) = P (A|B)P (B) + P (A|B c )P (B c )

P (A) = P (A ∩ B) + P (A ∩ B c )

Probability Mass Function (PMF) The PMF satisfies

X
pX (x) ≥ 0 and pX (x) = 1
x

8.2.5 Expected Value

Mathematical Expectation: If p is the probability that a person will re-

ceive a sum of money S. then the mathematical expectation or simply

168
Shiv Datt Kumar Statistics and Probability

expectation is defined as pS.

Expectation or Mean, expected value, or average is a weighted average of

the possible outcomes of our random variable.

Mathematically, if x1 , x2 , x3 , . . . are all of the distinct possible values that

X can take with probability p1 , p2 , . . . , pn respectively, then mathematical
expectation of X is

P P
E(X) = xi P (X = xi ) = p i xi
i i

Indicator Random Variable is a random variable that takes value 1 or 0.

It is always an indicator of some event: if the event occurs, the indicator
is 1; otherwise it is 0. They are useful for many problems about counting
how many events of some kind occur. Write


1 if A occurs,

IA =

0 if A does not occur.


Note that IA2 = IA , IA IB = IA∩B , and IA∪B = IA + IB − IA IB .

Variance and Standard Deviation

var(X) = E (X − µ(X))2 = E(X 2 ) − (E(X))2

p
SD(X) = var(X)

169
Shiv Datt Kumar Statistics and Probability

8.3 Probability distributions:

The distribution obtained by taking the possible values of random variable to-
gether with their respective probabilities is called the probability distribution.

Random Variable: A random variable X is function from the sample space

S to real numbers R.

Example 8.3.1. For sample space of tossing of two coins S = {HH, T H, HT, T T },
X : S → R, for the number of heads X(HH) = 2, X(T H) = X(HT ) = 1,
X(T T ) = 0.

If a random variable X takes finite or countably infinite number of values,

then X is called a discrete random variable i.e. a real valued function
X defined on a discrete sample space S is called a discrete random variable.
For example, sample space of tossing of two coins S = {HH, T H, HT, T T },
number of raining days in year, number of cars [passing through a lane.

Probability function of a discrete random variable

Let X be a discrete random variable which takes values x1 , x2 , . . . , xn . and

let P (X = x) = p, Then pi is called the probability function if it satisfies the
following conditions:

(i) pi ≥ 0,
P
(ii) pi = 1.

The collection of pairs (xi , pi ), i = 1, 2, 3 . . . , n

X x1 x2 . . . xn

P (X = xi ) p1 p2 . . . pn

170
Shiv Datt Kumar Statistics and Probability

is called discrete probability distribution of the discrete random variable X.

Continuous random variable: A random variable, if it can take any value

in a given interval or intervals is called continuous random variable. For e.g.
1
depth of grounded water, life span of electric bulb, y = sin x, (0 ≤ x ≤ π).
2
Continuous probability distribution

A probability function if it satisfies the following conditions:

(i) f (x) ≥ 0, ∀x∈R

R∞
(ii) −∞
f (x) dx = 1.

is called continuous probability distribution function.

If the function is continuous, then

R ∞
Mean = x f (x) dx
−∞
R ∞
Variance = (x − x̄)2 f (x) dx, where x̄ = mean.
−∞

Example 8.3.2. A function is defined by






 0 if x < 2,


f (x) = 2x + 3 if 2 ≤ x ≤ 4


 18


0
 if x > 4.

Show that it is a probability distribution function.

Solution: If f (x) is probability distribution function, then f (x) ≥ 0 and
R ∞
f (x) dx = 1.
−∞

171
Shiv Datt Kumar Statistics and Probability

8.3.1 Binomial distribution

It is a discrete distribution. Let A be an event or trial associated with the

random experiment E. Let probability of success of A = P (A) = p and prob-
ability of failure of A = 1 − p = q. Let X be the random variable with the
number of times that event A has occurred out of n trials. A random variable
X is said to follow the binomial distribution if the probability mass function
is given by
n
P (X = r) = p(r) = Cr pr q n−r , r = 0, 1, 2, . . . , n, and q = 1 − p.

Remark 8.3.1. 1. Assignment of probability in binomial distribution is

P (X = r) = nr=0 n Cr pr q n−r = (p + q)n = 1.

P

2. If the experiment is repeated N times, then the frequency function of

n
binomial distribution is defined by f (r) = N p(r) = N Cr pr q n−r ,

Mean of binomial distribution

Pn Pn n r n−r
Pn−1 n
E(x) = r=0 xr pr = r=0 r Cr p q = np r=1 Cr pr−1 q (n−1)−(r−1) =
n(p + q)n−1 = np

Variance of binomial distribution

E(x2 ) = nr=0 x2r pr = nr=0 r2 pr = nr=0 (r(r − 1) + r) pr = nr=0 r(r − 1) pr +
P P P P
Pn 2
r=0 r pr = n(n − 1)p + np.

Variance = E(x2 ) − [E(x)]2 = n(n − 1)p2 + np − (np)2

= np − np2 = np(1 − p) = npq.
√ √
Standard Deviation = variance = npq

Example 8.3.3. Two players A and B play tennis game. Their chances of
winning game are in the ratio of 3 : 2 respectively. Find A0 s chance of winning
game atleast twice out four game played.

172
Shiv Datt Kumar Statistics and Probability

Solution: Let X = random variable denoting the number of success,

p = probability that A win a game = 3/5, q = 1 − p = 2/5.
4
P (X = r) = p(r) = Cr (3/5)r (2/5)4−r , for r = 0, 1, 2, 3, 4.
Hence p(X ≥ 2) = 1 − P (X < 2) = 1 − [P (X = 0) + P (X = 1)]
513
= 1 − [4 C0 (3/5)0 (2/5)4 + 4 C1 (3/5)(2/5)3 ] = .
625

Example 8.3.4. An irregular six faced dice is thrown 12 times. The expecta-
tion that it will give six even number is twice the expectation that it will give 5
even number. If 1000 sets, each of exactly 12 trials are made. How many sets
are expected not to give any even number ?
Solution: Let X = number of even number obtained in 12 throws.
p = probability of getting an even number It is given that
P (X = 6) = 2 P (X = 5) or 12 C6 p6 q 6 = 2 12 C5 p5 q 7
7 12 7
p = 2q = 2(1 − p) ⇒ p = and q = .
6 19 19
P (probability of getting no even number) =P (X = 0) = q 12
 12
7
Number of sets having no even number = 1000 × P (X = 0) = 1000 ×
19
= 0.0063.

8.3.2 Poisson distribution

It a particular limiting case of binomial distribution when p or q is very small

and n is very large. It is a discrete distribution. Poisson distribution is given
mr e−m
by P (r) = , where m = mean of the distribution = np.
r!
In binomial distribution,
n
P (r) = Cr pr q n−r = n
Cr pr (1 − p)n−r , r = 0, 1, 2, . . . , n.
Here m = np ⇒ p = m/n.

173
Shiv Datt Kumar Statistics and Probability

n n(n − 1)(n − 2) · · · (n − (r − 1)) r n−r

P (r) = Cr pr q n−r = pq
r!
n(n − 1)(n − 2) · · · (n − (r − 1)) r
= p (1 − p)n−r
  r!
n r − 1  m r 
   
n n 1 n 2 m n
− − ··· − 1−
n n n n n n n n n
=  m r .
r! 1 −
n 
r−1
   
1 2  m
1− 1− ··· 1 − mr 1 −
n n n n
=  m  r .
r! 1 −
n
Taking limit n → ∞,
mr e−m
P (r) = ,
r!

Properties:
P∞ mr e−m P∞ mr
 
1. r=0 = em r=0 = e−m em = 1
r! r!
Px m r
 
Px −m
2. Distribution function: f (x) = P [X ≤ x] = r=0 P [X = r] = e r=0 ,
x!
x = 0, 1, 2, 3, . . . ,.

mx e−m
 
P∞ P∞
3. Mean = x=0 x P [X = x] = x=0 x
x!
2
√ f r2
P P
f r
4. Variance = σ 2 = m, Standard deviation = m= P − P .
f f

5. Mode = [m] (integral part of m) .

m
6. Recurrence formula = P (r + 1) = P (r).
r+1

7. Moment (quantitative measure of shape of a set of points) generating

t
function = em(e −1)

Example 8.3.5. Let the probability of an individual coal miner being killed in
1
a mine accident during a year is . Use appropriate statistical distribution
2400
174
Shiv Datt Kumar Statistics and Probability

to calculate the probability that in a mine employing 300 miners, there will be
atleast one fatal accident.
1 300 1
Solution: Let p = , n = 200, m = np = = .
2400 2400 8
mr e−m
Now P (r) = ,
r!
P (probability of atleast one killed) =P (1 or 2 or · · · 300)

=P (1) + P (2) + · · · + P (300) = 1 − P (0) = 1 − e−1/8

Example 8.3.6. In a certain factory of motor tyres, there is a chance of 1

tyre in every 500 tyre to be defective. Tyres are supplied in a lots of 10 using
Poisson distribution. Calculate approximate number of lots in a consignment
of 10, 000 lot containing no defective, one defective, and two defective tyres.
1 1
Solution: Let p = , n = 10, m = np = = 0.02. Now probability in
500 50
mr e−m
Poisson distribution is given by P (r) = ,
r!
(0.02)0 e−0.02
P (0) = = e−0.02 = 0.9802.
0!
The number of lots containing no defective = 10, 000 × 0.9802
(0.02)1 e−0.02
P (1) = = e−0.02 × 0.02 = 0.9802 × 0.02 = 0.019604
1!
The number of lots containing one defective =196.
(0.02)2 e−0.02
P (2) = = e−0.02 × 0.02 = 0.9802 × 0.02 = 0.019604
2!
The number of lots containing two defective =10, 000 × 0.00019604 = 2 lots.

8.3.3 Normal (Gaussian) distribution

It is a continuous distribution. It is most frequently used distribution to model

random phenomenon. Originally it was defined as limiting case of binomial dis-
tribution when p and q are very small and n is very large. Normal distribution

175
Shiv Datt Kumar Statistics and Probability

has two parameters µ, σ, where µ = Mean, σ= standard deviation.

PDF of normal distribution is given by

(x − µ)2
1 −
f (x) = P [X = x] = √ e 2σ 2 , −∞ < x, µ < ∞, σ > 0——(1)
σ 2π
R x2
(x − µ)2
1 −
P (X1 < x < x2 ) = √ e 2σ 2 dx.
x1 σ 2π
z2
x−µ 1 −
On substitution, z = in (1) we get f (z) = √ e 2 − − − − − −(2).
σ 2π
Here µ = 0, s.d. = 1. Equation (2) is known as standard form of normal
distribution. f (x) in (1) is symmetrical about the line x = µ, so we have
expectation E[x] = µ.

Relation between binomial distribution and normal distribution: If

N is large and if neither p nor q are too close to zero, the binomial can be
X − Np
closely distributed with standardized variable given by z = √ .
N pq
Approximation becomes better with increasing N and in the limiting case,
it is exact.
Properties of Normal Distribution:

176
Shiv Datt Kumar Statistics and Probability

1. Mean = µ

2. Variance = σ 2

3. Standard deviation = σ.

p
4. Mean deviation= σ 2/π

8.4 Markov and Chebyshev inequalities

Suppose distribution of random variable X is unknown but its expected value

µ is known, we may want to ask how likely it is for X to be ‘far’ from mean µ.
This gives some idea of the spread of the distribution.

Proposition 8.4.1. (Markov’s Inequality) Let X be a random variable that takes

only non-negative values. Then for any positive real number r,

E(X)
P (X ≥ r) ≤ , provided E(X) exists.
r

Proof. We prove it for discrete RVs, for continuous RVs, sums are replaced by
integrals.
P
By definition, E(X) = xP (X = x)
x
We split this sum into two pieces, depending on whether or not x ≥ r.
P P
E(X) = xP (X = x) + xP (X = x)
x≥r x<r
P
= r P (X = x) + 0 (since in the first sum we assumed x ≥ r.
r≥r
P
=r P (X = x) .
x≥r
P
=r P (X ≥ r).
x≥r

177
Shiv Datt Kumar Statistics and Probability

Remark 8.4.1. Markov’s inequality tells us that as long as X does not take
negative values, the probability that X is twice as large as its expected value
1
is at most , which we can see by setting r = 2E(X). More generally, the
2
probability that a random variable is at least k times its expected value is at
1
most . Notice that the only things we assumed about this random variable
k
are that it can not be negative and has finite mean. Knowledge of its variance
or it’s probability distribution, in general.

Example 8.4.1. A coin is weighted so that its probability of landing on heads

is 20% Suppose the coin is flipped 20 times. Find a bound for the probability it
lands on heads at least 16 times.
1
It is a binomial distribution with n = 20 and = . It’s expected value is 4.
5
E(X) 1
Markov’s inequality tells us that P (X ≥ 16) ≤ = .
16 4
Comparing this to the actual probability that this happens: P (X ≥ 16) =
20
   k  20−k
20 1 4
= 1.38 × 10−8 .
P
.
k=16 k 5 5
So it seems like this is not a very good estimate. We’ll see later that this
distribution (at least, for large n) is close to normal, and Markov’s inequality
doesn’t get close to the true value for such “compact” distributions.

Lemma 8.4.2. (Chebyshev’s Inequality) Let X be any random variable with

finite expected value and variance. Then for every positive real number a,
V ar(X)
P (|X − E(X)| ≥ a) ≤
a2
Proof. Let Y = (X − E(X))2 . Then Y is non-negative valued random variable
with expected value E(Y ) = E((X − E(X))2 = E(X 2 ) − (E(X))2 = V ar(X).
By Markov inequality
E(X) |V ar(X)|
P (Y ≥ r2 ) ≤ =
r r2
178
Shiv Datt Kumar Statistics and Probability

Notice that the event Y ≥ r2 ⇒ (X − E(X))2 ≥ r2 is same as |X − E(X)| ≥ r,

so we conclude that
|V ar(X)|
|P (X − E(X)| ≥ r) ≤ .
r2

Remark 8.4.2. Chebyshev’s Inequality gives a bound on the probability that

X is far from it’s expected value.

Example 8.4.2. Suppose a fair coin is flipped 100 times. Find a bound on the
probability that the number of times the coin lands on heads is at least 60 or at
most 40.

Proof. Let X be the number of times the coin lands on heads. We know X has
a binomial distribution with expected value 50 and variance 1000.50.5 = 25.
By Chebyshev, we have
25 1
P (X < 40 ∪ X > 60) = P (|X − µ| ≥ 10) ≤ 2
= . The actual
10 4
probability of this happening is close to 5%..

Example 4 (Markov’s Inequality is Tight). Consider a random variable X

that takes the P (X = 1) = P (X = −1) = 0.5. Its mean is zero, variance is 1
and P (|X| ≥ 1) = 1. For this random variable Chebyshev is tight (hold with
equality).
V ar(X)
P (|X| ≥ 1) ≤ = 1.
12

179