Vous êtes sur la page 1sur 12

Mathematical and Computer Modelling 55 (2012) 1833–1844

Contents lists available at SciVerse ScienceDirect

Mathematical and Computer Modelling


journal homepage: www.elsevier.com/locate/mcm

Applications of two-dimensional triangular functions for solving


nonlinear class of mixed Volterra–Fredholm integral equations
K. Maleknejad ∗ , Z. JafariBehbahani
Department of Mathematics, Karaj Branch, Islamic Azad University, Karaj, Iran

article info abstract


Article history: An efficient numerical method is proposed for solving nonlinear mixed type Volterra–
Received 13 March 2011 Fredholm integral equations, using two-dimensional orthogonal triangular functions
Accepted 10 November 2011 (2D-TFs) in a direct approach. Some properties of 2D-TFs are novelty determined and an
operational matrix for integration in mixed type is prepared. Since this approach does
Keywords: not need any integration, all calculations would be easily implemented, and it has several
Two-dimensional orthogonal triangular
advantages in reducing computational burden. Finally by comparison of numerical results,
functions
Nonlinear mixed Volterra–Fredholm
accuracy and efficiency of the method will be shown.
integral equations © 2011 Elsevier Ltd. All rights reserved.
Direct method
Operational matrix

1. Introduction

Various problems in physics, mechanics and biology arise to a nonlinear mixed type Volterra–Fredholm integral equation.
Such equations also appears in modeling of the spatio-temporal development of an epidemic, theory of parabolic initial-
boundary value problems, population dynamics, and Fourier problems [1–3].
In general form, a mixed Volterra–Fredholm integral equation can be written as
 s
f ( s, t ) = g ( s, t ) + U (s, t , x, y, f (x, y))dydx, (1)
0 Ω

where f (s, t ) is an unknown scalar valued function defined on


D = [0, T ] × Ω ,
and Ω is a closed subset of Rn , n = 1, 2, 3. The functions g (s, t ) and U (s, t , x, y, f ) are given functions defined on D and
s = {(s, t , x, y, f ) : 0 ≤ x ≤ s ≤ T , t ∈ Ω , y ∈ Ω },
respectively [4]. It is obvious that any finite interval [0, T ] can be transformed to [0, 1] by a linear map, so we suppose that
[0, T ] = [0, 1] and Ω = [0, 1] without loss of generality. Moreover, we assume U (s, t , x, y, f ) = k(s, t , x, y)[f (s, t )]p , for
convenience. p is a positive integer.
Actually few numerical methods have been known for approximating the solutions of Eq. (1). Kauthen presented
continuous time collocation method for linear case of Eq. (1) and analyzed the discrete convergence properties [5]. The
results of Kauthen had been extended to nonlinear case by Brunner [6]. Hacia used projection methods for solving linear

∗ Corresponding author. Tel.: +98 21 732 254 16; fax: +98 21 732 284 16.
E-mail addresses: maleknejad@iust.ac.ir (K. Maleknejad), zjbehbahani@kiau.ac.ir (Z. JafariBehbahani).
URL: http://webpages.iust.ac.ir/maleknejad (K. Maleknejad).

0895-7177/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mcm.2011.11.041
1834 K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844

case of Eq. (1) [7,8]. Guoqiang gave asymptotic error expansion for the trapezoidal Nystrom and Euler Nystrom method
[9,10]. Methods based on Adomian decomposition series for approximation the solution of Eq. (1) had been presented by
Maleknejad and Hadizadeh [11], Wazwaz [12]. Moreover nonlinear systems of mixed Volterra–Fredholm integral equations
had been solved numerically by Maleknejad and Fadaei Yami [13]. Hadizadeh et al. obtained a numerical solution of
linear Volterra–Fredholm integral equations of mixed type using the bivariate Chebyshev collocation approach [14]. Also
Banifatemi et al. introduced a method for solving Eq. (1) using two dimensional Legendre wavelets [15]. Yousefi et al. used
He’s variational iteration method for Eq. (1) [16]. The homotopy perturbation method for mixed Volterra–Fredholm integral
equations had been applied by Yildirim [17]. Recently, a direct method with two-dimensional block-pulse functions had
been presented for special case of Eq. (1) [18].
This paper uses two-dimensional triangular functions (2D-TFs) introduced by Babolian et al. [19] for solving Eq. (1). This
purpose needs some additional properties of 2D-TFs, and we obtain them in future sections. Furthermore an operational
matrix for approximating the mixed integral part in Eq. (1), is novelty presented in this article.
By our present method Eq. (1) can be easily reduced to a nonlinear system of algebraic equations. The properties of 2D-TFs
lead to this system being set up in very fast and simple manner.
Finally, the accuracy and efficiency of proposed method is checked on some numerical examples.

2. A brief review of triangular functions

2.1. One-dimensional triangular functions

In an m-set of one-dimensional triangular functions (1D-TFs) over interval [0, 1), the ith left hand and right hand
functions are defined as
s − ih

Ti1 (s) = 1 − h , ih ≤ s < (i + 1)h,
0, otherwise ,
s − ih

Ti2 (s) =
, ih ≤ s < (i + 1)h,
h
0, otherwise ,

where i = 0, 1, 2, . . . , m − 1 and h = 1
m
. We can also define
T
T1(s) = T01 (s), T11 (s), ..., Tm1 −1 (s) ,

T
T2(s) = T02 (s), T12 (s), ..., Tm2 −1 (s) ,
 

and

T1(s)
 
T(s) = ,
T2(s)

and the vector T(s) is called 1D-TFs vector.


It is obvious that {Ti1 (s)}m
i=0 and {Ti (s)}i=0 are disjoint. So
−1 2 m−1

T1(s) · T1T (s) ≃ diag(T1(s)) = T1


 (s),
T1(s) · T2T (s) ≃ 0m×m ,
(2)
T2(s) · T1T (s) ≃ 0m×m ,
T2(s) · T2T (s) ≃ diag(T2(s)) = T2
 (s),

 (s) and T2
where T1  (s) are m × m diagonal matrices [20]. Orthogonality of 1D-TFs are shown in [21], that is,
 1
p q
Ti (s)Tj (s)ds = ∆p,q δi,j , (3)
0

where δi,j denotes Kronecker delta function and

h

 ,
 p = q ∈ {1, 2}
∆p,q = 3
h

, p ̸= q.
6
K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844 1835

So,
 1  1
h
T1(t )T1T (t )dt = T2(t )T2T (t )dt = Im×m ,
0 0 3
 1  1
h
T1(t )T2T (t )dt = T2(t )T1T (t )dt = Im×m .
0 0 6
s s
T1(τ )dτ and T2(τ )dτ in terms of 1D-TFs follows

Expressing 0 0
 s
T1(τ )dτ ≃ P1s · T1(s) + P2s · T2(s), (4)
0
 s
T2(τ )dτ ≃ P1s · T1(s) + P2s · T2(s), (5)
0

where P1sm×m and P2sm×m , the operational matrices of integration in 1D-TF domain, can be represented as
0 1 1 ··· 1
 
0 0 1 ··· 1
h 0 0 0 ··· 1
P1s =  ,
2 .. .
..
.
..
.
..
.
.. 
.
0 0 0 ··· 1
1 1 1 ··· 1
 
0 1 1 ··· 1
h 0 0 1 ··· 1
P2s =  .
2 .. .
..
.
..
.
..
.
.. 
.
0 0 0 ··· 1
More details of 1D-TFs may be found in [20–23].

2.2. Two-dimensional triangular functions

An m1 × m2 -set of 2D-TFs on the region ([0, 1) × [0, 1)), are defined as


  
s − ih1 t − jh2
 1−
 1− , ih1 ≤ s < (i + 1)h1
1,1 h1 h2
Ti,j (s, t ) =
 jh2 ≤ t < (j + 1)h2 ,
0, otherwise ,

  
s − ih1 t − jh2
 1−
 , ih1 ≤ s < (i + 1)h1
1,2 h1 h2
Ti,j (s, t ) =
 jh2 ≤ t < (j + 1)h2 ,
0, otherwise ,

  
s − ih1 t − jh2

 1− , ih1 ≤ s < (i + 1)h1
2,1 h1 h2
Ti,j (s, t ) =
 jh2 ≤ t < (j + 1)h2 ,
0, otherwise ,

  
s − ih1 t − jh2

 , ih1 ≤ s < (i + 1)h1
2,2 h1 h2
Ti,j (s, t ) =
 jh2 ≤ t < (j + 1)h2 ,
0, otherwise ,

where i = 0, 1, 2, . . . , m1 − 1, j = 0, 1, 2, . . . , m2 − 1 and h1 = 1
m1
, h2 = 1
m2
[19]. m1 and m2 are arbitrary positive integers.
1,1 1,2 0,0 2,2
In Fig. 1 the functions T0,0 (s, t ), T0,0 (s, t ), Ti,j (s, t ) and T0,0 (s, t ) are illustrated for m1 = m2 = 2.
It is clear that
1 ,1
Ti,j (s, t ) = Ti1 (s) · Tj1 (t ),
1,2
Ti,j (s, t ) = Ti1 (s) · Tj2 (t ),
2,1
(6)
Ti,j (s, t ) = Ti2 (s) · Tj1 (t ),
2,2
Ti,j (s, t ) = Ti2 (s) · Tj2 (t ).
1836 K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844

Fig. 1. The 2D-TFs in sub-region [0, .5) × [0, .5) for m1 = m2 = 2.

On the other hand, if


 T
1 ,1 1,1 1 ,1
T11(s, t ) = T0,0 (s, t ), T0,1 (s, t ), ..., Tm1 −1,m2 −1 (s, t ) ,
 T
1 ,2 1,2 1 ,2
T12(s, t ) = T0,0 (s, t ), T0,1 (s, t ), ..., Tm1 −1,m2 −1 (s, t ) ,
 T
2 ,1 2,1 2 ,1
T21(s, t ) = T0,0 (s, t ), T0,1 (s, t ), ..., Tm1 −1,m2 −1 (s, t ) ,
 T
2,2 2 ,2 2 ,2
T22(s, t ) = T0,0 (s, t ), T0,1 (s, t ), ..., Tm1 −1,m2 −1 (s, t ) ,

then T(s, t ), the 2D-TFs vector, can be defined as

T11(s, t )
 
T12(s, t )
T(s, t ) =  . (7)
T21(s, t )
T22(s, t ) 4m
1 m 2 ×1

Some properties of 2D-TFs are as follows [19]

,j (s, t )}, {Ti,j (s, t )}, {Ti,j (s, t )} and {Ti,j (s, t )} are obviously disjoint
i. Each set of {Ti11 12 21 22

 p ,q
Ti11,j1 1 (s, t ), p1 = p2 q1 = q2
p ,q p ,q
Ti11,j1 1 (s, t ) · Ti22,j2 2 (s, t ) ≃ i1 = i2 j1 = j2 ,
0, otherwise ,

for p, q ∈ {1, 2}, i1 , i2 = 0, 1, 2, . . . , m1 − 1, and j1 , j2 = 0, 1, 2, . . . , m2 − 1. So


T11(s, t ) · T11T (s, t ) ≃ diag(T11(s, t )),
T11(s, t ) · T12T (s, t ) ≃ 0m1 m2 ×m1 m2 ,
T11(s, t ) · T21T (s, t ) ≃ 0m1 m2 ×m1 m2 ,
T11(s, t ) · T22T (s, t ) ≃ 0m1 m2 ×m1 m2 .
These relations also satisfy for T12(s, t ), T21(s, t ) and T22(s, t ), similarly. Hence

T(s, t ) · TT (s, t ) ≃ diag(T(s, t )) = T̃(s, t ).

ii. 2D-TFs are orthogonal, that is


 1  1
p ,q p ,q
Ti11,j1 1 (s, t ) · Ti22,j2 2 (s, t )dsdt = ∆p1 ,p2 δi1 ,i2 · ∆q1 ,q2 δj1 ,j2 ,
0 0
K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844 1837

where δ denotes Kronecker delta function, and


h

 ,
 α = β ∈ {1, 2},
∆α,β = 3
h

, α ̸= β.
6
iii. For every 4m1 m2 -vector X ,
T(s, t ) · TT (s, t ) · X ≃ X̃ · T(s, t ),
where X̃ = diag(X ).
iv. The disjoint property of T11(s, t ), T12(s, t ), T21(s, t ) and T22(s, t ) also implies that for every (4m1 m2 × 4m1 m2 )-matrix A,

TT (s, t ) · A · T(s, t ) ≃ Â · T(s, t ),


in which  is an 4m1 m2 -vector with elements equal to the diagonal entries of matrix A.
v. A function f (s, t ) defined over ([0, 1) × [0, 1)) may be extended by 2D-TFs as
f (s, t ) ≃ C 1T · T11 + C 2T · T12 + C 3T · T21 + C 4T · T22
= C T · T(s, t ), (8)
where C is a 4m1 m2 -vector given by
T
C = C 1T C 2T C 3T C 4T .

The coefficients in vectors C 1, C 2, C 3, and C 4 can be computed by sampling the function f (s, t ) at grid points si and tj
such that si = ih1 and tj = jh2 , for various i and j. Therefore,

C 1k = f (si , tj ),
C 2k = f (si , tj+1 ),
(9)
C 3k = f (si+1 , tj ),
C 4k = f (si+1 , tj+1 ),
in which k = im2 + j and i = 0, 1, . . . , m1 − 1, j = 0, 1, . . . , m2 − 1.
vi. Also, the positive integer powers of a function f (s, t ) may be approximated by 2D-TFs as
[f (s, t )]p ≃ CpT · T(s, t ),
where Cp is a column vector whose elements are pth power of the elements of the vector C .
vii. Let k(s, t , x, y) be a function of four variables on ([0, 1) × [0, 1) × [0, 1) × [0, 1)). It can be approximated with respect
to 2D-TFs as follows:
k(s, t , x, y) ≃ TT (s, t ) · K · T(x, y), (10)
where T(s, t ) and T(x, y) are 2D-TFs vectors of dimension 4m1 m2 and 4m3 m4 , respectively. K is a (4m1 m2 × 4m3 m4 )
2D-TFs coefficients matrix and may be computed by sampling the function k(s, t , x, y), [19].

3. Supplementary properties of 2D-TFs

In this section, we derive and present some new properties of 2D-TFs that need in solving mixed type Volterra–Fredholm
integral equations.

3.1. Expansion of 1D-TFs with respect to 2D-TFs

Using (8), each element of an m1 -set of 1D-TFs can be expanded with respect to an m1 × m2 -set of 2D-TFs. For T01 (s) we
have
T01 (s) ≃ C 1T C 2T C 3T C 4T · T(s, t ).
 

Since t not appears in T01 (s), Eqs. (9) lead to

1, k = 1, 2, . . . , m2 ,

C 1k = C 2k = T01 (si ) =
0, k = m2 + 1, . . . , m1 m2 ,
C 3k = C 4k = T01 (si+1 ) = 0, k = 1, 2, . . . , m1 m2 .
Other elements of T(s) may be expanded in the same manner. So
T(s) ≃ M2m1 ×4m1 m2 · T(s, t ),
1838 K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844

where
 
I Im1 ×m1 0m1 ×m1 0m1 ×m1
M = m1 ×m1 ⊗ 11×m2 ,
0m1 ×m1 0m1 ×m1 Im1 ×m1 Im1 ×m1
and ⊗ denotes the Kronecker product defined for two arbitrary matrices P and Q as
P ⊗ Q = (pij Q ).
Hence the Kronecker product of vector 11×m2 from right side in any matrix or vector, lead to m2 -repetition of each component
in its corresponding row.

3.2. Operational matrix for integration


s1
In order to approximate 0 0 u(x, y)dydx, first we should compute this integral for vector T(x, y). From Eqs. (6), the
integrals with respect to x and y can be computed separately. During this section, we construct an operational matrix for
s1
approximating the integration with respect to x in 0 0 T(x, y)dydx.
Let Ts(x) be the x-components of T(x, y). We may write
T01 (x)
  
..

.  2m2 times 
 

 T01 (x)
  

 T11 (x)
 

.. T1(x)
    
. 2m2 times 
T1(x)

Ts(x) =  ⊗ 1m2 ×1 .
 
=
T2(x)

 T11 (x)


.. T2(x) 4m ×1
 
.
 
  1

( )
 2  
T x 
 m1 −1 
.
 
..
 
 2m2 times

Tm2 1 −1 (x)

4m 1 m2 ×1

Using Eqs. (4) and (5), the integral of Ts(x) can be approximated as
  
P1s P2s
 s
P1s P2s
Ts(x)dx ≃  ⊗ 1m2 ×1  · T(s)

0
P1s P2s
P1s P2s
  
P1s P2s
P1s P2s
≃  ⊗ 1m2 ×1  · M · T(s, t )

P1s P2s
P1s P2s
= W · T(s, t ),
in which W is a (4m1 m2 × 4m1 m2 )-matrix as follows:
  
P1s P2s   
P1s P2s I I 0 0
W =  ⊗ 1m2 ×1  · ⊗ 11×m2

P1s P2s 0 0 I I
2m1 ×4m1
P1s P2s
 
P1s P1s P2s P2s
P1s P1s P2s P2s
= ⊗ 1m2 ×m2 . (11)
P1s P1s P2s P2s
P1s P1s P2s P2s

3.3. Product properties

Now we attempt to compute T(s, t ) · TT (s, t ) approximately. From Eqs. (6) each component of this matrix can be written
as production of two terms with respect to variables s and t, separately. So

T(s, t ) · TT (s, t ) = Aij (s)Bij (t ) i,j=1,2,...,4m m ,


 
(12)
1 2

where A(s) and B(t ) are (4m1 m2 × 4m1 m2 )-matrices and must be computed.
K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844 1839

m −1 m −1
It is immediately concluded from disjoint property of {Ti1 (s)}i=10 and {Ti2 (s)}i=10 , Eqs. (2) and (6) that

T1(s)
  
T1(s)
A(s) =  ⊗ 1m2 ×1 

T2(s)
T2(s)
· T1T (s), T1T (s), T2T (s), T2T (s) ⊗ 11×m2
  

 (s) T1 ( s)
 
T1 0 0
 (s) T1
T1  ( s) 0 0 
≃  ⊗ 1 m ×m . (13)
 0 0 T2(s) T2(s)
  2 2

0 0 T2(s) T2(s)
 

m −1 m −1
Moreover, disjoint property of {Tj1 (t )}j=20 and {Tj2 (t )}j=20 implies that
 
1m1 ×1 ⊗ T1
1m1 ×1 ⊗ T2
B(t ) = 
1m1 ×1 ⊗ T1
1m1 ×1 ⊗ T2
· 11×m1 ⊗ T1, 11×m1 ⊗ T2, 11×m1 ⊗ T1, 11×m1 ⊗ T2 ,
 

in which the (t ) term in T1(t ) and T2(t ) is canceled, for convenience. Therefore

Bq(t ) Bq(t )
 
B(t ) ≃ , (14)
Bq(t ) Bq(t )

where Bq(t ) is a (2m1 m2 × 2m1 m2 )-matrix of the form

1m1 ×m1 ⊗ (T1 ˜)


˜ · T1 1m1 ×m1 ⊗ (T1 ˜)
˜ · T2
 
Bq(t ) =
1m1 ×m1 ⊗ (T1 ˜)
˜ · T2 1m1 ×m1 ⊗ (T2 ˜) .
˜ · T2

Note that for future necessarily, the matrices A(s) and B(t ) in Eqs. (13) and (14) are approximated in different forms. Since
all blocks in B(t ) are diagonal matrices, Eq. (12) can be written as

11(s, t ) 12(s, t )
 
Q Q 0 0
Q 12(s, t ) 13(s, t )
Q 0 0
T(s, t ) · T (s, t ) ≃  ,
T
  
(15)
 0 0 21(s, t )
Q Q 22(s, t )

0 0 22(s, t )
Q Q23(s, t )

in which by considering

A1(s) = trace(T1
 (s) ⊗ 1m ×m ),
2 2

A2(s) = trace(T2
 (s) ⊗ 1m ×m ),
2 2

B1(t ) = trace(1m1 ×m1 ⊗ (T1


 (t ) · T1
 (t ))),
B2(t ) = trace(1m1 ×m1 ⊗ (T1
 (t ) · T2
 (t ))),
B3(t ) = trace(1m1 ×m1 ⊗ (T2
 (t ) · T2
 (t ))),

we have

(Q 11)k = (A1)k · (B1)k ,


(Q 12)k = (A1)k · (B2)k ,
(Q 13)k = (A1)k · (B3)k ,
k = 1, 2, . . . , m1 m2 . (16)
(Q 21)k = (A2)k · (B1)k ,
(Q 22)k = (A2)k · (B2)k ,
(Q 23)k = (A2)k · (B3)k
Now, let X be a 4m1 m2 -vector as
T
X = X 1T , X 2T , X 3T , X 4T ,

1840 K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844

where X 1, X 2, X 3 and X 4 are m1 m2 -vectors. It can be concluded from Eqs. (15) to (16) that
  
Q
11 Q
12 0 0

X1
Q
12 Q
13 0 0  X 2
T(s, t ) · T (s, t ) · X ≃ 
T

  
0 0 Q
21 22 X 3
Q
 
0 0 Q
22 Q
23 X4
   
1 · B1
X  2 · B2
X  0 0 A1
X1 · B2
 2 · B3
X  0 0
 A1
=
 0

  A2
0 3 · B1
X  4 · B2
X
0 0 X 3 · B2
  4 · B3
X  A2
     
B1
 B2
 0 0 X1 0 0 0 A1
B2
 B3
 0 0 0
  2 0
X 0 A1
=
0
 
3 0  A2
0 B1
   0
B2 0 X
0 0 B2
 B3
 0 0 0 X
4 A2

= V (t ) · X̃ · Ts(s), (17)

hence the s-components and t-components may be decomposed in vector Ts(s) and matrix V (t ), respectively. Note that
s-components also are aggregated in last vector.

4. Solving mixed type Volterra–Fredholm integral equations

In this section, we propose a reliable method for solving mixed type Volterra–Fredholm integral equations of the form
 s 1
f (s, t ) = g (s, t ) + λ k(s, t , x, y)[f (x, y)]p dydx, (18)
0 0

numerically. As usual, approximations of f (s, t ), g (s, t ) and [f (s, t )]p with respect to 2D-TFs may be written as

f (s, t ) ≃ C T · T(s, t ) = TT (s, t ) · C ,


g (s, t ) ≃ GT · T(s, t ) = TT (s, t ) · G, (19)
[f (s, t )] ≃ p
CpT · T(s, t ) = T (s, t ) · Cp ,
T

where T (s, t ) is defined in Eq. (7) and the 4m1 m2 -vectors C ,G and Cp are 2D-TF coefficients of f (s, t ), g (s, t ) and [f (s, t )]p ,
respectively. Elements of Cp are p-th powers of the elements of C . Also, we can approximate kernel k(s, t , x, y) using Eq. (10)
with m3 = m1 and m4 = m2 as follows:

k(s, t , x, y) ≃ TT (s, t ) · K · T(x, y), (20)

where the (4m1 m2 · 4m1 m2 )-matrix K is 2D-TF coefficients of k(s, t , x, y) [19].


From Eqs. (17), (19) and (20) the integral part in Eq. (18) can be approximated as
 s 1  s 1
k(s, t , x, y)[f (x, y)]p dydx ≃ TT (s, t ) · K · T(x, y) · TT (x, y) · Cp dydx
0 0 0 0
 s 1
= T ( s, t ) · K ·
T
T(x, y) · TT (x, y) · Cp dydx
0 0
 s 1
≃ T (s, t ) · K ·
T
V ( y) · C̃p · Ts(x)dydx
0 0
 1  s
= TT (s, t ) · K · V ( y)dy · C̃p · Ts(x)dx (21)
0 0

where (4m1 m2 × 4m1 m2 )-matrix V ( y) and 4m1 m2 -vector Ts(x) are defined in Eq. (17).
The first integral in Eq. (21) can be approximated as
 ( y)  ( y)
 
B1 B2 0 0
1 1
 ( y)
 ( y)
 
B2 B3 0 0
V ( y)dy ≃

 dy
0 0
 0 0 B1
 B2

0 0 B2
 B3

K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844 1841

h h2 
2
I1 I1 0 0
3 6 
 h2 h2 
 I1 I1 0 0 
≃6 3
 
 0 h2 h2 
0 I1 I1 

3 6 

h2 h2

0 0 I1 I1
6 3
= H, (22)
where we put I1 = Im1 m2 ×m1 m2 , for convenience.
For approximating the second integral in Eq. (21), we have
 s
Ts(x)dx ≃ W · T(s, t ),
0

in which (4m1 m2 × 4m1 m2 )-matrix W defined in (11). So,


 s 1  
k(s, t , x, y)[f (x, y)]p dydx ≃ TT (s, t ) · K · H · C̃p · W · T(s, t ),
0 0

≃ (KH C̃p W ) · T(s, t )


where (KH C̃p W ) is a 4m1 m2 -vector with components equal to the diagonal components of the matrix KH C̃p W . Since C̃p is a

diagonal matrix, we get

(KH C̃p W ) = Π · Cp ,

(23)
in which Π is a (4m1 m2 × 4m1 m2 )-matrix with components
Πi,j = (KH )i,j · Wj,i , i, j = 1, 2, . . . , 4m1 m2 .
Substituting Eqs. (19) and (23) in Eq. (18), and replacing ≃ with =, it follows that
TT (s, t ) · C = TT (s, t ) · G + TT (s, t ) · Π · Cp ,
or
C − Π · Cp = G. (24)
Eq. (24) is a nonlinear system of 4m1 m2 algebraic equations. Components of unknown 4m1 m2 -vector C can be obtained by
solving this system using Newton’s or other iterative methods. Hence, an approximate solution f (s, t ) ≃ C T T(s, t ) can be
computed for Eq. (18) without using any projection method.

5. Numerical examples

The method presented in this paper is used to find numerical solutions of four examples. Our results are compared with
the exact solutions by calculating the following error function:
e(s, t ) = |f (s, t ) − f¯m1 ,m2 (s, t )|,

where f (s, t ) and f¯m1 ,m2 (s, t ) are the exact and approximate solutions of the integral equation, respectively. The values of
e(s, t ) over the set
Dgrids = {(0.0, 0.0), (0.1, 0.1), (0.2, 0.2), . . . , (0.9, 0.9)},
are computed for different values of m1 and m2 , and collected in Tables 1–4.
Moreover, the values of e(s, t ) over the region ([0, 1) × [0, 1)) for Examples 1 and 4 are displayed in Figs. 1 and 2, with
m1 = m2 = 16.
The computations associated with the examples were performed using Matlab 7.0 software on a personal computer.

Example 1. For the following nonlinear Volterra–Fredholm integral equation:


 s 1
f (s, t ) = g (s, t ) + 16 e(s+t +x+y) [f (x, y)]3 dydx, 0 ≤ s < 1,
0 0

where g (s, t ) = es+t +4 − e5s+t − e5s+t +4 , with the exact solution f (s, t ) = es+t , Table 1 and Fig. 2 show the numerical results.
1842 K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844

Table 1
The numerical results for Example 1.
(s, t ) e(s, t ) m1 = m2 = 4 e(s, t ) m1 = m2 = 8 e(s, t ) m1 = m2 = 16

(0.0, 0.0) 0 0 0
(0.1, 0.1) 1.4853 × 10−2 1.5274 × 10−2 1.6348 × 10−3
(0.2, 0.2) 5.9763 × 10−2 1.0857 × 10−2 4.0535 × 10−3
(0.3, 0.3) 7.8689 × 10−2 1.0937 × 10−2 4.3510 × 10−3
(0.4, 0.4) 6.5780 × 10−2 2.2908 × 10−2 3.1225 × 10−3
(0.5, 0.5) 9.7749 × 10−2 2.5617 × 10−2 5.2446 × 10−3
(0.6, 0.6) 8.3208 × 10−2 3.0525 × 10−2 5.1789 × 10−3
(0.7, 0.7) 1.3690 × 10−1 2.9937 × 10−2 8.0677 × 10−3
(0.8, 0.8) 1.6545 × 10−1 3.5379 × 10−2 9.6129 × 10−3
(0.9, 0.9) 1.6955 × 10−1 5.3123 × 10−2 9.4423 × 10−3

Fig. 2. The error function graph for Example 1 with m1 = m2 = 16.

Table 2
The numerical results for Example 2.
(s, t ) e(s, t ) m1 = m2 = 4 e(s, t ) m1 = m2 = 8 e(s, t ) m1 = m2 = 16

(0.0, 0.0) 0 0 0
(0.1, 0.1) 7.7331 × 10−4 7.0520 × 10−5 1.9057 × 10−4
(0.2, 0.2) 5.4913 × 10−4 5.9322 × 10−4 5.0202 × 10−4
(0.3, 0.3) 9.0843 × 10−4 1.7033 × 10−3 7.7732 × 10−4
(0.4, 0.4) 1.1296 × 10−3 2.2460 × 10−3 2.0232 × 10−3
(0.5, 0.5) 6.5833 × 10−4 1.9259 × 10−4 5.2930 × 10−5
(0.6, 0.6) 6.9185 × 10−3 3.7710 × 10−3 3.9886 × 10−3
(0.7, 0.7) 5.1107 × 10−3 8.4427 × 10−3 3.8816 × 10−3
(0.8, 0.8) 1.2697 × 10−2 1.2200 × 10−2 4.6088 × 10−3
(0.9, 0.9) 1.8627 × 10−2 1.1523 × 10−2 9.3261 × 10−3

Example 2 ([14]). Consider the following linear Volterra–Fredholm integral equation:


 s 1
2 2
f (s, t ) = sin(s) + t − es + − yex f (x, y)dydx, 0 ≤ s < 1, (25)
3 3 0 −1

where (s, t ) ∈ ([0, 1) × [−1, 1)), with the exact solution f (s, t ) = sin(s) + t. Replacing y by (2y − 1), Eq. (25) can be
changed to
 s 1
1 1
f (s, t ) = sin(s) + t − es + − (2y − 1)ex f (x, y)dydx, 0 ≤ s < 1,
6 6 0 0

in which (s, t ) ∈ ([0, 1)×[0, 1)), and the presented method may be applied to it. The numerical results are shown in Table 2.

Example 3 ([8,14]). Consider the following linear Volterra–Fredholm integral equation:


 s 1
f (s, t ) = g (s, t ) − t 2 e−y f (x, y)dydx, 0 ≤ s < 1, (26)
0 0
K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844 1843

Table 3
The numerical results for Example 3.
(s , t ) e(s, t ) m1 = m2 = 4 e(s, t ) m1 = m2 = 8 e(s, t ) m1 = m2 = 16

(0.0, 0.0) 0 0 0
(0.1, 0.1) 1.6190 × 10−2 2.7778 × 10−3 8.5538 × 10−4
(0.2, 0.2) 1.2381 × 10−2 4.3333 × 10−3 6.5403 × 10−4
(0.3, 0.3) 1.3741 × 10−2 4.6691 × 10−3 7.2516 × 10−4
(0.4, 0.4) 2.0272 × 10−2 3.7878 × 10−3 1.0695 × 10−3
(0.5, 0.5) 6.8027 × 10−3 1.6922 × 10−3 3.5914 × 10−4
(0.6, 0.6) 2.3771 × 10−2 4.6505 × 10−3 1.2525 × 10−3
(0.7, 0.7) 2.0739 × 10−2 6.4046 × 10−3 1.0928 × 10−3
(0.8, 0.8) 2.2988 × 10−2 6.9586 × 10−3 1.2101 × 10−3
(0.9, 0.9) 3.0518 × 10−2 6.3168 × 10−3 1.6056 × 10−3

Table 4
The numerical results for Example 4.
(s , t ) e(s, t ) m1 = m2 = 4 e(s, t ) m1 = m2 = 8 e(s, t ) m1 = m2 = 16

(0.0, 0.0) 0 0 0
(0.1, 0.1) 1.5119 × 10−2 2.5073 × 10−3 9.3910 × 10−4
(0.2, 0.2) 1.0238 × 10−2 3.8017 × 10−3 6.3588 × 10−4
(0.3, 0.3) 1.0782 × 10−2 3.9091 × 10−3 6.6109 × 10−4
(0.4, 0.4) 1.6751 × 10−2 2.8581 × 10−3 1.0241 × 10−3
(0.5, 0.5) 2.7193 × 10−3 6.7929 × 10−4 1.6976 × 10−4
(0.6, 0.6) 2.0847 × 10−2 3.7462 × 10−3 1.2434 × 10−3
(0.7, 0.7) 1.8974 × 10−2 5.8418 × 10−3 1.1309 × 10−3
(0.8, 0.8) 2.4877 × 10−2 7.0708 × 10−3 1.4257 × 10−3
(0.9, 0.9) 3.8556 × 10−2 7.5793 × 10−3 2.1739 × 10−3

Fig. 3. The error function graph for Example 4 with m1 = m2 = 16.

where g (s, t ) = s2 et − 23 s3 t 2 , with the exact solution f (s, t ) = s2 et . The numerical results using presented method are
shown in Table 3.
The absolute error of numerical results obtained by the method proposed in [8] are almost equal to the ones obtained by
the present method.
Also a numerical method based on bivariate Chebyshev polynomials and collocation method has been proposed for
Eq. (26) [14]. Unfortunately, the absolute errors have been computed only in zeros of Chebyshev polynomials, which have
the least values of error. So, it seems that the errors in other points should be more. Although the comparison of results
does not show the categorical superiority of the proposed method from the viewpoint of accuracy, it seems that the number
of calculations in the current method is considerably less than that of the other methods. This is due to the fact that the
generation of the algebraic equations system in the present method needs only sampling of functions, multiplication and
addition of matrices, and needs no integration. So, the present method can be run very quickly for large values of m1 and m2 .

Example 4. For the following nonlinear Volterra–Fredholm integral equation:


 s 1
f (s, t ) = g (s, t ) + t 2 e−4x [f (x, y)]2 dydx, 0 ≤ s < 1,
0 0

where g (s, t ) = s2 e2t − 51 s5 + t 2 , with the exact solution f (s, t ) = s2 e2t , Table 4 and Fig. 3 show the numerical results.
1844 K. Maleknejad, Z. JafariBehbahani / Mathematical and Computer Modelling 55 (2012) 1833–1844

6. Concluding remarks

The presented method uses 2D-TFs and their operational integration matrix to transform a nonlinear mixed
Volterra–Fredholm integral equation to a system of nonlinear algebraic equations. The benefits of this method are lower
cost of setting up the system of equations without applying any projection method such as Galerkin, Collocation, etc., and
any integration. Moreover, operational matrix W and H in Eqs. (11) and (22) can be computed at once for large values of m1
and m2 , and stored for using in various problems. Therefore, Eq. (24) is set up only by sampling g and k in grid points, and
also computing Π . Thus the computational cost of operations is low. These advantages make the method very simple and
cheap as computational point of view.
Its accuracy and applicability were checked on some examples. The numerical results show that the accuracy of the
obtained solutions is good. Furthermore, the current method can be run with increasing m1 and m2 until the results settle
down to an appropriate accuracy.

References

[1] B.G. Pathpatte, On mixed Volterra–Fredholm integral equations, Indian J. Pure. Appl. Math. 17 (1986) 488–496.
[2] O. Diekmann, Thersholds and traveling for the geographical spread of infection, J. Math. Biol. 6 (1978) 109–130.
[3] H.R. Thieme, A model for the spatial spread of an epidemic, J. Math. Biol. 4 (1977) 337–351.
[4] H. Brunner, Collocation Methods for Volterra Integral and Related Functional Equations, Cambridge University Press, Cambridge, 2004.
[5] J.P. Kauthen, Continuous time collocation method for Volterra–Fredholm integral equations, Numer. Math. 56 (1989) 409–424.
[6] H. Brunner, On the numerical solution of nonlinear Volterra–Fredholm integral equations by collocation methods, SIAM J. Numer. Anal. 27 (4) (1990)
987–1000.
[7] L. Hacia, On approximate solution for integral equations of mixed type, ZAMM Z. Angew. Math. Mech. 76 (1996) 415–416.
[8] L. Hacia, Projection methods for integral equations in epidemic, J. Math. Model. Anal. 7 (2) (2002) 229–240.
[9] H. Guoqiang, Asymptotic error expansion for the trapzoidal Nystrom method of linear Volterra–Fredholm integral equations, J. Comput. Appl. Math.
51 (3) (1994) 339–348.
[10] H. Guoqiang, Asymptotic error expansion for the Nystrom method for a nonlinear Volterra–Fredholm integral equation, J. Comput. Appl. Math. 59
(1995) 49–59.
[11] K. Maleknejad, M. Hadizadeh, A new computational method for Volterra–Fredholm integral equations, J. Comput. Appl. Math. 37 (1999) 1–8.
[12] A.M. Wazwaz, A reliable treatment for mixed Volterra–Fredholm integral equations, Appl. Math. Comput. 127 (2002) 405–414.
[13] K. Maleknejad, M.R. Fadaei Yami, A computational method for system of Volterra–Fredholm integral equations, Appl. Math. Comput. 183 (2006)
589–595.
[14] M. Hadizadeh, M. Asgari, An effective numerical approximation for the linear class of mixed integral equations, Appl. Math. Comput. 167 (2005)
1090–1100.
[15] E. Banifatemi, M. Razzaghi, S. Yousefi, Two-dimensional Legender wavelets method for the mixed Volterra–Fredholm integral equations, J. Vibr.
Control 13 (2007) 1667–1675.
[16] S.A. Yousefi, A. Lotfi, Mehdi Dehghan, He’s variational iteration method for solving nonlinear mixed Volterra–Fredholm integral equations, J. Compute.
Math. Appl. 58 (2009) 2172–2176.
[17] A. Yildirim, Homotopy perturbation method for the mixed Volterra–Fredholm integral equations, Chaos, Solitons Fractals 42 (2009) 2760–2764.
[18] K. Maleknejad, K. Mahdiani, Solving nonlinear Mixed Volterra–Fredholm integral equations with two dimensional block-pulse functions using direct
method, Commun. Nonlinear Sci. Numer. Simulat. 16 (2011) 3512–3519.
[19] E. Babolian, K. Maleknejad, M. Roodaki, H. Almasieh, Two-dimensional triangular functions and its applications to nonlinear 2D Volterra–Fredholm
integral equations, Comput. Math. Appl. 60 (2010) 1711–1722.
[20] E. Babolian, R. Mokhtari, M. Salmani, Using direct method for solving variational problems via triangular orthogonal functions, Appl. Math. Comput.
202 (2008) 452–464.
[21] A. Deb, A. Dasgupta, G. Sarkar, A new set of orthogonal functions and its application to the analysis of dynamic systems, J. Franklin Inst. 343 (1) (2006)
1–26.
[22] E. Babolian, Z. Masouri, S. Hatamzadeh-Varmazyar, Numerical solution of nonlinear Volterra–Fredholm integro–differential equations via direct
method using triangular functions, Comput. Math. Appl. 58 (2009) 239–247.
[23] K. Maleknejad, H. Almasieh, M. Roodaki, Triangular functions (TF) method for the solution of Volterra–Fredholm integral equations, Commun.
Nonlinear Sci. Numer. Simulat. 15 (11) (2009) 3293–3298.

Vous aimerez peut-être aussi