Applied Mathematics and Computation 186 (2007) 212–218

www.elsevier.com/locate/amc

Numerical computational method in solving Fredholm integral

equations of the second kind by using Coifman wavelet
a,*
K. Maleknejad , T. Lotfi b, Y. Rostami c

a
School of Mathematics, Iran University of Science and Technology, Narmak, Tehran 16844, Iran
b
Department of Mathematics, Islamic Azad University (Hamadan Unit), Hamadan, Iran
c
Department of Mathematics, Islamic Azad University (Karaj Branch), Karaj, Iran

Abstract

In this paper, we bring three theorems that enable us to approximate the solution of Ferdholm integral equations of the
second kind. Then we use Coifman wavelets or Coiflets as scaling functions for projection that satisfied the conditions of
theorems for approximation. Also we use this projection to convert the integral equation to a Galerkin system, which is the
most important of the expansion methods for solving linear integral equations. Finally, by using numerical examples we
show that our estimation have a good degree of accuracy.
 2006 Elsevier Inc. All rights reserved.

Keywords: Integral equation; Coifman wavelet; Multiresolution; Scale function approximation; Galerkin system

1. Introduction

This section provide an overview of the topics that we need in this paper. A wave is usually defined as an
oscillating function of time or space, such as a sinusoid. Fourier analysis is wave analysis, it expands a signal
or function in term of sinusoid. A wavelet is a ‘‘small wave’’, which has its energy concentrated in time to give
a tool for the analysis of transient, nonstationary, or time-varying phenomena.
We will take wavelets and use them in series expansion of signals or functions as the same way a Fourier
series uses the wave or sinusoid to represent a signal or function [1].

1.1. Wavelet and wavelets expansion systems

A signal or function f(t) can be often better analyzed, described, or processed if expressed as a linear decom-
position by
X
f ðtÞ ¼ al wl ðtÞ; ð1Þ
l

*
Corresponding author.
E-mail addresses: maleknejad@iust.ac.ir (K. Maleknejad), lotfi@iauh.ac.ir (T. Lotfi).

0096-3003/$ - see front matter  2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2006.06.127
 K. Maleknejad et al. / Applied Mathematics and Computation 186 (2007) 212–218 213

where l is an integer index for the finite or infinite sum, al are the real valued expansion coefficients, and
wl(t) are a set of real valued functions of t called expansion set. If the expansion (1) is unique, the set is called
a basis for the class of function that can be so expressed. If the basis is orthogonal, meaning
Z
hwk ðtÞ; wl ðtÞi ¼ wk ðtÞwl ðtÞdt ¼ 0; k 6¼ l; ð2Þ

then the coefficients can be calculated by the inner product

Z
ak ¼ hf ðtÞ; wk ðtÞi ¼ f ðtÞwk ðtÞdt: ð3Þ

For the wavelet expansion, a two parameter system is constructed such that (1) becomes
XX
f ðtÞ ¼ aj;k wj;k ðtÞ; ð4Þ
k j

where both j and k are integer indices and the wj,k(t) are the wavelet expansion functions that usually form an
orthogonal basis. The set of expansion coefficients aj,k are called the discrete wavelet transform (DWT) of f(t)
and (4) is the inverse transform [2].

1.2. The discrete wavelet transform

Our goal is to generate a set of expansion functions such that any signal in L2 ðRÞ (the space of square inte-
gral functions) can be represented by the series
X
f ðtÞ ¼ aj;k 2j=2 wð2j t  kÞ; ð5Þ
j;k

now let wj;k ðtÞ ¼ 2j=2 wð2j t  kÞ; j; k 2 Z, then

X
f ðtÞ ¼ aj;k wj;k ðtÞ: ð6Þ
j;k

A more specific form indicating how the aj,k are calculated can be written using inner products if the wj,k(t)
form an orthonormal basis for the space of signals of interest, we can write
X
f ðtÞ ¼ hwj;k ðtÞ; f ðtÞiwj;k ðtÞ: ð7Þ
j;k

1.3. The scaling function

In order to use the idea of multiresolution, we will start by defining the scaling function and then define the
wavelet in terms of it [3]. As we described for the wavelet in previous section, we define a set of scaling func-
tions in terms of integer translates of the basic scaling function by
/k ðtÞ ¼ /ðt  kÞ; k 2 Z; / 2 L2 : ð8Þ
The subspace of L2 ðRÞ spanned by these functions is defined as
V 0 ¼ Spank2Z f/k ðtÞg: ð9Þ
The over-bar denotes closure. This means that
X
f ðtÞ ¼ ak /k ðtÞ 8f ðtÞ 2 V 0 : ð10Þ
k

One can generally increase the size of the subspace by changing the time scale of the scaling functions. A two-
dimensional family of function is generated from the basic scaling function by scaling and translation by
/j;k ¼ 2j=2 /ð2j t  kÞ; ð11Þ
214 K. Maleknejad et al. / Applied Mathematics and Computation 186 (2007) 212–218

whose span over k is

V j ¼ Spank f/k ð2j tÞg ¼ Spank f/j;k ðtÞg 8k 2 Z: ð12Þ
This means that if f(t) 2 Vj, then it can be expressed as
X
f ðtÞ ¼ ak /ð2j t þ kÞ: ð13Þ
k

1.4. Multiresolution analysis (MRA)

In order to agree with our intuitive ideas of scale or resolution, we formulate the basic requirement of
multiresolution analysis (MRA) by requiring a nesting of the spanned spaces as [1]:
    V 2  V 1  V 0  V 1  V 2      L2 ; ð14Þ
V j  V jþ1 8j 2 Z ð15Þ
with
V 1 ¼ f0g; V 1 ¼ L2 : ð16Þ
The space that contains high resolution signals will contain those of lower resolution also. Because of the def-
inition of Vj, the spaces have to satisfy a natural scaling condition:
f ðtÞ 2 V i () f ð2tÞ 2 V jþ1 ; ð17Þ
which ensures elements in a space are simply scaled version of the elements in the next space. The nesting of
the space of /(2jt  k), denoted by Vj and shown in (14) and (17), is achieved by requiring that /(t) 2 V1,
which means that if /(t) 2 V0, it is also in V1, the space spanned by /(2t). This means /(t) can be expressed
in terms of a weighted sum of shifted /(2t) as
X pffiffiffi
/ðtÞ ¼ hðnÞ 2/ð2t  nÞ; n 2 Z; ð18Þ
n

where the coefficients h(n) are a sequence of real or perhapspffifficomplex

ffi numbers called the scaling function coef-
ficient (or the scaling filter or the scaling vector) and the 2 maintains the norm of the scaling function with
the scale of two.

• This recursive equation is fundamental to the theory of the scaling function and is, in some ways, analogous
to a differential equation with coefficients h(n) and solution /(t) that may or may not exist or be unique [3].

1.5. The wavelet functions

The important features of a signal can be better described or parameterized, not by using /j,k(t) and increas-
ing j to increase the size of the subspace spanned by the scaling functions, but by defining a slightly different set
of functions wj,k(t) that span the differences between the spaces spanned by the various scales of the scaling
function. These functions are the wavelet [4].
There are several advantages to requiring that the scaling functions and wavelets to be orthogonal. Orthog-
onal basis functions allow simple calculation of expansion coefficients. The orthogonal complement of Vj in
Vj+1 is defined as Wj. This means that all members of Vj are orthogonal to all members of Wj. We require
Z
h/j;k ðtÞ; wj;l ðtÞi ¼ /j;k ðtÞwj;l ðtÞdt ¼ 0 ð19Þ

for all appropriate j; k; l 2 Z. The relationship of the various subspaces can be seen from the following expres-
sions. From (14) we see that we may start at any Vj, say at j = 0, and write
V 0  V 1  V 2      L2 : ð20Þ
 K. Maleknejad et al. / Applied Mathematics and Computation 186 (2007) 212–218 215

We now define the wavelet spanned subspace W0 such that

V 1 ¼ V 0  W 0; ð21Þ
which extends to
V 2 ¼ V 0  W 0  W 1: ð22Þ
In general this gives
L2 ¼ V 0  W 0  W 1     ð23Þ
when V0 is the initial space spanned by the scaling function /(t  k). Since these wavelets reside in the space
spanned by the next narrower scaling function, W0  V1, they can be represented by weighted sum of shifted
scaling function /(2t) defined in (18) by [1]:
X pffiffiffi
wðtÞ ¼ h1 ðnÞ 2/ð2t  nÞ ð24Þ
n2Z

for some set of coefficients h1(n) = (1)nh(1  n). The function generated by (24) gives the prototype or
mother wavelet w(t) for a class of expansion functions of the form
wj;k ¼ 2j=2 wð2j t  kÞ; ð25Þ
j j j/2 2
where 2 is the scaling of t, 2 t  k is the translation in t, and 2 maintain the L norm of the wavelet and
different scales [1].

2. Approximation of signals by scaling function projection

We define the kth moments of /(t) and w(t) as

Z Z
mðkÞ ¼ tk /ðtÞdt; m1 ðkÞ ¼ tk wðtÞdt; ð26Þ

and the discrete kth moments of h(n) and h1(n) as

X X
lðkÞ ¼ nk hðnÞ; l1 ðkÞ ¼ nk h1 ðnÞ: ð27Þ
n n

The orthogonal projection of a signal f(t) on the scaling function subspace Vj is given and denoted by
X
P j ff ðtÞg ¼ hf ðtÞ; /j;k ðtÞi/j;k ðtÞ; ð28Þ
k

which gives the component of f(t) which is in Vj and which is the best least squares approximation to f(t) in Vj.
We now state an important relation of the projection (28) as an approximation to f(t) in terms of the number
of zero wavelet moments and the scale [1].
Theorem 1. If m1(l) = 0 for l = 0, 1, . . . , L then the L2 error is
e1 ¼ kf ðtÞ  P j ff ðtÞgk2 6 C 1 2jðLþ1Þ ; ð29Þ
where C1 is a constant independent of j and L, but depend on f(t) and the wavelet system [1].
This theorem states [1,2]: A second approximation involves using the samples of f(t) as the inner product
coefficients in the wavelet expansion of f(t) in (28). We denotes this sampling approximation by
X
S j ff ðtÞg ¼ 2j=2 f ðk=2j Þ/j;k ðtÞ: ð30Þ
k

Theorem 2. If m(l) = 0 for l = 1, . . . , L then the L2 error is

e2 ¼ kf ðtÞ  S j ff ðtÞgk2 6 C 2 2jðLþ1Þ ; ð31Þ
where C2 is a constant independent of j and L, but depend on f(t) and the wavelet system [1].
216 K. Maleknejad et al. / Applied Mathematics and Computation 186 (2007) 212–218

This is a similar approximation or converges to the previous theorem results, but relates the projection of
f(t) on j-scale subspace to the sampling approximation in that same subspace. This ‘‘vector space’’ shows the
nature and relationship of the two types of approximations. The use of samples as inner products is an
approximation within the expansion subspace Vj. The use of a finite expansion to represent a signal f(t) is
an approximation from L2 onto subspace Vj. Theorems (1) and (2) shows the nature of those approximations,
which is very good for wavelet.
If we consider a wavelet system where the number of scaling functions and wavelets are set zero and this
number is as large as possible, then the following results is true [1].
Theorem 3. If m1(l) = m(l) = 0 for l = 1, . . . , L, and m1(0) = 0, then the L2 error is
e3 ¼ kf ðtÞ  S j ff ðtÞgk2 6 C 3 2jðLþ1Þ ; ð32Þ
where C3 is a constant independent of j and L, but depend on f(t) and the wavelet system.

3. Coifman wavelet-Galerkin method by scaling function projection

In this section, we make Galerkin system, that its projection uses Coifman wavelet as basis. We first
describe general scaling function projection, and then, discuss about special case, where say it Coifman
wavelet.
Consider a linear Fredholm integral equation of second kind:
Z b
xðsÞ ¼ yðsÞ þ kðs; tÞxðtÞdt; ð33Þ
a

where k 2 L2([a, b] · [a, b]), x 2 L2([a, b]), and s 2 [a, b]. Set operator K as below [6]:
Z b
ðKxÞðsÞ ¼ kðs; tÞxðtÞdt:
a

By this notation, we can write (33), as

ðI  KÞx ¼ y: ð34Þ

3.1. Wavelet-Galerkin method

Now we approximate solution (signal) x(s), and kernel k(s, t) by wavelet in m scale [7,8]:
X
xðsÞ ¼ ~xi 2m=2 /ð2m s  iÞ; ð35Þ
i2Z
XX
kðs; tÞ ¼ ~k i;l 2m=2 /ð2m s  iÞ2m=2 /ð2m t  lÞ; ð36Þ
i2Z l2Z

and
X
yðsÞ ¼ ~y i 2m=2 /ð2m s  iÞ; ð37Þ
i2Z

where ~xi ; ~y i and ~k i;l are the wavelet coefficient of x(s), y(s) and k(s, t), respectively. By setting u ¼ 2m s; v ¼
2m t; xi ¼ ~xi 2m=2 ; y i ¼ ~y i 2m=2 and k i;l ¼ ~k i;l 2m , from (35)–(37), we have [7–9]:
X
xðsÞ ¼ xi /ðu  iÞ; ð38Þ
i2Z
XX
kðs; tÞ ¼ k i;l /ðu  iÞ/ðv  lÞ; ð39Þ
i2Z l2Z

and
X
yðsÞ ¼ y i /ðu  iÞ: ð40Þ
i2Z
 K. Maleknejad et al. / Applied Mathematics and Computation 186 (2007) 212–218 217

By applying Eqs. (38)–(40) in integral equation (33), we have [9]:

X X Z bXX X
xi /ðu  iÞ ¼ y i /ðu  iÞ þ k i;l /ðu  iÞ/ðv  lÞ xi /ðv  iÞdv: ð41Þ
i2Z i2Z a i2Z l2Z i2Z

Now by orthonormality of bases, we have [2,4]:

Z b
/ðu  iÞ/ðu  lÞdu ¼ di;l ; ð42Þ
a

and therefore by taking the inner product of both sides of Eq. (41) with /(u  i), we have [2]:
X
xi  k i;l xl ¼ y i ; i 2 Z: ð43Þ
l2Z

We can write this system in compact form as below:

ðI  KÞx ¼ y; ð44Þ
where
I ¼ ½di;l ; K ¼ ½k i;l ; x ¼ ½xi ; and y ¼ ½y i :

4. Numerical performances

For showing efficiency of numerical method, we consider the following examples. We note that [6]:
Z 12 !12
1
1 XN
keN k ¼ e2N ðtÞdt  e2 ðxi Þ ;
1 N i¼0 N

where

eðsi Þ ¼ xðsi Þ  xN ðsi Þ; i ¼ 0; 1; . . . ; N :

Such that xN(si) and x(si) are, respectively the approximate and exact solutions of the integral equations.
Notation 1. In the following examples, we consider Coiflet scaling function and coefficient by N = 6 [1,5].

4.1. Examples
R p=2
Example 1. Consider xðsÞ ¼ sin s  s þ 0 stxðtÞdt with exact solution x(s) = sin s.
sþ1 R1
Example 2. Consider xðsÞ ¼ es  e sþ11 þ est xðtÞdt with exact solution x(s) = es.
0

R1 s; s 6 t
Example 3. Consider xðsÞ ¼ s þ 0 Kðs; tÞxðtÞdt; Kðs; tÞ ¼ with exact solution x(s) = sec 1 sin s.
t; s P t
Table 1 shows the computed error keNk for the before examples.

Table 1
Errors keNk at scale m = 4, and Coifman wavelet
N Example 1 Example 2 Example 3
2 2
2 3.2 · 10 4.1 · 10 5.7 · 102
3 5.2 · 103 3.7 · 103 3.7 · 103
4 4.7 · 106 6.3 · 105 5.9 · 104
5 1.3 · 109 9.4 · 107 4.3 · 107
6 2.1 · 1012 2.7 · 1010 8.1 · 109
218 K. Maleknejad et al. / Applied Mathematics and Computation 186 (2007) 212–218

5. Conclusion

We know that, the choice of basis is always important in determining the conditioning of the linear equa-
tions and hence the stability of the calculation against both quadrature and round-off errors [6]. Therefore we
use scaling function projection with Coifman wavelet to obtain orthonormal basis, which is very useful for
projection methods, since an orthonormal basis has the advantage that it guarantees the stability of the matrix
equations in Galerkin Method. Morever, in this paper we see that,

• The goal of most expansion of a function or signal is to have the coefficients of the expansion aj,k give more
useful information about the signal than is directly obvious from the signal itself [1–3].
• A second goal is to have most of the coefficients be zero or very small. This is what is called a sparse rep-
resentation and is extremely important in applications for statistical estimation and detection, data com-
pression, noise reduction, and fast algorithms [1–4].
• At any given scale, the projection of the signal on the subspace at that scale approaches the function itself as
the number of zero wavelet moments goes to infinity.
• For any given length, the projection goes to the function as the scale goes to infinity.
• These approximation converges exponentially fast.

References

[1] G.S. Burrus, R.A. Gopinath, H. Guo, Introduction to Wavelets and Wavelet Transform, Prentice Hall, 1998.
[2] G.G. Walter, X. Shen, Wavelets and Other Orthogonal Systems, second ed., Chapman and Hall (CRC), 2001.
[3] H.L. Resnikoff, R.O. Wells, Wavelet Analysis, Springer, 1998.
[4] Ingrid Daubechies, Ten Lecture on Wavelets, SIAM, Philadelphia, PA, 1992.
[5] Jun Tian, The Mathematical Theory and Applications of Biorthonormal Coifman Wavelet Systems, PhD thesis, Rice University,
February 1996.
[6] L.M. Delves, J.L. Mohamed, Computational Methods for Integral Equations, Cambridge University Press, 1988.
[7] K. Maleknejad, D. Rahbar, Numerical solution of Fredholm integral equation of the second kind by using B-spline function, Int. J.
Eng. Sci. 13 (5) (2000) 9–17.
[8] K. Maleknejad, F. Mirzaee, Using rationalized haar wavelet for solving linear integral equations, Appl. Math. Comput. 160 (2) (2005)
579–587.
[9] K. Maleknejad, H. Mesgarani, T. Nizad, Wavelet-Galerkin solution for Fredholm integral equation of the second kind, Int. J. Eng. Sci.
13 (5) (2002) 75–80.