Professional Documents
Culture Documents
Integral Equations
D.C. Sharma
Head, Department of Mathematics
Dean, School of Mathematics, Statistics and Computational Sciences
Central University of Rajasthan, Kishangarh, Ajmer
M.C. Goyal
Dean (Academics), Professor and Head of Department of Mathematics
Rajasthan College of Engineering for Women, Jaipur
Delhi-110092
2017
INTEGRAL EQUATIONS
D.C. Sharma and M.C. Goyal
© 2017 by PHI Learning Private Limited, Delhi. All rights reserved. No part of this book may
be reproduced in any form, by mimeograph or any other means, without permission in writing
from the publisher.
ISBN-978-81-203-5280-3
The export rights of this book are vested solely with the publisher.
Published by Asoke K. Ghosh, PHI Learning Private Limited, Rimjhim House, 111, Patparganj
Industrial Estate, Delhi-110092 and Printed by Mudrak, 30-A, Patparganj, Delhi-110091.
Contents
Preface................................................................................................................ ix
Acknowledgements.............................................................................................. xi
1. Basic Concepts.............................................................1–11
1.1 Introduction 1
1.2 Abel’s Problem 1
1.3 Initial Value Problem and Boundary Value Problem 2
1.4 Integral Equation 3
1.5 Special Kinds of Kernels 4
1.6 Classification of Integral Equation 5
1.7 Iterated Kernels 6
1.8 Reciprocal Kernal or Resolvent Kernel 7
1.9 Eigenvalues and Eigenfunctions 7
1.10 Solution of an Integral Equation 8
Exercise 1.1 11
by Using Laplace
D.C. Sharma
M.C. Goyal
ix
Acknowledgements
We wish to record our sincere thanks to the authors of many earlier standard
works, which have been consulted by us during the preparation of the book.
Our heartfelt thanks are due to Prof. G.C. Sharma and Prof. K.C. Gupta,
whose inspiration prompt us to take this project of writing on this topic.
We express our personal and sincere thanks to our senior colleagues and
students for their useful suggestions in preparing the blueprint and subject
matter of the book. We further reiterate that without fruitful discussion with
them, it would have been impossible to present this book.
We are also thankful to our family members for taking all the care during
long hours at the desk.
Finally, the authors are especially thankful to PHI Learning for its
cooperation in all aspects of the production of this book.
D.C. Sharma
M.C. Goyal
xi
Chapter 1
Basic Concepts
1.1 INTRODUCTION
Various physical problems in physics and other applied fileds culminate into
initial value problems or boundary value problems. Although it is equivalent to
frame the problem in the form of (ordinary and partial) differential equations
or in the form of integral equations, but it is preferred to choose the integral
form due to two main reasons. Firstly, the solution of integral equation is much
easier than the original boundary value or the initial value problems. The second
reason lies in the fact that integral equations are better suited to approximate
methods than differential equations. Moreover, integral equations develop as
representation formulae for the solution of differential equations. We will find
in the forthcoming chapters that differential equations can be replaced by an
integral equation with the help of initial and boundary conditions. As a result,
each solution of the integral equation satisfies the boundary conditions itself.
P ds
∴ T
O 2 g( x – x )
Fig. 1.1 Particle falling under gravity along the curve PQO.
Now, if the shape of curve is given, then s, and hence, ds can be expressed
in terms of x. We take
ds = u(x)dx, and then
x u(x ) dx
T
0 2 g( x x )
Here, if the curve is such that the time of descent T is a function of x,
say, f(x), then the above relation shapes as
x 1
f ( x) u(x ) dx
0 2 g( x – x )
This leads to find the unknown function u(x) and we get Abel’s integral
equation.
During the analysis, it is found that an initial value problem is always
converted to a Volterra integral equation, while a boundary value problem
is always converted to a Fredholm integral equation. Before we take the
classification of integral equations, it is worth to be familiar with the initial
value problems and the boundary value problems.
v( x ) u( x ) f ( x ) l K ( x, x ) u(x ) dx (1.2)
a
where the upper limit of the integral may be variable or constant. The functions
v, f, K are known functions and u is to be determined, l is a non-zero
parameter, which may be real or complex.
The importance of keeping l separate from K lies in the fact that it plays
an essential role in the theoretical arguments for the problem under context.
Kinds of linear integral equations
1. First kind, if v = 0, Eq. (1.2) reduces to
f ( x ) l K ( x, x ) u(x ) dx
0
a
f ( x ) l K ( x, x ) u(x ) dx
u( x )
a
K ( x, x ) K (x , x )
where the bar represents the complex conjugate. A real kernel is symmetric
x
if K(x, x) = K(x, x). For instance exx is symmetric, while tan 1 is not a
x
symmetric kernel.
n
2. Separable or degenerate kernel: If K ( x, x ) gi ( x ). hi (x ) means K
i 1
has been expressed as the sum of a finite number of terms, each of which is
the product of function of x only and x only, then such a kernel is called
separable or degenerate kernel. Obviously, gi(x) and hi(x) are linearly
independent (or else some of the terms will combine, and consequently, the
number of terms will reduce). A degenerate kernel has a finite number of
characteristic values.
3. Difference kernel: A kernel of the form K(x – x) is called difference
kernel.
Basic Concepts 5
where a and b both are constants is called Fredholm integral equation of the
third kind. Here, f(x), v(x) and K(x, x) are known functions, u(x) is unknown
function and l is real or complex parameter.
Now, if in Eq. (1.3), we set v(x) = 0, then we get
b
f ( x ) l K ( x, x ) u(x ) dx
0 (1.4)
a
x
( x ) f ( x ) l K ( x, x ) . u(x )dx (1.9)
u
a
then it is called Volterra integral equation of the second kind.
Further, if in Eq. (1.9), f(x) = 0, then it becomes
x
u( x ) l K ( x, x ) . u(x )dx (1.10)
a
and is called homogeneous Volterra integral equation of the second kind.
Singular integral equation
An integral equation is a singular integral equation, if either
1. One or both the limits of integration are infinite, or
2. The kernal becomes infinite at one or more points within the range of
integration. For instance,
x x
( x) f ( x) l
u e u(x )dx
x 1
and u( x ) = ∫ ⋅ u(ξ )dξ, 0 a 1
0 ( x − ξ )α
are both singular integral equations.
Convolution integral equations
If the kernel K(x, x) is a function of one variable and is of the type K(x, x) =
K(x – x), then the integral equations, i.e., Eqs. (1.5) and (1.9) take the shape
b
f ( x ) l K ( x x ) . u(x )dx (1.11)
u( x )
a
x
and f ( x ) l K ( x x ) . u(x )dx (1.12)
u( x )
a
x
and ( x ) f ( x ) l S ( x, x; l ) f (x )dx (1.17)
u
a
respectively, then R(x, x; l) and S(x, x; l) are called the reciprocal or resolvent
kernel of the respective integral equation.
If G(x) and H(x) are constant functions, then Eq. (1.20) reduces to
d H H F
dx G
F ( x, x )dx
G
dx
x (1.21)
2
EXAMPLE 1.1: Show that the function u( x ) e x 2 x is a solution of
1
3
Fredholm integral equation u( x ) 2 e x x u(x ) dx 2 xe x .
0
2
x
Solution: We have u( x ) e 2 x
3
2
u(x ) ex 2x ,
3
Then, the L.H.S. of the given integral equation
2 1 2
e x 2 x 2 e x x ex 2x dx
3 0 3
1
2 2x
e 2 x 2 e x x 2
x
3 3 0
2 2
e x 2 x 2e x 1 2 xe
x
3 3
Basic Concepts 9
\ L.H.S. = R.H.S.
Hence proved.
EXAMPLE 1.2: Show that the function u(x) = xex is a solution of the
Volterra integral equation.
x
sin x 2 cos( x x ) . u(x )dx
u( x )
0
Solution: We have u(x) = xex
\ u(x ) x ex
x
Then, the R.H.S. of given integral equation sin x 2 cos( x x ) x ex dx
x 0
sin x 2 x ex {cos(x x )}dx
0
sin x.cos x , 0 x x
where, K ( x, x ) (ii)
cos x sin x , x x p
Solution: We have u(x) = cos2x, which means u(x) = cos2x, and thus, the
R.H.S. of Eq. (i) shapes as [we use kernel as defined by Eq. (ii)]
10 Integral Equations
xx p
cos x 3 cos x sin x cos 2x dx 3 sin x cos x cos 2x dx
x 0x x
x p
cos x 3cos x (cos 2x sin x ) dx 3sin x (cos 2x cos x ) dx
0 x
3 x 3 p
cos x cos x (sin 3x sin x ) dx sin x (cos3x cos x ) dx
2 0 2 x
x p
3 1 3 1
cos x cos x cos3x cos x sin x sin 3x sin x
2 3 0 2 3 x
3 1 1
cos x cos x cos3 x cos x 1
2 3 3
3 1 1
sin x sin 3p sin p sin 3 x sin x
2 3 3
1 3
cos x (cos3 x cos x sin 3 x sin x ) (cos2 x sin 2 x ) cos x
2 2
1 3
= − cos 2 x + cos 2 x = cos 2 x = u( x )
2 2
Hence, u(x) = cos2x is a solution of the given integral equation.
px
EXAMPLE 1.4: Show that the function u( x ) sin is a solution of the
2
p2 1 x
Fredholm integral equation u( x )
4 0 K ( x, x )u(x ) dx 2
, where the kernel
is of the form
x (2 x )/2, 0 x x ,
K ( x, x )
x (2 x )/2, x x 1
px px
Solution: Since u( x ) sin , we have u(x ) sin
2 2
px p2 x
K ( x, x ) u(x ) dx K ( x, x ) u(x ) dx
1
sin
2
4 0 x
Now, substituting for kernel K(x, x) and u(x), we get
p x p 2 x 1 px 11 px
sin
2
4 0 2
x (2 x ) sin dx x (2 x ) sin dx
2 x 2 2
px p2 x px p2 1 px
sin (2 x ) x sin dx x (2 x )sin dx
2 8 0 2 8 x 2
p 2 x cos(px /2)
1
cos(px /2) 1
(2 x )
x ( 1) dx
8 p /2 x p /2
p x p 2 (2 x ) 2 x p x 2 2 px
x
sin cos sin
2 8 p 2 p p 2 0
p2x 2 p x 2 2 px
1
(2 x ) cos sin
8 p 2 p p 2 x
p x p 2 (2 x ) 2 x px 4 px
sin cos 2 sin
2 8 p 2 p 2
p2x 2 px 4 4 px
(2 x ) cos 2 1 2 sin
8 p 2 p p 2
p x (2 x ) x x x
sin 1
2 2 2 2 2
x
Thus, u( x ) is a solution of the given integral equation.
2
EXERCISE 1.1
Verify that the given functions are solutions of the corresponding integral
equations:
1 x x
(1 x 2 ) 3/2 , u( x )
1. u( x ) 2
u(x ) dx
1 x 0 1 x2
1
1, u( x ) x (e xx 1)u(x ) dx
2. u( x ) ex x
0
1 x u(x )
3. u( x ) , dx 1
p x 0 x x
1
4. u( x ) x , u( x ) K ( x, x ) u(x ) dx
0
x (2 x )
2 , 0 x x
x
x (4 x 3/2 7), K ( x, x )
15 x (2 x ) , x x 1
2
x3 x
5. g( x ) x ; g( x ) x 0 sinh( x t ) g(t ) dt
6
Chapter 2
Applications to Ordinary
Differential Equations
2.1 INTRODUCTION
The quest for establishing a representation formula to replace an ordinary
differential equation (with initial value problem or boundary value problem)
always leads to an integral equation. More specifically, it is found that an initial
value problem is converted to a Volterra integral equation, while a boundary
value problem is converted to a Fredholm integral equation. Finally, it eases
our work of finding the solution of integral equation thus obtained.
Recalling once more, in an initial value problem, the boundary conditions
are for the same value of independent variable, while in the case of boundary
value problem, the boundary conditions are for different values of independent
variable.
d n 1 y x
or
dx n 1
a u( x) dx qn 1 [using Eq. (2.2)] (2.4)
d n 1 y x
or
dx n 1 a
u(t ) dt qn 1 (2.5)
Now, using Eq. (1.19) (i) for the double integral, we get
d n 2 y x
dx n 2
a ( x t ) u(t ) dt ( x a) qn 1 qn 2 (2.7)
Now, integrating Eq. (2.6) with respect to x from a to x
x
d n 3 y x x x
a u( x)(dx) qn 1 ( x a)dx qn 2 dx
3
n 3
dx a
a a
x
d n 3 y x ( x a) 2
a
( n 3) 3 x
or y ( a) u ( x )( dx ) q n 1 qn 2 [ x ]a
dx n 3 2 a
d n 3 y x ( x a)2
a
3
or u ( x )( dx ) q n 1 qn 2 ( x a) qn 3 (2.8)
dx n 3 2!
d n 3 y x ( x a)2
a
3
or u(t )( dt ) q n 1 ( x a)qn 2 qn 3
dx n 3 2!
Now, applying Eq. (1.19) for the triple integral, we get
d n 3 y x ( x t )2 ( x a)2 ( x a)
dx n 3
a 2!
u(t )dt qn 1
2!
qn 2
1!
qn 3 (2.9)
14 Integral Equations
dny d n 1 y d n 2 y dy
n
a1 ( x ) n 1
a 2 ( x) n 2
a n 1 ( x ) a n ( x) y
dx dx dx dx
u( x ) [qn 1a1 ( x ) qn 2 ( x a)qn 1 a 2 ( x )
( x a) n 1
q0 q1 ( x a) qn 1 a n ( x )]
(n 1)!
x ( x t )2 ( x t )n 1
a1 ( x ) ( x t )a 2 ( x ) a3 ( x) a n ( x ) u(t )dt (2.12)
a
2! (n 1)!
Now for L.H.S., we use Eq. (2.1) and let
y ( x ) qn 1a1 ( x ) qn 2 ( x a)qn 2 a 2 ( x )
( x a) n 1
q0 q1 ( x a) qn 1 a n ( x)
(n 1)!
( x t )n 1
and K ( x, t ) a1 ( x ) ( x t )a 2 ( x ) a n ( x)
(n 1)!
so that Eq. (2.12) reduces to
x
φ ( x ) u( x ) y ( x ) K ( x, t ) u(t )dt (2.13)
a
Again, let φ ( x ) y ( x )
f ( x ) , then
x
f ( x ) K ( x, t ) u(t )dt (2.14)
u( x )
a
which is Volterra integral equation of the second kind. Thus, we have found a
relation between a linear differential equation (Eq. 2.1) and a Volterra integral
equation, which establishes that an initial value problem is converted to a
Volterra integral equation.
dy x
or u(t )dt (v)
dx 0
Now, integrating Eq. (iv) with respect to x and using Eq. [ii(a)], we obtain
x x
0 u( x)(dx) 0 u(t )(dt )
2 2
y( x )
y(0)
y( x )
which on applying Eq. (1.19) gives
x
y
( x) 0 ( x t )u(t )dt (vi)
d2 y
Finally, putting from Eq. (iii) and y from Eq. (vi) into Eq. (i), we get
dx 2
x
u( x ) x ( x t )u(t )dt
x
1 1 ( x t )u(t )dt
or u( x )
0 0
Solution:
The given equation is
d3y
2 xy
0 (i)
dx 3
d3y
Let u( x ) (ii)
dx 3
Integrating Eq. (ii) with respect to x from 0 to x, we get
x
d2 y x x
dx 0
2 0 u( x ) dx or y ( x ) y (0) 0
u( x ) dx
x
or y ( x ) 1 u( x ) dx (iii)
0
1 x2 x ( x t )2
or y( x ) x u(t )dt [using Eq. (1.19) (v)
2 2 0 2!
Now, putting the values from Eqs. (ii) and (v) in Eq. (i), we get
1 x2 x ( x t )2
u( x ) 2 x x 2 x u(t )dt 0
2 2 0 2!
x
or u( x ) x ( x 1)2 x ( x t )2 u(t )dt, which is the required equation.
0
Note: In this example, we have not changed the variable of integration from
x to t for y¢¢(x) and y¢(x), since this change was not required.
dy x
or 1 u( x )dx (iii)
dx 0
dy x
or 1 u(t )dt (iv)
dx 0
x
or y( x ) 1 x ( x t )u(t )dt (v)
0
dy d2 y
Putting the values of y, and from Eqs. (v), (iv) and (ii), respectively,
dx dx 2 x
in Eq. (i), we get u( x ) x sin x e x (1 x ) [sin x e x ( x t )]u(t ) dt , which
0
is non-homogeneous Volterra integral equation of second kind.
EXERCISE 2.1
Answers:
x
u( x ) 3 (5 x 3t ) u(t )dt
(a)
0
x
u( x ) 1 (2 x t ) u(t )dt
(b)
0
x 1
) xe x 1 x ( x 2 1) x ( x 2 x )( x t )2 u(t )dt
u( x
(c) 0 2
x
u( x ) F ( x ) c1a1 ( x ) (c0 c1 x )a 2 ( x ) a1 ( x ) a 2 ( x ).( x t ) u(t )dt
(d) 0
x2 x
or y ( x ) 1 sin x y( x ) (e x cos x ) y( x ) dx (iii)
2 0
20 Integral Equations
x3 x
or y( x ) 1 x [sin t ( x t )(et cos t )]y(t ) dt
6 0
x3 x
or ) x 1 [sin t ( x t )(et cos t )]y(t ) dt (iv)
y( x
6 0
which is a Volterra integral equation of the second kind.
Now, we differentiate Eq. (iv) with respect to x and obtain
x2 d x
y ( x )
2
1
dx 0
{sin t ( x t )(et cos t )}y(t ) dt
or y ( x ) sin x y ( x ) e x y( x )
x (vi)
which is same as the given differential equation. To get the initial conditions,
we put x = 0 in Eq. (iv) and (v), and obtain
y(0) + 1 and y¢(0) = –1 (vii)
Thus Eqs. (vi) and (vii) together provide the original initial value problem.
Applications to Ordinary Differential Equations 21
EXERCISE 2.2
1 1 t 2 (1 x ), t x
3. y( x
) x (1 x ) K ( x, t ) y(t ) dt; K ( x, t
)
2 0
xt (1 t ), t x
x3 x 1 t , t x
4. y( x ) K ( x, t ) y(t ) dt; K ( x, t )
6 2 0
x, t x
Solution:
The given conditions are
u(0) = 0 [i(a)]
u(l) = 0 [i(b)]
The given differential equation is
u(x) = –lu(x)
Integrating it with respect to x from 0 to x, we get
x x
0 u "( x) dx l u( x ) dx
0
x
or l u( x ) dx
u ( x ) u (0)
0
x
Let u¢(0) = c so that u ( x ) c l u( x ) dx.
0
Again, on integration, we get
x x x x
0 u ( x) dx
c dx l u( x )(dx ) cx l u(t )(dt )2
2
0 0 0
x
or u( x ) u(0) cx l ( x t ) u(t ) dt
0
l l
l 0
c (l t )u(t ) dt
l l x (l t ) x
u( x ) 0 l
u(t ) dt l ( x t ) u(t ) dt
0
x l x (l t ) l l x (l t ) x
or u( x ) 0 l
u(t ) dt
x l
u(t ) dt l( x t ) u(t ) dt
0
x x (l t ) l x (l t )
u( x ) l
or ( x t ) u(t ) dt l u(t ) dt
0 l x l
x t (l x ) l x (l t )
or u( x ) l u(t ) dt u(t ) dt
0 l x l
Applications to Ordinary Differential Equations 23
l
or u( x ) l K ( x, t ) u(t ) dt (iii)
0
t (l x )
l , if 0 t x
or where, K ( x, t ) (iv)
x (l t ) , if x t l
l
Equation (iii) is the required integral equation, whose kernel K(x, t) is defined
by Eq. (iv).
0 y ( x) dx 0 y( x) dx
0 x dx
x2 x
or y ( x ) y 0) y( x ) dx
2 0
x3 x
y( x ) cx ( x t ) y(t ) dt [iv(a)]
6 0
1 1
0 c y(t ) dt
2 0
[Here, we apply Eq. (1.20)]
1 1
c y(t ) dt
2 0
and then, Eq. [iv(a)] is
3
1 1 x x
y( x ) x y(t ) dt ( x t ) y(t ) dt [iv(b)]
2 0 6 0
y( x )
x x3
2 6
0
x
x
1
x
x.y(t ) dt x. y(t ) dt ( x t ) y(t ) dt
0
3
x 3x x 1
y (
x) t y(t )dt x y(t ) dt
6 0 x
1 3 1
or y ( x ) ( x 3 x ) K ( x, t ) y(t ) dt (v)
6 0
x, x t
where, K ( x, t ) (vi)
t , x t
Converse: We take the integral equation, i.e., Eq. [iv(b)].
x x3 1 x
y( x )
x y(t ) dt ( x t ) y(t ) dt (vii)
2 6 0 0
1 x2 1 x
y ( x )
y(t ) dt y(t ) dt (viii)
2 2 0 0
x
y ( x ) x 0 {y(t )} dt y( x )
0 x
y ′′( x ) = x − 0 − y( x )
or y¢¢(x) + y = x (ix)
Also from Eqs. (vii) and (viii)
y(0) = 0 and y¢(1) = 0
So, Eq. (ix) is the original differential equation with the boundary conditions.
Applications to Ordinary Differential Equations 25
0 y ( x) dx 0 dx 0 x y( x) dx
x
x x y( x ) dx
y ( x ) y (0)
0
x x 1 x
y( x ) (1 x ) xt (1 t ) y(t ) dt xt (1 t ) y(t ) dt t ( x t ) y(t ) dt
2 0 x 0
x x 1
y( x ) (1 x ) t 2 (1 x ) y(t ) dt xt (1 t ) y(t ) dt
2 0 x
26 Integral Equations
x x
or y( x ) (1 x ) K ( x, t ) y(t ) dt (v)
2 0
t 2 (1 x ), when t x
where, K ( x, t ) (vi)
xt (1 t ), when t x
Equation (v) is the required integral equation, whose kernel K(x, t) is defined
by Eq. (vi).
1 (1 x ) x, 0 x x
u
( x) 0 l K ( x, x ) u(x ) dx , where
K ( x, x )
(1 x )x , x x 1
,
then prove that m(x) is also a solution of the boundary value problem
d 2u
l 0, u(1)
u 0, u(0) 0
dx 2
Solution: Here, from the integral equation,
1
u( x ) l K ( x, x ) u(x ) dx (i)
0
(1 x ) x, 0 x x
where, K ( x, x ) (ii)
(1 x )x , x x 1
We have to find the differential equation satisfying u(x) with the corresponding
boundary conditions. The kernel may be followed by Figure. 2.1 given below:
u( x ) l K ( x, x ) u(x ) dx K ( x, x ) u(x ) dx
x 1
0 x
du d x d 1
dx dx
0
l (1 x )x u(x ) dx
dx x
l x (1 x ) u(x ) dx
du x dx d
dx
0 x l (1 x)x u(x )dx l (1 x) x u( x) dx 0 dx (0)
1 d dx
l x(1 x ) u(x ) dx 0 (1) l x(1 x) u( x)
x x dx dx
du x 1
or
dx 0 l (1) x u(x ) dx x l (1 x ) u(x ) dx (v)
Differentiating Eq. (v) with respect to x using Leibnitz’s rule, we get
d 2u x dx
{lx u(x )} dx l x u( x ) 0
dx 2
x
0 dx
1 dx
{l (1 x ) u(x )}dx 0 l (1 x ) u( x )
x x dx
d 2u d 2u
or
l x u( x ) l (1 x ) u( x ) 0 (vi)
lu
dx 2 dx 2
which is the required equation with boundary conditions given by Eq. [iv(a)]
and [iv(b)].
EXERCISE 2.3
Answer:
p
y( x ) l K ( x, t ) y(t )dt
0
(t 1)(p x 1)
,0t x
p
where, K ( x, t )
( x 1)(p t 1) , x t p
p
vvv
Chapter 3
Solution of Homogeneous
Fredholm Integral
Equations of the Second Kind
3.1 INTRODUCTION
In the previous chapters, we have learnt the basic terminology and suitability
of integral equations over differential equations, and then, the classification
of the integral equations. Although, there have been many developments in
the theory, the basic division as initial value problems into volterra integral
equations and boundary value problems into Fredholm integral equations is
a must to follow. As we approach to simplify these equations, it is found
convenient to focus the kernel, kind and the homogeneous nature of these
equations. In this chapter, we shall restrain ourselves only for the homogeneous
Fredholm integral equations of the second kind. The chapter begins with the
discussion of an essential part called eigenvalue, eigenfunction, and then
related theorems are explained.
Note:
1. If kernel K(x, t) is continuous for a x b, a t b, for the finite
values of a and b, then corresponding to every eigenvalue of l there
exists a finite number of linearly independent eigenfunctions; the number
of such functions is called index of the eigenvalue. Different eigenvalues
have different indices.
2. As seen above, along with f(x), cf(x) is also an eigenfunction for l0,
where c is an arbitrary constant (though not L.I.).
3. The number l = 0 is not taken as eigenvalue, since it provides trivial
solution.
4. A homogeneous Fredholm integral equation may not have eigenvalues
and eigenfunctions if the kernel is not symmetric.
EXAMPLE 3.1: Find the eigenvalue and eigenfunction of the homogeneous
1
integral equation g( x ) l 0 e g(t ) dt .
x t
l c 2t 1 l c 2
c
e (e 1)
2 0 2
l
which is c 1 (e2 1)
0
2
Now, for non-trivial solutions,
l 2
1 (e 1)
0
2
2
l 2
(iv)
e 1
n b
u( x ) l fi ( x ) gi (t )u(t )dt (3.6)
a
i 1
[By interchanging the order of integration and summation]
b
Let Ci g (t )u(t )dt , i
a i
1, 2, ..., n (3.7)
. . . .
. . . .
. . . .
b n b
a gn ( x)u( x) dx l
i 1
Ci gn ( x ) fi ( x ) dx [3.9(n)]
a
b
We define a ji a g j ( x) fi ( x)dx (i, j =1, 2, ..., n) (3.10)
Now using Eqs. (3.7) and (3.10), Eq. [3.9(a)] becomes
n
C1 l Cia1i
i 1
or
C1 l C1a11 C2a12 Cna1n
Similarly, we derive other equations from Eq. (3.9), and finally, obtain
following simultaneous linear equations to establish Ci, which are
The eigenvalues of Eq. (3.3) are obtained by D(l) = 0, and we get the
maximum n eigenvalues.
Here, kernal K(x, t) is symmetric and let f0(x) and f1(x) be eigenfunctions of
K(x, t) corresponding to eigenvalues l0 and l1(l0 ≠ l1). Then, we are required
to prove that f0(x) and f1(x) are orthogonal functions on the interval [a, b], i.e.,
b
a f0 ( x) f1 ( x)dx 0 (3.15)
Since f0(x) and f1(x) are eigenfunctions, by definition, these will satisfy
Eq. (3.14). Thus,
b
f0 ( x ) l0 K ( x, t )f0 (t ) dt (3.16)
a
and
b
f1 ( x ) l1 K ( x, t )f1 (t )dt (3.17)
a
b
l0 f0 (t )
a
a
b
K ( x, t ) f1 ( x )dx dt
34 Integral Equations
b
l0 f0 (t )
a a
b
K (t , x ) f1 ( x )dx dt (3.19)
a f0 ( x) f1 ( x) dx 0
This proves the proposition.
b b
a f0 ( x) f 0 ( x)dx l0 a f 0 ( x) a K ( x, t ) f0 (t )dt
b
dx
b
a f0 ( x) f 0 ( x)dx l0 a f0 (t )
b
K ( x, t ) f 0 ( x )dx dt a
b
By changing the order of integration, and now since K(x, t) = K(t, x), we have
b b
a f0 ( x) f 0 ( x)dx l0 a f0 (t ) a K (t, x) f 0 ( x)dx
b
dt (3.23)
Now, interchanging the variables (x t), Eq. [3.22(b)] takes the shape as
b
f 0 (t ) l 0 K (t , x ) f 0 ( x )dx
a
and then, Eq. (3.23) is
b b 1
a f0 ( x)f 0 ( x)dx l0 a f0 (t ) l 0 f 0 (t ) dt
b b
or l 0 f0 ( x ) f 0 ( x )dx l0 f0 ( x ) f 0 ( x )dx
a a
(Using the property of definite integrals)
l
b
or 0 l0 f0 ( x ) f 0 ( x )dx
0 (3.24)
a
Using Eq. (3.21), we get
b
2ib u2 v 2 dx
0
a
a u
2 b
v 2 dx 0
and thus, we conclude b = 0, which proves the proposition.
1
f ( x ) l t x x t f (t )dt
0
has no real eigenvalue and no eigenfunction.
Solution: The given equation is
1
f ( x ) l t x x t f (t )dt
0
1 1
or f ( x ) l x tf (t )dt l x t f (t )dt (i)
0 0
1
Let c1 tf (t )dt (ii)
0
1
and c2 t f (t )dt (iii)
0
36 Integral Equations
Thus,
f (t ) l c1 t l c2 t (iv)
0 t l c1
1
and then, by Eq. (ii), c1 t l c2 t dt
1 1
t 5/2 t3
c1 l c1 l c2
5/2 0 3 0
2l l
c1 1 c 0 (v)
5 3 2
Similarly, by Eq. (iv) and (iii)
1
c2 0 t l c1 t l c2 t dt
1 1
t2 t 5/2
c2 l c1 l c2
2 0 5 / 2 0
l 2l
c1 1 c2
0 (vi)
2 5
The system of Eqs. (v) and (vi) will have a non-zero solution if
2l l
1
5 3
D( l ) 0
l 2l
1
2 5
4l 2 l 2 l2
1 0 1 0 l i 150
25 6 150
which means that the given integral equation does not possess any real
eigenvalue.
c2 0 sin t{l c1 sin t l c2 cos t} dt
2p 2p
l c1 sin 2t l c cos 2t
c2 t 2
2 2 0 2 2 0
or c2 l c1p l c1p c2 0 (vi)
For non-zero solution of this system of linear equations,
1 lp
D( l ) 0 0 1 l 2p 2 0
lp 1
1
or l
p
1 1
Hence, the eigenvalues are given by l1 and l2 .
p p
Now, determining eigenfunction,
1
(a) For l , Eqs. (v) and (vi) are c1 – c2 = 0; hence, from Eq. (iv),
p
1 1
g(s) c1 sin s cos s c1
p p
c1
g(s) (sin s cos s)
p
c
or g(s) (sin s cos s) , taking 1 1
1 p
(b) For l , Eqs. (v) and (vi) are c1 + c2 = 0,
p
Hence, from Eq. (iv)
1 1
g(s) c1 sin s ( c1 ) cos s
p p
38 Integral Equations
c1
g(s) (sin s cos s)
p
c1
or
g(s) (sin s cos s) , taking 1
p
1
Finally, for the eigenvalue l , the eigenfunction is g(s) = sin s + cos s
1 p
and for eigenvalue l , the eigenfunction is g(s) = sin s – cos s.
p
EXAMPLE 3.5: Find the eigenvalues and eigenfunctions of the following
homogeneous integral equation:
p
( x ) l (cos2 x cos 2t cos3 x cos3 t ) f (t ) dt
f
0
p 1 cos 2t p1
or c1 1 l cos 2t dt l c2 (cos 5t cos t ) dt
0
0 2 0 2
p
p cos 2t 1 sin 5t sin t
or c1 1 l (1 cos 4t ) dt l c2
0
0 2 4 10 2 0
sin 2t t sin 4t
p
or c1 1 l l c2 0 0
4 4 16 0
lp
or c1 1 0 c2 0 (v)
4
Similarly, by Eqs. (iii) and (iv),
Solution of Homogeneous Fredholm Integral Equations of the Second Kind 39
p
c2 cos3 t (l c1 cos2 t l c2 cos3t )dt
0
p
0 cos
5
We find that t0
p p 1
0 cos 0
3
and t cos3t
dt (cos3t 3cos t ) cos3t dt
4
1 p 1 cos 6t 3 p
4 0 2
dt cos t cos3t dt
4 0
p
1 sin 6t 3 p
t (cos 4t cos 2t ) dt
8 6 0 8 0
p
p 3 sin 4t sin 2t p
8 8 4 2 0 8
p lp
\ c2 0 c1 l
c2 or 0c1 c2 1 0 (vi)
8 8
For non-zero solution of system of Eqs. (v) and (vi), D(l) = 0.
lp
1 0
4
0
lp
0 1
8
or lp lp
1 1 0
4 8
8 4
l ,
p p
which are the eigenvalues.
Now, determining eigenfunctions,
(a) For l 4/p ; by Eq. (v), c1 0 c2 0 0
4 p c2
0, c1 is arbitrary
and by Eq. (vi), c2 1 0
p 8
Hence, by Eq. (iv),
f(x) = lc1 cos2 x + 0
4
or f ( x) c1 cos2 x
p
4c
or f(x) = cos2 x taking 1 1
p
40 Integral Equations
p 8
(b) For l 8/p ; by Eq. (v), c1
1 4 p 0 c1 0
c1
0 and c2 is arbitrary
p 8
and by Eq. (vi), 0c1 c2 1 0
8 p
8 8
f ( x) 0 cos2 x c2 cos3 x
p p
8c2
f(x) = cos 3x taking 1
p
EXERCISE 3.1
2 1
g(s) l st g(t ) dt
3. 1 st
1
( x ) l (5 xt 3 4 x 2 t 3 xt ) f (t ) dt
4. f
1
p
5. g( x ) l 0 [cos x cos 2t cos3 x cos t ] g(t ) dt
2 3
Answers:
1. No eigenvalue
2.
3, 3, ( x 2 x 2 )
1
1 1
3. 17 265 , s 2 2732 , s 0.4399
2 s s
1 3x
4. ) x2
, f ( x
4 2
4 8 2
5.
, , g1 ( x ) cos x, g2 ( x ) cos3 x
p p
Chapter 4
4.1 INTRODUCTION
Like the previous chapter, this chapter is also devoted to Fredholm integral
equations of the second kind. But, now the equation is not homogeneous,
i.e., F(x) ≠ 0 in general. Furthermore, the nature of kernel is separable. This
chapter shows that the role of characteristic function with determinant D(l) is
of vital importance. All interrelated possible cases are included with suitable
examples.
b
We prescribe gi (t
)u(t )dt c
i, (i 1, 2, ..., n) (4.4)
a
which shapes Eq. (4.3) as below:
n
( x ) F ( x ) l ci fi ( x ) (4.5)
u
i 1
Now, for determining ci’s, we multiply Eq. (4.5) by g1(x), g2(x), ..., gn(x)
successively and integrate over the interval [a, b], and frame
b b b
∫a g1 ( x )u( x )dx = ∫a F ( x ) g1 ( x )dx + λ ∑ ci ∫a g1 ( x ) ⋅ fi ( x )dx [4.6(a)]
b b b
a g2 ( x)u( x)dx a F ( x)g2 ( x)dx l ci a g2 ( x) fi ( x)dx [4.6(b)]
………………………………………………………………
………………………………………………………………
………………………………………………………………
b b b
a gn ( x)u( x)dx a F ( x)gn ( x)dx l ci a gn ( x) fi ( x)dx [4.6(n)]
We now define
b
a ji g j ( x ) fi ( x )dx , (i, j = 1, 2, ..., n) (4.7)
a
b
and b j g j ( x )F ( x )dx , (j = 1, 2, ..., n) (4.8)
a
Now, using Eqs. (4.4), (4.7), (4.8) for [4.6(a)], we obtain
n
c
1 b1 l cia1i
i 1
D( l )
la 21 1 la 22 la 2 n
la n1 1 la nn
la n 2
D(l) = 0 is a polynomial of utmost degree n in l. D(l) does not vanish
identically, since D(0) = 1. We consider the following three cases:
Fredholm Integral Equations with Separable Kernels 43
Case 1: When F(x) = 0, then by Eq. (4.8), bj = 0, which makes the system
[Eq. (4.9)] a homogeneous linear equation. We consider the following two
situations:
(a) If D(l) ≠ 0, then we get a unique zero solution, i.e., c1 = c2 = ...
= cn = 0 for Eq. (4.9). So Eq. (4.1) has the solution u(x) = 0.
(b) If D(l) = 0, those values of l for which D(l) = 0 are the eigen values
and any non-zero solution of homogeneous [ F(x) = 0] Fredholm
integral equation is known as eigenfunction of the integral equation.
Case 2: When F(x) ≠ 0, but
b
a g j ( x)F ( x) 0, j = 1, 2, ..., n,
i.e., F(x) is orthogonal [see Eq. (3.3)] to all functions gj(x), then by Eq. (4.8),
bj = 0. (j = 1, 2, ..., n), and hence, the system [Eq. (4.9)] reduces to
homogoneous linear equations. We find the following two situations:
(a) If D(l) ≠ 0, then the unique zero solution is ci = 0 for Eq. (4.9),
which provides the solution of Eq. (4.1) as u(x) = F(x), [here, we
use Eq. (4.5)).
(b) If D(l) = 0, then the system [Eq. (4.9)] provides infinite non-zero
solutions, and thus Eq. (4.1) has infinite non-zero solutions. The
resulting solution is the sum of F(x) together with the arbitrary
multiples of eigenfunctions.
Case 3: When at least one of bj is not zero, we have the following two
situations:
(a) If D(l) ≠ 0, we get a unique non-zero solution of Eq, (4.9), and
hence, a unique non-zero solution of Eq. (4.1) is obtained by Eq. (4.5).
(b) If D(l) = 0, we get either no solution or infinite solutions, and thus,
Eq. (4.1) has no solution or infinite solutions.
1
(s) f (s) l st g(t )dt .
EXAMPLE 4.1: Solve g
0
so that g
(s) f (s) l sc (iii)
or g
(t ) f (t ) l tc (iv)
1
by which Eq. (ii) provides c 0 t[ f (t ) ltc]dt
44 Integral Equations
1
1 t3
or c t f (t )dt l c
0
3 0
1 lc
or c t f (t )dt
0 3
l 1 3 1
or c
1
3 0 t f (t )
dt c
3 l 0
t f (t )dt , l 3
c 0 (cos t l c sin t )dt
or
c sin t p0 l c cos t p0
0 l c[ ( 1) 1]
c 2l c
c(1 2l )
0
1
or
c 0 if l
2
1
By Eq. (iii), the solution is f(x) = cos x, provided l .
2
1
1 (1 e x t ).y(t )dt.
EXAMPLE 4.3: Solve y( x )
0
1
and c2 et y(t )dt (iii)
0
or c1 1 c1 c2 (e 1)
1
or c2 (vi)
Now, using Eq. (v), 1e
1 t
0 e [1 c1 c2 e ] dt
t
c
2
1
c
c2 = et c1et 2 e2t
2 0
c2 2
c2 (e 1)(1 c1 ) (e 1)
2
Putting for c2,
1 (e 1)
2
e 1 (e 1)(1 c1 )
1 2
(e2 2e 3)
or c1 (vii)
2(e 1)2
Putting c1 and c2 is Eq. (iv), we get the required solution.
2p
Let c1 cos t g(t )dt (ii)
0
2p
and c2 0 sin t g(t ) dt (iii)
so that Eq. (i) reduces as
sin s cos s
g(s) f (s) c1 c2 (iv)
p p
46 Integral Equations
As per the given value of f(s), we take the following two cases:
Case 1: When f(s) = s, so that
c c
g(s) s 1 sin s 2 cos s (v)
p p
c1 c2
Hence, g(t )
t sin t cos t (vi)
p p
which provides by Eq. (ii)
2p c c
c1 0 (cos t ) t 1 sin t 2 cos t dt
p p
2p c1 2p c 2p
0 t cos t dt
or c1
p 0
sin t cos t dt 2
p 0 cos2 t dt
2p 2p
2p c1 1 1 c sin 2t
c1 = t sin t 0
2p
sin t dt . cos 2t 2 t 2
0 p 2 2 0 2p 0
c2
or c1 0 cos t 0 0
2p
[2p 0]
2p
or c1 – c2 = 0 (vii)
Similarly, by Eqs. (iii) and (iv),
2p c c
c2 sin t t 1 sin t 2 cos t dt
0 p p
2p c 2p c 2p
or c
2 0 t sin t dt 2p1 0 (1 cos 2t )dt 2p2 0 sin 2t dt
2p 2p
2p c1 sin 2t c2 cos 2t
t cos t 0
c2
2p
( cos t )dt
2p t 2 2p 2
0 0 0
c1
2p sin t 0 2p 0 0 (viii)
2p
or c1
2p
or c1 c2
2p
Looking at Eqs. (vii) and (viii), we find that this system is inconsistent,
and hence, possesses no solution.
Case 2: When f(s) = 1, then by Eqs. (v) and (vi),
c c
1 1 sin s 2 cos s (ix)
g( s )
p p
c c
Hence, 1 1 sin t 2 cos t (x)
g(t )
p p
Fredholm Integral Equations with Separable Kernels 47
2p c c
Then by Eq. (ii),
c1 0 cos.t 1 1 sin t 2 cos t dt (xi)
p p
which provides c1 = c2.
Again, by Eqs. (iii) and (x) ,
2p c c
c
2 0 sin t 1 1 sin t 2 cos t dt (xii)
p p
which provides c1 = c2.
Thus, Eqs. (xi) and (xii) provide c1 = c2 = c0 (arbitrary constant), and
hence, by Eq. (ix), we have
c
g(s) 1 0 (sin s cos s)
p
or g(s) 1 c(sin s cos s)
It is the required solution. Clearly, c can have arbitrary value; hence, in case
when f(s) = 1, we find infinitely many solutions to Eq. (i).
x cos t t
p
f ( x) l 2
sin x cos x sin t f (t )dt
x
p
p
c3 p f (t ) sin t dt (iv)
so that Eq. (i) can be expressed as
f ( x)
x l c1 x l c2 sin x l c3 cos x (v)
\ f (t )
t l c1t l c2 sin t l c3 cos t (vi)
p
Thus c1 p (t l c1t l c2 sin t l c3 cos t ) cos t dt
p p p
(1 l c1 ) t cos t dt l c2 sin t cos t dt l c3 cos2 t dt
or c1 p p p
p 1
c1 0 0 2 l c3 (1 cos 2t )dt
0 2
48 Integral Equations
a
[ f ( x )dx 0 if f ( x ) is an odd function]
a
p
sin 2t
or c l c3 t c1 lp c 0 (vii)
2 0
1 3
By Eqs. (iii) and (vi),
p
p [t l c1t l c2 sin t l c3 cos t ]t
2
c2 dt
p p p
(1 l c1 )t dt l c2 t sin tdt l c3 t cos t dt
3 2 2
or c2
p p p
c2 0 0 2l c3 t 2 sin t 2t sin t dt
p p
0 0
c3 p [t l c1 t l c2 sin t l c3 cos t ] sin t dt
p p
2(1 l c1 ) t sin tdt 2l c2 sin 2 tdt 0
c3
0 0
2l c2
c3 2(1 l c1 ) t ( cos t )0 ( cos t )dt
p p p
0
2 0 (1 cos 2t )dt
p
1
c3 2(1 l c1 ) p 0 l c2 t sin 2t
2 0
or c3 2p (1 l c1 ) l c2p
or 2pl c1 l c2p c3 2p (ix)
Solving Eqs. (vii), (viii) and (ix), we get
2p 2 l 8p 2 l 2p
c1 , c2 , c3
1 2l p 2 2 1 2l p 2 2
1 2 l 2p 2
Putting these values of c1, c2 and c3 in Eq. (v), we get the required solution as
2pl
f ( x)
x (lp x 4 lp sin x cos x )
1 2 l 2p 2
or g( x )
f ( x ) (l sin x )c1 (l cos x )c2 (ii)
2p
where, c1 cos t g(t ) dt (iii)
0
2p
and c2 0 sin t g(t ) dt (iv)
Now, from Eq. (ii),
g(t )
f (t ) (l sin t )c1 (l cos t )c2 (v)
Putting Eq. (v) in Eq. (iii), we have
2p
c
1 0 cos t[ f (t ) (l sin t )c1 (l cos t )c2 ] dt
2p 2p sin 2t 2p 1
c1 0 cos t f (t ) dt l c1
0 2
dt l c2
0 2
(1 cos 2t ) dt
2p
c1 0 cos t f (t ) dt 0 l c2 p
2p
c1 lp c2 0 cos t f (t ) dt (vi)
1 1
We now find eigenfunctions corresponding to l , .
p p
1
For l , using Eq. (vi) and (vii) with f(x) = 0, these equations give
p
c1 – c2 = 0, and therefore, Eq. (ii) becomes
c
g( x ) 1 (sin x cos x ) A(sin x cos x )
c p
where, A 1 is an arbitrary constant.
p
Thus, (sin x + cos x) or any non-zero constant multiple of (sin x + cos x)
1
will be the eigenfunction corresponding to eigenvalue l .
1 p
Similarly, corresponding to l , Eq. (vi) and (vii) with f(x) = 0 give
p
c1 + c2 = 0, and therefore, Eq. (ii) becomes
c
g( x ) 2 (sin x cos x ) B(sin x cos x )
p
c2
where, B = is an arbitrary constant.
π
Thus, (sin x – cos x) or any non-zero constant multiple of (sin x – cos x)
1
will be the eigenfunction corresponding to eigenvalue l .
p
Case 2: When f(x) ≠ 0, but
2p 2p
0 cos t f (t ) dt
0 and 0 sin t f (t )
0
i.e., f(t) is orthogonal to cos t and sin t, then Eqs. (vi) and (vii) provide
homogeneous linear equations. We have the following two situations:
(a) If D(l) ≠ 0, Eqs. (vi) and (vii) provide c1 = 0 = c2. So, the solution
of Eq. (i) is g(x) = f(x).
1
(b) If D(l) = 0, then l and as found in Case 1, we have infinite
p
non-zero solutions. The resulting solution of (1) is g(x) = f(x) + the
1
arbitrary multiples of (sin x + cos x) if l , and (sin x – cos x) if
1 p
l .
p
2p 2p
Case 3: When at least one of the cos t f (t )dt 0 or 0 sin t f (t )dt 0
0
we have the following two situations:
(a) If D(l) ≠ 0, we will have a unique non-zero solution of Eq. (i).
1
(b) If D(l) = 0 and l , then Eq. (vi) and (vii) become
p
Fredholm Integral Equations with Separable Kernels 51
cos t f (t ) dt
2p
c1
c2
[x(a)] 0
2p
sin t f (t )dt
c1 c2 [x(b)]
0
When l = –1/p, then Eqs. (vi) and (vii) provide
cos t f (t ) dt
2p
c1
c2 0
2p
[xi(a)]
c2 sin t f (t )dt
c1 [xi(b)]
0
Now, Eqs. [x(a)] and [x(b)] are incompatible unless the function f(t) satisfies
the condition.
2p 2p
0 sin t f (t ) dt
cos t f (t ) dt
0
2p
or 0 (cos t sin t ) f (t ) dt
0 (xii)
Similarly, Eqs. [xi(a)] and [xi(b)] are incompatible unless the function f(t)
satisfies the condition
2p
0 (sin t cos t ) f (t ) dt
0 (xiii)
When Eqs. (xii) and (xiii) are satisfied, the Eqs. (x) and (xi) become
redundant and we have infinitely many solutions.
1
Another case: When l and Eq. (xii) is satisfied, then by Eq. (x)
p
2p
c2
c1 cos t f (t ) dt
0
2p
Thus, if l = –1/p and 0 (sin x cos x ) f ( x ) dx
0 , Eq. (i) possesses
infinitely many solutions.
EXAMPLE 4.7: Solve the following integral equation and discuss its all
possible cases:
1
φ ( x ) F ( x ) l (1 3 xt )φ (t )dt
0
Solution:
1
The given integral equation is f(x) = F(x) + l (1 3 xt ) f (t )dt (i)
0
It may be expressed as
1 1
F ( x ) l f (t ) dt 3 x l t φ (t )dt (ii)
φ( x)
0 0
1
Let c1 φ (t )dt (iii)
0
1
and c2 t φ (t )dt (iv)
0
so that Eq. (ii) shapes as below:
φ ( x ) F ( x ) l c1 3 x l c2 (v)
so that φ (t ) F (t ) c1l 3l c2 t (vi)
1
Thus, c
1 0 [F (t ) c1l 3l c2t ]dt
1 1
or
c1 0 F (t )dt c1l 3l c2 2 (vii)
Similarly, by Eqs. (iv) and (vi),
1
c2 0 t[F (t ) c1l 3l c2t ]dt
1 1 1
or c2 0 tF (t )dt c1l 2 3c2 l 3 (viii)
Equations (vii) and (viii) may be expressed as
3l 1 1
(1 l )c1
c
2 2 0 F (t )
dt F ( x )dx (ix)
0
l 1 1
and
2
c1 (1
l )c2 0 t F
(t )dt x F ( x )dx (x)
0
The determinant
1 l 3l /2 1
(l )
D (4 l 2 ) (xi)
l / 2 1 l 4
Case 1: A unique solution of the system of equations i.e., Eqs. (ix) and
(x) will exist if and only if D(l ) 0 l 2 . As discussed in Section 4.2
Fredholm Integral Equations with Separable Kernels 53
(Case 3), the values of c1 and c2 can be determined by solving this system,
which in turn, after putting in Eq. (v) gives the solution of integral equation,
i.e., Eq. (ii) Particularly, if F(x) = 0 and l ≠ ±2, we derive c1 = 0 = c2,
which leads to trivial solution f(x) = 0. Clearly, the numbers l = ±2 are the
eigenvalues for the problem.
If l = 2, Eqs. (ix) and (x) shape as
1
c1 3c2 0 F ( x)dx 1
and
1
0 (1 x )F ( x )dx
0
c1 3c2 0 xF ( x ) dx
This means that for l = 2, the system of equations becomes incompatible
1
(possessing no solution) unless (1 x )F ( x )
0 . If this is true, then we get
0
same set from both the equations, which means one relation in c1 and c2, and
hence, infinitely many solutions will exist.
If l = –2, a similar argument follows.
Case 2: Let F(x) = 0, then Eq. (i) is homogeneous integral equation;
1
f ( x ) l (1 3 xt ) φ (t ) dt (xii)
0
If l ≠ ±2, then Eq. (i) has the trivial solution f(x) = 0, as mentioned in
Case 1.
For non-trivial solution l = ±2 are the eigenvalues, we find eigen- function
for l = 2 first. Eqs. (ix) and (x) for F(x) = 0 reduce to
c1 = 3c2
and then, by Eq. (v),
f(x) = 2(3c2 – 3xc2) = A(1 – x)
where, A = 6c2 is an arbitrary constant.
Thus, in this case, for l = 2, the eigenfunction is A(1 – x).
Similarly, for l = –2, Eq. (ix) and (x) provide c1 = c2.
Again, by Eq. (v),
f ( x)
2c1 (1 3 x )
B(1 3 x )
where B = –2c1 is an arbitrary constant.
Clearly, in this case, when l = –2, the eigenfunction is B(1 – 3x).
Case 3: When F(x) ≠ 0, then the given integral equation, i.e., Eq. (i) is
non-homogeneous. We consider the following three situations.
(a) When l ≠ ±2: This situation has already been dealt in Case 1.
(b) When l = 2: The compatibility is there if F(x) is orthogonal to
1
(1 – x). We refer back to Case 1 [if l = 2, and (1 x )F ( x ) dx
0]
0
and we have found
54 Integral Equations
1
c1 3c2 F ( x ) dx
0
where, A = 6c2
1
Finally, if l = 2 and (1 x ) F ( x ) 0 , Eq. (xiii) gives infinitely
0
many solutions.
(c) When l = –2: As before, the compatibility is there if F(x) is orthogonal
with (1 – 3x).
1
i.e., 0 (1 3x) F ( x) dx
0 , and we find
1 1
3 0
c
1 c2 F ( x ) dx
which by Eq. (v), gives
1 1
φ ( x ) F ( x ) l c2
3 0
F ( x ) dx 3 xc2 l
2 1
3 0
or φ( x) F ( x) F ( x ) dx B(1 3 x ) (xiv)
where, B
2c2 2c1
1
Finally, if l = –2 and 0 (1 3x)F ( x)
0, Eq. (xiv) provides infinitely
many solutions.
EXERCISE 4.1
p
cos l sin( x t ) φ (t ) dt
5. φ ( x ) 0
Fredholm Integral Equations with Separable Kernels 55
1
f (s) l st (1 st ) g(t ) dt . Find its resolvent kernel also.
6. g(s) 1
1
x l ( xt 2 x 2 t ) g(t ) dt
8. Solve g(s) 0
p /2
1 l cos( x t ) g(t ) dt , and find its eigenvalues.
9. Solve g( x ) 0
Answers
ex
1. u( x ) , l 1/(e2 1)
1 l (e2 1)
2 2
2. f ( x ) 2 x p (p sin x )/(p 1)
2 2
3. g(s) [(240 60 l ) 80 l s ]/(240 120 l l )
1
3st 5s 2 t 2
f (s) l
6. g(s) f (t ) dt
1 3 2 l 5 2 l
3st 5s 2 t 2
resolvent kernel R(s
, t; l )
3 2l 5 2l
7. g(s) = {2/(2 – l)} sin s, l ≠ 2
(240 60 l ) x 80 l x 2
8. g( x )
240 120 l l 2
l (cos x sin x )
9. g( x )
1 ,l
4/(p 2)
1 p (p 2)/4
Chapter 5
5.1 INTRODUCTION
In the previous chapters, Fredholm integral equations of the second kind have
been considered for any given kernal K(x, t) by having the eigenvalues and
corresponding eigenfunctions. In this chapter, the same equation is the main
motive, but now, the kernel is symmetric.
K1 ( x, t ) K ( x, t ) [5.4(a)]
b
K n ( x, t ) K ( x, z ). K n 1 ( z, t ) dz, n = 2,3, ... [5.4(b)]
a
b
Also, K n ( x, t ) K n 1 ( x, z ). K ( z, t ) dz , n = 2, 3, ... [5.4(c)]
a
56
Integral Equations with Symmetric Kernels 57
We shall use the principle of mathematical induction to prove the required
results.
By Eq. [(5.4(b)],
b
K 2 ( x, t ) K ( x, z ). K1 ( z, t ) dz
a
By Eq. [(5.4(a)],
b
K 2 ( x, t ) a K (t, z) K1 (z, x) dz
Thus, K 2 ( x, t ) K 2 (t , x )
which shows that K2(x, t) is symmetric by definition, and the required result
is true for n = 1, 2.
Let Kn(x, t) be symmetric for n = m. Then, by definition,
K m ( x, t ) K m (t , x ) [5.4(d)]
We shall show that Kn(x, t) is also symmetric for n = m + 1, i.e.,
K m 1 ( x, t ) K m 1 (t , x ) [5.4(e)]
Now, by Eq. [5.4(b)],
b
K m
1 ( x , t ) a K ( x, z) K m (z, t ) dz
Now, using Eqs. (5.3) and [5.4(d)],
b
K m
1 ( x , t ) a K (z, x) K m (t, z) dz
b
K
m 1 ( x , t ) a K m (t, z) K (z, x) dz
which upon using Eq. [5.4(c)] becomes
K m 1 ( x, t ) K m 1 (t , x ) , which is the R.H.S. of Eq. [5.4(e)].
Thus, the iterated kernel Kn(x, t) is symmetric for n = 1, 2; and it is also
symmetric for n = m + 1 whenever it is true for n = m. Hence, by mathematical
induction, Kn(x, t) is symmetric for n = 1, 2, ... .
Kf a K ( x, t ) f (t ) dt
The operator adjoint to K is
b
Ky a K (t, x) y (t ) dt
The two operators Kf and Kψ are connected as follows:
(Kf , y ) (f , K y )
For a symmetric kernel, it reduces to (Kf , y ) (f , K y ) , which means that a
symmetric kernel is self-adjoint. Now, since the permutation of factors in a scalar
product is equivalent to taking the complex conjugate, so (f , K f ) (Kf , f ).
Combining this with (Kf ,y ) (f , Ky ) , we find that for a symmetric kernel,
the inner product (Kf,f) is always real. The converse of this is also true.
Integral Equations with Symmetric Kernels 59
|| f ( x ) a i .fi ( x ) || 0 as n ¥
i 1
or, f ( x ) l Kf ( x ) 0
Multiplying by f ( x ) and integrating with respect to x from a to b, we obtain
|| f ( x ) ||2 l (Kf , f ) 0
or l || f ( x ) ||2 /(Kf , f )
r r
Since both N and D for R.H.S. are real, hence l is real.
60 Integral Equations
l1l2 f2 ( x ) K ( x, t ) f1 (t ) dt dx
b b
(l2 l1 )(f1 , f2 )
a a
b b (5.8)
f ( x ) K ( x, t ) f2 (t ) dt dx
a a 1
Since kernel K(x, t) is symmetric, we have
K ( x, t ) K (t , x ) (5.9)
Using Eq. (5.9), we find that R.H.S. of Eq. (5.8) vanishes, and so, we get
(l2 l1 )(f1 , f2 )
0 (5.10)
Now, since l2 ≠ l1, Eq. (5.10) reduces to (f1,f1) = 0,
which means f1 and f2 are orthogonal.
Property 3: The multiplicity of any non-zero eigenvalue is finite for every
b b
a a | K ( x, t ) | dx dt is finite.
2
symmetric kernel for which
Property 4: The eigenvalues of a symmetric L2 kernel forms a finite or an
infinite sequence {ln} with no finite limit point.
Property 5: The set of eigenvalues of the second iterated kernel coincide
with the set of squares of the eigenvalues of the given kernel.
Property 6: The sequence of eigenfunctions of a symmetric kernel can be
made orthonormal.
0 | l1 | | l2 | | ln | | ln 1 |
Let f1 ( x ), f2 ( x ), , fn ( x ) (5.17)
be the sequence of eigenfunctions corresponding to the eigenvalues given by
the sequence shown in Eq. (5.16) and arranged in such a way that they are
no longer repeated and are linearly independent in each group corresponding
to the same eigenvalue.
Thus, to each eigenvalue lk in Eq. (5.16), there corresponds just one
eigenfunction fk(x) in Eq. (5.17). Further, we suppose eigenfunctions fk(x) in
Eq. (5.12) have been orthonormalised.
Now, the Fourier coefficients fn of the function f(x) with respect to the
orthonormal system {fn(x)} are [given by Eq. (5.13)]
f n (
f , fn ) (Kh
, fn ) (h, K fn )
1 h
fn (hfn ) n , using the self-adjoint property of the operator and the
ln ln
relation lnKfn = fn.
Hence, the Fourier series for f(x) is given by Eq. (5.12) and is expressed as
hn
l n f
f ( x) ~
( x ) fnfn ( x) (5.18)
n 1n n 1
62 Integral Equations
We now estimate the remainder term of the series Eq. (5.18), as shown
below:
2
n p
fk ( x ) h p 2 h p | fk ( x ) |2
k lh hk
k k lk2
k
n 1 k n 1 n 1
h p f 2 ( x)
hk2 k 2 (5.19)
k
n 1 1 lk
k
Using Bessel’s inequality,
|fn ( x ) |2 b
ln2
| K ( x t ) |2 dt C12
a
n1
We find that the above series is bounded. Moreover, since h(x) is an
L2 function, it follows that the series hk2 is convergent and the partial
nP k 1
sum hk2 can be made arbitrarily small. Hence, Eq. (5.18) converges
k n 1
absolutely and uniformly.
We now proceed to show that Eq. (5.18) converges to f(x) in the mean.
For this purpose, let us denote its partial sum as
n
h
y n ( x ) m fm ( x ) (5.20)
m 1 lm
1 (h, K 2( n 1) h)
ln 2 1 . Again, max (5.23)
ln21 (h, h)
where we have omitted the modulus sign from the scalar product (h,K2(n+1)h)
because it is a positive quantity.
Integral Equations with Symmetric Kernels 63
Combining Eq. (5.22) and (5.23), we find
(h, h)
2
|| f ( x ) y n ( x ) || (h, K 2( n 1) h) (5.24)
ln21
Since ln+1 ¥, Eq. (5.24) gives
|| f ( x ) y n ( x ) || 0 as n ¥ (5.25)
Now, we use the relation
|| f ( x ) y ( x ) || || f ( x ) y n ( x ) || || y n ( x ) y ( x ) || (5.26)
where y(x) is the limit of the series with partial sum yn.
As shown above, the first term on the R.H.S. of (5.26) tends to zero, and to
show that the second term of R.H.S. of (5.26) also tends to zero, we proceed
as follows:
Since Eq. (5.18) converges uniformly, we have, for an arbitrarily small
and positive quantity , | y n ( x ) y ( x ) | , when n is sufficiently large.
|| y n ( x ) y ( x ) || (b a)1/2
and hence || y n ( x ) y ( x ) || 0
Hence, Eq. (5.26) shows that f(x) = y(x), and thus, the result follows.
The function F(x) can be expressed over the interval (a, b) by a linear
combination of the normalised eigenfunctions of homogeneous integral equation
b
y( x ) l K ( x, t ) y(t )dt (5.29)
a
having K(x, t) as its kernel.
64 Integral Equations
Procedure of solution
From Eq. (5.27), we have
b
( x ) l K ( x, t ) y(t )dt (5.30)
y( x ) f
a
Since this equation is of the form Eq. (5.28), it follows from Hilbert–Schmidt
theorem
am fm ( x) , a x b (5.31)
y( x ) f ( x )
m 1
b
and f m f ( x ) fm ( x ) dx (5.35)
a
Now, using Eqs. (5.34) and (5.35) and interchanging the order of integration
C
m fm l
b
a
y(t ) a
b
K (t , x ) fm ( x )dx dt (5.37)
fm
or y( x
) f ( x) l S f ( x ) (5.41)
lm l m
Further, using Eq. (5.35) for fm, we get (variable of integration changed)
fm ( x ) b
( x) f ( x) l
y a f (t ) fm (t ) dt
m lm l
f ( x )fm (t )
b
or ( x ) f ( x ) l m
y f (t ) dt
m lm l
a
b
or ( x ) f ( x ) l R ( x, t; l ) f (t ) dt (5.42)
y
a
66 Integral Equations
Case 3: Infinitely many solutions: Let l = lk, and fk = 0. Then by Eq. (5.39)
lCm C 0 l C Ck Ck
C fm , k k
m
lm l
This being an identity does not impose any restriction on Ck, with the result ak
[refer to Eq. (5.40)] is of form zero/zero and becomes arbitrary. In this situation,
we express Eq. (5.41) as follows:
lm
y( x ) f ( x ) A fk ( x ) l ' fm ( x ) (5.45)
m lm l
where primed S implies that we shall omit m = k in the summation and A is
an arbitrary constant. The solution given in Eq. (5.45) due to arbitrary nature
of A shows that given Eq. (5.27) possesses infinitely many solutions.
1
Solution: The given equation is y( x ) = ( x + 1)2 + ∫ ( xt + x 2 t 2 ) y(t )dt (i)
−1
Comparing Eq. (i) with Eq. (5.27), i.e.,
b
) f ( x) l
y( x K ( x, t ).y(t )dt (ii)
a
( x 1)2 , l
we get f ( x ) ( xt x 2 t 2 ) (iii)
1, K ( x, t )
First, we determine the eigenvalues and the corresponding normalised
eigenfunctions of Eq. (i) after deleting its non-homogeneous part. So, let
Integral Equations with Symmetric Kernels 67
1
) l
y( x ( xt x 2 t 2 ) y(t )dt (iv)
1
1 1
y( x ) l x t .y(t ) dt l x 2 t 2 y(t ) dt (v)
1 1
1
Let 1 t y(t )dt c1 (vi)
1
1 t
2
and .y (t ) dt c2 (vii)
y ( x ) l c1 . x l c2 . x 2 (viii)
So, Eq. (v) reduces to
and y (t ) l c1 . t l c2 . t 2 (ix)
1
1 t (l c1 . t l c2 . t
2
Then, by Eq. (vi) c1 ) dt
1 1
t3 t4
or c1 l c1 l c2
3 1 4 1
2
c1 l c1 0.c2
3
2l
or 1 c1 0.c2 0 (x)
3
Similarly, by Eq. (ix) and (vii)
1
c2 1 t 2 (l c1 .t l c2 t 2 )dt
1 1
t4 t5
or c2 c1l c2 l
4 1 5 1
2l
or 0. c1 1 c2
0 (xi)
5
Equation (x) and (xi) will provide non-trivial values of c1 and c2 only if
2λ
1− 0
3
D( λ ) = = 0 (xii)
2λ
0 1−
5
Giving l = 3/2, 5/2 as the required eigenvalues.
Determination of eigenfunction corresponding to l1 = 3/2
3
Putting
l ( l1) in Eq. (x) and (xi), we get
2
2 3
0 and 0. c1 1 . c2
0. c1 0. c2 0
5 2
we find that c2 = 0 and c1 is arbitrary.
68 Integral Equations
3
Then, by Eq. (viii), one eigenfunction is y1 ( x ) c1 x or y1(x) = x by taking
2
3c1
1
2
Now, the corresponding normalised eigenfunction f1(x) is given by
y1 ( x ) x x 6
f1 ( x ) (xiii)
1 1 2
1 {y ( x )}2 dx 2 1 x 2 dx 2
1 1 1
5
Determination of eigenfunction corresponding to l 2
2
5
Putting
l ( l
2 ) in Eqs. (x) and (xi), we get
2
2 5
0. c1 1 . c2 0 0. c1 0. c2 0
5 2
and
2 5 5
1 . c1 0. c2 0 1 c1 0. c2 0
3 2 3
We find that here, c1 = 0 and c2 is arbitrary.
Now, substituting these c1 and c2 in Eq. (viii), the eigenfunction
5
corresponding to l2 is
2
5
y2 ( x ) 0 c2 x 2
2
2 5
y2(x) = x , (by taking c2 1 )
2
Further, the corresponding normalised eigenfunction f2(x) is given by
y2 ( x ) x2 10 2
f2 ( x ) 1
1
x
2 (xiv)
1 y ( x ) 2 dx 2 1 x 4 dx 2
1 2 1
Now, by Eqs (iii) and (xiii) [also refer to Eq. (5.35)],
1 1 6x
f1 1 f ( x ). f1
( x )dx 1 ( x 1)2
2
dx
1
6 1 6 x 4 2 x3 x2 2 6 (xiv)
1
3 2
f1 ( x 2 x x )dx
2 2 4 3 2 3
1
Similarly, by Eqs. (iii) and (xiv),
1 1 10 2
1 f ( x) f2 ( x
)dx ( x 1) 2
f2 x dx
1 2
Integral Equations with Symmetric Kernels 69
1
10 x5 2 x 4 x3 8
f2 10 (xv)
2 5 4 3 1 15
3
Also, from Eq. (iii) l = 1 and l1 , l2 5 ;
2 2
Thus, we see that l l1 l2 , the integral equation, i.e., Eq. (i) will possess
a unique solution, which is given by
2
fm
y
( x) f ( x) l f ( x ) (xvi)
lm l m
m 1
f f f f
y( x ) ( x 1)2 1 1 1 2 2
l1 l l2 l
2 x 6 8 10 10 2
6 x
2
3 2 15 2
y( x ) ( x 1)
3 5
1 1
2 2
25 2
or y( x
) x 6 x 1 , is the solution of Eq. (i).
9
EXAMPLE 5.2: Using Hilbert–Schmidt theorem, find the solution of the
following symmetrical integral equation
3 1
y( x ) ( x 2 1) 1 ( xt x
2 2
t ) y(t ) dt
2
Solution: The given equations is
3 1
2 1
y( x ) ( x 2 1) ( xt x 2 t 2 ) y(t )dt (i)
Equation (iv) is same as Eq. (iv) of Example 5.1. So, we take the eigenvalues
and the corresponding normalised eigenfunctions and these are
70 Integral Equations
3 5 x 6 10 x 2
l1 , l2 , f
1 ( x) , f
2 ( x)
2 2 2 2
1 1 x 6
1 f ( x)f1 ( x)dx
1 ( x 1) 2 dx
2
Now, f1 ( x ) 0 (v)
1
and f2 ( x ) f ( x )f2 ( x )dx
1
1
1 10 x 2 10 x 5 x 3 8 10
f2 ( x ) 1 ( x 2 1)
2
dx 2
2 5 3 15
(vi)
0
3
Here, we find that l l1 and l ≠ l2, so here, infinitely many
2
solutions will exist [case 3], and then the solution is given by
2
fm
y( x ) f ( x ) Af1 ( x ) l ' .f ( x ) (vii)
m 1 lm l m
Here, S' means that the term for m = 1 must be neglected. So, we get
f2
y( x ) f ( x ) Af1 ( x ) l f ( x)
l2 l 2
(8 10)
x 6 3 2
15 x 10
y( x ) ( x 2 1) A
2 2 53 2
2 2
A 6
or y( x ) x 2 1 c1 x 4 x 2 , c1
2
2
or y( x ) 5 x c1 x 1 is the required solution, c1 being an arbitrary constant.
c1l p c l p
c1
2 0
(1 cos 2t ) dt 2 sin 2 t dt
2 0
p p
c1l sin 2t c2 l cos 2t
c1 t
2 2 0 2 2 0
c1l
c1 p
2
c1 (2 l p ) 0 c2 (x)
0
Similarly, using Eq. (ix), (vii) gives
p
c2 sin t c1l cos t c2 l sin t dt
0
c1l p c l p
c2
2 0
sin 2t dt 2 (1 cos 2t ) dt
2 0
p p
c1l cos 2t lc sin 2t
or c2 2 t 2
2 2 0 2 o
c l
0 2 (p 0 )
c2
2
or 0 c1 (2 l p ) c2
0 (xi)
Equations (x) and (xi) will have a non-trivial solution if
72 Integral Equations
2 lp 0
D( l ) 0
0 2 lp
(2 lp )(2 lp )
0
2 2 2 2
l ,
,(say) l1 , l2
p p p p
So, l1 and l2 are the eigenvalues.
2 2c1
y1 ( x )
c1 cos x cos x , by taking 1
p p
The corresponding normalised eigenfunction f1(x) is given by
y1 ( x ) cos x cos x
f1 ( x ) 1/2
p p 1 cos 2 x
{y1 ( x )}2 dx
p
cos x dx
2
0 0 0 2
dx
1/2
cos x cos x 2
f1 ( x ) cos x (xii)
1 sin 2 x
p p / 2 p
x
2 2 0
–2
Determination of eigenfunction corresponding to l 2 =
p
2
For this value of
l l2 , Eqs. (x) and (xi) give
p
4c1 + 0.c1 = 0 and 0.c1 + 0.c2 = 0
This means c1 = 0 and c2 is arbitrary.
2
Putting these values in Eq. (viii) and recollecting that l , we have
the eigenfunction p
y2 ( x ) 0 ( 2 / p ) c2 sin x
2c
or y2(x) = sin x, Taking 2 1
p
Integral Equations with Symmetric Kernels 73
y2 ( x ) sin x 2
f2 ( x ) sin x (xiii)
p y ( x ) 2 dx
1/2 p p
0 2
sin x dx
0
2
p p 2
Now, f1 ( x ) 0
f ( x ).f1 ( x ) dx
0
1.
p
cos x dx 0 (xiv)
p p 2 2
and f ( x) f2 ( x) dx
f2 ( x )
0 1 0 p
sin x dx
2
p
(xv)
Case 1: Let l ≠ l1 and l ≠ l2, then Eq. (i) will possess unique solution
given by
2
fm
( x) f ( x) l
y fm ( x )
l m l
m 1
f f
1 l . 1 f1 ( x ) l . 2 f2 ( x )
y( x )
l1 l l2 l
Now, substituting the values of l1, f1, f1 and l2, f2, f2 , we get
1/2 1/2
l 2 2
y( x ) 1 0 2
l
p
sin x since f1 0
2
p l
4l sin x
or y( x ) 1 (xvi)
2 pl
2
Case 2: Let l l2 . Since f2 ≠ 0, so Eq. (i) possesses no solution.
p
[Note: f1 = 0, and denominator of last term becomes zero.]
Case 3: Let l = l1 = 2/p. Since f1 = 0, there exists infinitely many solutions
given by
2
fm
y( x ) f ( x ) Af1 ( x ) l ' fm ( x ) (xvii)
m 1 l m l
2 sin x
y( x )
1 c cos x
p
2A
where, c is an arbitrary constant.
p
EXAMPLE 5.4: Using Hilbert–Schmidt method, solve
1
x l K ( x, t ) y(t ) dt
y( x )
0
x (t 1), 0 x t
where, K ( x, t )
t ( x 1), t x 1
x (t − 1), 0 ≤ x ≤ t
where K = (ii)
t ( x − 1), t ≤ x ≤1
Comparing Eq. (i) with Eq. (5.27), we get
f(x) = x (iii)
First, we begin with the determination of eigenvalues and the corresponding
eigenfunctions of
1
y( x ) l K ( x, t ) y(t ) dt (iv)
0
( x ) l K ( x, t ) y (t ) dt K ( x, t ) y (t ) dt
x 1
or y
0 x
Now, from Eq. (ii),
x 1
y( x ) l t ( x 1) y(t ) dt l x (t 1) y (t ) dt (v)
0 x
d H ( x) H ( x ) F dH dG
dx G ( x )
F ( x, x ) dx
G ( x ) x
dx F [ x, H ( x )]
dx
F[ x, G( x )]
dx
x d d
y
'( x ) 0 x {lt ( x 1). y (t )}dt l x ( x 1). y( x). dx ( x) l.0 ( x 1).y(0) dx (0)
1 d d
{l x (t 1). y (t )}dt l x (1 1). y(1). (1) l .x ( x 1).y( x ) ( x )
x x dx dx
Integral Equations with Symmetric Kernels 75
x
or y '( x ) l t . y (t ) dt l x ( x 1).y ( x ).1 0
0
1
l (t 1).y(t ) dt 0 l x ( x 1).y ( x ).1
x
x 1
or y
( x ) 0 lt .y (t ) dt x l (t 1). y(t ) dt (vi)
Again, differentiating Eq. (vi) with respect to x, we get
x d d
y "( x ) 0 x [lt . y(t )] dt l x. y ( x) dx ( x) l .0. y (0) dx (0)
1 d d
[ l (t 1).y (t ) ]dt l (1 1). y (1) (1) l ( x 1) y( x ). ( x )
x x dx dx
or y "( x ) 0 l x. y( x ) 0 0 0 l ( x 1). y( x )
or y "( x ) l .y ( x )
0 (vii)
by Eq. (v) y(0) = 0 (viii)
and y(1) = 0 (ix)
*
Equation (vii) with Eqs. (viii) and (ix) is Sturm–Liouville problem.
[refer to {r ( x ). y '}' {q( x ) l p ( x )}y
0 a x b
satisfying k1y + k2y¢ = 0 at x = a and l1y + l2y¢ = 0 at x = b]
We consider the different cases depending upon l
Case 1: Let l = 0, The solution of Eq. (vii) is
y(x) = Ax + B (x)
Using Eqs. (viii) and (ix), we get
B = 0, A + B = 0
A = 0 = B, so for l = 0, the solution is y(x) = 0.
Thus, l = 0 is not an eigenvalue and y(x) = 0 is not an eigenfunction.
Case 2: Let l = m2 (m ≠ 0). Then, the solution of Eq. (vii) is
y Ae m x Be m x ;
Using Eq. (viii) and (ix), we get
A + B = 0, Ae m Be m
0
giving A = 0 = B.
Thus y(x) = 0 is not eigenfunction.
ln n2 p 2, n = 1, 2, 3, ...
and
y( x ) B sin np x
Taking B = 1, we have the eigenfunction
yn ( x ) sin np x , n = 1, 2, 3, ... (xi)
The normalised eigenfunction fn(x) is given by
yn ( x ) sin np x
fn ( x ) 1/2 1/2
1 y ( x ) 2 dx 1 sin 2 np x dx
0 n 0
or, fn(x) = 2 sin np x (xii)
1
Now, fn = 0 f ( x).fn ( x) dx ; f(x) = x [By Eq. (iii)]
1
fn = 0 x. 2 sin np x dx
cos np x 1 1 cos np x
fn = 2 x. 0 dx
np 0 np
cos np 1 sin np x
1
fn = 2
np np np 0
( 1) n 1 ( 1) n 1 2
fn = 2 2 2 .0 (xiii)
np n p np
Now, two cases arise,
Case 1: Let l ≠ ln, n = 1,2,3,... i.e., l is not an eigenvalue. Then Eq. (i)
will possess a unique solution, given by
Integral Equations with Symmetric Kernels 77
fn
( x) f ( x) l
y fn ( x )
n 1 l n l
( 1) n 1 2 1
y(x) = x l 2 2 2 sin np x
n 1 n p n p l
2 l 1 sin np x
n
y(x) = x (xiv)
p n 1 n n2 p 2 l
Case 2: Let l = ln = –n2p2, n = 1, 2, 3, ...
b
= K(x) [By taking C / {K (t )}2 dt 1 ]
a
K ( x)
so f1 ( x ) 1/2
b {K ( x )}2 dx
a
1
b b b 2
f ( x ). f1 ( x )dx a f ( x ). K ( x )dx / a {K ( x )} dx
2
f1
a
f ( x)
y(x)
l
.
a f ( x). K ( x)dx . K ( x)
1 1 1
b {K ( x )}2 dx
a l b {K ( x )}2 dx 2 b {K ( x )}2 dx 2
a a
b
l . K ( x ) f ( x ). K ( x )dx
a
f ( x)
y(x) b
1 l {K ( x )}2 dx
a
78 Integral Equations
f ( x ) A K ( x ) {K ( x )}2 dx
b 2
y(x)
a
f ( x ) c. K ( x )
y(x)
1
2p 2p
where, c1 cos t .y(t )dt and c2 sin t .y (t )dt
0 0
Clearly,
y(t ) l c1 sin t l c2 cos t
c
1 l p c2 c1 lp c
2 0
and c
2 l p c1 lp c1 c
2 0
1 lp 1 1
D (l ) 0 l1 , l2
lp 1 p p
Integral Equations with Symmetric Kernels 79
1
Determination of eigenfunction for l1
p
1
Taking
l l 1 , we get c1 = c2 so that
p
c
y1 ( x ) 1 (sin x cos x ) (sin x cos x )
p
c1
(By taking 1)
p
The corresponding normalised eigenfunction f1(x) is given by
1
2p 2p x (sin x cos x )
Now, 0 f ( x) f1 ( x)dx
f1 0 2p
2p
dx
2p 2p x (sin x cos x )
and f2 0 f ( x ). f2 ( x )dx
2p
0
dx 2p
In the question, it is given that l ≠ l1, l2; hence, Eq. (i) will possess
unique solution, which is given by
2
fm
y( x ) x l l fm ( x )
m 1 m l
f .f ( x ) f .f ( x )
xl 1 1
y( x ) l 2 2
l1 l l2 l
y( x )
x
l 2p (sin x cos x ) / 2p
l 2p (sin x cos x )/ 2p
1 1
l l
p p
2l 2p 2 sin x 2lp cos x
or y( x )
x
1 l 2p 2 1 l 2p 2
80 Integral Equations
1
) e x l K ( x, t ). y(t )dt
y( x
0
sinh x .sinh (t 1)
, 0 xt
sinh 1
where, K ( x, t )
sinh t . sinh ( x 1) , t x 1
sinh 1
sinh x .sinh (t − 1)
, 0≤ x≤t [ii(a)]
sinh 1
where, K ( x, t ) =
sinh t . sinh ( x − 1) , t ≤ x ≤ 1 [ii(b)]
sinh 1
Here, f(x) = ex
First, we determine the eigenvalues and eigenfunction for the homogeneous
integral equation
1
y( x ) l K ( x, t ). y(t )dt (iii)
0
1 1 x
Then, fn
0
f ( x ) . fn ( x )dx e fn ( x ) dx
0
1
fn e x . 2 sin np x dx
0
2 e x (sin np x np cos np x )
1
fn 2 2 0
1 n p
2
fn e np cos np ( np )
1 n2p 2
np 2
or fn 1) n , n 1, 2,3 (ix)
1 e (
1 n2p 2
Here, we consider two situations:
(a) Let l ≠ ln (means l is not an eigenvalue). Then Eq. (i) will possess
a unique solution given by
fn
( x) f ( x) l
y f ( x)
n 1 ln l n
np 2 2 sin np x
y( x ) e x l 2 2
[1 e( 1)n ]
n 1 1 n p (1 n2p 2 ) l
n[1 e( 1) n ]sin np x
or, ) e x 2pl
y( x
n 1 [1 n2p 2 ][1 l n2p 2 ]
1 x 3 6 4 3 ( x t) y(t)dt
1
(b) y( x )
0
3x
[ Ans : y( x ) 1 x 3 C 1 3 1 , C being an arbitrary constant.]
2
Integral Equations with Symmetric Kernels 83
EXERCISE 5.1
Using Hilbert–Schmidt theorem, solve the following symmetric integral
equations:
1 1
1. y( x ) x y(t ) dt ,
2 0
1
2. y( x ) x l 0 xs y(s)ds,
1
x l y(s)ds, l 1
3. y( x )
0
1
4. y( x ) (1 x 3) (6 4 2) 0 ( x t ) y(t )dt
cos p x l K ( x, x ) y(x )dx
5. y( x )
x ( x 1) 0 x x
where, k ( x, x )
Answers: x (x 1) x x 1
1
1. [y(x) = – x + c]
2
3. [y(x) = x + l/{2(1 – l)]
4. [ y( x ) (1 x 3) (1 3 x / 2) c(1 x 3) ]
Chapter 6
Solution of Integral
Equations of the Second Kind
by Successive Approximation
6.1 INTRODUCTION
In the previous chapters of the book, the solution of the integral equations
was mainly focused upon Fredholm integral equation. In this chapter the
development is for Volterra integral equation also. Moreover, where the
solution is not possible in closed form, successive approximate method is
also discussed. Apart from the various theorems, the chapter includes three
parts—(a) iterated kernels, (b) resolvent kernel, and (c) solution of integral
equations by applying resolvent kernel.
K1 ( x, t ) K ( x, t ) [6.2(a)]
b
14243
14243
and
K ( x, z) K n 1 (z, t )dz,
K n ( x, t )
t
n
2,3, …
x
[6.4(b)]
or
K n 1 ( x, z) K (z, t )dz,
K n ( x, t )
t
2,3, …
n
K m ( x, t ) a Kr ( x, y) K m r ( y, t ) dy
where, r is any positive integer less than m.
* A convergent infinite series has a sum. But, if we differentiate (or integrate) its all the
terms, then this sum may not be equal to the derivative (or integral) of sum. However,
if the series is absolutely and uniformly convergent, then it is possible. To check this
characteristics, we check if the modulus of the nth term is less than a definite quantity.
86 Integral Equations
b b
b
l 2 K ( x, t ) K (t , t1 ) f (t1 ) l K (t1 , t2 ) y(t2 ) dt2 dt1dt
a a a
b b b
or y( x ) f ( x ) l K ( x, t ) f (t )dt l 2 K ( x, t ) K (t , t1 ) f (t1 ) dt1dt
a a a
b b b
l 3 K ( x, t ) K (t , t1 ) K (t1 , t2 ) y(t2 ) dt2 dt1 dt (6.13)
a a a
Rn 1 ( x ) (6.14)
b b b
where, Rn+1(x) l n 1 K ( x, t ) K (t , t1 ) K (tn 1 , tn ) y(tn )dtn dt1dt (6.15)
a a a
Solution of Integral Equations of the Second Kind by Successive Approximation 87
We now consider the following infinite series:
b b b
f ( x ) l K ( x, t ) f (t )dt l 2 a K ( x, t )a K (t, t1 ) f (t1 ) dt1dt (6.16)
a
Following the assumptions (1) and (2), each term of Eq. (6.16) is continuous
in I; and it is thus obvious that Eq. (6.16) is also continuous, provided it
converges uniformly in I.
Let Un(x) represent the general term of Eq. (6.16), which we express as
b b b
U n ( x ) l n K ( x, t ) K (t , t1 ) K (tn 2 , tn 1 ) f (tn 1 ). dtn 1 dt1dt (6.17)
a a a
Therefore,
| U n ( x ) | | l |n N M n (b a) n [By the conditions of theorem] (6.18)
And Eq. (6.15) converges only when
| l | M ( b a) 1
or | l | 1 / M (b a) (6.19)
and it holds good due of condition 3.
It thus means that if Eq. (6.7) has a continuous solution, it is given by
Eq. (6.14). If y(x) is continuous in I, |y(x)| must have a maximum value, say
Y. Thus,
|y(x)| Y (6.20)
Now, from Eq. (6.15),
b b b
| Rn 1 ( x ) | | l n 1 K ( x, t ) K (t , t1 ) K (tn 1 , tn ) y(tn ) dtn dt1dt |
a a a
or | Rn 1 ( x ) | | l |n 1 Y M n 1 (b a)n 1
Using condition 1 and Eq. (6.20) and now due to condition 3
lim Rn 1 ( x ) 0
n
It thus means that function y(x) satisfying Eq. (6.14) is the continuous
function given by Eq. (6.16) or Eq. (6.8).
x
t
f ( x ) l K ( x, t ) f (t ) l K (t , t1 ) y(t1 ) dt1 dt
y( x )
a a
x x t
f ( x ) l K ( x, t ) f (t )dt l 2 K ( x, t )
or y( x ) a K (t, t1 ) y(t1 )dt1 dt (6.27)
a a
Now, we find y(t1) to put in Eq. (6.27), and for this, we replace t1 by t2, and then,
t by t1 in Eq. (6.26), we get
t1
f (t1 ) l K (t1 , t2 ) y(t2 )dt2
y(t1 )
a
a a a
x x t
f ( x ) l K ( x, t ) f (t )dt l 2 K ( x, t ) K (t , t1 ) f (t1 ) dt1 dt
or y( x )
a a a
x t t1
l 3 K ( x, t ) K (t , t1 ) K (t1 , t2 ) y(t2 )dt2 dt1dt (6.28)
a a a
Now, due to conditions (1) and (2), each term of Eq. (6.31) is continuous
in I. It makes Eq. (6.31) also continuous in I, provided it converges uniformly
in I. Let Vn(x) be the general term of Eq. (6.31), given by
x t tn 2
Vn ( x ) l n K ( x, t ) K (t , t1 ) K (tn 2 , tn 1 ). f (tn 1 ) dtn 1 dt1dt (6.32)
a a a
n
Then, | Vn ( x ) | | l |n N M n ( x a)
n!
[Here, we have applied conditions 1 and 2 over mod of Eq. (6.32)]
(b a) n
or | Vn ( x ) | | l |n N M n , a xb
n!
or | Vn ( x ) | | l |n N [ M (b a)]n /n! (6.33)
Now, from Eq. (6.32), it is clear that Eq. (6.31) is convergent for all l, N,
M, (b – a); thus, from Eq. (6.33), it is followed that Eq. (6.31) is convergent
obsolutely and uniformly.
So, if Eq. (6.21) has a continuous solution, it must be expressed by
Eq. (6.29). If y(x) is continuous in I, |y(x)| must have a maximum value, say Y.
or |y(x)| Y (6.34)
( x a) n 1
Now, from Eq. (6.30), | Rn 1 ( x ) | | l |n 1 Y M n 1
(n 1)!
(b a) n 1
| Rn 1 ( x ) | | l |n 1 Y M n 1 axb
(n 1)!
Hence, lim
n Rn 1 ( x ) 0
It thus follows that function y(x) satisfying Eq. (6.29) is the continuous
function given by Eq. (6.24) or Eq. (6.31). It thus proves the theorem.
For the required solution y(x), we begin with zero order approximation.
So, let
y0(x) = f(x) (6.36)
Now, if yn(x) and yn–1(x) represent the nth and (n – 1)th order approximations
respectively, then these are connected by
b
f ( x) l
yn ( x ) K ( x, t ) yn 1 (t ) dt (6.37)
a
Now, since y0(t) = f(t) by [using Eq. (6.36), we have [by Eq. (6.39)]
b
( x ) f ( x ) l K ( x, t ) f (t )dt (6.40)
y1
a
f ( x ) l K ( x, z ) f ( z ) l K ( z, t ). f (t )dt dz
b b
y2 ( x )
a a
b b b
f ( x ) l K ( x, z ) f ( z )dz l 2 K ( x, z ) K ( z, t ) f (t )dt dz (6.42)
y2 ( x )
or a a a
y2 ( x ) f ( x ) l K ( x, t ) f (t ) dt l 2 f (t ) K ( x, z ) K ( z, t )dz dt
b b b
a a a
Now, using Eqs. [6.38(a)] and [(6.38(b)] in the second and third term,
respectively, we obtain
b b
y2 ( x ) f ( x ) l K1 ( x, t ) f (t )dt l 2 K 2 ( x, t ) f (t )dt
a a
2 b
or f ( x ) l m K m ( x, t ) f (t )dt (6.43)
y2 ( x )
a
m 1
Solution of Integral Equations of the Second Kind by Successive Approximation 91
Proceeding similarly, we generalise it and write
n b
f ( x ) l m K m ( x, t ) f (t )dt (6.44)
yn ( x )
a
m 1
Now, upon taking the limit n ¥, we find the Neumann series
b
lim
y( x ) f ( x ) l m K m ( x, t ) f (t )dt (6.45)
n yn ( x ) a
m 1
Putting values in R.H.S. using Eq. [6.51(a)] for K1(x, t) and [6.51(b)] for
Km(x, t), we get
b
K ( x, t ) l m 1 K ( x, z ) K m 1 ( z, t ) dz
R( x , t; l )
a
m 2
Setting n for m – 1,
b
K ( x, t ) l n K ( x, z ) K n ( z, t ) dz
R( x , t; l )
a
n 1
b
K ( x, t ) l l m 1 K ( x, z ) K m ( z, t ) dz
R( x , t; l )
a
m 1
[Replacing n by m]
Now, interchanging the signs of summation and integration
b
, t; l) K ( x, t ) l l m 1K m ( z, t ) K ( x, z ) dz
R( x
a
m 1
Using Eq. (6.48) in the integrnal part,
b
K ( x, t ) l [ R( z, t; l )] K ( x, z ) dz
R( x , t; l )
a
b
or K ( x, t ) l K ( x, z ) R ( z, t; l ) dz
K ( x , t; l )
a
Here,
a 0,
b 2p .
So, K1 (
x, t ) K (
x, t ) sin( x 2t ) (iii)
Let n = 2 in Eq. (ii), we have
2p
K 2 ( x, t ) 0 K ( x, z ) K1 ( z, t ) dz
K 2 (
x, t ) 0 sin ( x 2 z ) sin( z 2t ) dz
1 2p
2 0
K 2 ( x, t ) [cos( x 2t 3z ) cos( x 2t z )] dz
2p
1 1
K 2 ( x, t ) sin( x 2t 3z ) sin( x 2t z )
2 3 0
So, K 2 ( x, t ) 0 (iv)
Let n = 3 in Eq. (ii). Then, we have
2p
K3 ( x, t ) 0 K ( x, z ) K 2 ( z, t )dz 0 [due to Eq. (iv)]
K 2 ( x, t ) 0 K ( x, z) K1 (z, t )dz
for K1(z, t), we use Eq. (v).
p x
K 2 ( x, t ) 0 e cos z e z cos t dz
p
e x cos t . e x cos z dz
0
p
e z
e x cos t 2 2 [cos z sin z ]
1 1 0
1 ep x
or ( 1)1
K 2 ( x, t ) (e cos t ) (vi)
2
94 Integral Equations
(1 ep ) x 1 ep
e cos t ( 1)1
2 2
2
1 ep x 2
or K3(x, t)
( 1) e cos t (vii)
2
Looking the form of Eqs. (v), (vi) and (vii), we write
n 1
1 ep
n 1
K n ( x, t )
( 1) n e x cos t;
1, 2,3,...
2
(c) Here, with kernel K(x, t) = x + sin t, we have a = –p and b = p.
Now, using Eq. (i),
K1(x, t) = K(x, t) = (x + sin t) ... (viii)
For n = 2, Eq. (ii) gives
p
K 2 ( x, t ) p K ( x, z) K1 (z, t ) dz
p
or K 2 ( x, t ) ( x sin z )( z sin t ) dz
p
p p p p
x z dz sin t sin z dz x sin t dz z sin z dz
K 2 ( x, t )
p p p p
p
K 2 ( x, t ) x 0 sin t 0 x sin t ( z )pp [ z ( cos z )]pp 1 ( cos z ) dz
p
K ( x, z) K2 (z, t )dz
K3 ( x, t )
p ( x sin z){2p (1 z sin t )}dz
p
( x, t ) 4p 2 ( x sin t )
K3 (x)
Further, for K4(x, t), let n = 4 in Eq. (ii) and using the previons results,
p
K 4 ( x, t ) p K ( x, z) K3 (z, t )dz
Solution of Integral Equations of the Second Kind by Successive Approximation 95
p
p ( x sin z) {4p
2
K 4 ( x, t ) ( z sin t )}dz
K 4 ( x, t ) 4p 2 2p (1 x sin t ) (xi)
or , t ) 4p 2 K 2 ( x, t ) (xii)
K 4 ( x
Similarly, p 2 K3 ( x, t ) 16p 4 K1 ( x, t ) (xiii)
K 5 ( x, t ) 4
p 2 K 4 ( x, t ) 16p 4 K 2 ( x, t ) (xiv)
K 6 ( x, t ) 4
In general,
If n (2p )2 m 2 ( x sin t ), m = 1, 2, 3,...
2m 1, K 2 m 1 ( x, t )
K 2 m ( x, t ) (2p )2 m 1 (1 x sin t ), m = 1, 2, 3, ...
n 2m,
(d) Here, we have K(x, t) = x – t, a = 0 and b = 1.
Applying Eq. (i),
K1 ( x, t ) K ( x, t ) x t (xv)
and by Eq. (ii) for n = 2,
1
K 2 ( x, t ) 0 K ( x, z) K1 (z, t )dz
1
or K 2 ( x, t ) 0 ( x z) (z t )dz [By Eq. (xv)]
x t 1
K 2 ( x, t
) xt (xvi)
2 3
Next, for n = 3, by Eq. (ii),
1
K3 ( x, t ) 0 K ( x, z) K2 (z, t )dz
Using Eqs. (xv) and (xvi), we have
1 z t
K3 ( x, t
) 0 ( x z) 2
zt dz
3
After simplification,
( x t) 1
K3 ( x, t )
K1 ( x, t ) (xvii)
12 12
Further for n = 4, by Eq. (ii),
1
K 4 ( x, t ) 0 K ( x, z) K3 (z, t )dz
which upon using Eqs. (xv) and (xvii) becomes
1 ( z t )
K 4 ( x, t
) 0 ( x z) 12
dz
96 Integral Equations
Upon solving,
1 x t 1 1
K
4 ( x, t ) xt K ( x, t ) (xviii)
12 2 3 12 2
Proceeding similarly, we find
2
1 1
K 5 ( x, t )
K3 ( x, t )
K1 ( x, t )
12 12
2
1 1
and K 6 ( x, t )
K 4 ( x, t )
K 2 ( x, t )
12 12
Generalising, if n = 2m – 1, then
( 1) m 1
K 2 m 1 ( x t ), m = 1,2,3...
12 m 1
and if n = 2m, then
( 1) m 1 x t 1
K2m m 1
xt , m = 1,2,3...
12 2 3
m 1 2
l (e2 1) l (e2 1) l (e2 1)
We find 2
1
2
2
+...
m 1
l (e2 1)
As it is an infinite geometric series with the common ratio ,
2
m 1
l (e 1)
2
1 2
so 2
l (e 1) 2 l (e2 1)
2
(vii)
m 1
1
2
l (e2 1) 2
Provided 1 | l | 2 (viii)
2 e 1
Finally, using Eq. (vii) in Eq. (vi), the required resolvent kernel is
2
l ) e x t
R( x, t; 2 , provided Eq. (viii) is satisfied.
2 l (e 1)
(b) Using Eq. (6.2), for K ( x, t ) (1 x )(1 t ), a 0, b 1,
we have K1 ( x, t ) K ( x, t ) (1 x )(1 t ) (ix)
1
and
K m ( x, t ) 0 K ( x, z) K m 1 (z, t )dz
1
Let m = 2, So,
K 2 ( x, t ) 0 K ( x, z) K1 (z, t )dz
1
K 2 ( x, t ) 0 (1 x)(1 z) (1 z)(1 t )dz
1 2
or (1 x )(1 t ) (1 z 2 ) dz (1 x )(1 t ) (x)
0 3
Now, let m = 3, So,
1
K3 ( x, t ) 0 K ( x, z) K2 (z, t )dz
2
1 2 2
K3(x, t) 0 (1 x)(1 z) 3 (1 z)(1 t )dz 3 (1 x)(1 t ) (xi)
98 Integral Equations
m 1 2
2l 2l 2l
We find that 3
1
3 3
+....
m 1
2l
is an infinite G.P.* with common ratio . Hence, this sum
3
m 1
2l 1 3 , provided 2l 1 or | l | 3 (xiv)
2l 3 2l 3 2
m 1 3 1
3
Summing up, the required resolvent kernel using Eqs. (xiii) and (xiv) is
3(1 x )(1 t ) 3
R( x , t; l ) , provided | l | (xv)
3 2l 2
Type 3: Solution of Fredholm integral equation with the help of resolvent kernel
Let
b
f ( x ) l K ( x, t ) y(t )dt (6.54)
y( x )
a
be given Fredholm integral equation.
Let Km(x, t) be the mth iterated kernel and let R(x,t;l) be the resolvent
kernel for Eq. (6.54), which is given by [refer to Eq. (6.48)]
R( x , t; l ) l m 1km ( x, t ) (6.55)
m 1
Now, if the sum of the infinite series [Eq. (6.55) exists, i.e., R(x, t; l) can be
obtained in closed form (as found in Example 6.2), then the required solution
of Eq. (6.54) is given by
b
( x ) f ( x ) l R( x, t; l ) f (t ) dt (6.56)
y
a
a
* For G.P. a + ar + ar + ...∞, S∞ = , provided | r | < 1 .
2
1− r
Solution of Integral Equations of the Second Kind by Successive Approximation 99
EXAMPLE 6.3: Solve the integral equation
5x 1 1
y( x ) x.t y(t )dt
6 2 0
by the method of successive approximation.
f (
x ) 1,
l l , K ( x
, t ) sin( x t )
th
The m iterated kernel Km(x, t) following Eq. (6.2) is given by
K1 (
x, t ) K (
x, t ) sin( x t ) (ii)
p
K m ( x, t ) 0 K ( x, z) K m 1 (z, t )dz
p
Let m = 2. So,
K 2 ( x, t ) 0 K ( x, z) K1 (z, t )dz
In the present case,
p
K2(x, t) 0 sin( x z) sin(z t )dz
1 p
2 0
K 2 ( x, t ) [cos( x t ) cos(2 z x t )]dz
or
p
1 1
K
2 ( x, t ) z cos( x t ) sin(2 z x t )
2 2 0
1 1 1 p
K 2 (
x, t ) p cos( x t ) sin( x t ) sin( x
t) cos( x t ) (iii)
2 2 2 2
Now, let m = 3, So,
p
K3 ( x, t ) 0 K ( x, z) K2 (z, t )dz
Putting for K(x, z), and K2(x, t)
p p
K3
(x, t) 0 sin( x z) 2 cos(z t )dz
2
p
which simplifies to K3 ( x, t ) sin( x t ) (iv)
2
3
p
By m = 4,
K 4 ( x, t ) cos( x t )
2
4
p
and by m = 5, K 5 ( x, t ) sin ( x t ) (v)
2
Solution of Integral Equations of the Second Kind by Successive Approximation 101
Now, looking at Eq. (ii) to (v), the symmetry is among odd and even m, and
we express the resolvent kernel R(x,t;l) by
R( x , t; l ) l m 1K m ( x, t )
m 1
R( x, t; l ) K1 ( x, t ) l K 2 ( x, t ) l 2 K3 ( x, t ) + ...
R( x , t; l ) {K1 ( x, t ) l 2 K3 ( x, t ) l 4 K 5 ( x, t ) }
l {K 2 ( x, t ) l 3 K 4 ( x, t ) l 5 K 6 ( x, t ) }
lp 2 lp 4
) sin( x t ) 1
R( x , t; l
2 2
lp lp 2 lp 4
cos( x t ). 1
2 2 2
2
lp 1 lp
R( x, t;
l ) sin( x t ) cos( x t ) 2 , provided 2 1
2 1 (lp / 2)
2
R( x , t; l ) [2 sin ( x t ) lp cos( x t )] (vi)
4 l 2p 2
Finally, the required solution is given by [refer to Eq. (6.56)]
p
y( x ) f ( x ) l R( x, t; l ) f (t )dt
0
2l p
2 2 0
y( x ) 1 [2 sin( x t ) lp cos( x – t )]dt
4l p
4l 2
y( x )
1 (2 cos x lp sin x ),| l | [After simplification]
4l p 2 2 p
EXERCISE 6.1
1/2
1. y( x ) x y(t ) dt
0
e 1 1 1
2 2 2 0
2. y( x ) e x y(t ) dt
1
x l xt y(t ) dt
3. y( x )
0
x 1 p /2
4 4 0
4. y( x
) sin x xt y(t ) dt
1
5. y( x ) f ( x ) l e x t y(t ) dt
0
102 Integral Equations
3 x 1 x 1 1 1
6. y( x ) e xe t y(t ) dt
2 2 2 2 0
1
1 l (1 3 xt ) y(t ) dt
7. y( x )
0
Answers:
1
1. x
4
2. ex
3. (3 x ) / (3 l ),| l | 3
4. sin x.
l 1 x t
1 l 0
5. f ( x ) e f (t )dt ,| l | 1
3 x 1 x e
6. e xe 1
2 2 3
4 2l (2 3 x )
7. ,| l | 2
4 l2
Then, with the help of Eqs. (6.58) and (6.59), we find y1(x), y2(x) and y3(x).
Note: In case zero order approximation is provided with the problem, we
modify our results, i.e., Eqs. (6.58) and (6.59).
1
or y1 ( x ) 2 x l x
2
Next, let n = 2. Then, by Eq. (iii),
1
y2 ( x ) 2 x l ( x t ) y1 (t ) dt
0
1 1
y2 ( x ) 2 x l ( x t ) 2t l t dt ; [By Eq. (iv)]
0
2
2 7
or y2 ( x ) 2 x l x l 2 x , [After simplification] (v)
3 12
Next, let n = 3. Then, by Eq. (iii),
1
y3 ( x ) 2 x l ( x t ) y2 (t )dt
0
1 2 7
y3 ( x ) 2 x l ( x t ) 2t l t l 2 t dt
0
3 12
2 7x 2 13 x 5
or y3 ( x ) 2 x l x l 2 l3 (vi)
3 6 3 12 8
EXERCISES 6.2
1
1. y( x ) 1 l ( x t ) y(t )dt
0
Answer:
1 2 7 3 13 x 5
1. y3 (n)
1 lx l x l
2 12 12 8
104 Integral Equations
k ( x, t ) K1 ( x , t ) K 2 ( x, t ) K3 ( x, t ) K n ( x, t )
For K1(x, t), we use Eq. [6.60(a)] and express this as [using Eq. (6.62)]
k ( x, t ) K ( x
, t ) K 2 ( x, t ) K3 ( x, t ) K n ( x, t ) (6.63)
b b
a K1 ( x, z). K1 (z, t )dz a K1 ( x, z) K2 (z, t )dz
b
a K1 ( x, z)[K1 (z, t ) K2 (z, t ) ]dz
which upon using Eqs. [6.60(a)] and (6.61), becomes
b
K ( x, z ) k ( z, t )dz (6.64)
k ( x, t ) K ( x, t )
a
a [K1 ( x, z) K2 ( x, z) ] K1 (z, t )dz
which upon using Eqs. [6.60(a)] and (6.61), becomes
b
k ( x.z ) K ( z, t )dz (6.65)
k ( x, t ) K ( x, t )
a
Now, from Eqs. (6.64) and (6.65), we have
b
k ( x, t ) K ( x, t ) a K ( x, z) k (z, t )dz [6.66(a)]
b
and k ( x , t ) K ( x , t ) a k ( x, z) K (z, t )dz [6.66(b)]
Solution of Integral Equations of the Second Kind by Successive Approximation 105
Two functions K(x, t) and k(x, t) are said to be reciprocal if they both are
real and continuous in R and satisfy the conditions given in Eq. [6.66(a)] or
Eq. [6.66(b)].
Theorem: If K(x, t) is real and continuous in R, there exists a reciprocal
function k(x, t) given by
k ( x
, t ) K1 ( x, t ) K 2 ( x, t ) K n ( x, t ) (6.67)
where K1(x, t), K2(x, t),... are iterated functions (or kernels), provided that
M(b – a) < 1, where M is the maximum value of |K(x, t)| in R for which
a x b a t b.
(Proof of this theorem has already been covered under section 6.9)
b b b
b
a k ( x, t ) y(t )dt a k ( x, t ) f (t )dt a y(z) a k ( x, t ) K (t, z)dt dz (6.71)
Now, since k(x, t) and K(x, t) are reciprocal functions, we have [by (6.66)b]
b
a k ( x, t ). K (t
, z )dt k ( x, z ) K ( x, z ) (6.72)
1 1 1 1 x t 1
t ) (e x t ) 2 3
k ( x, 2 .e e x t
2 2 2 1
1
2
So, k(x, t) = – ex–t (ix)
Now, by Eq. (6.75), the solution of Eq. (i) is given by
1
y( x ) f ( x ) e x t f (t ) dt (x)
0
and this is the required solution.
Note: The practice questions given in Exercise 6.1 can also be solved by
Volterra method.
and let the zero order approximation for the required solution y(x) be
y0(x) = f(x) (6.77)
Further, let yn(x) and yn–1(x) be nth and (n – 1)th order approximations and
these are connected by
x
f ( x ) l K ( x, t ) yn 1 (t ) dt (6.78)
yn ( x )
a
We also define [refer to Eqs. [6.4(a)] and [6.4(b)] the iterated kernels as
K1 ( x, t ) K ( x, t ) [(6.79(a)]
x
and
K n ( x, t ) t K ( x, z ) K n 1 ( z, t )dz [6.79(b)]
x
( x ) f ( x ) l K ( x, t ) . f (t )dt [using Eq. (6.77)] (6.80)
y1
a
108 Integral Equations
z
or f ( z ) l K ( z, t ) f (t )dt [Replacing x by z] (6.81)
y1 ( z )
a
f ( x ) l K ( x, z ) f ( z ) l K ( z, t ). f (t )dt dz
x z
y2 ( x )
a a
x
K ( x, z )
K ( z, t ). f (t )dt dz
x z
( x ) f ( x ) l K ( x, z ). f ( z ) dz l 2
y2
or
a z a t a
(6.83)
We now change the order of integration. Figure [6.2(a)] shows the last double
integral of Eq. (6.83). For changing the order, we refer Figure [6.2(b)] for the
same area of integration.
(a) (b)
Figure 6.2 Supporting diagrams for change of order of integration.
f ( x ) l K ( x, z ). f ( z )dz l 2 f (t ) K ( x, z ). K ( z, t )dz dt
x x x
or, y2 ( x )
a a t
z
(6.84)
Now, by [6.79(b) for n = 2,
x x
K 2 ( x, t ) K ( x, z ) K1 ( z, t )dz K ( x, z ) K ( z, t )dz [Using (6.79(a)]
t t
Now, using this result of K2(x, t) in R.H.S. of Eq. (6.84) for double integral
part, we have [replacing z by t in the second integral, along with Eq. [6.79(a)]
x x
f ( x ) l K1 ( x, t ). f (t )dt l 2 f (t ). K 2 ( x, t )dt
y2 ( x )
a a
2 x
or ( x ) f ( x ) l m K m ( x, t ). f (t )dt (6.85)
y2
a
m 1
Proceeding similarly, we can express Eq. (6.85) as
Solution of Integral Equations of the Second Kind by Successive Approximation 109
n x
( x ) f ( x ) l m K m ( x, t ). f (t )dt (6.86)
yn
a
m 1
K1 ( x, t ) K ( x, t ) [6.93(a)]
x
K m ( x, t ) K ( x, z ). K m 1 ( z, t )dz
and t [6.93(b)]
Now, from Eq. (6.89), we have
R(
x, t; l ) K1 ( x, t ) l m 1K m ( x, t )
m 2
x
R( x
, t; l ) K ( x , t ) l m 1 t K ( x, z ). K m 1 ( z, t )dz
m 2
110 Integral Equations
x
, t; l ) K ( x , t ) l n
R( x K ( x, z ). K n ( z, t )dz
t
n 1
x
, t; l ) K ( x, t ) l l m 1 K ( x, z ). K m ( z, t )dz
R( x
t
m 1
K1 ( x, t ) K ( x, t ) (i)
x
K n ( x, t ) K ( x, z ). K n 1 ( z, t )dz , n = 2,3, (ii)
t
x 2 cos x 2 cos z
or K3 ( x, t ) t .
2 cos z 2 cos t
. ( z t )dz [Here, we have used Eq. (iii), and (v)]
x
2 cos x ( z t )2 2 cos x ( x t )2
K3 ( x, t ) (vi)
2 cos t 2 2 cos t 2!
t
Similarly, by n = 4, we find
2 cos x ( x t )3
K 4 ( x, t ) (vii)
2 cos t 3!
After observing the values of K1(x, t), K2(x, t), K3(x, t), K4(x, t), we mention
2 cos x l ( x t ) l ( x t )
2 2
l 3 ( x t )3
R( x, t; l) 1
2 cos t 1! 2! 3!
2 cos x l ( x t ) . This is the required resolvent kernel.
R( x , t; l ) .e
2 cos t
Type 2: Solution of Volterra integral equation with the help of resolvent kernel
Let the Volterra integral equation be
x
( x ) f ( x ) l K ( x, t ). y(t )dt (6.96)
y
a
For its kernel K(x, t), let Km(x t) be the iterated kernels as already given in
Eqs. [6.93(a)] and [6.93(b)]. We have also learnt the method of finding the
corresponding resolvent kernel in Type 1.
R( x , t; l ) l m 1K m ( x, t ) (6.97)
m 1
Let the sum of infinite series [Eq. (6.97)] be found in closed form. Then, the
required solution of Eq. (6.96) is given by {also refer Section 6.11}
x
( x ) f ( x ) l R( x, t; l ). f (t )dt (6.98)
y
a
EXAMPLE 6.9: With the help of resolvent kernel, find the solution of the
following integral equation:
x 1 x2
y( x ) 1 x 2 . y(t )dt
0 1 t2
Solution:
The given equation is
x 1 x2
y( x ) 1 x 2 . y(t )dt (i)
0 1 t2
Type 3: Solution of Volterra integral equation when the sum of the infinite
series occuring in the formula for resolvent kernel cannot be found
Let the Volterra integral equation be
x
( x ) f ( x ) l K ( x, t ). y(t )dt (6.99)
y
0
With the problems mentioned above, we use the following formula, known
as Neumann series:
x
( x ) f ( x ) l m K m ( x, t ). f (t )dt (6.100)
y
0
m 1
where, Km(x, t) is the mth iterated kernel.
114 Integral Equations
EXAMPLE 6.10: Find the Neumann series for the solution of integral
equation:
x
y( x ) 1 x l ( x t ). y(t )dt
0
Solution:
The given equation is
x
y( x ) 1 x l ( x t ). y(t )dt
0 (i)
Comparing with Eq. 6.99),
f(x) = 1 + x, l = l, K(x, t) = (x – t).
Also, let Km(x, t) be the mth iterated kernel, then by Eqs. [(6.93(a)] and
[6.93(b)],
K1 ( x, t ) K ( x, t ) x t (ii)
x
and K m ( x, t ) K ( x, z ). K m 1 ( z, t )dz (iii)
t
x
Let m = 2, So, K 2 ( x, t ) K ( x, z ). K1 ( z, t )dz
t
x
K 2 ( x, t ) t ( x z )( z t )dz [by Eq. (ii)]
x
( z t )2 x ( 1)( z t )2 ( x t )3
K 2 ( x, t )
( x z ). t dz
0 (iv)
2 2 3!
t
Let m = 3 in Eq. (iii), we get
x
K3 ( x, t ) K ( x, z ). K 2 ( z, t )dz
t
( z t )3 x( x t )5
or
t ( x z)
K3 ( x, t )
3!
dz (v)
5!
and so on. Now, the Neumann series [shown in Eq. (6.100) is
x
( x ) f ( x ) l m K m ( x, t ). f (t )dt
y
0
m 1
x x
y( x ) (1 x ) l K1 ( x, t ).(1 t )dt l 2 K 2 ( x, t ).(1 t )dt + ...
0 0
x x ( x t )3
y( x ) (1 x ) l ( x t ) (1 t )dt l 2 (1 t )dt + ...
0 0 3!
x2 x3 x4 x5
or y( x ) 1 x l l2 + ... (vi)
2! 3! 4! 5!
It is obtained after simplication, and this is the required solution of Eq. (i).
Solution of Integral Equations of the Second Kind by Successive Approximation 115
Remarks: 1. In this question, if l = 1, then Eq. (vi) can be summed equal to ex.
2. In this question, if we find the resolvent kernel, then
R( x, t; l ) l m 1K m ( x, t ) K1 ( x, t ) l K 2 ( x, t ) l 2 K3 ( x, t )
m 1
( x t )3 ( x t )5
l2
R( x , t; l ) ( x t ) l
3! 5!
Since the sum of this series cannot be found, therefore we cannot go like
Type 2.
Another example is given here in which we have adopted the procedure
of Type 3.
1
Similarly, K 4 ( x, t
) ( x10 t 3 x 7t 4 3 x 4 t 7 xt10 ) (vii)
162
and so on.
Now, it can be verified that for the pattern of iterated kernels, the sum
of the series for the resolvent kernel cannot be ascertained; hence, we use
Neumann’s series, which is given by Eq. (6.100)
x
( x ) f ( x ) l m K m ( x, t ). f (t )dt (viii)
y
0
m 1
116 Integral Equations
Now, expanding for m, putting the values of K1(x, t), K2(x, t), ... as found
above, and simplifying, we get
x3 x6 x9
y( x )
1 + ... (ix)
2 2.5 2.5.8
This is the solution of Eq. (i).
Type 4: Method of successive approximation for solving Volterra integral
equation of the second kind
Let the integral equation be
x
( x ) f ( x ) l K ( x, t ). y(t )dt (6.101)
y
0
Also, let f(x) be continuous is [0, a] and K(x,t) be continuous for 0 x a,
0 t x.
We begin with some given function y0(x) continuous in [0, a]. Then,
replacing y(t) on R.H.S. of Eq. (6.101) by y0(x), we get
x
( x ) f ( x ) l K ( x, t ). y0 (t )dt (6.102)
y1
0
Since y1(x) given by Eq. (6.102) is itself continuous in [0, a], we proceed
similarly and arrive at a sequence of functions y0(x), y1(x), y2(x), ..., yn(x), ...
x
where, ( x ) f ( x ) l K ( x, t ). yn 1 (t ) dt
yn (6.103)
0
f ( x
) 1, l 1, K ( x, t
) ( x t)
The nth order approximation,
x
( x ) f ( x ) l K ( x, t ). gn 1 (t ). dt
gn
0
x
1 ( x t ). gn 1 (t ). dt (ii)
gn ( x )
0
Solution of Integral Equations of the Second Kind by Successive Approximation 117
x
Let n = 1, 1 ( x t ).0 dt
g1 ( x ) 1
0
Next, let n = 2 in Eq. (ii); using g1(x) = 1,
x x2 x2
g2 ( x ) 1 ( x t ).1 dt 1 x 2
1
0 2 2! ,
Then, putting n = 3 in Eq. (ii) and using g2(x), we get
x x2 x2 x4
1 ( x t ) 1 dt
g3 ( x ) 1
0 2! 2! 4!
In general, we have
x2 x4 x 2n 2
gn ( x ) 1
2! 4! (2n 2)!
Note: As n ¥, we get the infinte term to get cosh x.
Hence, the required sum is nLim
gn ( x ) cosh x
.
and g 0 (t ) = 0 (ii)
Comparing Eqs. (i), (ii) with Eq. (6.101),
[here, y( x ) g( x ) ], f ( x )
x, l
1, K ( x, t )
x t , g0 ( x )
0
The nth order approximation is given by
x
( x ) f ( x ) l K ( x, t ). gn 1 (t ) dt
gn
0
x
or gn ( x ) x ( 1) ( x t ). gn 1 (t ) dt (iii)
0
x3 x5 x 2 n 1
In general, gn ( x ) x ( 1) n 1
3! 5! (2 n 1)!
As the required solution is Lim
n gn ( x ) , we get
x3 x 2 n 1
Lim
n gn ( x ) x ( 1) n sin x
3! (2n 1)!
x1 gn21 (t )
gn ( x ) dt (ii)
0 1 t2
x 1
so that
g1 ( x )
0 1 t2
dt tan 1 x
x 1 (tan 1 t )2 1
g2 ( x ) 0 1t
dt tan 1 x (tan 1 x )3
2 3
2
1
1 tan 1 t (tan 1 t )3
x 3 dt
g3 ( x )
0 1 t2
1 2 1
g3 ( x ) tan 1 x (tan 1 x )3 (tan 1 x )5 (tan 1 t ) 7
3 35 79
Lim 1
Now, g( x ) n
gn ( x )
tan(tan x) x
(In the expansion of tan x1, if we put x1 = tan–1 x, we get it)
(b) If we take g0 ( x ) x, g1 ( x ) x g2 ( x ) g3 ( x ) gn ( x )
Solution of Integral Equations of the Second Kind by Successive Approximation 119
Lim
So, g( x ) n
( x ) x
EXERCISE 6.3
1. Find the resolvent kernel, given (a) K(x, t) = 1, (b) K(x, t) = ex–t.
2. With the help of resolvent kernel, solve
x
( x ) f ( x ) l e x t y(t )dt
(a) y
0
x
( x ) sin x 2 e x t y(t )dt
(b) y
0
x
(c) y( x ) 1 y(t )dt ,
0
x
x (t x ).y(t )dt
(d) y( x ) 0
x
1 (t x ).y(t )dt
(e) y( x )
0
x
(f) g( x ) cos x x 2 (t x ).g(t )dt
0
3. Using the method of successive approximation, find the solution of
x
1 y(t ) dt , y0 ( x )
(a) y( x )
0 [RU 94, 03]
0
x
(b) y( x ) 1 x y(t ) dt , y0 ( x ) 1
0
x
x.2 2 x t . g(t )dt , g0 ( x )
(c) y( x ) x
x.2 x
0
Recollect the following expansions:
2 3 n n
(a) e x 1 x x x x nLim x
1
2! 3! n 0 n ! n
x2 x3 x 4 xn
(b) log (1 x ) x ( 1) n 1
2 3 4 n
n
x3 x5 np .x ( 1) k 2 k 1
(c) sin x x sin , x
3! 5! 2 n! k 0 (2 k 1)!
n
x2 x4 x6 np x x 2n
(d) cos x
1 cos ( 1)n
2! 4! 6! 2 n! n 0 (2n)!
3 5
(e) tan x x x 2 x for | x | p
3 15 2
x3 x5
x 2 n 1
(f) sinh x x for all x
3! 5! n 0 (2 n 1)!
120 Integral Equations
x2 x4 x 2n
(g) cosh x 1 for all x
2! 4! n 0 (2 n)!
x 3 2 x 5 17 x 7 p
(h) tanh x x for | x |
3! 15 315 2
1 x 3 1.3 x 5 1.3.5 x 7
(i) sin 1 x x . . ( 1 x 1)
2 3 2.4 5 2.4.6 7
3 5
(j) tan 1 x x x x ( 1 x 1)
3 5
Answers:
1 3x 2 1
(b) y( x ) e sin x cos x
5 5 5
x
(c) y = e
(d) y(x) = sin x
(e) y(x) = cos x
1
(f) g( x )
cos x sin x x sin x
2
3. (a) y(x) = ex
(b) y(x) = 1
(c) g(x) = 2x(1–e–x)
Chapter 7
7.1 INTRODUCTION
The solution of the Fredholm integral equation of the second kind, i.e.,
b
( x ) f ( x ) l K ( x, t ) g(t ) dt (7.1)
g
a
where the functions f(x) and K(x, t) are integrable, has a unique solution
b
( x ) f ( x ) l R( x, t; l ) f (t ) dt (7.3)
g
a
where the resolvent kernel R(x, t; l) is a meromorphic function of parameter
l defined by
D( x , t; l )
R( x , t; l ) ; [D(l) ≠ 0] (7.4)
D( l )
121
122 Integral Equations
( l )m z , ..., zm
and D(l ) 1 K 1 dz1 dzm (7.6)
m 1 m ! z1 , ..., zm
K ( x1 , t1 ) K ( x1 , t2 ) K ( x1 , tn )
K ( x2 , t1 ) K ( x2 , t2 ) K ( x2 , t n ) x , x , … , xn
K 1 2 (7.7)
: : : : t1 , t2 , … , tn
K ( xn , t1 ) K ( x n , t2 ) K ( xn , tn )
Since Eq. (7.10) holds for all values of x in the internal (a, b), in particular,
it must be satisfied at the n points of division xi (i = 1, 2, ..., n), and thus, the
following system of equations is obtained:
n
( xi ) f ( xi ) l h K ( xi , x j ) g( x j ), (i = 1, 2, ..., n) (7.11)
g
j 1
Now, we express
g( xi ) g
i , f ( xi ) fi , K
( xi , x j ) Kij (7.12)
The values of g1, g2, ..., gn obtained by solving the algebraic system Eq. (7.14)
are approximate solutions of Eq. (7.2) at the points x1, x2, ..., xn. These solutions
g1, g2, ..., gn can be plotted as ordinates, and by interpolation, we can draw a
curve g(x) which is an approximation to the actual solution.
The solutions g1, g2, ..., gn obtained by solving the algebraic system of
equations [i.e., Eq. (7.14)] may be expressed in the form of the ratios of certain
determinants, with the resolvent determinant Dn(l) of the above algebraic
system [i.e., Eq. (7.14)], where
1 l hK11 l hK12 l hK1n
l hK 21 1 l hK 22 l hK 2 n
Dn (l ) (7.15)
l hK n1 l hK n 2 1 l hK nn
Similarly, the term containing the factor (–lh)3 is the sum of the
determinants of the form
K pp K pq K pr
( l h)3 K qp K qq K qr
Krp Krq Krr
where (p, q, r) is an arbitrary triplet of integers taken from the sequence
1, 2, ..., n with p < q < r.
Proceeding likewise, we obtain the remaining terms in the expansion of
Dn(l). Thus, the determinant [Eq. (7.15)] may be expressed in the following
form:
n
( l h)2 n K pp K pq
1 l h K ss
Dn (l )
s 1 2! p, q 1 K qp K qq
K pp K pq K pr
( l h )3 n
3!
K qp K qq K qr
p, q , r 1
Krp Krq Krr
K p1 p1 K p1 p2 K p1 pn
K p2 p1 K p2 p2 K p2 pn
( l h) n n
n!
p1 , p2 ,..., pn 1
K pn p1 K pn p2 K pn pn
n
( l h)2 n xp xq
1 lh K(x p , x p )
Dn (l ) K
2 ! p, q 1 x p xq
m 1
( l h )3 n x p , x q , xr
K (7.16)
3! p, q , r 1 x p , xq , xr
where D(l) is called Fredholm’s determinant and the series occurring on the
R.H.S. of Eq. (7.17) is called Fredhlom’s first series.
It may be pointed out that Hilbert gave a rigorous proof of the fact that
sequence Dn(l) D(l) in the limit and Fredholm proved the convergence of
Eq. (7.17) for all values of l by using the fact that kernel K(x,t) is bounded
and is an integrable function. Thus, D(l) is an entire function of the complex
parameter l.
Now, we can find the solution of Eq. (7.2) in the form given by Eq. (7.3),
where the resolvent kernel is the quotient of D(x, t; l) and D(l). For this, we
determine D(x, t; l) as the sum of certain functional series. It is known that
the resolvent kernel R(x, t; l) satisfies the following relation:
b
, t; l ) K ( x, t ) l K ( x, z ) R ( z, t; l ) dz (7.18)
R( x
a
From Eqs. (7.6) and (7.18), it follows that
b
D ( x , t; l ) D ( z, t; l )
K ( x, t ) l K ( x, z ) dz, {D(l) ≠ 0}
D (l ) a
D (l )
b
or D ( x, t; l ) K ( x, t ) D(l ) l K ( x, z ) D( z, t; l ) dz (7.19)
a
Now, the form of the series shown in Eq. (7.17) for D(l) suggests that we
search the solution of Eq. (7.19) in the form of a power series in parameter
l, i.e.,
( l )m
x, t; l ) B0 ( x, t )
D ( Bm ( x, t ) (7.20)
m 1 m !
For this, rewriting Eq. (7.17) as
( l )m
D (l ) 1 µm (7.21)
m 1 m !
b b
x , x ,…, xm
where, mm K 1 2 dx1 dxm (7.22)
a a
x1 , x2 ,…, xm
Now, substituting the series for D(x, t; l) and D(l) from Eqs. (7.20) and (7.21)
in Eq. (7.19) and comparing the coefficients of equal powers of l, we derive
the following recursion relations:
B0 ( x, t ) K ( x, t ) (7.23)
b
and Bm ( x, t ) µm K ( x, t ) m K ( x, z ) Bm 1 ( z, t ) dz
(7.24)
a
Now, we shall show that for each m(m = 1, 2, 3, ....),
126 Integral Equations
b b
x, z , z ,…, zm
Bm ( x, t ) K 1 2 dz1 … dzm (7.25)
a a
t , z1 , z2 ,…, zm
First, we see that for m = 1, Eq. (7.24) takes the following form:
b
B1 ( x, t ) µ1 K ( x, t ) K ( x, z ) B0 ( z, t ) dz
a
b b
or B1 ( x, t ) K ( x, t ) K ( z, z ) dz K ( x, z ) K ( z, t ) dz
a a
b
x, z
or B1 ( x, t ) K dz (7.26)
a
t, z
It shows that Eq. (7.25) holds for m = 1.
To prove that Eq. (7.25) holds for general m, we expand the determinant
under the integral sign by the relation
K ( x, t ) K ( x, z1 ) K ( x, zm )
K ( z1 , t ) K ( z1 , z1 ) K ( z1 , zm )
x , z , z ,..., zm
K 1 1 2
t , z1 , z2 ,..., zm (7.27)
K ( zm , t ) K ( zm , z1 ) K ( zm , zm )
with respect to the elements of the given row, transposing in turn the first
column one place to the right, integrating both sides and using Eq. (7.22), the
result, i.e., Eq. (7.25) follows by mathematical induction.
Now, from Eqs. (7.21), (7.23) and (7.25), we derive the so–called
Fredholm’s second series.
b b
( l )m x, z ,..., zm
, t; l ) K ( x , t )
D( x ... K 1 dz1 ... dzm (7.28)
m 1 m ! a a t , z1 ,..., z m
Multiplying both sides of Eq. (7.29) by R(x, z; l), and then, integrating both
sides with respect to ‘z’ from a to b, we get
b b b b
R ( x, z; l ) g ( z ) dz R ( x , z ; l ) f ( z ) dz l
R ( x , z ; l ) K ( z , t ) dz g(t ) dt (7.30)
a a a a
R ( x, z; l ) g
( z )dz R ( x, z; l ) f (z) dz [ R( x, t; l ) K ( x, t )] g(t ) dt
a a a
b b b b
or R ( x, t; l ) g(t )dt R ( x, t; l ) f (t ) dt R( x, t; l ) g(t ) dt K ( x, t ) g(t ) dt
a a a a
b b
or K ( x, t ) g(t ) dt R ( x, t; l ) f (t ) dt (7.32)
a a
( l )m
and D (l ) 1 µm (7.39)
m 1 m !
The function D(x, t; l) is called Fredholm minor and D(l) is the Fredholm
determinant.
In case, kernel K(x, t) is bounded or the integral
b b
K
2
( x, t ) dx dt
a a
has a finite value, Eqs. (7.38) and (7.39) converge for all values of l, and
therefore, are entire functions for l.
Also, the resolvent kernel R(x, t; l) is an analytic function of l, except for
those values of l which are the zeros of the function D(l). Obviously, the
latter are poles of the resolvent kernel R(x, t; l).
Alternative method for calculating Bm(x,t) and mn
The coefficients mn and the function Bm(x, t) are also found from the following
recurrence relations:
We have
b
m0 = 1, µn Bn 1 (s, s) ds, (7.43)
a
b
and Bn ( x, t ) µn K ( x, t ) K ( x, z ) Bn 1 ( z, t ) dz, m 1
(7.44)
a
Since m0 = 1 and B0(x, t), is directly known, therefore we can use Eqs. (7.43)
and (7.44) to find, in succession µ1 , B1 ( x, t ); µ2 , B2 ( x, t ), and so on. Continuing
in this way, all the coefficients can be calculated. In certain cases, depending
on the explicit form of the kernel, Eqs. (7.38) and (7.39) contain only a finite
number of terms.
It is to be kept in mind that one distinct advantage of Fredholm method
is that Eq. (7.37) is uniformly convergent for all values of l unless D(l) = 0.
EXAMPLE 7.1: Find the resolvent kernel of the following kernels by using
Fredholm determinants:
(a) K(x, t) = xet; a = 0, b = 1
(b) K(x, t) = 2x – t; 0 x 1, 0 t 1,
(c) K(x, t) = sin x cos t; 0 x 2p, 0 t 2p,
Solution: (a) Here, K(x, t) = xet
Now, by Eq. (7.24), B0(x, t) = K(x, t) = xet
1
K ( x, t ) K ( x, z1 )
Also, by Eq. (7.26) B1 ( x, t ) dz1
0
K ( z1 , t ) K ( z1 , z1 )
1
xet xe z1
B1 ( x, t ) dz1 [Using Eq. (7.40)]
t z1
z
0 1 e z1 e
B1(x, t) = 0
130 Integral Equations
(since the columns of the determinat under the integral sign are identical)
11 K ( x, t ) K ( x, z1 ) K ( x, z2 )
Similarly, B2 ( x, t ) K ( z1 , t ) K ( z1 , z1 ) K ( z1 , z2 ) dz1dz2
00
K ( z2 , t ) K ( z2 , z1 ) K ( z2 , z2 )
xet xe z1 xe z2
11
z1e z1e z1e dz1dz2
t z1 z2
B2 ( x, t ) 0
00
z2 et z2 e z1 z2 e z2
Since B1(x, t) = B2(x, t) = 0, it follows that Bn(x, t) = 0, for n 1. Thus, from
Eq. (7.42), we have
1 1
K ( z1 , z1 )dz1 z1e z1
m1 dz1
0 0
1
m1 [ z1e z1 ]10 e z1 dz1
0
m1 e [e z1 ]10 e (e 1) 1
11
z1e z1 z1e z2
and m2 z ez1 z ez2 dz1dz2 0
00 2 2
[2z1 z1 (2 x t 4 x t ) 2 xt ] dz1
2
or B1 ( x
, t)
0
1
2z3 z 2 2
or B1 ( x, t ) 1 1 ( 2 x 2t ) 2 xtz1 x t 2 xt
3 2 0 3
11 K ( x, t ) K ( x, z1 ) K ( x, z2 )
Also, B2 ( x, t ) K ( z1 , t ) K ( z1 , z1 ) K ( z1 , z2 ) dz1dz2
00
K ( z2 , t ) K ( z2 , z1 ) K ( z2 , z2 )
11 2x t 2 x z1 2 x z2
B2 ( x, t ) 2z1 t 2 z1 z1 2 z1 z2 dz1dz2
00 2 z2 t 2 z2 z1 2 z2 z2
or B2(x, t) = 0. (After little simplification)
Hence, Bp(x, t) = 0 for all p 2.
1 1 1
1 1
Again, m1 K ( z1 , z1 ) dz1 (2 z1 z1 )dz1 z12
0 0 2 0 2
11
K ( z1 , z1 ) K ( z1 , z2 )
and m2 dz dz
00
K ( z2 , z1 ) K ( z2 , z2 ) 1 2
11
2 z1 z1 2 z1 z2 1
m2
2z2 z1 2 z2 z2 dz1dz2 3
(Upon simplification)
00
111 K ( z1 , z1 ) K ( z1 , z2 ) K ( z1 , z3 )
and m3 K ( z2 , z1 ) K ( z2 , z2 ) K ( z2 , z3 ) dz1dz2 dz3
000 K ( z3 , z1 ) K ( z3 , z2 ) K ( z3 , z3 )
111 2 z1 z1 2 z1 z2 2 z1 z3
m3 2 z2 z1 2 z2 z2 2 z2 z3 dz1dz2 dz3
000 2 z3 z1 2 z3 z 2 2 z3 z 3
m3 = 0 (Upon simplification).
Hence, mp = 0 for all p 3.
Thus, we have upon using Eq. (7.5),
132 Integral Equations
( l )m
K ( x, t )
D( x , t; l ) Bm ( x, t )
K ( x, t ) l B1 ( x, t )
m 1 m !
2
D( x, t; l ) 2 x t l x t 2 xt
3
and by using Eq. (7.6),
( l )m l2 1 l2
1
D(l ) m m
1 lm m2
1 l
m 1 m ! 2! 2 6
The Fredholm resolvent kernel is given by Eq. (7.4) providing
2
2 x t l x t 2 xt
D( x , t; l ) 3
R ( x , t; l )
D( l ) 1 (l /2) (l 2 /6)
(c) Here, B0(x, t) = K(x, t) = sin x cos t. As before
2p
K ( x, t ) K ( x, z1 )
B1 ( x, t ) K ( z1 , t )
dz
K ( z1 , z1 ) 1
0
2p
sin x cos t sin x cos z1
B1 ( x, t ) dz = 0
sin z1 cos t sin z1 cos z1 1
0
D( x , t; l ) K ( x, t ) l B1 ( x, t )
D(x, t; l) = sin x cos t (By substituting the above values)
By using Eq. (7.6),
( l )m
D(l ) 1 mm
m 1 m !
D(l ) 1 lm1 1
(Upon substituting the above values)
D( x , t; l )
R ( x , t; l ) sin x cos t
D( l )
EXAMPLE 7.2: Find the resolvent kernel of the following kernels by using
the Fredholm determinants:
K(x, t) = 1 + 3xt, 0 x 1, 0 t 1
(a)
K(x, t) = x2t – xt2, 0 x 1, 0 t 1
(b)
Solution: (a) Here, B0(x, t) = K(x, t) = 1 + 3xt. Now,
1
K ( x, t ) K ( x, z1 )
B1 ( x, t ) dz
K ( z1 , t ) K ( z1 , z1 ) 1
0
1 1 3 xt 1 3 xz1
B1 ( x, t ) dz1
1 3z1t 1 3z12
0
3( x t )
B1 ( x, t )
3 xt 1 (Upon simplification)
2
1 K ( x, t ) K ( x, z1 ) K ( x , z2 )
Also, B2 ( x, t ) K ( z1 , t ) K ( z1 , z1 ) K ( z1 , z2 ) dz1 dz2
0 K ( z2 , t ) K ( z2 , z1 ) K ( z2 , z2 )
1 3 xt 1 3 xz1 1 3 xz2
11
B2 ( x, t ) 1 3z1t 1 3z12 1 3z1 z2 dz1 dz2
00
1 3z2 t 1 3z2 z1 1 3z22
B2(x, t) = 0 (Upon simplification)
Hence, Bp(x, t) = 0 for all p 2.
1 1
K (z1 , z1 )dz1
(1 3z1 ) dz1
2
Next, m1 2
0 0
11
K ( z1 , z1 ) K ( z1 , z2 )
m2 dz dz
K ( z2 , z1 ) K ( z2 , z2 ) 1 2
00
11
1 3z12 1 3z1 z2
m2 dz1dz2
00 1 3z2 z1 1 3z22
1
m2 , (upon simplification)
2
Also, m3 = 0 (can be verified by readers)
Hence, mp = 0 for all p 3.
134 Integral Equations
( l )m
Now, we have , t; l ) K ( x , t )
D( x Bm ( x, t )
m 1 m !
3( x t )
D( x, t; l )
1 3 xt l 3 xt 1
2
( l )m
and D(l ) 1 mm
m 1 m !
l2 l2
D(l ) 1 lm1
m2 1 2l
2! 4
The Fredholm resolvent kernel is given by
D( x , t; l )
R( x , t; l )
D( l )
3( x t )
1 3 xt l 3 xt 1
2
R( x , t; l )
1 2l (1 / 4)l 2
(b) Here, B0(x, t) = K(x, t) = x2t – xt2. Then, as before
1
K ( x, t ) K ( x, z1 )
B1 ( x, t ) dz
K ( z1 , t ) K ( z1 , z1 ) 1
0
1
x 2 t xt 2 x 2 z1 xz12
B1 ( x, t ) dz1
0 z12 t z1t 2 z12 z1 z1 z12
x t xt 1
B1 ( x, t )
xt
4 3 5
Also, B2(x, t) = 0
Hence, Bp(x, t) = 0 for all p 2.
1 1
Next, m1 K ( z1 , z1 )dz1 ( z13 z13 )dz1 0
0 0
11
K ( z1 , z1 ) K ( z1 , z2 )
m2 dz dz
K ( z2 , z1 ) K ( z2 , z2 ) 1 2
00
11
0 z12 z2 z1z22
m2 dz1dz2
00 z22 z1 z2 z12 z22 z2 z2 z22
1
m2 , (Upon simplification).
120
Classical Fredholm Theory 135
Also, we find m3 = 0.
Hence, mp = 0 for all p 3.
Now, we have
( l )m
, t; l ) K ( x , t )
D( x Bm ( x, t )
m 1 m !
D( x , t; l ) K ( x, t ) l B1 ( x, t )
x t xt 1
D( x, t; l ) x 2 t xt 2 l xt
4 3 5
( l )m l2
and 1
D( l ) mm
1 lm1 m2
m 1 m ! 2!
l2
D(l ) 1
240
x t xt 1
x 2 t xt 2 l xt
D( x , t; l ) 4 3 5
Finally, R ( x , t; l )
D( l ) 2
1 (l /240)
EXAMPLE 7.3: Using the recurrence relations, find the resolvent kernels of
the following kernels:
(a) K(x, t) = x – 2t; 0 x 1, 0 t 1
(b) K(x, t) = sin x cos t; 0 x 2p, 0 t 2p
(c) K(x, t) = 4xt – x2; 0 x 1, 0 t 1
Solution: (a) Here, K(x, t) = x – 2t.
The resolvent kernel R(x, t; l) is given by Eq. (7.4) as
D( x , t; l )
R( x , t; l )
D(l ) (i)
( l )m
, t; l ) K ( x , t )
where by Eq. (7.5), D( x Bm ( x, t ) (ii)
m 1 m !
( l )m
and by Eq. (7.6), D(l ) 1 m m (iii)
m 1 m !
1
and Bp ( x, t ) m p K ( x, t ) p K ( x, z ) Bp 1 ( z, t ) dz , p 1
(vi)
0
Letting p = 1 in Eq. (v), we obtain
1 1
1
B0 (s, s) ds
m1 (s 2s)ds
2
0 0
1 2 1
B1 ( x, t ) ( x 2t ) ( x 4t ) 2 xt
2 3 2
2
B1 ( x, t )
2 xt x t
3
Now, taking p = 2 in Eq. (v), we obtain
1 1
2
B1 (s, s) ds 3 2s
2
m2 2s ds
0 0
1
2 2s3 1
m2 s s2
3 3 0 3
Next, taking p = 2 in Eq. (vi), we obtain
1
B2 ( x, t ) m2 K ( x, t ) 2 K ( x, z ) B1 ( z, t ) dz
and
0
1
1 2
B2 ( x, t
) ( x 2t ) 2 ( x 2 z ) 2 zt z t dz
3 3
0
1
1 2x 4
B2 ( x, t ) ( x 2t ) 2 2 xzt xz xt z 4 z 2 t 2 z 2 2 zt dz
3 0
3 3
Classical Fredholm Theory 137
1
1 2 xz xz 2 2 4 2
B2 ( x, t ) ( x 2t ) 2 xtz 2 xtz z 2 z 3t z 3 z 2 t
3 3 2 3 3 3
0
1 4x 4 8 4
B2 ( x, t ) ( x 2t ) 2 xt x 2 xt t 2t
3 3 3 3 3
B2 ( x, t ) 0
Since B2(x, t) = 0, therefore by Eq. (vi),
Bp(x, t) = 0 and mp = 0 for all p 3.
Substituting the above values in Eq. (ii) and (iii), we have
D( x , t; l ) K ( x, t ) l B1 ( x, t )
2
D( x, t; l ) x 2t l 2 xt x t )
3
l2
and D( l )
1 lm 1 m
2! 2
l l2
D(l ) 1
2 6
2
x 2t l 2 xt x t
3
Thus, by Eq. (i), R( x, t; l )
l l2
1
2 6
(b) Here, K(x, t) = sin x cos t.
Now, B0(x, t) = K(x, t) = sin x cos t
2p
Also, m0 = 1, m p Bp1 (s, s)ds, p 1 (vii)
0
2p
and Bp ( x, t ) m p K ( x, t ) p K ( x, z ) Bp 1 ( z, t )dz , p 1
(viii)
0
Let p = 1, in Eq. (vii). We have
2p 2p
m1
B0 (s, s)ds sin s cos s ds
0 0
2p 2p
1 1 cos 2s
m1
2 0
sin 2 s ds 0
2 2 0
Now, taking p = 1 in Eq. (viii), we obtain
138 Integral Equations
2p
B1 ( x, t ) m1K ( x, t ) K ( x, z)B0 (z, t ) dz
0
2p
B1 ( x, t ) (sin x cos z )(sin z cos t )dz
0
2p
sin x cos t sin z cos z dz
B1 ( x, t ) 0
0
D( x
, t; l ) K
( x, t ) sin x cos t
and D(l) = 1
Hence, R(x, t; l) = sin x cos t
(c) Here, K(x, t) = 4xt – x2.
We have B0(x, t) = K(x, t) = 4xt – x2
1
m0 = 1 and m p Bp 1 (s, s)ds , p 1 (ix)
0
1
and Bp ( x, t ) m p K ( x, t ) p K ( x, z ) Bp 1 ( z, t )dz , p 1
(x)
0
Taking p = 1 in Eq. (ix), we obtain
1 1
m1 B0 (s, s)ds (4s 2 s 2 )ds [ s 3 ]10 1
0 0
3 0
1
B1 ( x, t ) 4 xt x 2 x ( x 2 16 xt ) 2 x 2 t
3
4 4
B1 ( x, t) 2 x 2 t x 2 x xt
3 3
Classical Fredholm Theory 139
Next taking, p = 2 in Eq. (ix), we obtain
1 1
4 2 4
B1 (s, s 2s
3
m2 )ds s s s 2 ds
3 3
0 0
1
s 4 4
s 2 4s3 1
m2 s 3
2 9 2 9 0 9
and now putting p = 2 in Eq. (x), we obtain
1
B2 ( x, t ) m2 K ( x, t ) 2 K ( x, z ) B1 ( z, t ) dz
0
1
1 4 4
B2 ( x
, t) (4 xt x 2 ) 2 (4 xz x 2 ) 2 z 2 t z 2 z zt dz
9 3 3
0
1
1 4 4t
B2 ( x, t
) (4 xt x 2 ) 2 (4 xz x 2 ) z 2 2t z 1 dz
9 0 3 3
1
1 4 4t 4 4t
B2 ( x, t ) (4 xt x 2 ) 2 4 x 2t z 3 z 2 4 x 1 x 2 2t x 2 1 z dz
9 3 3 3 3
0
1
1 4 z 3 4t 4 x 2 4t
B2 ( x, t ) = (4 xt − x 2 ) − 2 x 2t − z 4 + 4 x 1 − − x 2 2t − − 1 − z 2
9 3 3 3 3 2 3 0
1 4 1 4t 4 x2 4t
B2 ( x, t
) (4 xt x 2 ) 2 x 2t 4 x 1 x 2 2t 1
9 3 3 3
3 2 3
B2(x, t) = 0
Since B2(x, t) = 0, therefore Bp(x, t) = 0 and mp = 0 for all p 2.
Substituting these values in Eq. (ii) and (iii), we have
4 4
D( x, t; l ) K ( x, t ) l B1 ( x, t ) 4 xt x 2 l 2 x 2 t x 2 x xt
3 3
l2 l2
and D(l ) 1 lm1 m2 1 l
2! 18
4 4
4 xt x 2 l 2 x 2 t x 2 x xt
3 3
and thus, R ( x, t; l )
1 l (l 2 / 18)
EXAMPLE 7.4: Determine D(l) and D(x, t; l) for the following kernels for
the prescsibed limits a and b:
(a) K(x, t) = 1; a = 0, b = 1
(b) K(x, t) = sin x; a = 0, b = p
140 Integral Equations
B0 (s, s) ds
µ1 sin s ds
[ cos s ]p0
2
0 0
Classical Fredholm Theory 141
Taking p = 1 in Eq. (vii), we obtain
p
B1 ( x, t ) µ1K ( x, t ) K ( x, z ) B0 ( z, t ) dz
0
p
( x, t ) 2 sin x sin x sin z dz
B1
0
B1 ( x, t ) 2 sin x 2 sin x 0
D ( x , t; l )
K ( x, t )
sin x, and D (l )
1 l µ1
1 2l
0 0 3 0 3
142 Integral Equations
D( x
, t; l ) K
( x, t ) xt
103
D( l ) 1 lm1 1 l
3
xt
R ( x , t; l )
1 l (103 /3)
Thus, the required solution of Eq. (i) is given by
10
( x ) f ( x ) l R( x, t; l ) f (t ) dt
g
0
10
lx
) ex te dt
t
g( x 3
1 l (10 / 3)
0
lx
ex
g( x ) [tet et ]10
0 (Integrating by parts)
1 l (103 / 3)
3l x
g( x ) e x (10e10 e10 1) (Upon simplification)
3 1000 l
3l x
ex
g( x ) (1 9e10 )
3 1000 l
(b) Here, we have a = 0, b = 1 and K(x, t) = x + t
Again, B0(x, t) = K(x, t) = x + t
1
m0 = 1 and m p Bp 1 (s, s)ds , p 1 (iv)
0
1
and Bp ( x, t ) m p K ( x, t ) p K ( x, z ) Bp 1 ( z, t )dz (v)
0
1 1
m1 B0 (s,
s)ds 2
s ds s 2 ]10 1
[
0 0
1 1 1 1 1 1 x t 1
B2 ( x, t ) ( x t ) 2 t t xt x
6
3 2
2 2 3 2 2 3
B2 ( x, t ) = 0 (Upon simplification)
144 Integral Equations
1 1
D( x, t; l ) x t l ( x t ) xt
2 3
l2 1
and D(l ) 1 lm1 m2 1 l l 2
2! 12
1 1
x t l ( x t ) xt
2 3
and thus, R ( x , t; l ) 2
1 l (l / 12)
Hence, the solution of the integral equation is given by
1
( x ) f ( x ) l R( x, t; l ) f (t )dt
g
0
1 1
( x t ) l ( x t ) xt
1
( x) f ( x) l
g 2 3
f (t )dt .
or 2
0 1 l ( l / 12)
K ( x, t ) xt ( xt )1/2 , m0
B0 ( x, t ) 1
1 1 1
s3 s 2 5
and m1 B0 (s, s) ds (s s) ds
2
0 0 3 2 6
0
Classical Fredholm Theory 145
1
B1 ( x, t ) m1K ( x, t ) K ( x, z ) B0 ( z, t ) dz
Next,
0
1
5
B1 ( x, t ) {xt ( xt )1/2 } {xz ( xz )1/2 }{zt ( zt )1/2 }dz
6
0
1
5 xtz 3 z 5/2 z 2
B1 ( x, t ) {xt ( xt )1/2 } (x t t x ) ( xt )1/2
6 3 5/2 2
0
5 xt 2 1
B1 ( x, t ) {xt ( xt )1/2 } ( x t t x ) ( xt )1/2
6 3 5 2
1 1 2
B1 ( x, t ) xt ( xt )1/2 ( x t t x )
2 3 5
1
Then, m2 B1 (s, s) ds
0
1
1 1 2
2 s
2
m
2 s (s s s s) ds
3 5
0
1
1 1 4
2 s
2
m
2 s s 3/2 ds
3 5
0
1
s 3 s 2 4 s 5/2 1
m2
6 6 5 5 / 2 75
0
1
B2 ( x, t ) m2 K ( x, t ) 2 (K ( x, z ) B1 ( z, t ) dz
Further,
0
1
1 1 1 2
B2 ( x, t ) {xt ( xt )1/2 } 2 {xz ( xz )1/2 } zt ( zt )1/2 ( z t t z ) dz
75 0 2 3 5
1 xt 2 x t 2 x t 4 xt t x ( xt )1/2 4
B2 ( x, t ) {xt ( xt )1/2 } 2 ( xt )1/2 t x
75 6 15 15 25 5 6 25
B2 ( x, t ) = 0 (Upon simplification)
146 Integral Equations
D( x
, t; l ) K ( x, t ) l B1 ( x, t )
1 1 1
xt ( xt )1/2 l xt ( xt )1/2 ( x t t x )
D( x , t; l )
2 3 2
and
l2 5 1 2
D (l )
1 l c1 c2
1 l l
2! 6 150
Therefore, the resolvent kernel is given by
1 1 2
xt ( xt )1/2 l xt ( xt )1/2 ( x t t x
2 3 5
R ( x , t; l )
1 (5/6)l (1/150)l 2
Hence, the required solution of the given equation is
1
( x ) f ( x ) l R( x, t; l ) f (t ) dt
g
0
1 1 2
xt ( xt )1/2 l xt ( xt )1/2 ( x t t x )
1
2 3 5 dt
or g( x ) x l 2
0 1 (5 / 6)l (1 / 150)l
l
g( x ) x
1 (5 / 6)l (1/150)l 2
1
xt 3 2 xt 5/2 l xt 3 2 x l t 5/2 4 x l t 5/2 2 x l t 3
3 5 6 15 25 15
0
l x 2 x l x 2l x 4 x l 2l x
g( x ) x 2 3
1 (5/6)l (1/150)l 5 6 15 25 15
1
( x ) f ( x ) l R( x, t; l ) f (t ) dt
g
0
2
1 2x t l x t 2 xt
x 3 t
or g( x
) l dt
6 0 1 (l / 2) (l 2 / 6) 6
1
x l 2
6 l 2 3l 6 0
2 2
g( x ) 2 x t l t xt t 2 xt dt
3
1
x l 2 t3 1 2 xt 2 t 3 xt 3
g( x ) 2 xt l t
6 l 3l 6 3 3 2 3 3
0
x l 1 1 x 1 x
g( x ) 2 x l
6 l 3l 6 3 3 2 3 3
x l 1 lx
g( x ) 2 x
6 l 3l 6 3 6
1 l
g( x
) x 2 (6 x 2 l x )
6 l 3l 6
1 l (6 x 2) l 2 x
g
( x) x
6 l 2 3l 6
where {xi} and {ti}, (i = 1,2,...,n) are two sequences of arbitrary variables,
Eq. (7.47) converges for all values of l, and hence, it is an entire function of l.
Proof: First, it will be proved that every zero of D(l) is a pole of the
resolvent kernel R(x, t; l) given by
D( x , t; l )
R( x , t; l ) (7.48)
D( l )
The order of its pole is at most equal to the order of the zero of D(l).
The Fredholm’s first series is given by
(l ) j
b b x1 ,..., x j
D(l ) 1 ... K dx1 ...dx j (7.49)
j 1 j ! a a x1 ,..., x j
and the Fredholm's second series is given by
(l ) j
b b x, x1 ..., x j
, t; l ) K ( x , t )
D( x ... K dx1 ...dx j (7.50)
j 1 j ! a a t , x1 ,..., x j
Differentiating both sides of Eq. (7.49) with respect to l and interchanging
the indices of the variables of integration, it can be expressed as
b
D1 (l ) D( x, x; l ) dx (7.51)
a
From this relation, it follows that if l0 is a zero of order p of D(l), then it is
a zero of order (p – 1) of D¢(l), and consequently, l0 may be a zero of order
at most (p – 1) of the entire function D(x, t; l). Thus, l0 is the pole of order
at most p, particularly, if l0 is a simple pole of D(l), then D(l0) = 0, D¢(l0)
≠ 0 and l0 is a simple pole of the resolvent kernel. Moreover, it follows
from Eq. (7.51) that D(x, t; l) ≠ 0. In this particular case, it is observed from
the following equation:
b
D( x, t; l ) K ( x, t ) D(l ) l K ( x, z ) D( z, t; l ) dz
a
K ( x1 , t1 ) K ( x1 , t2 )... K ( x1 , tn ) K ( x1 , z1 )... K ( x1 , z j )
K ( x2 , t1 ) K ( x2 , t2 )... K ( x2 , tn ) K ( x2 , z1 )... K ( x2 , z j )
K ( xn , t1 ) K ( xn , t2 )... K ( xn , tn ) K ( xn , z1 )... K ( xn , z j ) (7.56)
K ( z1 , t1 ) K ( z1 , t2 )... K ( z1 , tn ) K ( z1 , z1 )... K ( z1 , z j )
K ( z j , t1 ) K ( z j , t2 )... K ( z j , tn ) K ( z j , z1 )... K ( z j , z j )
150 Integral Equations
in terms of the elements of the first row and integrating j times with respect
to z1, z2, ..., zj for j 1, we obtain
b b x1 ,..., xn , z1 ,..., z j n
1t ,..., t , t
l 1 l 1 ,..., t ,
n 1z ,..., z j l 1
a a
b b x2 ,..., xn , z1 , z2 ,..., zl ,..., z j
... K ( x1 , zl )K dz1 ...dz j (7.57)
a a t1 ,..., tn 1 , tn , z1 , z2 ,..., zl 1 , zl 1 ,..., z j
Here, it is noted that the symbols for the determinant K on the right-hand
side of Eq. (7.57) do not contain the variable xi in the upper sequence and the
variables t1 or zl in the lower sequence. Further, by transposing the variable xl
in the upper sequence to the first place by means of l + n – 2 transpositions,
it is followed that all the components of the second sum on the right side are
equal. Therefore, Eq. (7.57) can be writtern as
b b x1 ,..., xn , z1 ,..., z j n
x1 , x2 ,..., xn n x ,..., xn
Dn l
(1)l 1 K ( x1 , tn )Dn 1 t ,..., t 2
t1 , t2 ,..., tn l 1 1 l 1 , tl 1 ,..., t n
b
z, x2 ,..., xn
l K ( x1 , z )Dn l dz (7.59)
a t1 , t2 ,..., tn
Expanding the determinant shown in Eq. (7.56) with respect to the first column
and proceeding as above, we obtain the integral equation:
Classical Fredholm Theory 151
The relations shown in Eqs. (7.59) and (7.60) hold for all values of l.
With the help of (7.59), one can find the solution of Eq. (7.45) for the special
case when l = l0 is an eigenvalue of the kernel, by supposing that l = l0
is a zero of multiplicity m of D(l). Then, as mentioned earlier, the minor
Dm ≠ 0 and even the minors D1, D2 ..., Dm–1 may not identically vanish.
Suppose Dr is the first minor in the sequence D1, D2 ..., Dm–1 such that Dr ≠ 0.
Then number r must be between 1 and m and is the index of eigenvalue l0.
It thus follows that Dr–1 = 0. Then Eq. (7.59) shows that
x1 , x2 ,..., xr
g1 ( x ) Dr l0 (7.61)
t1 , t2 ,..., tr
is a solution of Eq. (7.45). Substituting x at different points of the upper
sequence in the minor Dr, we obtain r non -trivial solutions gi(x), (i = 1,2, ..., r)
of Eq. (7.45) which are often written as
x1 ,..., xi 1 , x, xi 1 ,..., xr
D l0
t1 ,................., tr
fi ( x ) ; (i 1,2,..., r ) (7.62)
x1 ,..., xi 1 , xi , xi 1 ,..., xr
Dr l0
t1 ,................., tr
in which the denominator is non–zero.
We now establish that solutions fi given by Eq. (7.62) are linearly independent.
In the determinant shown in Eq. (7.56), if we put two of the arguments xi equal,
this is equivalent to having two rows equal, and consequently, the determinant
vanishes. Thus in Eq. (7.62), we see that fk(xi) = 0 for i ≠ k, whereas fk (xk) = 1.
Now, if we have a relation of the form ck fk 0 , then putting x = xi, we
k
get ci = 0 and so the solution fi are linearly independent. This system of
solutions fi is known as fundamental system of eigenfunctions of l0, and any
linear combination of these functions gives a solution of Eq. (7.45).
Conversely, it can be shown that any solution of Eq. (7.45) must be a
linear combination of f1(x), f2(x), ..., fv(x). For this purpose, we define a kernel
H(x, t; l), which corresponds to the resolvent kernel R(x, t; l) of Section 7.2.
x, x1 ,..., xr
Dr 1 l0
t , t1 ,..., tr
H ( x , t; l ) (7.63)
x1 ,..., xr
Dr l0
t1 ,..., tr
152 Integral Equations
x, x1 ,..., xr x1 ,..., xr r
D
r 1 l 0 K ( x , t ) Dr 1 l0 ( 1)l K ( xl , t )
t , t1 ,..., tr t1 ,..., tr l 1
b
x, x1 ,..., xl 1 , xl 1 ,..., xr x, x1 ,..., xr
Dr l0 l0 K ( z, t ) Dr 1 l0 (7.64)
t1 , t2 ..., tr a z, t1 ,..., tr
In each minor Dr, of Eq. (7.64), we transpose the variable x from the first
place to the place between the variables xl–1 and xl+1 and divide both sides
by the constant
x1 ,..., xr
Dr l0 0
t1 ,..., tr
to get
b r
K ( xl , t ) fl ( x ) (7.65)
H ( x, t; l ) K ( x, t ) l0 H ( x, z; l )K ( z, t )dz
a l 1
Now suppose that g(x) is an arbitrary solution of Eq. (7.45). Multiplying both
sides of Eq. (7.65) by g(t), and then, integrating both sides with respect to t
from a to b, we get
b b r
g( x ) g( xl )
g(t ) H ( x, t; l ) dt
g( z ) H ( x, z; l ) dz fl ( x ) (7.66)
a
l0 a l 1 l 0
where we have used Eq. (7.45) in all terms except the first. Further, we have
also taken
b
K ( xl , t )g(t )dt g( xl )
l0
a
Cancelling the equal terms in Eq. (7.66), we obtain
r
g( x ) g( xl ) fl ( x )
l 1
It, thus, completes the proof.
x, x1 , x2 ,..., xr
b
Dr 1 l 0
t , t1 , t2 ,..., tr r
f ( x) l
g( x ) f (t )dt clfl ( x ) (7.68)
a
x1 , x2 ,..., xr l 1
Dr l0
t1 , t2 ,..., tr
b
( x ) f ( x ) l K ( x, t ) g(t )dt .
Proof: Consider g
a
b b t , z ,..., z
(l ) j
, x; l ) K (t , x )
1 j
and D(t
j 1
... K dz ....dz j (7.71)
j ! a a x, z1 ,..., z j 1
respectively. From this fact, it follows that kernels of Eq. (7.67)and its transpose
in Eq. (7.69) have the same eigenvalues. Further, the corresponding resolvent
kernel for Eq. (7.69) is
R(t , x; l ) D(t , x; l )/D(l ) (7.72)
and hence, the solution of Eq. (7.69) is
b
D(t , x; l )
( x) f ( x) l
h f (t ) dt (7.73)
a
D( l )
provided l is not an eigenvalue.
Next, it is obvious that not only the transposed kernel has the eigenvalues
as the original kernel of Eq. (7.67) but also the index r of each of the
eigenvalues is equal. Moreover, the eigenfunctions of the transposed equation
for an eigenvalue l0 are given by
x1 ,...., xr
Dr
t1 ,..., ti 1 , t , ti 1 ,..., tr
hi (t ) (7.74)
x1 ,...., xr
Dr
t ,..., t , t , t ,..., t
1 i 1 i i 1 r
where the values (x1, ...,xr) and (t1, ..., tr) are to be so chosen that the denominater
does not vanish. Substituting r in different places in the lower sequence of
154 Integral Equations
f ( x )h p ( x ) dx g( x)hp ( x) dx l K ( x, t )g(t )hp ( x)dx dt
a a aa
b b b
f ( x )h p ( x ) dx p
g ( x ) h ( x ) l K (t, x)hp (t ) dt dx (7.75)
a a a
Since hp(x) is an eigenfunction of the transposed homogeneous equation,
therefore
b
h p ( x ) l K (t , x )h p (t ) dt (7.76)
a
where λ is the corresponding eigenvalue. Now, from Eqs. (7.75) and (7.76),
we have
b
f ( x)hp ( x) dx 0 (7.77)
a
Thus, it follows that a necessary condition for Eq. (7.77) to have a solution
is that the non-homogeneous term f(x) be orthogonal to each solution of the
transposed homogeneous equation.
Conversely, now it will be proved that the condition Eq. (7.77) of
orthogonality is sufficent for the existence of a solution. In what follows, we
shall also obtain an explicit solution in such a case. At this stage, we define
a kernel H(t, x; l) as follows:
x, x1 ,...., xr
Dr 1 l0
t , t1 ,..., tr
H (t , x; l ) (7.78)
x1 ,...., xr
Dr l0
t1 ,..., tr
wherein it is assumed that Dr ≠ 0 and that r is the index of eigenvalue l0.
To prove the required result, we show that if the orthogonality condition is
satisfied, then the function
b
( x ) f ( x ) l0 H ( x, t; l ) f (t ) dt (7.79)
g0
a
is a solution.
Substituting this value for g(x) in Eq. (7.67), we have [by referring to Eq. (7.65)]
b b b
f ( x ) l0 H ( x, t; l ) f (t ) dt f ( x ) l0 K ( x, t ) f (t ) l0 H (t , z; l ) f ( z ) dz dt
a a
a
Classical Fredholm Theory 155
b b
or f (t ) dt H (t , z; l ) K ( x, t ) l0 K ( x, z ) H ( z, t; l )dz 0 (7.80)
a a
Proceeding similarly, we obtain its transpose as
b r
K ( x, t p )h p (t )
H ( x, t; l ) K ( x, t ) l0 K ( x, z ) H ( z, t; l )dz
p 1
a
EXERCISE 7.1
compute D(l) and D(x, t; l) for the following kernels for the prescribed
limits a and b:
(a) K(x, t) = sin (x + t), a = 0, b = p
(b) K(x, t) = ex–t, a = 0, b = 1
4. Determine the resolvent kernel, and hence solve the following integral
equations:
1
) e x e x t g(t ) dt
(a) g( x
0
2p
(b)
g( x ) cos 2 x sin x cos t g(t ) dt
0
156 Integral Equations
Answers:
sin x sin t lp (1 2 sin x sin t )
1. (a) R( x, t; l )
1 4p 2 l 2
2 e x et
(b) R( x, t; l )
1 l (e2 1)
x + t + 1 + 2λ ( xt + 1 / 3)
2. (a) R( x, t ; λ ) =
1 − 2λ − (4 / 3)λ 2
sin( x t ) pl cos( x t )
(b) R( x, t; l )
1 p 2l2
3. (a) D( x, t ; λ ) = sin( x + t ) + ( λπ /2) cos( x − t ), D( λ ) = 1 − (π 2 /4)λ 2
(b) D( x, t; l ) x, D(l ) 1 42l
1 x
4. (a) g ( x ) = e
2
(b) g( x ) cos 2 x
Chapter 8
8.1 INTRODUCTION
The integral transform methods provide a useful tool for the solution of integral
equations of various special forms. Let the following double integral exist:
bb
g( x ) F ( x, t1 )K (t1 , t ) g(t ) dt dt1 (8.1)
aa
This double integral can be evaluated as an iterated integral.
b
If we take f ( x ) K ( x, t ) g(t ) dt, (8.2)
a
1
e
ist
and has g(s) f (t ) dt
2p
The function f(s) is known as Fourier transform of g(t) and g(s) as the
inverse Fourier transform of f(t), and vice versa.
2
2. Considering the double integral g(s)
p (sin sx sin xt ) g(t ) dx, we
0 0
2
p 0
find that this leads to the sine transform f (s) (sin st ) g(t ) dt and its
2
p 0
inverse as, g(s) (sin st ) f (t ) dt, respectively.
f ( x ) e xt g(t ) dt (8.5)
0
In Eqs. (8.4) and (8.5), the range of integration is infinite, while in integral
x
g(t )
f ( x) dt (8.6)
0 x t
the range of integration is finite, but the kernel is discontinuous.
The Abel’s Integral Equation
One of the simplest form of singular integral equation, which appears in
mechanics, is the Abel’s integral equation, given as,
x
g(t )
f ( x) ( x t )a dt, 0 a 1 (8.7)
0
e
px
L{ f ( x
); p} F
( p) f ( x ) dx (8.8)
0
provided that the integral exists. It is to be recollected that the Laplace transform
of f(x) exists if the integral in Table 8.1 shows Laplace transform for same
elementary functions. Eq. (8.8) is convergent for some designated values of p.
L{c1 f1 ( x ) cn
f n ( x ); p} c1 L{ f1 ( x ); p} cn L{ f n ( x ); p}
L{c1 f1 ( x ) cn f n ( x );
p} c1F1 ( p) cn Fn ( p)
* For details of Laplace transform and its inverse, the reader can refer to the recent book
on Integral Transforms published by RBD.
160 Integral Equations
10. Final value theorem: Let f(x) be continuous for all x 0 and be of
exponential order as x ¥. Also, suppose that f¢(x) is of class A. Then
lim f ( x ) lim pL{ f ( x ); p}
x p 0
f ( x ) L1{F ( p); x}
where L–1 is known as the inverse Laplace transformation operator.
Table 8.2 shows inverse Laplace transform of some elementary functions.
L1{c1F1 ( p) cn Fn ( p);
x} c1 f1 ( x ) cn f n ( x )
x
d2g
g( x ) sin x cos( x u) g(u) du
dx 2
0
L
{g( x ); p} L{ f ( x ); p} L{K ( x ) * g( x )}
or G(p) = F(p) + K(p)G(p) [using Eq. (8.11) and the convolution theorem]
F ( p)
or G( p) (8.14)
1 K ( p)
Integral Transform Methods 165
Now, applying the inverse Laplace transform to both sides of Eq. (8.14),
we obtain.
F ( p)
g( x ) L1 ; x
1 K ( p)
x
or g( x ) f ( x ) R( x t ) f (t ) dt (8.20)
0
Applying Laplace transform to both sides of Eq. (8.20), we get
L
{g( x ); p} L{ f ( x ); p} L{R( x ) * f ( x )}
or G
( p) F ( p) R( p) F ( p) (8.21)
{g( x ); p}, F ( p) L{ f ( x ); p} and R( p) L{R( x ); p} (8.22)
where, G( p) L
Using Eq. (8.14) for G(p) in Eq. (8.22), we get
F ( p)
F ( p)[1 R( p)]
1 K ( p)
1 K ( p)
or
R( p) 1 (8.23)
1 K ( p) 1 K ( p)
Now, applying the inverse Laplace transform to both sides of Eq. (8.23),
we get
K ( p)
L1
R( x t )
1 K ( p)
K ( x t ) g(t ) dt,
2 2
Let f ( x) ( x 0) (8.24)
0
1 1/2 1/2 1/2
1/2
x u , t v1/2 , g
1 ( v) v g(v ), and f1 (u) f (u ) (8.25)
2
Then, Eq. (8.24) takes the form
u
K (u v) g1 (v) dv,
f1 (u) (u 0)
0
Solution: (i) The given integral equation is of convolution type, and therefore
expressing it as:
f(x) = g(x)*x–a (i)
Taking Laplace transform of both sides of Eq. (i) and applying the
convolution theorem, we have
(1 a )
or F ( p) G( p)
p1a
p1a F ( P) p
or
G( p) [ (a ) p a F ( p)]
(1 a ) (a )(1 a )
p
G ( p) [ (a ) p a F ( p)]
(p / sin pa )
(∵ (a ) (1 a ) p / sin pa )
p sin pa
G( p) L{x a 1 * f ( x )},
p
sin pa x
G( p) pL ( x t )a 1 f (t ) dt (By convolution theorem) (ii)
p 0
168 Integral Equations
x
Let h
( x) ( x t )a 1 f (t ) dt (iii)
0
Now, L h '( x
); p pL {h( x ); p}
h(0) pL{( x ); p} {∵h(0) 0}
x
or pL ( x t )a 1 f (t ) dt; p L{h '( x ); p} [by Eq. (iii)] (iv)
0
Using Eq. (iv) in Eq. (ii), we get
sin pa
G( p)
L{h '( x ); p}
p
Now, taking inverse, we have
x
sin pa d ( x t )a 1 f (t ) dt
g( x ) L 1
{G( p); x}
p dx 0
(1 / 2) 1 1 2!
or G( p) 1/2
2 3
p p p p
1 1 1 2
or G
( p) 1/2 3/2 5/2 (vi)
p p p p
Now, applying the inverse Laplace transform to both sides of Eq. (vi),we
obtain
1 1 1 1 1 1
1
g( x) L 1/2 L 3/2 2 L 5/2
p p p p
1 x 1/2 x1/2 2 x 3/2
g
( x)
p (1 / 2) (3 / 2) (5 / 2)
1 1/2
g( x
) [x 2 x1/2 (8 / 3) x 3/2 ], (Upon simplification)
p
x
EXAMPLE 8.2: Solve the integral equation
sin x J0 ( x t ) g(t ) dt.
0
Integral Transform Methods 169
Solution: The given integral equation may be expressed in convolution form
as
sin
x = g(x)*J0(x) (i)
Taking Laplace transform of both sides of Eq. (i) and using the convolution
theorem, we have
L{sin x} L{g( x )} L{J 0 ( x )}
1 1
or 2
G( p)
p 1 p 12
1
or G( p) =
p2 + 1
Now, taking the inverse Laplace transform, we have
1
g( x ) L {G( p); x} J 0 ( x )
x
EXAMPLE 8.3: Solve the integral equation x e
x t
g(t ) dt .
0
Solution: The integral equation may be written in convolution form as
x = ex * g(x) (1)
Now, taking Laplace transform of both sides of Eq. (i) and using the
convolution theorem, we get
L{x} L{e x * g
( x )} L{e x } L{g( x )}
1 1
or G( p)
p 2 p 1
p 1 1 1
or G( p
) (ii)
p2 p p2
Taking Laplace transform of both sides of Eq. (i) and using the convolution
theorem, we get
L{g(
x )} L{1} L{g( x )} L{sin x}
1 1
or G( p
) G ( p) 2
p p 1
1 1
or 1 2 G ( p)
p 1 p
p2 1 1 1
or G( p
)
p3 p p3
Applying Laplace inversion, we get
x2 x 2 (ii)
g( x )
1 1
2! 2
Verification of solution: N
ow, we show that Eq. (ii) satisfies the given
integral equation.
(iii)
From Eq. (ii), g(x) = 1 + (x2/2)
Thus, the R.H.S. of Eq. (iii) provides
x
1 sin ( x t )(1 t 2 /2) dt
0
x
t 2 x
1 1 cos( x t ) t cos( x t ) dt
2 0 0
x2 x
11 cos x [ t sin( x t )]0 sin ( x t ) dt
x
2 0
x2
2 cos x [cos( x t )]0x
2
x2 x2
2 cos x (1 cos x ) 1 g( x );
2 2
= L.H.S. of Eq. (iii)
EXAMPLE 8.5: Solve the following Volterra integral equation of the first kind:
x
g( x )} L{e x } 2 L [ g( x )] L{cos x}
L{
1 p
or G
( p) 2G( p) 2
p 1 p 1
p2 1
or G( p)
( p 1)3
[( p 1) 1]2 1
or G( p) (ii)
( p 1)3
Now, applying inversion of Eq. (ii), we get
[( p 1) 1]2 1
g( x ) L1
( p 1)3
172 Integral Equations
( p 1) 1
2
g( x ) e x L1 (By first shifting theorem)
p3
p2 2 p 2 x 1 1 2 2
g( x ) e x L1
3
e L 2 3
p p p p
1 1 1
g( x ) e x L1 2 L1 2 2 L1 3
p p p
g( x ) e x [1 2 x 2( x 2 / 2!)]
g( x ) e x (1 2 x x 2 )
or g( x ) e x (1 x )2
(b) Rewriting the given integral equation, we have
1 3
g( x ) x
x * g( x ) (iii)
6
Applying Laplace transform to both sides of Eq. (iii) and using the
convolution theorem, we have
1
x )} L{x} L{x 3}. L{g( x )}
L{g(
6
1 1 3!
or G(
p) G( p)
p2 6 p 4
1 1
or G( p) 1 4
p p2
1 1 1
or
G ( p) 2 2
2 p 1 p 1
Now, applying inversion, we obtain
1
g( x ) (sinh x sin x ).
2
EXAMPLE 8.7: Solve the following integro-differential equations
x
'( x ) sin x g(t ) cos t dt , where g(0) = 0.
(a) g
0
x
(b) g '( x ) 5 cos 2( x t ) g(t )dt
10, where g(0) = 2.
0
Integral Transform Methods 173
Solution: (a) The given integral equation can be written as
g'(x) = sin x + g(x)*cos x (i)
Also, given that g(0) = 0. (ii)
Applying Laplace transform to both sides of Eq. (i) and using the
convolution theorem, we obtain
L
{g '( x )} L{sin x} L{g( x )}L{cos x}
1 p
or p G( p) g(0) 2
G( p) 2
p 1 p 1
1 1
or 1 p G( p) [Using Eq. (ii)]
2 2
p 1 p 1
1
or G( p) (iii)
p3
Now, inverting Eq. (iii) to get
1 x
2
x2
x ) L1 3
g( .
p 2! 2
(b) The given integral equation may be written as
g'(x) + 5[cos2x*g(x)] = 10 (iv)
Also, given that g(0) = 2. (v)
Applying Laplace transform to both sides of Eq. (iv) and using the
convolution theorem, we get
5 10
or 1 2 pG( p) p
2, [Using Eq. (v)]
p 4
2 p 10 p2 4 2 p3 10 p2 8 p 40
or G( p) 2
p2 p 9 p2 ( p2 9)
or 8 40 10 p 50 (vi)
G( p) 2
9p 9p 9( p2 9)
Now, taking inverse Laplace transform of Eq. (vi), we have
8 40 10 p 50
) L1 2
g( x
9 p 9 p 9( p2 9)
174 Integral Equations
8 1 40 1 1 10 1 p 50 1 1
g( x ) L1 L 2 L 2 L 2
9 p 9 p 9 p 9 9 p 9
8 40 10 50
g( x ) x cos3 x sin 3 x
9 9 9 27
1
or g( x ) (24 120 x 30 cos3 x 50 sin 3 x )
27
EXAMPLE 8.8: Find the resolvent kernel of the following Volterra integral
equations, and hence, find their solutions:
x
(a) g( x ) f ( x ) ( x t ) g(t ) dt
0
x
(b) g ( x ) f ( x ) e( x t ) g(t ) dt
0
Solution: (a) The given integral equation can be written as
g(x) = f(x) + g(x)*x (i)
Applying Laplace transform to both sides of Eq. (i) and using the
convolution theorem, we have
L ( g( x ) L{ f ( x )} L{g( x )} L{x}
or 1
G( p) F ( p) G( p) 2
p
1
or
1 2 G( p) F ( p)
p
p2
or G ( p) 2 F ( p) (ii)
p 1
Let R(x – t) be the resolvent kernel of the given integral equation. The
required solution is given by
x
g( x ) f ( x ) R( x t ) f (t ) dt (iii)
0
or g
( x ) f ( x ) R( x ) * f ( x ) (iv)
Applying Laplace transform to both sides of Eq. (iv) and using the
convolution theorem, we get
or ( p) F ( p) R( p) F ( p) [where R( p) L{R( x )} ]
G
p2
or F
( p) F ( p) R( p) F ( p) [Using Eq. (ii)]
p2 1
Integral Transform Methods 175
p2 1
or R
( p) 2
1 2
p 1 p 1
Now, inverting, R(x) = L–1{R(p)} = sinh x
so that R(x – t) = sinh (x – t)
giving the required resolvent kernel.
Now, substituting the above value of R(x – t), the required solution by
Eq. (iii) is
x
g( x ) f ( x ) sinh ( x t ) f (t ) dt
0
(b) The given integral equation can be written as
g(x) = f(x) + g(x)*ex (v)
Applying Laplace transform to both sides of Eq. (v) and using the
convolution theorem, we have
{g( x )} L{ f ( x )} L{g( x )}. L{e x }
L
1
or G( p) F ( p) G( p) G( p)
p 1
1
or 1 G ( p)
F ( p)
p 1
p 1
or G( p) F ( p) (vi)
p2
Let R(x – t) be the resolvent kernel of the given integral equation. Then,
we know that the required solution is given by
x
g( x ) f ( x ) R( x t ) f (t )dt (vii)
0
or g
( x ) f ( x ) R( x ) * f ( x ) (viii)
Applying Laplace transform to both sides of Eq. (viii) and using the
convolution theorem, we get
L
{g( x )} L{ f ( x )} L{R( x )}. L{ f ( x )}
or G
( p) F ( p) R( p) F ( p), where R( p) L{R( x )}
p 1
or F ( p) F ( p)[1 R( p)], [Using Eq. (vi) for G(p)]
p2
p 1 1
or R(
p)
1
p2 p2
176 Integral Equations
1 1 2x
Now inverting, R( x ) L e
p 2
Hence, e2( x t )
R( x t )
It is the required resolvent kernel.
And now, substituting the value of R(x – t), the required solution by Eq. (vii) is
x
( x ) f ( x ) e2( x t ) f (t ) dt
g
0
1 1/2 1/2
1/2
x u , t n 1/2 , g
1 (n ) n g(n ), and f1 (u) f (u1/2 ) (i)
2
Then, the given integral equation takes the form
u
g1 (u )
f1 (u ) = ∫ dv (ii)
0 (u − v )α
Now, taking Laplace transform of Eq. (ii), we get
1 1 sin ap a 1
h(x) L a ; x x
p a ) p
Remarks: (a) If the kernel of the Fourier transform is taken as eipx, then
Eqs. (8.31) and (8.32) take the following forms:
1
e e
ipx ipx
F ( p) f ( x ) dx and f ( x ) F ( p)dp
2p
(b) If the kernel of the Fourier sine transform is taken as sin px, then
Eqs. (8.33) and (8.34) take the forms as:
2
Fs ( p)
p 0
f ( x )sin ( px ) dx and f ( x ) Fs ( p)sin px dp
0
(c) If the kernel of the Fourier cosine transform is taken as cos px, then
Eqs. (8.35) and (8.36) take the forms as
2
Fc ( p)
f ( x ) cos( px ) dx and f ( x )
p Fc ( p) cos( px) dp
0 0
F{c1 f1 ( x ) cn
f n ( x ); p} c1F{ f1 ( x ); p} cn F{ f n ( x ); p}
c1F1 ( p) cn Fn ( p)
This property holds good for sine and cosine transforms also.
5. Change of scale property:
1
(a) If F{ f ( x ); p} F
( p), then F{ f (ax ); p} F ( p /a) .
a
1
(b) If Fs { f ( x ); p} Fs ( p), then Fs { f ( ax ); p} Fs ( p /a) .
a
1
(c) If Fc { f ( x ); p} F
c ( p), then Fc { f ( ax ); p} Fc ( p /a) .
a
6. Shifting property:
If F x a); p} eipa F ( p).
{ f ( x ); p} F ( p), then F{ f (
7. Convolution or faltung of two functions: The convolution of two
integrable functions f(x) and g(x), where –¥, < x < ¥, is expressed and defined
as
1
f *g
2p
f (u) g( x u) du
Integral Transform Methods 179
8. The convolution theorem for Fourier transforms–Statement: Let
(a) f(x) and g(x) and their first order derivatives be continuous on (¥, ¥),
(b) f(x) and g(x) be absolutely integrable on (¥, ¥),
(c) F(p) and G(p) be Fouier transforms of f(x) and g(x); respectively.
Then, the Fourier transform of the convolution of f(x) and g(x) exists and is
the product of the Fourier transforms of f(x) and g(x), i.e.,
F{ f * g} F{ f ( x ); p} . F{g( x ); p}
2
(1 p), 0 p 1
Then, Fc ( p) p
0, p 1
Hence, by the Fourier cosine inversion formula,we have
1
2 2 2
f ( x)
p Fc ( p
) cos px dp
p p
(1 p) cos px dp
0 0
2(1 cos x )
f(x)
p x2
This is the required solution.
2 1 cos x 2 sin 2 ( x / 2) p
p0 x 2
dx 1or x 2
dx
2
0
1 , 0 p 1
f ( x )sin px
dx 2 , 1 p 2
0 0 , p 2
2
Solution: Let
p f ( x)sin
px dx F
s { f ( x )} Fs ( p).
0
1 , 0 p 1
2
Then,
Fs ( p) 2 , 1 p 2
p
1 , p 2
Hence, by the Fourier sine inversion formula, we get
2
f ( x)
x Fs ( p)sin px dp
0
1 2
2 2 2
p 0
f ( x) 1.sin px dp 2sin px dp 0sin px dp
p1 p2
1 2
2 cos px 4 cos px
f ( x)
p x 0 p x 1
2 4
f ( x ) [ cos x 1] { cos 2 x cos x}
px px
2
f ( x) (1 cos x 2 cos 2 x )
px
EXERCISE 8.1
g(t )sin ( x t ) dt
g( x )
0
f ( x) cos px dx e
p
if
g(0) g
'(0) 0
Answers
1
g(t )
2.
2
x2
) x2
3. (a) g( x (b) g(x) = cos x
12
x
4. (a) g( x ) cxe
(b) g( x ) x 2 2( x 1)e x
(c) g(x) = 1
x x
(d) g( x ) J1 ( x ) J 0 (t )dt or g( x ) 2d ( x ) J1 ( x ) J 0 (t ) dt
0 0
x2 16 x 3/2
6. (a) g( x ) 1 (b) g ( x ) = ±
2 π
(c) g(x) = 0,
5 2 1 4
(d) g ( x ) = 4 + x + x
2 24
2
7. f ( x ) =
π (1 + x 2 )
Integral Transform Methods 183
∫ (sin ) ( )
x
λ J1 (t )
8. 1− λ2 (x − t) dt + λ cos x 1 − λ 2
(1 − λ )2 0
t
+
λ2
1− λ 2 (
sin x (1 − λ 2 )
x
g ( x ) = f ( x ) + ∫ R( x, t ) f (t ) dt
9.
0
∞
1
where, R( x, t ) = ∑
( x − t )Γ[n(1 − α )] n =1
[λΓ (1 − α )( x − t )1−α ]n
g ( x ) = xe x − e x + 1
10.
Index
Kernel, 3
of Fourier cosine transform, 177 Real eigenvalues, 34
of Fourier sine transform, 177, 178 Reciprocal functions, 104
Kernel of Fourier transform, 178 Reciprocal kernal, 7, 85
Kernel of Fourier transformation, 177 Regularity condition, 57
Kinds of linear integral equations, 4 Resolvent determinant, 123
Resolvent kernel, 7, 84, 85