Accepted Manuscript

Numerical solution of Volterra–Fredholm integral equations using

Legendre collocation method

S. Nemati

PII: S0377-0427(14)00437-3
DOI: http://dx.doi.org/10.1016/j.cam.2014.09.030
Reference: CAM 9815

To appear in: Journal of Computational and Applied

Mathematics
Received date: 16 April 2014
Revised date: 30 July 2014

Please cite this article as: S. Nemati, Numerical solution of Volterra–Fredholm integral
equations using Legendre collocation method, Journal of Computational and Applied
Mathematics (2014), http://dx.doi.org/10.1016/j.cam.2014.09.030

This is a PDF file of an unedited manuscript that has been accepted for publication. As a
service to our customers we are providing this early version of the manuscript. The manuscript
will undergo copyediting, typesetting, and review of the resulting proof before it is published in
its final form. Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
Manuscript
Click here to view linked References

Numerical solution of Volterra-Fredholm integral

equations using Legendre collocation method

S. Nemati
Department of Mathematics, Faculty of Mathematical Sciences, University of Mazandaran,
Babolsar, Iran

Abstract

A numerical method for solving the Volterra-Fredholm integral equations is

presented. The method is based upon shifted Legendre polynomials approx-
imation. The properties of shifted Legendre polynomials are first presented.
These properties together with the shifted Gauss-Legendre nodes are then uti-
lized to reduce the Volterra-Fredholm integral equations to the solution of a
matrix equation. An estimation of the error is given. Illustrative examples are
included to demonstrate the validity and applicability of the technique.
Keywords: Volterra-Fredholm integral equations, shifted Legendre
polynomials, Gauss-Legendre nodes, matrix equation
2010 MSC: 65R20, 42C10

1. Introduction

One of the fundamental classes of equations is integral equations. Such

equations arise in many areas of science and engineering (see, [1–3]) and play an
important role in the modeling of real-life phenomena in other fields of science
5 (see, [4]). For these reasons, integral equations have received much attention in
the last decades. Some of integral equations cannot be solved by the well-known
exact methods. Hence, it is desirable to introduce numerical methods with high
accuracy to solve these equations numerically.

Email address: [email protected] (S. Nemati)

Preprint submitted to Journal of Computational and Applied Mathematics July 30, 2014
10 Over the years, it was found that the spectral methods are a valid method
to obtain approximations for integral equations (see for example, [5–11]). Our
aim in this paper is to propose a method to approximate a class of Volterra-
Fredholm integral equations on the interval [0, l] by using the shifted Legendre
polynomials.
15

Consider the following Volterra-Fredholm integral equation

Z h(x) Z l
A(x)y(x)+B(x)y(h(x)) = f (x)+λ1 k1 (x, t)y(t)dt+λ2 k2 (x, t)y(h(t))dt
0 0
(1)
where the functions k1 (x, t) and k2 (x, t) are known kernel functions on the in-
terval [0, l] × [0, l] and the functions A(x), B(x), h(x) and f (x) are known
functions defined on the interval [0, l] and 0 ≤ h(x) < ∞, y(x) is the unknown
function and λ1 , λ2 are real constants such that λ1 2 + λ2 2 6= 0. When h(x) is
20 a first-order polynomial, equation (1) is called functional integral equation with
proportional delay. Numerical solution of this class of integral equations has
been introduced using Lagrange collocation method by K. Wang and Q. Wang
in [12]. Also, they have applied Taylor collocation method to solve equation
(1) numerically. In this paper, we will apply the shifted Legendre collocation
25 method to approximate the solution of equation (1).
The paper is organized as follows: Preliminaries needed hereafter are given
in Section 2. In Section 3, we design the shifted Legendre collocation technique
to solve equation (1). An estimation of the error is provided in Section 4, when
a sufficiently smooth function is expanded in terms of the shifted Legendre
30 polynomials. In Section 5, we present three numerical examples to illustrate
the validity and applicability of the method. Concluding remarks are given in
Section 6.

2
2. Preliminaries and notations

The well-known Legendre polynomials are defined on the interval [−1, 1] and
can be determined with the aid of the following recurrence formulae:

L0 (x) = 1,
L1 (x) = x,
Lm+1 (x) = m+1 xLm (x)
2m+1
− m+1 Lm−1 (x),
m
m = 1, 2, . . . , −1 ≤ x ≤ 1.

In order to use these polynomials on the interval [0, l] we define the so-called
shifted Legendre polynomials of degree i as follows:
2
ψi (x) = Li ( x − 1), i = 0, 1, 2, · · · .
l
We consider the space L2 [0, l] equipped with the following inner product and
norm: Z l 1
hf, gi = f (x)g(x)dx, kyk2 = hy, yi 2 .
0
The set of shifted Legendre polynomials forms a complete L2 [0, l]-orthogonal
system such that the orthogonality condition is

Z l  l , i = j,
2i+1
ψi (x)ψj (x)dx =
0  0, i=
6 j.

A function y(x) ∈ L2 [0, l], may be expressed in terms of shifted Legendre poly-
nomials as
∞
X
y(x) = ci ψi (x),
i=0
where the coefficients ci are given by
Z
2i + 1 l
ci = y(x)ψi (x)dx, i = 0, 1, 2, · · · .
l 0

Considering only the first (n + 1)-terms shifted Legendre polynomials we have

n
X
y(x) ≃ ci ψi (x) = C T ψ(x),
i=0

where the shifted Legendre coefficient vector C and the shifted Legendre vector
ψ(x) are given by
C = [c0 , c1 , . . . , cn ]T , (2)

3
ψ(x) = [ψ0 (x), ψ1 (x), . . . , ψn (x)]T . (3)

3. Numerical solution

35 In this section, we approximate the solution of equation (1) using shifted

Legendre polynomials. We assume that the known functions in equation (1)
satisfy the conditions that this equation has a unique solution [5, 14, 15, 16, 17].
We approximate the function y(x) using the way mentioned in the previous
section as follows:
n
X
y(x) ≃ ci ψi (x) = C T ψ(x), (4)
i=0
where the coefficients ci , i = 0, 1, . . . , n are the unknowns to be determined and
C and ψ(x) are defined as (2) and (3), respectively.
Using (4) we can consider that
n
X
y(h(x)) ≃ ci ψi (h(x)). (5)
i=0

Substituting equations (4) and (5) into equation (1) yields:

n n Z h(x) n
!
X X X
A(x) ci ψi (x)+B(x) ci ψi (h(x)) = f (x)+λ1 k1 (x, t) ci ψi (t) dt
i=0 i=0 0 i=0

Z n
!
l X
+ λ2 k2 (x, t) ci ψi (h(t)) dt. (6)
0 i=0
Suppose that:
Z h(x) Z l
gi (x) = A(x)ψi (x)+B(x)ψi (h(x))−λ1 k1 (x, t)ψi (t)dt−λ2 k2 (x, t)ψi (h(t))dt,
0 0

then, equation (6) can be rewritten as:

n
X
ci gi (x) = f (x). (7)
i=0

Collocating equation (7) in n + 1 roots of the shifted Legendre polynomial

ψn+1 (x), the shifted Gauss-Legendre nodes, we obtain:
n
X
ci gi (xj ) = f (xj ), f or j = 0, 1, . . . , n, (8)
i=0

4
which can be written in the following matrix form:

GT C = F,

where
F = [f (x0 ), f (x1 ), . . . , f (xn )]T ,

and
G = (gij ), i, j = 0, 1, . . . , n,

in which the elements of the matrix G are determined as follows:

gij = gi (xj ), i, j = 0, 1, . . . , n.

Finally, the unknown vector C can be computed by:

−1
C = (GT ) F.

Therefore, the approximate solution of equation (1) is given by y(x) = C T ψ(x).

4. Error estimate

40 In this section we are concerned with the convergence analysis of the pro-
posed method.
We assume that y(x) is a sufficiently smooth function on [0, l] and pn (x) is
the interpolating polynomial to y at points xi , where xi , i = 0, 1, . . . , n are the
roots of the (n + 1)-degree shifted Chebyshev polynomial in [0, l], then we have
n
y (n+1) (η) Y
y(x) − pn (x) = (x − xi ), η ∈ [0, l],
(n + 1)! i=0

therefore, we obtain
n+1
Mn (l)
|y(x) − pn (x)| ≤ , (9)
22n+1 (n + 1)!

where Mn = max0≤x≤l |y (n+1) (x)|.

Using (9) we obtain the following result.
Theorem 4.1. Suppose that the known functions in equation (1) are real

5
(n + 1)-times continuously differentiable functions on the bounded interval [0, l]
and yn (x) = C T ψ(x) be the shifted Legendre polynomials expansion of the
exact solution, y(x), where

C = [c0 , c1 , . . . , cn ]T ,

and
ψ(x) = [ψ0 (x), ψ1 (x), . . . , ψn (x)]T .
Pn
Let ȳn (x) = i=0 c̄i ψi (x) be the approximate solution obtained by the method
proposed in Section 3 and Mn = max0≤x≤l |y (n+1) (x)|, then, there exist real
numbers α and βn such that
n+1
Mn (l)
ky(x) − ȳn (x)k2 ≤ α + βn kC − C̄k2 , (10)
22n+1 (n + 1)!

where
C̄ = [c̄0 , c̄1 , . . . , c̄n ]T ,

and the norm on the right hand side is the usual Euclidian norm for vectors.
Proof. Let Rn [x] be the space of all real-valued polynomials of degree≤ n.
Using the definition, yn (x) and ȳn (x) are in Rn [x] and yn (x) is the best approx-
imation of y(x) in Rn [x]. We can write

ky(x) − ȳn (x)k2 ≤ ky(x) − yn (x)k2 + kyn (x) − ȳn (x)k2 . (11)

Using (9) we obtain

Z ! 12
l
2
ky(x) − yn (x)k2 = |y(x) − yn (x)| dx
0
 " #2  12
Z l n+1
Mn (l)
≤ dx
0 22n+1 (n + 1)!
√ Mn (l)n+1
= l 2n+1 . (12)
2 (n + 1)!

6
Also, we have
  21
Z l "X
n
#2
kyn (x) − ȳn (x)k2 =  (ci − c̄i )ψi (x) dx
0 i=0

Z l "X
n
#" n
X
# ! 21
≤ |ci − c̄i | 2
|ψi (x)| 2
dx
0 i=0 i=0
" n
# 21 n Z
! 21
X X l
= |ci − c̄i | 2
|ψi (x)| dx2

i=0 i=0 0

n
! 12
X 1
= kC − C̄k2 l . (13)
i=0
2i + 1

Finally from (11)–(13) we conclude that (10) is valid with

v !
u n
√ u X 1
α = l, βn = t l .
i=0
2i + 1

5. Numerical examples

In this section, some examples are given to demonstrate the applicability

of the proposed method. In order to demonstrate the error of the method we
introduce the notations:

ēn (x) = |y(x) − ȳn (x)|,

δ̄n = kēn (x)k2 ,

n+1
(l)
αn = ,
22n+1 (n + 1)!
where y(x) and ȳn (x) are the exact solution and the computed solution by the
45 presented method, respectively. Also, we use the notations βn and Mn which
were defined in Section 4. The computations were performed on a personal
computer using a 3.00 GHz processor and the codes were written in Mathematica
8. In all the examples of this section we have l = 1. The computing times
(seconds) to obtain the numerical solution ȳn (x) are given for these examples.

7
Example 5.1. Consider the following Volterra-Fredholm integral equation [13]
Z ex Z 1
(sin x)y(x) + (cos x)y(ex ) = f (x) + ex+t y(t)dt − ex+t y(et )dt,
0 0
x
50 where f (x) = 13 ex −1 + e3 +ex 2 − ee [2 + ex (−2 + ex )] +e2x cos x+x2 sin x.
The exact solution of this equation is y(x) = x2 . Tables 1 and 2 and Figure 1
show the numerical results for this example. In Table 1 the numerical results
for the error δ̄n of the proposed method with n = 2, 3, 4 (Method 1) are dis-
played together with the results obtained by the Taylor collocation method [13]
55 (Method 2), the Taylor polynomial method [10] (Method 3) and the Lagrange
collocation method [12] (Method 4).

Table 1: Comparison of the errors for Example 5.1.

n Method 1 Method 2 Method 3 Method 4
2 1.19 × 10 −15
7.64 × 10−15
1.96 × 10 −15
2.82 × 10−15
3 8.53 × 10−16 1.22 × 10−14 2.75 × 10−15 1.36 × 10−14
4 5.20 × 10−15 3.41 × 10−14 4.42 × 10−15 1.91 × 10−13

Table 2: Numerical results for Example 5.1.

n Mn αn βn kC − C̄k2 Computing time
2 0 5.20833 × 10 −3
1.23828 3.59035 × 10 −15
1.671
3 0 3.25521 × 10−4 1.29468 1.15678 × 10−15 2.672
4 0 1.62760 × 10 −5
1.33690 9.08850 × 10 −15
4.359

Example 5.2. Consider the Volterra-Fredholm integral equation

Z 2x Z 1
x2 y(x) + ex y(2x) = f (x) + ex+t y(t)dt − ex−2t y(2t)dt,
0 0
x
where f (x) = − e4 − 1 −2+x
4e cos 2 + 1 3x
2e cos 2x − 1 −2+x
4e sin 2 + x2 sin x +
60 ex sin 2x − 12 e3x sin 2x.

8
en HxL

8. ´ 10-15
n=2

6. ´ 10-15
n=3
-15
4. ´ 10

n=4
2. ´ 10-15

x
0.2 0.4 0.6 0.8 1.0

Figure 1: Plot of the ēn (x), with n = 2, 3, 4 for Example 5.1.

Its exact solution is y(x) = sin x. The numerical results of this example are
displayed in Tables 3, 4 and Figure 2. In Table 3, the error δ̄n of the proposed
method with n = 2, 5, 8, 9 (Method 1) are compared with the results obtained by
the Taylor collocation method [13] (Method 2), the Taylor polynomial method
65 [10] (Method 3) and the Lagrange collocation method [12] (Method 4).

Table 3: Comparison of the errors for Example 5.2.

n Method 1 Method 2 Method 3 Method 4
2 1.46 × 10−2 7.87 × 10−2 3.41 × 10−2 7.87 × 10−2
5 2.93 × 10−5 6.23 × 10−5 3.68 × 10−4 6.23 × 10−5
8 3.94 × 10−8 1.89 × 10−8 1.24 × 10−5 1.77 × 10−7
9 2.29 × 10−9 2.35 × 10−8 3.46 × 10−7 7.21 × 10−6

Example 5.3. As a final test problem, consider the following Volterra-Fredholm

9
Table 4: Numerical results for Example 5.2.
n Mn αn βn kC − C̄k2 Computing time
2 1.00000 5.20833 × 10 −3
1.23828 1.76483 × 10 −2
1.689
5 0.84148 6.78168 × 10 −7
1.37048 7.85257 × 10 −5
4.938
8 1.00000 2.10246 × 10−11 1.44244 3.03147 × 10−7 11.906
9 0.84148 5.25614 × 10 −15
1.46057 5.08242 × 10 −9
15.328

integral equation
Z h(x) Z 1
y(x) = f (x) + k1 (x, t)y(t)dt − k2 (x, t)y(h(t))dt,
0 0

where h(x) = Ln(x + 1), k1 (x, t) = ex+t , k2 (x, t) = ex+h(t) , f (x) = e−x −
ex (h(x) − 1).
70 The exact solution of this problem is y(x) = e−x . The numerical results of this
example are displayed in Tables 5, 6 and Figure 3. In Table 5, the error δ̄n
of the proposed method with n = 2, 5, 8, 9 (Method 1) are compared with the
results obtained by the Taylor collocation method [13] (Method 2), the Taylor
polynomial method [10] (Method 3) and the Lagrange collocation method [12]
75 (Method 4). Figure 3 displays the exact solution and approximate solution with
n = 1, 2.

Table 5: Comparison of the errors for Example 5.3.

n Method 1 Method 2 Method 3 Method 4
2 1.99 × 10 −3
3.27 × 10 −3
3.59 × 10 −2
3.27 × 10−3
5 2.56 × 10−7 4.30 × 10−7 3.05 × 10−4 4.30 × 10−7
8 2.78 × 10−8 5.96 × 10−8 5.61 × 10−7 5.78 × 10−7
9 2.35 × 10−8 8.84 × 10−8 1.41 × 10−7 1.86 × 10−5

10
Table 6: Numerical results for Example 5.3.
n Mn αn βn kC − C̄k2 Computing time
2 2.71828 5.20833 × 10 −3
1.23828 9.13839 × 10 −4
3.406
5 2.71828 6.78168 × 10 −7
1.37048 9.06751 × 10 −8
14.313
8 2.71828 2.10246 × 10−11 1.44244 7.71625 × 10−9 33.938
9 2.71828 5.25614 × 10 −15
1.46057 7.71737 × 10 −9
43.985

6. Conclusions

80 In this paper, we have proposed a collocation method based on the shifted

Legendre polynomials to numerically solve a spacial class of Volterra-Fredholm
integral equations. This method uses the shifted Gauss-Legendre nodes to re-
duce the considered Volterra-Fredholm integral equations to the solution of a
matrix equation. It is seen from the numerical experiments that the convergence
85 rate of the numerical solution is similar to the convergence rate of the best ap-
proximation of the solution by shifted Legendre polynomials, when n increases.
The results in Tables 2, 4 and 6 confirm the theoretical analysis of error. The
comparison of the obtained results with those based on other methods shows
that the present method is a powerful tool for finding the numerical solutions
90 of such equations.

References

[1] F. Bloom, Asymptotic bounds for solutions to a system of damped integro-

differential equations of electromagnetic theory, J. Math. Anal. Appl. 73
(1980) 524–542.

95 [2] M.A. Abdou, Fredholm-Volterra integral equation of the first kind and
contact problem, Appl. Math. Comput. 125 (2002) 177–193.

[3] S.A. Isaacson, R.M. Kirby, Numerical solution of linear Volterra integral
equations of the second kind with sharp gradients, J. Comput. Appl. Math.
235 (2011) 4283–4301.

11
100 [4] J. Frankel, A Galerkin solution to regularized Cauchy singular integro-
differential equations, Q. Appl. Math. 24 (1995) 145–258.

[5] H. Brunner, on the numerical solution of Volterra-Fredholm integral equa-

tion by collocation methods, SIAM J. Numer. Anal. 27 (4) (1990) 87–96.

[6] K. Maleknejad, H. Almasieh,M. Roodaki, Triangular functions (TFs)

105 method for the solution of nonlinear Volterra-Fredholm integral equations,
Commun. Nolin. Sci. Numer. Simula. 15 (2010) 3293–3298.

[7] Y. Ordokhani, Solution of nonlinear Volterra-Fredholm-Hammerstein inte-

gral equations via rationalized Harr functions, Appl. Math. Comput. 180
(2006) 436–443.

110 [8] Y. Ordokhani, M. Razzaghi, Solution of nonlinear Volterra-Fredholm-

Hammerstein integral equations via a collocation method and rationalized
Harr functions, Appl. Math. Lett. 21 (2008) 4–9.

[9] S. Yalcinbas, Taylor polynomial solution of nonlinear Volterra-Fredholm

integral equations, Appl. Math. Comput. 127 (2002) 195–206.

115 [10] K. Maleknejad, Y. Mahmoudi, Taylor polynomial solution of high-order

nonlinear Volterra-Fredholm integro-differential equations, Appl. Math.
Comput. 145 (2003) 641–653.

[11] S. Yousei, M. Razzaghi, Legendre Wavelets method for the nonlinear

Volterra-Fredholm integral equations, Math. Comput. Simula. 70 (2005)
120 419–428.

[12] K.Y. Wang, Q.S. Wang, Lagrange collocation method for solving Volterra-
Fredholm integral equations, Appl. Math. Comput. 219 (2013) 10434–
10440.

[13] K. Wang, Q. Wang, Taylor collocation method and convergence analysis

125 for the Volterra-Fredholm integral equations, J. Comput. Appl. Math. 260
(2014) 294–300.

12
[14] L.M. Delves, J.L. Mohamed, Computational Methods for Integral Equa-
tions, Cambridge University Press, Cambridge, 1985.

[15] O. Dickman, Thresholds and traveling waves for the geographical spread
130 of infection, J. Math. Biol. 6 (2) (1978) 109–130.

[16] H. Guoqiang, Z. Liqing, Asymptotic expansion for the trapezoidal Nystrom

method of linear Volterra-Fredholm equation, J. Comput. Appl. Math. 51
(3) (1994) 339–348.

[17] L. Hacia, On approximate solution for integral equation of mixed type,

135 ZAMM 76 (1996) 415–428.

13
e2 HxL

0.020

0.015

0.010

0.005

x
0.2 0.4 0.6 0.8 1.0
e5 HxL
0.0001

0.00008

0.00006

0.00004

0.00002

x
0.2 0.4 0.6 0.8 1.0
e8 HxL
1.2 ´ 10-8

1. ´ 10-8

8. ´ 10-9

6. ´ 10-9

4. ´ 10-9

2. ´ 10-9

x
0.2 0.4 0.6 0.8 1.0
e9 HxL

8. ´ 10-10

6. ´ 10-10

4. ´ 10-10

2. ´ 10-10

x
0.2 0.4 0.6 0.8 1.0

Figure 2: Plot of the ēn (x), with n = 2, 5, 8, 9 for Example 5.2.

14
1.0

0.9
y1 @xD
0.8

0.7
y2 @xD
0.6

0.5 y@xD

0.4

x
0.0 0.2 0.4 0.6 0.8 1.0

Figure 3: Plot of the exact solution and approximate solution with n = 1, 2 for Example 5.3.

15