MA209 Integral Equations and Green’s Function

Dr. Anjana Pradhan Ghorai

([email protected])

Department of Mathematics,
BIT, Mesra,
Ranchi.

MODULE -IV

Dr. Anjana Pradhan Ghorai MA209 IE and GF 1 / 73

This module is consisting of:
I Symmetric kernel
I Orthonormal system of function
I Fundamental properties of eigenvalues and eigenfunctions for
symmetric kernels
→ expansion of symmetric kernel in eigen function.
I Hilbert-Schmidt theorem
→ solution of symmetric integral equation by Hilbert- Schmidt
theorem.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 2 / 73

Introduction

Symmetric Kernels:
Definition A Kernel is said to be symmetric if it coincides with its
own complex conjugate.It is characterized by the the identity:
K (x , t) = K (t, x ),
where the bar denotes the complex conjugate.
If the Kernel is real, then its symmetry is defined by the identity:
K (x , t) = K (t, x ).
An integral equation with a symmetric Kernel is called symmetric in-
tegral equation.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 3 / 73

Continued..

Previously, we have discussed about the eigenvalues and eigenfunc-

tions for integral equation of different kind. We have also established
that the eigenvalues of the integral equations are the zero of certain
determinant. We have also seen that there are many kernels for which
there are no eigenvalues of the integral equation.
However, in this chapter, we will prove that for a symmetric Kernel,not
identically zero, at least one eigenvalue will always exist. This is an
important characteristic of the symmetric Kernels.
An eigenvalue is simple if there is only one corresponding eigenfunction,
otherwise, the eigenvalues are called degenerate.
The spectrum of the kernel K (x , t) is the set of all its eigenvalues.
Thus the spectrum of the symmetric kernel is never empty.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 4 / 73

Regularity conditions
Here we shall mainly deal with functions which are either continuous,
or integrable or square integrable.
A function f (x ) is said to be square integrable function if
Z b
| f (x ) |2 dx < ∞ [1]
a
A square integrable function f (x ) is also called L2 function.
The regularity conditions on the kernel K (x , t) are identical.
The kernel K (x , t) is said to be an L2 function if the following three
conditions
Z b Z bare satisfied:
(i) | K (x , t) |2 dx dt < ∞, ∀ x ∈ [a, b], ∀ t ∈ [a, b], [2]
Za b a
(ii) | K (x , t) |2 dx < ∞, ∀ t ∈ [a, b], [3]
Za b
(iii) | K (x , t) |2 dt < ∞, ∀ x ∈ [a, b]. [4]
a

Dr. Anjana Pradhan Ghorai MA209 IE and GF 5 / 73

The Inner or Scalar product of two functions
The Inner product of two complex L2 functions φ and ψ of real vari-
ables x , aZ ≤ x ≤ b, is denoted by:
b
hφ, ψi = φ(x )ψ̄(x )dx . [5]
a
Two functions are called orthogonal if their inner product is zero, that
is the functions φ and ψ are orthogonal if,
Z b
hφ, ψi = φ(x )ψ̄(x )dx = 0. [6]
a
The norm of a function φ(x ) is denoted by kφ(x )k and is defined by
the relation:
"Z #1/2 "Z #1/2
b b
2
kφ(x )k = φ(x )φ̄(x )dx = | φ(x ) | dx . [7]
a a
A function φ(x ) is said to be normalized if kφ(x )k = 1.
A non null function can always be normalized by dividing it by its norm.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 6 / 73

Continued...

Schwarz Inequality :
If φ(x ) and ψ(x ) are L2 functions,
then | hφ(x ), ψ(x )i |≤ kφ(x )k kψ(x )k . [8]
Minkowski Inequality :
If φ(x ) and ψ(x ) are L2 functions,
then kφ(x ) + ψ(x )k ≤ kφ(x )k + kψ(x )k . [9]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 7 / 73

Complex Hilbert Space

Complex Hilbert Space:

A linear space of infinite dimension with inner product(or scalar prod-
uct) hx , y i which is complex number, is called complex Hilbert Space
if it satisfies the following three axioms:
(i) the definiteness axiom: hx , y i > 0 ∀ x 6= 0,
(ii) the linearity axiom: hαx1 + βx2 , y i = αhx1 , y i + βhx2 , y i,
where α and β are the arbitrary complex numbers.
(iii) the axiom of Hermitian symmetry: hy , x i = hx , y i,
where bar denotes the complex conjugate.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 8 / 73

Complex Hilbert Space continued..

Let H be the set of complex-valued function φ(x ) defined in the interval

(a, b) such that:
Z b
| φ(x ) |2 dx < ∞,
a
Then H is a linear and complex Hilbert space L2̄ (a, b) or L2̄ .
The norm of the function generates the natural metric, defined as:
dhφ, ψi = kφ − ψk = hφ − ψ, φ − ψi1/2 . [10]
A metric space is called complete if every Cauchy sequence of functions
in this space is a convergent sequence.
A Hilbert space is as inner-produce linear space that is complete in its
natural metric.
The completeness of L2̄ spaces plays an important role in the theory
of linear operator.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 9 / 73

Fredholm operator

Fredholm operator K :
The Fredholm operator K is defined as:
Z b
Kφ = K (x , t)φ(t)dt. [11]
a
The operator adjoint to K is defined as :
Z b
Kψ = K (t, x )ψ(t)dt. [12]
a
The operators defined in (11) and (12) are connected as follows:
hK φ, ψi = hφ, K ψi. [13]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 10 / 73

Fredholm operator continued...
Theorem:
Prove that hK φ, ψi = hφ, K ψi.
Proof:
hK φ, ψi !
Z b Z b
= K (x , t)φ(t)dt ψ(x )dx , using (5) and (11),
a a
Z b Z b !
= K (x , t)ψ(x )dx φ(t)dt, changing the order of integration,
a a
Z b Z b !
= K (t, x )ψ(t)dt φ(x )dx , renaming the variables x and t
a a
as t and x respectively, !
Z b Z b
= K (t, x )ψ(t)dt φ(x )dx ,
a a
= hφ, K ψi, where bar denotes the complex conjugate.
Dr. Anjana Pradhan Ghorai MA209 IE and GF 11 / 73
Fredholm operator continued...

For the symmetric kernel, (13) reduces to:

hK φ, ψi = hφ, K ψi. [14]
which implies that, a symmetric operator ia a self-adjoint. Futher, we
find that permutation of factors in a scalar product is equivalent to
taking the complex conjugate, that is,
hφ, K φi = hK φ, φi.
Combining this fact with (14), we find that, for a symmetric kernel,
the inner product hK φ, φi is always real.
The converse of this is also true.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 12 / 73

An orthonormal system of functions

An orthonormal system of functions:

A finite or an infinite set {φk (x )} defined on an interval a ≤ x ≤ b is
said to be an orthogonal set if:
Z b
hφi , φj i = 0 or φi (x )φj (x )dx = 0, i 6= j, [1]
a
If none of the elements of this set is a zero vector, then it is called a
proper orthogonal set.
The set {φk (x )} is said to be an  orthonormal set if:
Z b 0, i 6= j,
hφi , φj i = φi (x )φj (x )dx =  [2]
a 1, i = j.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 13 / 73

An orthonormal system of functions

Gram-Schmidt method of ortho-normalization:

Given a finite or an infinite independent set of functions {ψ1 , ψ2 , · · · , ψk , ·
By Gram-Schmidt method of ortho-normalization, we can construct an
ortho-normal set {φ1 , φ2 , · · · , φk , · · · , · · · }.
ψ1
The method is as follows: Set φ1 = .
kψ1 k
To obtain φ2 , lets define w2 (x ) = ψ2 (x ) − hψ2 , φ1 iφ1 .
It can be easily proved that w2 is orthogonal to φ1 .
w2
Hence set φ2 = .
kw2 k

Dr. Anjana Pradhan Ghorai MA209 IE and GF 14 / 73

An orthonormal system of functions continued...

wk
Proceding in this manner, we obtain φk = ,
kwk k
k−1
X
where, wk (x ) = ψk (x ) − hψk , φi iφi .
i=1
Again, if we have a set of orthogonal functions, we can convert it into
an orthonormal set by dividing each function by its norm.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 15 / 73

An orthonormal system of functions continued...

Riesz-Fischer Theorem:
If {φi (x )} ia a given orthonormal system of functions in L2̄ space and
{αi } is a given sequence of complex numbers such that
∞
| αi |2 converges,
X

i=1
then there exists a unique function f (x ) for which αi are the Fourier
coefficients with respect to the orthonormal system {φi (x )} and to
which the Fourier series converges in the mean, that is,
n
X
f (x ) − αi φi (x ) → 0 as n → ∞.
i=1

Dr. Anjana Pradhan Ghorai MA209 IE and GF 16 / 73

Some useful results
If an orthonormal system of functions {φi } exists in L2̄ space in such a
manner that every other element in this space can be represented lin-
early in terms of this system, then it is known as an orthonormal basis.
Also, if any of the following conditions are satisfied, the orthonormal
set {φi } is complete. X X
(i) For every function ψ in L2̄ space, ψ = hψ, φi iφi = αi φi ,
i i
∞
| hψ, φi i |2 ,
X
(ii) For every function ψ in L2̄ space kψk =
i=1
which is known as Parseval’s identity.
(iii) The only function ψ in L2̄ space for which the Fourier coefficients
vanish is the trivial function.
(iv) There exists no function ψ in L2̄ space such that {ψ, φ1 , φ2 , · · · , φk , · ·
is an orthonormal set .
Dr. Anjana Pradhan Ghorai MA209 IE and GF 17 / 73
Some useful results

Fourier series of general character:

Let r (t) be a continuous, real and non-negative function in the interval
(a, b). The functions φ(t) and ψ(t) are orthogonal with respect to the
weight function r (t) if,
Z b
r (t)φ(t)ψ(t)dt = 0.
a
The set of functions {φ1 , φ2 , · · · , φk , · · · , · · · } will be said to be or-
thonormal with respectto the weight function r (t) if,
Z b 0, i 6= j,
r (t)φi (t)φj (t)dt =
a 1, i = j.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 18 / 73

Some useful results

And in such case, for the Fourier expansion in terms of such functions,
the introduction of a new inner product is required and is defined as
follows: Z
b
hφ, ψi = r (t)φ(t)ψ(t)dt,
a
with the corresponding norm:
(Z )1/2
b
kφkr = r (t)φ(t)φ(t)dt ,
a
and the space of function for which kφkr < ∞ is a Hilbert space.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 19 / 73

Some important examples of the complete
orthogonal and orthonormal system of functions

Some important examples

(i) The system φn (x ) = (2π)−1/2 e inx , where n takes every integer in
(−∞, ∞), is orthonormal in the interval (−π, π).
(ii) The functions 1, cosx , cos2x , ............... form an orthogonal system
in the interval (0, π).
(iii) The functions sinx , sin2x , sin3x ............... form an orthogonal sys-
tem in the interval (0, π).
(iv) The Legendre polynomial given by:
1 d n (x 2 − 1)n
P0 (x ) = 1, Pn (x ) = n , n = 1, 2, 3.........
2 n! dx n
form an orthogonal system in the interval (−1, 1).

Dr. Anjana Pradhan Ghorai MA209 IE and GF 20 / 73

Some important examples continued...
Also it can be easily shown
 that
Z 1 0, m 6= n,

Pm (x )Pn (x )dx = 2
−1 
 , m = n.
2n + 1
(v) The Chebychev polynomial given by:
Tn (x ) = cos(ncos −1 x ), n = 0, 1, 2, 3.........
form an orthogonal system with respect to the weight function
1
r (x ) = in the interval (−1, 1).
(1 − x 2 )1/2
Also it can be easily shown that 

 0, m 6= n,
Z 1 π

2 −1/2
(1 − x ) Tm (x )Tn (x )dx = , m = n 6= 0,
−1 

 2
 π, m = n = 0

Dr. Anjana Pradhan Ghorai MA209 IE and GF 21 / 73

Some important examples continued...

(vi) The system of functions

Jn (αi,n x ), i = 1, 2, 3, .... is orthogonal with respect to the weight
function x in the interval (0, 1).
The functions satisfy the relation:

Z 1 0, i 6= j,
xJn (αi,n x )Jn (αj,n x )dx =
0 J 2 (αi,n ), i = j
n+1
when Jn (x ) denote the Bessel’s function of first kind of order n(> −1),
and αi,n , i = 1, 2, 3, ... denotes the positive zeros of Jn (x ) = 0.
The systems (1)-(vi) are complete and play an important role in many
practical applications.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 22 / 73

Continued...

Two-dimensional orthonormal system:

Theorem:
If {φi (x )} is a complete orthonormal set over a ≤ x ≤ b, and {ψi (t)}
is a complete orthonormal set over c ≤ t ≤ d,
then the set
{φ1 (x )ψ1 (t), φ2 (x )ψ2 (t), φ3 (x )ψ3 (t), .................}
form a complete two-dimensional orthonormal set over the rectangle
a ≤ x ≤ b, c ≤ t ≤ d.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 23 / 73

Some fundamental properties of eigen-values and
eigen-functions for symmetric kernels

Theorem:1 If a kernel is symmetric then all its iterated kernels are

symmetric.
Proof: Let the Kernel K (x , t) is symmetric. then by definition:
K (x , t) = K̄ (t, x ) [1]
and the iterated kernels Kn (x , t), n = 1, 2, 3.... are defined as follows:
K1 (x , t) = K (x , t), [2]
Z b
Kn (x , t) = K (x , z)Kn−1 (z, t))dz, n = 2, 3, .... [3]
a
and Z b
Kn (x , t) = Kn−1 (x , z, )K (z, t))dz, n = 2, 3, .... [3a]
a

Dr. Anjana Pradhan Ghorai MA209 IE and GF 24 / 73

Continued...
we shall use the method of Mathematical induction to prove the result.
Now,
K2 (x , t)
Z b
= K (x , z)K1 (z, t))dz
Za b
= K (x , z)K (z, t)dz
Za b
= K̄ (z, x )K̄ (t, z)dz
Za b
= K̄ (t, z)K̄ (z, x )dz
Za b
= K̄ (t, z)K̄1 (z, x )dz
a
= K̄2 (t, x )
which implies that K2 (x , t) is symmetric.
Hence the result is true for n = 1 and n = 2.
Dr. Anjana Pradhan Ghorai MA209 IE and GF 25 / 73
Continued...
Let Kn (x , t) be symmetric for n = m, that is,
Km (x , t) = K̄m (x , t) [4]
we shall now prove that Kn (x , t) is also symmetric for n = m + 1,
i.e., Km+1 (x , t) = K̄m+1 (t, x ) [5]
now,
Km+1 (x , t)
Z b
= K (x , z)Km (z, t)dz,
Za b
= K̄ (z, x )K̄m (t, z)dz,
Za b
= K̄m (t, z)K̄ (z, x )dz,
a
= K̄m+1 (t, x ).
Hence, by the method of Mathematical induction, Kn (x , t) is symmet-
ric for all n.
Dr. Anjana Pradhan Ghorai MA209 IE and GF 26 / 73
Continued...
Theorem:II Every symmetric kernel with non-zero norm has at least
one eigenvalue.
Theorem:II A Every real symmetric kernel with non-zero norm has at
least one eigenvalue.
Theorem:III The eigenvalues of a symmetric kernel are real.
Theorem:IIIA If the kernel K (x , t) is real, symmetric, continuous and
identically not equal to zero, then all the eigenvalues are real.
Proof: Let λ and φ(x ) be an eigenvalue and a corresponding eigen-
function of the kernel K (x , t). Then by definition:
Z b
φ(x ) = λ K (x , t)φ(t)dt, [1]
a
Then multiplying (1) by φ̄(x ) and integrating with respect to x from
a to b, we have, ( )
Z b Z b Z b
φ(x )φ̄(x )dx = λ K (x , t)φ(t)dt φ̄(x )dx , [2]
a a a
Dr. Anjana Pradhan Ghorai MA209 IE and GF 27 / 73
Continued...

Now, by the definition of Fredholm operator K we have:

Z b
Kφ = K (x , t)φ(t)dt. [3]
a
Also, the norm of φ(x ) as
"Z #1/2
b
kφ(x )k = φ(x )φ̄(x )dx . [4]
a
Using (3), (4) and the definition of inner product, (2) reduces to
kφ(x )k2 = λhK φ, φi [5]
2
kφ(x )k
=⇒ λ = [6]
hK φ, φi
Since both the numerator and denominator of (6) are real, it follows
that the eigenvalue λ is real and the theorem is hence proved.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 28 / 73

Continued...
Theorem:IV The eigenfunctions of a symmetric kernel, corresponding
to different eigenvalues are orthogonal.
Proof: Let φm (x ) and φn (x ) are eigenfunctions corresponding to eigen-
values λm and λn respectively, where λm 6= λn . Then by definition, we
have : Z b
φm (x ) = λm K (x , t)φm (t)dt, [1]
a
and, Z b
φn (x ) = λn K (x , t)φn (t)dt, [2]
a
Since λn is real, (2) may be re-written as:
Z b
φ̄n (x ) = λn K̄ (x , t)φ̄n (t)dt, [3]
a
Since K (x , t) is symmetric, we have:
K̄ (x , t) = K (t, x ) [4]
Dr. Anjana Pradhan Ghorai MA209 IE and GF 29 / 73
Continued...
Using (4), (3) may be re-written as:
Z b
φ̄n (x ) = λn K (t, x )φ̄n (t)dt, [5]
a
Interchanging x and t in (5), we have:
Z b
φ̄n (t) = λn K (x , t)φ̄n (x )dx , [6]
a
Then, multiplying (1) by φ̄n (x ) and integrating with respect to x from
aZ to b, we have,
b
φm (x )φ̄n (x )dx
a
Z b (Z b )
= λm K (x , t)φm (t)dt φ̄n (x )dx ,
a
Z b ( Za )
b
= λm K (x , t)φ̄n (x )dx φm (t)dt, changing the order of inte-
a a
gration
Dr. Anjana Pradhan Ghorai MA209 IE and GF 30 / 73
Continued...

λm Z b
= φm (t)φ̄n (t)dt, by (6)
λn a Z b
=⇒ (λm − λn ) φm (x )φ̄n (x )dx =0
a
i.e., (λm − λn )hφm (x )φn (x )i = 0 [7]
Since, (λm 6= λn ), so (7) reduces to:
hφm (x )φn (x )i = 0 [8]
which shows that the eigenfunctions φm (x ) and φn (x ) are orthogonal.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 31 / 73

Continued...

Theorem:V The multiplicity of ZanyZ nonzero eigenvalus is finite for

b b
every symmetric kernel for which | K (x , t) |2 dx dt is finite.
a a
Theorem:VI The sequence of eigenfunctions of a symmetric kernel
can be made orthonormal.
Theorem:VII The eigenvalues of a symmetric kernel form a finite or
infinite sequence {λn } with no finite limit point.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 32 / 73

Continued...
Theorem:VIIA: The set of eigenvalues of the second iterated kernel
coincide with the set of square of the eigenvalues of the given kernel.
Proof Let λ and φ(x ) be an eigenvalue and a corresponding eigen-
function of the kernel K (x , t). Then by definition:
Z b
φ(x ) = λ K (x , t)φ(t)dt,
a
i.e., φ(x ) = λK φ
or, (I − λK )φ = 0 [1]
where I is the identity operator.
Operating both sides of (1) with the operator (I + λK ), we obtain:
(I − λ2 K 2 )φ = 0
Z b
or, φ(x ) = λ2 K 2 φ = λ2 K2 (x , t)φ(t)dt [2]
a
which implies that λ2 is an eigen value of the kernel K2 (x , t).
Dr. Anjana Pradhan Ghorai MA209 IE and GF 33 / 73
Continued...
Remark:1 The above result can be extended to the n-th iterated ker-
nel. The set of eigenvalues of the kernel Kn (x , t) coincide with the
set of n-th powers of the eigenvalues of the kernel K (x , t).
Remark:2 While proving the above theorem, the symmetry of the
kernel K (x , t) has not been assumed.
Theorem:IX If λ1 is the smallest eigenvalue of the kernel K (x , t),
then: " #
1 hkφ, φi
= max
| λ1 | kφk
Rb Rb
| K (x , t)φ(t)φ̄(x )dtdx |
= max a a [1]
kφk
or, equivalently,
1
= max hkφ, φi, kφk = 1 [2]
| λ1 |
Dr. Anjana Pradhan Ghorai MA209 IE and GF 34 / 73
Expansion in Eigenfunctions

Expansion in Eigenfunctions and Bilinear form:

Let K (x , t) be a nonnull, symmetric kernel which has a finite or infinite
number of eigenvalues (always real and nonzero). we order of the
sequence of eigenvalues as
λ1 , λ2 , λ3 , ....., λn ... [1]
in such a manner that each eigenvalue is repeated as many times as
its multiplicity and is arranged according to their absolute values , i.e.,
| λ1 |≤| λ2 |≤ ... ≤| λn |≤ λn+1 ≤ ......
Lets consider the sequence of ortho-normalized eigenfunctions corre-
sponding to the eigenvalues given by the sequence (1) as:
φ1 (x ), φ2 (x ), φ3 (x ), ....., φn (x ), ... [2]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 35 / 73

Expansion in Eigenfunctions continued....

The eigenfunction are arranged in such a that they are not repeated
and are linearly independent in each group corresponding to the same
eigenvalues. which implies that to each eigenvalues λk in (1), there
corresponds just one eigenfunction φk (x ) in (2)
suppose that a symmetric L2 kernal has at least one eigenvalue, say,
λ1 with corresponding eigenfunction φ1 (x ).
Lets consider the second truncated symmetric kernel as:
φ1 (x )φ̄1 (t)
K (2) (x , t) = K (x , t) − [3]
λ1
It can be proved that the second truncated symmetric kernel K (2) (x , t)
is non-null and also possess at least one eigenvalue λ2 (smallest) with
corresponding normalized eigenfunction φ2 (x ).

Dr. Anjana Pradhan Ghorai MA209 IE and GF 36 / 73

Expansion in Eigenfunctions continued....
Also the functions φ1 (x ) 6= φ2 (x ) even if λ1 = λ2 .
since:
Z b
K (2) (x , t)φ1 (t)dt
a
Z b
φ1 (x ) Z b
= K (x , t)φ1 (t)dt − φ1 (t)φ̄1 (t)dt
a λ1 a
φ1 (x ) φ1 (x )
= − =0
λ1 λ1
and,
Z b
K (2) (x , t)φ2 (t)dt
a
Z b
φ1 (x ) Z b
= K (x , t)φ2 (t)dt − φ2 (t)φ̄1 (t)dt
a λ1 a
φ2 (x ) φ2 (x )
= −0=
λ2 λ2
Dr. Anjana Pradhan Ghorai MA209 IE and GF 37 / 73
Expansion in Eigenfunctions continued....

Z b
or, φ2 (x ) = λ2 K (2) (x , t)φ2 (t)dt
a
Similarly, the third truncated symmetric kernel:
K (3) (x , t)
φ2 (x )φ̄2 (t)
= K (2) (x , t) −
λ2
2
X φi (x )φ̄i (t)
= K (x , t) − [4]
i=1 λi
yields the third eigenvalue λ3 (smallest) with corresponding normalized
eigenfunction φ3 (x ).

Dr. Anjana Pradhan Ghorai MA209 IE and GF 38 / 73

Expansion in Eigenfunctions continued....

Proceeding in this manner we finally arrive at the two following possi-

bilities:
(i) The above process terminates after n steps,
that is K (n+1) (x , t) = 0
and the kernel K (x , t) is a degenerate kernel given by:
n
X φi (x )φ̄i (t)
K (x , t) = [5]
i=1 λi
(ii) The above process can be continued indefinitely and there are
infinite number of eigenvalues.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 39 / 73

Expansion in Eigenfunctions continued....

Theorem:I If the sequence {φn (x )} be all the eigenfunctions of a

symmetric L2 kernel with {λn } as the sequence of corresponding
eigenvalues,
∞
X | φn (x ) |2
then the series converges and is bounded by C which is
n=1 λ2n
Z b
the upper bound of the integral | k(x , t) |2 dt .
a

Dr. Anjana Pradhan Ghorai MA209 IE and GF 40 / 73

Expansion in Eigenfunctions continued....

Theorem:II If the sequence {φn (x )} be all the eigenfunctions of a

symmetric L2 kernel with {λn } as the sequence of corresponding
eigenvalues, then the truncated symmetric kernel
n
φi (x )φ̄i (t)
K (n+1) (x , t) = K (x , t) −
X

i=1 λi
has the eigenvalues λn+1 , λn+2 , λn+3 , .... with corresponding eigenfunc-
tions φn+1 , φn+2 , φn+3 , .... .
The kernel K (n+1) (x , t) has no other eigenvalues or eigenfunctions.
Theorem:III A necessary and sufficient condition for a symmetric L2
kernel to be separable is that it must possess a finite number of eigen-
values.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 41 / 73

Hilbert-Schmidt Theorem

Hilbert-Schmidt Theorem:
If f (x ) can be written in the form:
Z b
f (x ) = K (x , t)h(t)dt [1]
a
where K (x .t) is a symmetric L2 kernel and h(t) is a L2 function,
then f (x ) can be expanded in an absolutely and uniformly convergent
Fourier series with respect to the orthonormal system of eigenfunctions:
φ1 (x ), φ2 (x ), φ3 (x ), ..., φn (x ), . of the kernel K (x , t) as:
∞
X
f (x ) = fn φn (x ), [2]
n=1
where fn = hf (x ), φn (x )i. [3]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 42 / 73

Hilbert-Schmidt Theorem continued...

The Fourier coefficients of the function f (x ) are related to the Fourier

coefficients hn of the function h(x ) by the relations:
hn
fn = [4]
λn
and,
hn = hh(x ), φn (x )i, [5]
∞
X hh(x ), φn (x )i
=⇒ f (x ) = φn (x ) [6]
n=1 λn
where λn are the eigenvalues of the kernel K (x , t).

Dr. Anjana Pradhan Ghorai MA209 IE and GF 43 / 73

Definite kernel and Mercer’sTheorem

Nonnegative-definite kernel:
A symmetric L2 kernel K (x .t) is said to be nonnegative-definite kernel
if: hK φ, φi ≥ 0 for every L2 function φ.
positive-definite kernel:
A symmetric L2 kernel K (x .t) is said to be positive-definite kernel
if hK φ, φi ≥ 0 and hK φ, φi = 0 =⇒ φ is a null function.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 44 / 73

Continued...

Nonpositive-definite kernel:
A symmetric L2 kernel K (x .t) is said to be nonnegative-definite kernel
if: hK φ, φi ≤ 0 for every L2 function φ.
Negative-definite kernel:
A symmetric L2 kernel K (x .t) is said to be negative-definite kernel
if: hK φ, φi ≤ 0 and hK φ, φi = 0 =⇒ φ is a null function.
Indefinite kernel:
A symmetric kernel that does not fall into any of the above types of
kernels, is known as indefinite kernel.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 45 / 73

Some important Theorems

The following theorem is an immediate consequence of the Hilbert-

Schmidt theorem.
Theorem (i):
A nonnull, symmetric L2 kernel K (x .t) is non-negative if and only if
all its eigenvalues are positive.
Theorem (ii):
A nonnull, symmetric L2 kernel K (x .t) is positive if and only if all
its eigenvalues are positive and some or every orthonormal system of
eigenfunctions of K (x , t) is complete.
Proof: Usng Hilbert-Schmidt theorem, we have:
∞
X hφ, φn (x )i
f (x ) = K φ = φn (x ) [1]
n=1 λn

Dr. Anjana Pradhan Ghorai MA209 IE and GF 46 / 73

Continued...
Taking inner product of both sides of (1) with φ, we have:
∞
X | hφ, φn (x )i |2
hK φ, φi = [2]
n=1 λn
Now , if λn > 0 for each n, then, (2) =⇒ hK φ, φi ≥ 0, ∀φ. And if
any eigenvalue say, λp < 0, then;
hK φp , φp i
∞
X | hφp , φn (x )i |2
=
n=1 λn
| hφp , φp i |2
=
λp
1
= <0
λp
Hence theorem (i) proved.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 47 / 73

Continued...

Now, from (2), we have:

hK φ, φi = 0 =⇒ hφ, φn i = 0 ∀n,
or, hφ, φn i = 0 =⇒ φ = 0 ∀φ.
which implies that every orthonormal system of eigenfunctions of K (x , t)
is complete.
Mercer’sTheorem:
If a nonnull, symmetric L2 kernel K (x .t) is quasi-definite( that is
, when all but a finite number of eigenvalues are of one/same sign)
∞
X 1
and continuous, then the series is convergent and the series
n=1 λn
∞
X φn (x ), φ̄n (x )
k(x , t) = is uniformly and absolutely convergent.
n=1 λn

Dr. Anjana Pradhan Ghorai MA209 IE and GF 48 / 73

Schmidt’s solution of non-homogeneous Fredholm
integral equation of second kind
Schmidt’s solution:
The non-homogeneous Fredholm integral equation of second kind is
given by: Z b
y (x ) = f (x ) + λ k(x , t)y (t)dt [1]
a
where K (x .t) is continuous,real and symmetric and λ is not an eigen-
value.
Hilbert-Schmidt Theorem states that any function F (x ) can be gener-
ated from a continuous function y (x ) by the operator:
Z b
λ k(x , t)y (t)dt,
a Z b
i.e., F (x ) = λ k(x , t)y (t)dt, [2]
a
where K (x .t) is continuous,real and symmetric.
Dr. Anjana Pradhan Ghorai MA209 IE and GF 49 / 73
Schmidt’s Theorem Continued...

Then, F (x ) can be generated over the interval (a, b) by a linear combi-

nation of normalized eigenfunctions of homogeneous integral equation:
Z b
y (x ) = λ k(x , t)y (t)dt, [3]
a

Dr. Anjana Pradhan Ghorai MA209 IE and GF 50 / 73

Schmidt’s solution Continued...

Method of solution :
Re-writing (1, we have,:
Z b
y (x ) − f (x ) = λ k(x , t)y (t)dt [4]
a
Since (4) is of the
X form (2), it follows by Hilbert-Schmidt Theorem:
y (x ) − f (x ) = αm φm (x ), a ≤ x ≤ b [5]
m=1
where, , φm (x ), m = 1, 2, 3.... are the normalized eigenfunctions and
λm are the corresponding eigenvalues of IE (3).
Let λm 6= λ, ∀ m = 1, 2, 3... [6]
Since φm (x ), m = 1,2, 3.... are normalized, we have:
Z b 0, m 6= n,
φm (x )φn (x )dx = [7]
a 1, m = n

Dr. Anjana Pradhan Ghorai MA209 IE and GF 51 / 73

Schmidt’s solution Continued...
Multiplying both sides of (5) by φm (x ) and integrating between the
limits a and b with Zrespect to x we have:
Z b b
y (x )φm (x )dx − f (x )φm (x )dx
a Z b a Z b
= α1 φ1 (x )φm (x )dx + α2 φ2 (x )φm (x )dx
a Z b a

+..... + αm φm (x )φm (x )dx + ..... [8]

a
now, lets
Z denote: b
Cm = y (x )φm (x )dx [9]
a
and Z
b
fm = f (x )φm (x )dx [10]
a
Then, using (7), (9) and (10), (8) reduces to:
Cm − fm = αm [11]
Dr. Anjana Pradhan Ghorai MA209 IE and GF 52 / 73
Schmidt’s solution Continued...

Again, Multiplying both sides 0f (1) by φm (x ) and integrating between

the limits a and b with respect to x we have:
Z b
y (x )φm (x )dx
a
Z b Z b (Z b )
= f (x )φm (x )dx + λ K (x , t)y (t)dt φm (x )dx
a a
Z b (Z a )
b
=⇒ Cm = fm + λ y (t) K (x , t)φm (x )dx dt
a
Z b ( Za )
b
=⇒ Cm = fm + λ y (t) K (t, x )φm (x )dx dt [12]
a a
Since K (x , t) is symmetric.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 53 / 73

Schmidt’s solution Continued...
Again, since φm (x ) is the eigenfunction corresponding to the eigenvalue
λm of IE (3), we have:
Z b
φm (x ) = λm K (x , t)φm (t)dt
Za b
= λm K (t, x )φm (x )dx
a
Z b
φm (t)
=⇒ = K (t, x )φm (x )dx [13]
λm a
Using (13), (12) reduces to:
Z b
y (t)φm (t)
Cm = fm + λ dt
a
Z b
λm
λ
Cm = fm + y (t)φm (t)dt
λm a
λ
or, Cm = fm + Cm [14]
λm
Dr. Anjana Pradhan Ghorai MA209 IE and GF 54 / 73
Schmidt’s solution Continued...

Eliminating Cm from (11 and (14), we have

λ
αm = fm [15]
λm − λ
and since,λm 6= λ, αm is well defined as denoted by (15).
Substituting the value of αm in (5), the required solution of IE (1) as:
X λfm
y (x ) − f (x ) = φm (x ),
m=1 λm − λ
X fm
or,y (x ) = f (x ) + λ φm (x ), [16]
m=1 λm − λ

Dr. Anjana Pradhan Ghorai MA209 IE and GF 55 / 73

Schmidt’s solution Continued...

Using (10), (16) reduces to

X φm (x ) Z b
y (x ) = f (x ) + λ f (t)φm (t)dt,
m=1 λm − λ a
Z b"X #
φm (x )φm (t)
or, y (x ) = f (x ) + λ f (t)dt,
a m=1 λm − λ
Z b
or, y (x ) = f (x ) + λ R(x , t : λ)f (t)dt, [17]
a
where the resolvent kernel R(x , t : λ) is given by
X φm (x )φm (t)
R(x , t : λ) = [18]
m=1 λm − λ

Dr. Anjana Pradhan Ghorai MA209 IE and GF 56 / 73

Schmidt’s solution Continued...

Three important cases may arise:

case I: Unique solution: Since, λm 6= λ and (15) gives a well defined
value of αm to be substituted in (5), the solution (16) or (17) exists
uniquely if and only if λ does not take on an eigenvalue.
case II: No solution: Let, λ = λi , where λi is the i-th eigenvalue and
also, let fi 6= 0 i.e.,
Z b
f (x )φi (x )dx 6= 0,
a
i.e., eigenfunction φi (x ) is not orthogonal to f (x ), the term
λfi
φi (x )
λi − λ
in (16) becomes unbounded and hence no solution of (1) exists.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 57 / 73

Schmidt’s solution Continued...

case III: Infinitely many solution:

Let, λ = λi , where λi is the i-th eigenvalue and also, let fi = 0 i.e.,
Z b
f (x )φi (x )dx = 0,
a
i.e., eigenfunction φi (x ) is orthogonal to f (x ), then for (14) reduces
to (for m = i )
λ
Ci = fi + Ci or, Ci = Ci for any arbitrary i.
λ
It implies that the coefficient αi of φi assumes to be the form of 0/0,
truely arbitrary.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 58 / 73

Schmidt’s solution Continued...

Therefore, the solution (16) of (1) can be re-written as:

0
X λfm
y (x ) = f (x ) + Aφi (x ) + φm (x ), [19]
m=1 λm − λ
0
X
where implies that it does not include the term when m = i and
m=1
A is an arbitrary constant.
Therefore, (19) represents infinitely many solutions of (1).

Dr. Anjana Pradhan Ghorai MA209 IE and GF 59 / 73

Schmidt’s solution Continued...
Example:1 Using Hilbert-Schmidt’s theorem solve the symmetric
integral equation ofZ second kind:
1
y (x ) = (1 + x )2 + (xt + x 2 t 2 )y (t)dt
−1
Solution: Given IEZis:
1
y (x ) = (1 + x )2 + (xt + x 2 t 2 )y (t)dt [1]
−1
Z b
Comparing (1) with y (x ) = f (x ) + λ k(x , t)y (t)dt, [2]
a
we have: f (x ) = (1 + x )2 , k(x , t) = (xt + x 2 t 2 ), λ = 1, a = −1
and b = 1 [3]
Now , we start with obtaining the eigenvalue and normalized eigen-
functions Zof corresponding symmetric IE:
1
y (x ) = λ (xt + x 2 t 2 )y (t)dt [4]
−1

Dr. Anjana Pradhan Ghorai MA209 IE and GF 60 / 73

Schmidt’s solution Continued...

Re-writing (4) as:

Z 1 Z 1
y (x ) = λ xty (t)dt + λ x 2 t 2 y (t)dt
−1 Z 1 −1 Z 1
or, y (x ) = λx ty (t)dt + λx 2 t 2 y (t)dt
−1 −1
or, y (x ) = λC1 x + λC2 x 2 [5]
where,Z
1
C1 = ty (t)dt [6]
Z−1
1
C2 = t 2 y (t)dt [7]
−1

Dr. Anjana Pradhan Ghorai MA209 IE and GF 61 / 73

Schmidt’s solution Continued...
Using (5), (6) becomes:
Z 1
C1 = t(λC1 t + λC2 t 2 )dt
−1 !
2λ
or, 1 − C1 + 0.C2 = 0 [8]
3
Using (5), (7) becomes:
Z 1
C2 = t 2 (λC1 t + λC2 t 2 )dt
−1 !
2λ
or, 0.C1 + 1 − C2 = 0 [9]
5
Equations (8) and (9) will have a non-trivial solution iff:
1 − (2λ/3) 0
D(λ) = = 0 =⇒ λ = 3/2 or λ = 5/2.
0 1 − (2λ/5)
Hence the required eigenvalues are:
λ1 = 3/2 or λ2 = 5/2. [10]
Dr. Anjana Pradhan Ghorai MA209 IE and GF 62 / 73
Schmidt’s solution Continued...
Determination of eigenfunction corresponding to λ1 = 3/2
Putting λ = λ1 = 3/2 in (8) and (9), we have :
C2 = 0 and C1 is arbitrary.
Putting these values of C1 and C2 with λ = 3/2 in(5), we have the
first eigenfunction y1 (x ) as:
3
y1 (x ) = C1 x = x ,
2
seting C1 = 2/3
Now, the corresponding normalized eigenfunction φ1√ (x ) as:
y1 (x ) x x 6
φ1 (x ) = " #1/2 = " #1/2 =
R1 R1 2
2 2
−1 {y1 (x )} dx −1 x dx

[11]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 63 / 73

Schmidt’s solution Continued...
Determination of eigenfunction corresponding to λ2 = 5/2
Putting λ = λ2 = 5/2 in (8) and (9), we have :
C1 = 0 and C2 is arbitrary.
Putting these values of C1 and C2 with λ = 5/2 in(5), we have the
first eigenfunction y2 (x ) as:
5
y2 (x ) = C2 x 2 = x 2 ,
2
seting C2 = 2/5
Now, the corresponding normalized eigenfunction φ2 (x √) as:
2 2
y2 (x ) x x 10
φ2 (x ) = " #1/2 = " #1/2 =
R1 R1 2
2 4
−1 {y2 (x )} dx −1 x dx

[12]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 64 / 73

Schmidt’s solution Continued...

Also, √ ! √
Z 1 Z 1
2 6x 2 6
f1 (x ) = f (x )φ1 (x )dx = (1 + x ) dx =
−1 −1 2 3
[13]
and, √ √
10x 2
Z 1 Z 1 !
8 10
f2 (x ) = f (x )φ2 (x )dx = (1 + x )2 dx =
−1 −1 2 15
[14]
Now,here from (3) and (10), λ = 1, λ1 = 3/2 and λ2 = 5/2
Hence λ 6= λ1 and λ 6= λ2 .

Dr. Anjana Pradhan Ghorai MA209 IE and GF 65 / 73

Schmidt’s solution Continued...
Therefore, (1) will possess a unique given by:
2
X fm
y (x ) = f (x ) + λ φm (x ),
m=1 λm − λ
2
2
X fm
= (1 + x ) + φm (x ),
m=1 λm − 1
f1 f2
= (1 + x )2 + φ1 (x ) + φ2 (x ),
λ1√ −1 √ λ2 − 1 √ √
2 (2 6/3)( 6/2)x (8 10/15)( 10/2)x 2
= (1 + x ) + +
3/2 − 1 5/2 − 1
25 2
= (x + 6x + 1)
9
25
=⇒ y (x ) = (x 2 + 6x + 1) [15]
9
Dr. Anjana Pradhan Ghorai MA209 IE and GF 66 / 73
Schmidt’s solution Continued...

Example:2 Using Hilbert-Schmidt’s theorem solve the symmetric

integral equation
Z π of second kind:
y (x ) = 1 + λ cos(x + t)y (t)dt
0
Solution: GivenZ π IE is:
y (x ) = 1 + λ cos(x + t)y (t)dt [1]
0 Z b
Comparing (1) with y (x ) = f (x ) + λ k(x , t)y (t)dt, [2]
a
we have: f (x ) = 1, k(x , t) = cos(x + t), a = 0 and b = π [3]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 67 / 73

Schmidt’s solution Continued...

Now , we start with obtaining the eigenvalue and normalized eigen-

functions Zof corresponding symmetric IE:
π
y (x ) = λ cos(x + t)y (t)dt [4]
0
Re-writingZ (4) as:
π Z π
y (x ) = λ cosx cost y (t)dt − λ sinx sint y (t)dt
0 Z π 0 Z π
or, y (x ) = λ cosx cost y (t)dt − λ sinx sint y (t)dt
0 0
or, y (x ) = λC1 cosx − λC2 sinx [5]
where,Z π
C1 = cost y (t)dt [6]
Z0π
C2 = sint y (t)dt [7]
0

Dr. Anjana Pradhan Ghorai MA209 IE and GF 68 / 73

Schmidt’s solution Continued...

Using Z(5), (6) becomes:

π
C1 = cost(λC1 cost − λC2 sint)dt
0
or, (2 − λπ)C1 + 0.C2 = 0 [8]
Using Z(5), (7) becomes:
π
C2 = sint(λC1 cost − λC2 sint)dt
0
or,0.C1 + (2 + λπ)C2 = 0 [9]
Equations (8) and (9) will have a non-trivial solution iff:
(2 − λπ) 0
D(λ) = = 0 =⇒ λ = 2/π or λ = −2/π.
0 (2 + λπ)
Hence the required eigenvalues are:
λ1 = 2/π or λ2 = −2/π. [10]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 69 / 73

Schmidt’s solution Continued...
Determination of eigenfunction corresponding to λ1 = 2/π
Putting λ = λ1 = 2/π in (8) and (9), we have :
C2 = 0 and C1 is arbitrary.
Putting these values of C1 and C2 with λ = 2/π in(5),
we have the first eigenfunction y1 (x ) as:
2
y1 (x ) = C1 cosx = cosx ,
π
seting C1 = π/2
Now, the corresponding normalized eigenfunction φ1 (x ) as:
!1/2
y1 (x ) cosx 2
φ1 (x ) = " #1/2 = " #1/2 = cosx
Rπ Rπ π
2 2
0 {y1 (x )} dx 0 cos xdx

[11]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 70 / 73

Schmidt’s solution Continued...

Determination of eigenfunction corresponding to λ2 = −2/π

Putting λ = λ2 = −2/π in (8) and (9), we have :
C1 = 0 and C2 is arbitrary.
Putting these values of C1 and C2 with λ = −2/π in(5),
we have the first eigenfunction y2 (x ) as:
y2 (x ) = −(−2/π)C2 sinx = sinx .
seting C2 = π/2
Now, the corresponding normalized eigenfunction φ2 (x ) as:
!1/2
y2 (x ) sinx 2
φ2 (x ) = " #1/2 = " #1/2 = sinx
Rπ Rπ π
2 2
0 {y2 (x )} dx 0 sin xdx

[12]

Dr. Anjana Pradhan Ghorai MA209 IE and GF 71 / 73

Schmidt’s solution Continued...
Z π Z π !1/2
2
Also, f1 (x ) = f (x )φ1 (x )dx = cosx dx = 0
0 0 π
[13]
Z π Z π !1/2 !1/2
2 2
and, f2 (x ) = f (x )φ2 (x )dx = sinx dx = 2
0 0 π π
[14]
Three cases arise:
case I:: If λ 6= λ1 and λ 6= λ2 . Then, (1) will possess a unique
solution given by:
2
X fm f1 f2
y (x ) = f (x )+λ φm (x ) = 1+ φ1 (x )+ φ2 (x ),
m=1 λm − λ λ1 − λ λ2 − λ
!1/2 !1/2
λ 2 2 4λsinx
=1+ 2 sinx = 1 −
[−(2/π) − λ] π π 2 + λπ
[15]
Dr. Anjana Pradhan Ghorai MA209 IE and GF 72 / 73
Schmidt’s solution Continued...

case II:: If λ 6= λ1 and λ = λ2 = −2/π. Then, since f2 6= 0 (1)

possess no solution.
case III:: If λ 6= λ1 = 2/π and λ 6= λ2 . Then, since f1 = 0 there
exists infinitely many solutions for (1) given by:
f2
y (x ) = f (x ) + Aφ1 (x ) + λ φ2 (x ),
λ2 − λ
where A is an arbitrary constant.
!1/2 !1/2 !1/2
2 2/π 2 2
or, y (x ) = 1+A cosx + 2 sinx
π [−(2/π) − (2/π)] π π
= 1 + Ccosx − (2/π)sinx , [16]
where C is an arbitrary constant.

Dr. Anjana Pradhan Ghorai MA209 IE and GF 73 / 73