Journal of Computational and Applied Mathematics 315 (2017) 182–194

Contents lists available at ScienceDirect

Journal of Computational and Applied

Mathematics
journal homepage: www.elsevier.com/locate/cam

Solving system of Volterra–Fredholm integral equations with

Bernstein polynomials and hybrid Bernstein Block-Pulse
functions
Esmail Hesameddini ∗ , Mehdi Shahbazi
Department of Mathematics Sciences, Shiraz University of Technology, P. O. Box 71555-313, Shiraz, Iran

article info abstract

Article history: This work approximates the unknown functions based on the Bernstein polynomials and
Received 8 September 2016 hybrid Bernstein Block-Pulse functions, in conjunction with the collocation method for the
numerical solution of system of Fredholm–Volterra integral equations. In both methods,
MSC: the system of integral equations is approximated by the Gauss quadrature formula with
45G10 respect to the Legendre weight function. The proposed methods reduce the system of
65R20
integral equations to a system of algebraic equations that can be easily solved by any usual
68U20
65C20
numerical methods. Moreover, the convergence analysis of these algorithms will be shown
by preparing some theorems. Numerical experiments are presented to show the superiority
Keywords: and efficiency of proposed methods in comparison with some other well known methods.
Bernstein polynomials © 2016 Elsevier B.V. All rights reserved.
System of Volterra–Fredholm integral
equations
Convergence analysis
Numerical algorithm
Gaussian quadrature

1. Introduction

Integral equations, depending on the structure of integrals, have different types, for example Fredholm integral equations,
Volterra integral equations and Fredholm–Volterra integral equations. Many scientific and engineering problems such as
scattering in quantum mechanics, water waves and spatial–temporal development of an epidemic, can be modeled as
integral equations [1–3]. The last arises from parabolic boundary value problems, various physical and biological models
[4–6] and mixed problems of mechanics of continuous media [7]. Many numerical methods have been extended for solving
these equations such as collocation method based on the radial basis functions [8], Legendre collocation method [9] and
Chebyshev polynomial approach [10]. System of integral equations also appear in many real world problems. So finding
approximate solutions for these systems is a challenging problem for many researchers. In [11], the discrete collocation
method was used for solving system of nonlinear Fredholm–Volterra integro differential equation. In [12], they presented
a Taylor-series expansion method for a second kind system of Fredholm integral equations with smooth or weakly singular
kernels. Babolian et al. [13], solved system of Fredholm integral equations and system of Volterra integral equations via hat
basis functions. In [14], system of Fredholm–Volterra type integral equations and in [15] system of linear Fredholm integral
equations were solved by He’s homotopy perturbation method and sinc collocation method, respectively.
Bernstein polynomials recently have been used to solve some linear as well as nonlinear differential equations
approximately by Bhatti and Bracken [16], Bhatta and Bhatti [17] and references therein. These polynomials are defined

∗ Corresponding author. Fax: +98 7117354523.

E-mail addresses: hesameddini@sutech.ac.ir (E. Hesameddini), me.shahbazi@sutech.ac.ir (M. Shahbazi).

http://dx.doi.org/10.1016/j.cam.2016.11.004
0377-0427/© 2016 Elsevier B.V. All rights reserved.
 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194 183

on an interval which forms a complete basis. Each of these polynomials is positive and their sum is unit. There exist in
the literature a number of approximate methods for solving numerically various classes of integral equations. Bernstein
polynomials have been used for solving problems formulated by mixed Volterra–Fredholm integral equations [18]. There
are some other works that used Bernstein polynomials as a basis such as Bernstein operational matrices method for solving
Fredholm and Volterra integral equations of the second kind [19].
Also, hybrid functions have been applied to solve the linear and nonlinear integral equations which is available in the
literature. Linear Fredholm integral equations of the second kind have been solved by using hybrid Legendre–Block-Pulse
functions in [20]. In [21], the second Chebyshev polynomials and hybrid Block-Pulse functions have been used to solve
system of integro-differential equations. Variational problems have been applied by using Bernoulli polynomials and hybrid
of Block-Pulse functions [22]. In [23], authors applied hybrid of Fourier and Block-Pulse Functions for solving integro-
differential equations.
In this work, we apply the Bernstein polynomials method (BPM) and hybrid Bernstein Block-Pulse functions method
(HBBPFM) for the following system of Volterra–Fredholm integral equations
 t  b
λX(t ) = F(t ) + K1 (t , s, X(s))ds + K2 (t , s, X(s))ds, t ∈ [a, b], (1)
a a

where λ ∈ R \ {0}, N ≥ 1, X = [X1 , X2 , . . . , XN ]T : [a, b] −→ RN is the unknown vector-valued function satisfying (1),
(1) (1) (1) (2) (2) (2)
F = [f1 , f2 , . . . , fN ]T : [a, b] −→ RN , K1 = [k1 , k2 , . . . , kN ]T : [a, b]2 × RN −→ RN and K2 = [k1 , k2 , . . . , kN ]T :
[a, b] × R −→ R are continuous functions in such a way that K1 and K2 are assumed to satisfy a global Lipschitz condition
2 N N

at their third variable, i.e., there exist M1 , M2 > 0 such that for all t ; s ∈ [a, b], u, v ∈ RN and l = 1, 2 there holds
∥Kl (t , s, u) − Kl (t , s, v)∥∞ ≤ Ml ∥u − v∥∞ . (2)
According to the Banach fixed point theorem, system (1) has one and only one vector-valued continuous solution, provided
that [24],
(M1 + M2 )(b − a) < |λ|. (3)
Recently, numerical methods have been proposed for solving (1). In this paper, we will extend HBBPFM and BPM to
approximate the solution of this system. The properties of Bernstein polynomials are used to reduce the problem into a
system of algebraic equations. Besides, an estimation of error bound for these methods will be given. Finally, we apply these
methods to several examples in order to show the efficiency of presented methods. These results confirm that, HBBPFM is
very effective and more accurate than BPM and some other methods. The rest of this paper is organized as follows.
In Section 2, some preliminaries are presented. In Section 3, the Bernstein polynomials method and its convergence
analysis for approximating the solution of (1) will be discussed. In Section 4, we design the hybrid Bernstein Block-Pulse
functions method to solve (1) and also its convergence analysis will be considered. Section 5 offers some numerical examples
to illustrate the efficiency of these algorithms.

2. Preliminaries

In this section, some preliminaries and notations of Bernstein polynomials method which are necessary for later are
recalled.

Definition 1 ([25]). The Bernstein polynomials of degree n are defined on the interval [0, 1] as follows:
n
Bn,k (t ) = t k (1 − t )n−k , k = 0, 1, 2, . . . , n. (4)
k

For convenience, we set Bn,k (t ) = 0, for k < 0 or k > n. One can approximate the function f : [0, 1] → R with the Bernstein
polynomials as:
n  
 k
Bn (f ) = f Bn,k (t ), (5)
k=0
n

where Bn,k , s are polynomials of degree n, defined by (4). The Bernstein polynomials of nth-degree are defined on the interval
[a, b] as
 n  (t − a)k (b − t )n−k
Bn,k (t ) = , k = 0, 1, 2, . . . , n, (6)
k ( b − a) n
and the function f : [a, b] → R can be approximated by the Bernstein polynomials as follows:
n
(b − a)k
  
Bn (f ) = f a+ Bn,k (t ). (7)
k=0
n
184 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194

b
Suppose that H = L2 [a, b) is a Hilbert space with the inner product defined by ⟨f , h⟩ = a f (t )h(t )T dt and V =
Span{Bn,0 (t ), Bn,1 (t ), . . . , Bn,n (t )}, since V is a finite-dimensional subspace of H, it is closed and convex. Thus, for f ∈ H,
there exists a unique best approximation out of V such as g, such that [25]
∥f − g ∥ ≤ ∥f − h∥, ∀h ∈ V .
Since g ∈ V , there exist unique coefficients c0 , c1 , . . . , cn such that
n

f ≃g = c0 Bn,k = CT ψ,
k=0

where ψ = [Bn,0 , Bn,1 , . . . , Bn,n ] and CT = [c0 , c1 , . . . , cn ]. Then, CT can be obtained by

T

CT = D−1 ⟨f , ψ⟩, (8)

where
 b
⟨f , ψ⟩ = f (t )ψ(t )T dt ,
a

and D is an (n + 1) × (n + 1) matrix which is said the dual matrix of ψ(t ) as well as will be introduced in the following form
 b
D = ⟨ψ, ψ⟩ = ψ(t )ψ(t )T dt .
a

3. Bernstein polynomials method

3.1. Method of solution

In this section, the Bernstein polynomials method to approximate the solution of (1) will be discussed [26].
Any unknown vector-valued function X can be expanded in the Bernstein basis as
Bn (X) = [Bn (X1 ), Bn (X2 ), . . . , Bn (XN )]T , (9)
where
n

Bn (Xj ) = Xj (rk ) Bn,k (t ), t ∈ [a, b], j = 1, 2, . . . , N , (10)
k=0
(b−a)k
in which rk = a + n . Xj (rk ) , k = 0, 1, 2, . . . , n are the unknown Bernstein coefficients, n is chosen as any positive
integer and Bn,k (t ), k = 0, 1, 2, . . . , n are the Bernstein polynomials defined in the previous section.
In order to approximate the solution of (1), we consider this system of Volterra–Fredholm integral equations as the
following form:
N N
 t  b N
(1) (2)
 
λij Xj (t ) = fi (t ) + kij (t , s, Xj (s))ds + kij (t , s, Xj (s))ds, i = 1, 2, . . . , N , (11)
j =1 a j =1 a j =1

in which
λ if i = j,

λij =
0 if i ̸= j.
Substituting (10) in (11), one obtains
N N
 t  b N
(1) (2)
 
λij Bn (Xj (t )) = fi (t ) + kij (t , s, Bn (Xj (s)))ds + kij (t , s, Bn (Xj (s)))ds. (12)
j =1 a j =1 a j =1

Collocating (12) in n Legendre–Gauss collocation points tl , l = 0, 1, . . . , n − 1, on interval [a, b], we get:

N  tl N
(1)
 
λij Bn (Xj (tl )) = fi (tl ) + kij (tl , s, Bn (Xj (s)))ds
j =1 a j =1
 b N
(2)

+ kij (tl , s, Bn (Xj (s)))ds, i = 1, 2, . . . , N , l = 0, 1, . . . , n − 1. (13)
a j =1

In our approach, we use the quadrature rule to approximate the integrals involved in this system, which achieves zero error
integration for polynomial integrands of degree less than or equal to 2n + 1 with n Legendre–Gauss nodes. For this purpose,
 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194 185

a simple linear transformation must be made. So, the intervals [a, tl ], l = 0, 1, . . . , n − 1 are transferred to a fixed interval
[a, b] by means of the transformation s = tbl−
−a
a
τ + bb−
−a
tl
a. Hence, employing this transformation on the first term on the right
hand side of (13), results in
N  b N    
 tl − a  (1) tl − a b − tl tl − a b − tl
λij Bn (Xj (tl )) = fi (tl ) + kij tl , τ+ a, Bn Xj τ+ a dτ
j =1
b−a a j =1
b−a b−a b−a b−a
 b N
(2)

+ kij (tl , τ , Bn (Xj (τ )))dτ , i = 1, 2, . . . , N , l = 0, 1, . . . , n − 1. (14)
a j =1

Applying the Gaussian quadrature, system (14) is converted to

N N n −1 
 tl − a   (1) tl − a
λij Bn (Xj (tl )) = fi (tl ) + ωk kij tl , τk
j=1
b − a j=1 k=0 b−a
   n −1
N 
b − tl tl − a b − tl
ωk k(ij2) (tl , τk , Bn (Xj (τk ))),

+ a, Bn Xj τk + a + (15)
b−a b−a b−a j=1 k=0

where τk , k = 0, 1, . . . , n − 1 are the n Legendre–Gauss nodes and ωk , k = 0, 1, . . . , n − 1 are the weights on [a, b]. Using
this method, the system of Volterra–Fredholm integral equation (1) is reduced to a system of nonlinear algebraic equations.
Therefore, one can obtain the approximate solution of system (1) as follows:
 T
n n
 n

Bn (X(t )) =
 ŵ1k Bn,k (t ), ŵ2k Bn,k (t ), . . . , ŵNk Bn,k (t ) . (16)
k=0 k=0 k=0

3.2. Convergence of the method

In this section, we will obtain an estimation of error bound for our numerical method. To do this, let us consider the
Banach space C = C [a, b] × · · · × C [a, b] (N times) endowed with the following norm
∥X(t )∥ = max max |Xj (t )|.
1≤j≤N a≤t ≤b

Theorem 1. Consider the system of Volterra–Fredholm integral equations (1). Assume that Bn (X(t )) = k=0 w 1k Bn,k (t ),
n
T
w 2k Bn,k (t ), . . . , wNk Bn,k (t )
n n
k=0 k=0 is the best approximation out of V and Bn (X (t )) be the approximate solution of (1).

Then,

Bn (X (t )) ≤ max max w jk − ŵjk  ,

Bn (X (t )) − 
   
(17)
1≤j≤N 0≤k≤n

where w jk , j = 1, 2, . . . , N , k = 0, 1, . . . , n is defined by (8) and ŵjk , j = 1, 2, . . . , N , k = 0, 1, . . . , n be the solution

of (15) computed through our presented method.
Proof. We can write
 
 n 
Bn (X (t )) − 
Bn (X (t )) = max max  w jk − ŵjk Bn,k (t )
     

1≤j≤N a≤t ≤b  
k=0
n

≤ max max max wjk − ŵjk  Bn,k (t ) ,

(18)
1≤j≤N a≤t ≤b 0≤k≤n
k =0

where
n

Bn,k (t ) = 1, a ≤ t ≤ b.
k=0

This implies that

Bn (X (t )) ≤ max max w jk − ŵjk  ,
Bn (X (t )) − 
   
(19)
1≤j≤N 0≤k≤n

and the proof is completed.

In the following theorem, we will discuss about the convergence of Bernstein polynomials method for system of
Volterra–Fredholm integral equation (1).
186 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194

Theorem 2. Considering the assumptions of Theorem 1, let F ∈ C ([a, b], RN ) and K1 , K2 ∈ C ([a, b]2 , RN ) satisfy the Lipschitz
condition (2) and (3) with Lipschitz constants M1 and M2 such that (M1 + M2 )(b − a) < |λ|, If Bn (X (t )) be the approximate
solution of (1), then for sufficiently large n we have
 |λ| + (M1 + M2 )(b − a)
X (t ) − 
Bn (X (t )) ≤ max max w jk − ŵjk  .
  
(20)
|λ| − (M1 + M2 )(b − a) 1≤j≤N 0≤k≤n
Proof. The system of Volterra–Fredholm integral equation (1) can be written as
 t  b
λX (t ) = F(t ) + K1 (t , s, X (t ))ds + K2 (t , s, X (t ))ds. (21)
a a

Bn (X (t )) in this system, results in

Substituting 
 t  b
λ F(t ) +
Bn (X (t )) =  K1 (t , s, 
Bn (X (s)))ds + K 2 ( t , s, 
Bn (X (s)))ds. (22)
a a

Since Bn (X (t )) is the best approximation, we have

 t  b
λBn (X (t )) = F(t ) + K1 (t , s, Bn (X (s)))ds + K2 (t , s, Bn (X (s)))ds. (23)
a a

Subtracting (21) from (22), we get

 t
λ X (t ) − 
Bn (X (t )) = F(t ) − 
F(t ) + K1 (t , s, X (t )) − K1 (t , s, 
Bn (X (s))) ds
   
a
 b
K2 (t , s, X (t )) − K2 (t , s, 
Bn (X (s))) ds,
 
+ (24)
a
so
Bn (X (t )) ≤ F(t ) − 
|λ| X (t ) −  F(t ) + M1 (b − a) X (t ) − 
Bn (X (s)) + M2 (b − a) X (t ) − 
Bn (X (s)) ,
       
(25)
or
(|λ| − (M1 + M2 )(b − a)) X (t ) − 
Bn (X (t )) ≤ F(t ) − 
F(t ) ,
   
(26)
then, we can write
1
X ( t ) − 
Bn (X (t )) ≤ F(t ) − 
F(t ) .
   
(27)
(|λ| − (M1 + M2 )(b − a))
In order to find a bound on F(t ) −  F(t ), subtracting (22) from (23), one obtains
 

 t
F(t ) − F(t ) = λ Bn (X (s)) − 
Bn (X (t )) − K1 (t , s, Bn (X (s))) − K1 (t , s, 
Bn (X (s))) ds
   
a
 b
K2 (t , s, Bn (X (s))) − K2 (t , s, 
Bn (X (s))) ds.
 
− (28)
a

Using the Lipschitz condition (2), results in

F(t ) − 
F(t ) ≤ |λ| Bn (X(t )) − 
Bn (X (t ))
   

+ M1 (b − a) Bn (X(t )) −  Bn (X (s)) + M2 (b − a) Bn (X(t )) − 

Bn (X (s)) ,
   
(29)
or
F(t ) − 
F(t ) ≤ (|λ| + (M1 + M2 )(b − a)) Bn (X (t )) − 
Bn (X (t )) .
   
(30)
According to Theorem 1, one can write
F(t ) − 
F(t ) ≤ (|λ| + (M1 + M2 )(b − a)) max max w jk − ŵjk  .
   
(31)
1≤j≤N 0≤k≤n

Substituting this bound in (27), the purpose will be achieved.

4. Hybrid Bernstein Block-Pulse functions method

4.1. Hybrid Bernstein Block-Pulse functions

In this section, we are going to approximate the solution of (1) by hybrid Bernstein Block-Pulse functions [27–29].
 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194 187

Definition 2. An M-set of Block-Pulse functions is defined over the interval [0, 1) as

m−1 m

1, ≤t≤ ,
bm ( t ) = M M
0, otherwise,

where bm , m = 1, 2, . . . , M, is the mth Block-Pulse function with a positive integer value for M. A set of Block-Pulse functions
bm (x), for m = 1, 2, . . . , M on the interval [a, b) are defined as follows:
(M − m + 1)a + (m − 1)b (M − m)a + mb

1, ≤t≤ ,
bm ( t ) = M M
0, otherwise.

Definition 3. Hybrid Bernstein polynomials and Block-Pulse functions bm,k (t ) for m = 1, 2, . . . , M and k = 0, 1, . . . , n are
defined on the interval [a, b) as

 n (t − rm−1 )k (rm − t )n−k

 

bm,k (t ) =
, rm−1 ≤ t ≤ rm ,
 k (rm − rm−1 )n
0, otherwise,
(M −m)a+mb
where rm = M
, m = 0, 1, 2, . . . , M .

4.2. Method of solution

Any function f (t ) ∈ L2 [a, b) can be expressed by hybrid Bernstein Block-Pulse functions as

∞ 
 ∞
f (t ) ≃ cmk bm,k (t ). (32)
m=1 k=0

Truncating (32) at some values of n and M, one obtains

M 
 n
f (t ) ≃ cmk bm,k (t ) = CT H(t ), (33)
m=1 k=0

where

C = [c10 , . . . , c1n , c20 , . . . , c2n , . . . , cMn , . . . , cMn ]T , (34)

and

H(t ) = [b1,0 (t ), . . . , b1,n (t ), b2,0 (t ), . . . , b2,n (t ), . . . , bM ,n (t ), . . . , bM ,n (t )]T . (35)

The constant coefficients cmk are ⟨y(x), bm,k (x)⟩ for k = 0, 1, 2, . . . , n and m = 1, 2, . . . , M . Let us approximate Xj (t ) by
Eq. (33) as

Xj (t ) ≃ XTj H(t ), j = 1, 2, . . . , N , (36)

where H(t ) is m(n + 1) column vectors similar to (35) and Xj be as follows:

j j j j
Xj = [X10 , . . . , X1n , X20 , . . . , X2n , . . . , Xmn
j
, . . . , Xmn
j
]T , j = 1, 2, . . . , N . (37)

In order to apply our method, substituting (36) in (11), results in

N N
 t  b N
(1) (2)
 
λ T
ij Xj H (t ) = fi (t ) + kij (t , s, (s))ds +
XTj H kij (t , s, XTj H(s))ds, i = 1, 2, . . . , N . (38)
j =1 a j =1 a j =1

Collocating (38) in M (n + 1) Legendre–Gauss collocation points tl , l = 0, 1, . . . , M (n + 1) − 1, on interval [a, b], we arrive

at the following system:
N  tl N  b N
(1) (2)
  
λij XTj H(tl ) = fi (tl ) + kij (tl , s, XTj H(s))ds + kij (tl , s, XTj H(s))ds, (39)
j=1 a j =1 a j =1

for i = 1, 2, . . . , N and l = 0, 1, . . . , M (n + 1) − 1. Now, we use the quadrature rule to approximate the integral involved
in (39), which achieves zero error integration for polynomial integrands of degree less than or equal to 2M (n + 1) + 1 with
M (n + 1) Legendre–Gauss nodes. For this purpose, a simple linear transformation must be made. So, M (n + 1) intervals
188 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194

[0, tl ] are transferred to a fixed interval [a, b] by means of the transformations s = tl − a

b−a
τ+ b−tl
b −a
a. Hence, employing this
transformation on the first term on the right hand side of (13), one obtains
N  b N  b N
 tl − a  (1)
 (2)
λij XTj H(tl ) = fi (tl ) + kij (tl , τ , XTj H(τ ))dτ + kij (tl , s, XTj H(s))ds. (40)
j =1
b−a a j=1 a j =1

Applying the Gaussian quadrature, system (40) is converted to

N N M (n+1)−1 N M (n
+1)−1
tl − a  
ωk k(ij1) (tl , ξk , XTj H(ξk )) + ωk k(ij2) (tl , ξk , XTj H(ξk )),
 
λij XTj H(tl ) = fi (tl ) + (41)
j =1
b − a j =1 k=0 j=1 k=0

where ξk for k = 0, 1, . . . , M (n + 1) − 1 are the M (n + 1) Legendre–Gauss nodes and ωk for k = 0, 1, . . . , M (n + 1) − 1

are the weights on [a, b]. Using this method, the system of Volterra–Fredholm integral equations (1) is reduced to a
system of nonlinear algebraic equations. By solving this system, the unknown values X̂Tj will be obtained, where X̂Tj are
the approximate values of XTj for j = 1, 2, . . . , N. Therefore, the approximate solution of system (1) will be determined.

4.3. Convergence analysis

In this section, we will obtain an estimation error bound for our numerical method.

Theorem 3. Consider the system of Volterra–Fredholm integral equations (1). Suppose that the known functions F =
[f1 , f2 , . . . , fN ]T ∈ C ([a, b], RN ) and K1 , K2 ∈ C ([a, b]2 , RN ) satisfy the Lipschitz condition (2) and (3) with Lipschitz constants
M1 and M2 such that (M1 + M2 )(b − a) < |λ|. Let ∥e(x)∥ be the error term computed through our presented method, then
lim ∥e(x)∥ = 0.
n→∞

Proof. We can write the integral equation (11) for selected i, j ∈ 1, 2, . . . , N as follows:
 t  b
(1) (2)
λij Xj (t ) = fi (t ) + kij (t , s, Xj (s))ds + kij (t , s, Xj (s))ds. (42)
a a

Substituting X̂Tj H(t ) in (42), results in

 t  b
(1) (2)
λ T
ij X̂j H ( t ) = fi ( t ) + kij (t , s, X̂Tj H (s))ds + kij (t , s, X̂Tj H(s))ds, (43)
a a

then, we have
  t
(1) (1)
 
λij Xj (t ) − X̂Tj H (t ) = kij (t , s, Xj (s)) − kij (t , s, X̂Tj H(s)) ds
a
 b
(2) (2)
 
+ kij (t , s, Xj (s)) − kij (t , s, X̂Tj H(s)) ds.
a

Using the Lipschitz condition (2), one obtains

   t 
 (1) (1)
|λ| · Xj (t ) − X̂j H(t ) =
T
kij (t , s, Xj (s)) − kij (t , s, X̂Tj H(s)) ds
  
a
 b 
 (2) (2)
+ kij (t , s, Xj (s)) − kij (t , s, X̂Tj H(s)) ds

a
 t   b 
≤ M1 Xj (s) − X̂Tj H(s) ds + M2 Xj (s) − X̂Tj H(s) ds
   
a a
 
≤ (M1 + M2 )(b − a) Xj (s) − X̂Tj H(s) , (44)
 

considering β = λ−1  (M1 + M2 )(b − a), we have

 
 
(1 − β) Xj (s) − X̂Tj H(s) ≤ 0. (45)
 

Since 0 < β < 1, so

 
lim Xj (s) − X̂Tj H(s) = 0.
 
n→∞
 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194 189

Table 1
Comparison between the exact, HBBPFM and BPM results of X1 (t ) for Example 1.
t BPM HBBPFM Ref. [12] Exact solution
n = 10 n = 5, M = 32 n = 6, M = 15 n = 10

0.1 0.0096534 0.0098655 0.00999012 0.0097452 0.01

0.2 0.0395216 0.0397225 0.03999237 0.0389454 0.04
0.3 0.0894715 0.0895871 0.08999635 0.0884424 0.09
0.4 0.1592075 0.1599325 0.15999867 0.158487 0.16
0.5 0.2491759 0.2499397 0.24999306 0.248533 0.25
0.6 0.3595264 0.3597958 0.35999623 0.357366 0.36
0.7 0.4896894 0.4899075 0.48999354 0.483707 0.49
0.8 0.6399057 0.6399396 0.63999120 0.627518 0.64
0.9 0.8092635 0.8099618 0.80999386 0.793446 0.81

Table 2
Comparison between the exact, HBBPFM and BPM results of X2 (t ) for Example 1.
t BPM HBBPFM Ref. [13] Exact solution
n = 10 n = 6, M = 25 n = 7, M = 14 n = 10

0.1 −0.0880342 −0.0888724 −0.0889944 −0.086771 −0.089

0.2 −0.1511753 −0.1517053 −0.1519901 −0.129285 −0.152
0.3 −0.1823645 −0.1825123 −0.1829906 −0.164652 −0.183
0.4 −0.1755401 −0.1759033 −0.1759965 −0.176745 −0.176
0.5 −0.1246323 −0.1249264 −0.2499930 −0.148173 −0.125
0.6 −0.0230764 −0.0238335 −0.0239990 −0.0585809 −0.024
0.7 0.1322590 0.1326450 0.1329963 0.123668 0.133
0.8 0.3516301 0.3516453 0.3519992 0.4630928 0.352
0.9 0.6381065 0.6387998 0.6899521 0.70009 0.639

Let ej (x) = Xj (s) − X̂Tj H(s) and e(x) = ej (x), then,


j

lim ∥e(x)∥ = 0,
n→∞

therefore, the proof is completed.

5. Numerical examples

In this section, we apply our methods on some examples and compare the quality of the computed solutions with some
other well known methods.

Example 1. Consider the following nonlinear system of Fredholm integral equations

 1  1
3X1 (t ) = f1 (t ) + (t − s)3 X1 (s)ds + (t − s)2 X2 (s)ds,
0 0
 1  1
3X2 (t ) = f2 (t ) + (t − s)4 X1 (s)ds + (t − s)3 X2 (s)ds, (46)
0 0

where f1 (x) = 3
20
− 11
10
t + 5t 2 − t 3 and f2 (x) = − 10
1
− 41
20
t + 9 2
20
t + 23 3
4
t − t 4.
The exact solutions of this system are X1 (t ) = t 2 and X2 (t ) = t 3 + t 2 − t [12]. Tables 1 and 2 show the numerical results for
X1 (t ) and X2 (t ) by HBBPFM and BPM, respectively. These tables exhibit the approximate solutions by HBBPFM for different
values of n = 6, M = 25 and n = 7, M = 14 also by BPM for value of n = 10. These results are compared with [12], which
they used m = 10 terms by the Taylor series expansion method. The outcomes reveal that the results by HBBPFM, with
using only a small number of bases, are very promising and superior to BPM and [12].

Example 2. Consider the following linear system of Volterra integral equations

t
(t − s)3 t
( t − s) 2
 
X1 (t ) = f1 (t ) + X1 (s)ds + X2 (s)ds,
0 2 0 2
t
(t − s)4 t
(t − s)3
 
X2 (t ) = f2 (t ) + X1 (s) + X2 (s)ds,
0 2 0 2
 
t 4 +t 3 t3
where f1 (t ) = t 2 + 1 − 6
, f2 (t ) = 1 − t 3 + t − t 4 840
+ t
8
+ 1
8
, with the exact solutions X1 (t ) = t 2 + 1 and
X2 (t ) = t − t 3 + 1.
190 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194

Table 3
Comparison of the absolute errors for X1 (t ) in Example 2.
t BPM HBBPFM Ref. [13]
n = 32 n = 10, M = 35 n = 12, M = 15 n = 32

0 9.57 × 10−7 1.07 × 10−7 1.35 × 10−11 0

0.1 7.21 × 10−7 2.82 × 10−7 2.16 × 10−11 1.80 × 10−4
0.2 6.23 × 10−7 3.16 × 10−7 4.28 × 10−11 1.80 × 10−4
0.3 8.16 × 10−7 7.99 × 10−7 3.96 × 10−10 3.10 × 10−4
0.4 9.93 × 10−7 7.02 × 10−8 5.82 × 10−10 2.70 × 10−4
0.5 4.72 × 10−7 6.42 × 10−8 7.74 × 10−10 1.60 × 10−4
0.6 3.77 × 10−7 1.15 × 10−8 5.29 × 10−10 3.70 × 10−4
0.7 7.82 × 10−8 1.33 × 10−8 4.92 × 10−10 5.10 × 10−4
0.8 3.23 × 10−8 2.41 × 10−8 9.19 × 10−10 5.80 × 10−4
0.9 2.78 × 10−7 9.42 × 10−8 7.31 × 10−10 5.90 × 10−4
1 8.49 × 10−7 4.22 × 10−7 6.61 × 10−9 5.20 × 10−4

Table 4
Comparison of the absolute errors for X2 (t ) in Example 2.
t BPM HBBPFM Ref. [13]
n = 32 n = 10, M = 35 n = 12, M = 15 n = 32

0 7.89 × 10−6 4.11 × 10−6 4.18 × 10−10 0

0.1 4.67 × 10−6 1.82 × 10−6 2.73 × 10−10 1.80 × 10−4
0.2 2.77 × 10−6 8.14 × 10−7 7.65 × 10−10 1.80 × 10−4
0.3 1.54 × 10−6 8.78 × 10−7 6.48 × 10−10 3.10 × 10−4
0.4 8.12 × 10−6 2.77 × 10−6 2.97 × 10−10 2.70 × 10−4
0.5 3.62 × 10−6 1.39 × 10−6 8.06 × 10−10 1.60 × 10−4
0.6 4.46 × 10−6 9.23 × 10−7 1.44 × 10−10 3.70 × 10−4
0.7 9.11 × 10−6 9.24 × 10−7 1.27 × 10−10 5.10 × 10−4
0.8 9.88 × 10−7 5.41 × 10−7 1.19 × 10−10 5.80 × 10−4
0.9 4.34 × 10−6 3.42 × 10−7 9.54 × 10−11 5.90 × 10−4
1 5.56 × 10−6 2.21 × 10−6 7.19 × 10−12 5.20 × 10−4

We selected our example from [13], in which they solved this system of integral equations by the hat basis functions method.
Tables 3 and 4, exhibit the absolute errors of X1 (t ) and X2 (t ) by HBBPFM for different values of n = 10, M = 35 and
n = 12, M = 15 also by BPM for value of n = 32. These results have been compared with the method in [13]. These tables
reveal that the evaluated absolute errors by HBBPFM for X1 (t ) and X2 (t ), will be decreased rapidly in comparison with BPM
and [13].

Example 3. Consider the following nonlinear system of Volterra–Fredholm integral equations

 t  1
X1 (t ) = f1 (t ) + 16ts (X1 (s)) ds + 2
t 2 s3 X2 (s)ds,
−1 −1
 t  1
X2 (t ) = f2 (t ) + (3t − 2s) (X2 (s)) ds +
2 2
(2t − 4)X1 (s)ds, (47)
−1 −1

where f1 (t ) = t 5 − 52 t 2 − t
2
and f2 (t ) = −t 5 − 25 t 4 − 53 t 5 + t + 16 .

The exact solutions of this system are X1 (t ) = 2t and X2 (t ) = t + 1 [30]. The approximate solutions of X1 (t ) and X2 (t ) by
HBBPFM for different values of n and M have been compared with the solutions by BPM for different values of n and the
Chebyshev polynomials method [30] in Tables 5 and 6. One can observe that HBBPFM is very effective and more accurate
than BPM and the Chebyshev polynomials method.

Example 4. Consider the following nonlinear system of Volterra–Fredholm integral equations

 t  1
X1 (t ) = f1 (t ) + (s − t )X1 (s)ds +
2
(ts2 X1 (s) + t (s + 1)X22 (s))ds,
−1 −1
 t  1
X2 (t ) = f2 (t ) + 2X2 (s)ds + 3tX12 (s)ds, (48)
−1 −1
4 5t 3
where f1 (t ) = −4t + 6
− t2 − t
10
− 5
12
and f2 (t ) = − 23 t 3 + 2t 2 − 9t − 53 .

The exact solution of this system is X1 (t ) = t + 1 and X2 = t 2 − t [30]. Tables 7 and 8 show the absolute errors of X1 (t )
and X2 (t ) by HBBPFM for different values of n = 5, M = 30, n = 6, M = 20 and n = 7, M = 15 also by BPM for values of
 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194 191

Table 5
Comparison of the absolute errors of X1 (t ) for different values of n in Example 3.
t BPM HBBFPM Ref. [30]
n = 10 n = 12 n = 6, M = 30 n = 7, M = 15 N = 10 N = 12

−1 3.67 × 10−4 4.88 × 10−6 3.14 × 10−6 2.01 × 10−7 0.01622 0.0009
−0.75 3.92 × 10−4 4.98 × 10−6 2.09 × 10−6 8.93 × 10−8 0.00901 0.00032
−0.5 1.22 × 10−4 5.66 × 10−6 1.54 × 10−6 5.34 × 10−8 0.00437 0.00001
−0.25 9.09 × 10−5 1.22 × 10−6 6.28 × 10−7 1.49 × 10−8 0.0014 0.00005
0 7.33 × 10−5 7.53 × 10−7 1.32 × 10−7 0 0 0
0.25 1.15 × 10−5 4.33 × 10−7 2.69 × 10−7 8.36 × 10−9 0.00016 0.00005
0.5 4.59 × 10−5 5.91 × 10−7 3.94 × 10−7 1.12 × 10−8 0.00188 0.00001
0.75 8.71 × 10−5 2.35 × 10−6 7.07 × 10−7 5.77 × 10−8 0.005278 0.00032
1 9.96 × 10−5 3.28 × 10−6 1.22 × 10−6 7.44 × 10−8 0.01122 0.0009

Table 6
Comparison of the absolute errors of X2 (t ) for different values of n in Example 3.
t BPM HBBFPM Ref. [30]
n = 10 n = 12 n = 6, M = 30 n = 7, M = 15 N = 10 N = 12

−1 8.23 × 10−4 9.24 × 10−7 3.12 × 10−7 9.47 × 10−8 0.03068 0.00045
−0.75 6.52 × 10−4 4.34 × 10−7 2.07 × 10−7 7.25 × 10−8 0.02497 0.00044
−0.5 5.71 × 10−4 6.70 × 10−7 1.33 × 10−7 5.19 × 10−8 0.01926 0.00043
−0.25 5.53 × 10−4 9.01 × 10−7 8.23 × 10−8 1.55 × 10−8 0.01355 0.00041
0 5.43 × 10−4 1.10 × 10−6 4.46 × 10−8 1.02 × 10−8 0.00784 0.00040
0.25 3.13 × 10−4 9.11 × 10−7 5.62 × 10−8 2.87 × 10−8 0.00212 0.00039
0.5 2.55 × 10−5 8.23 × 10−7 2.75 × 10−7 5.42 × 10−8 0.00359 0.00037
0.75 1.84 × 10−4 9.55 × 10−7 4.39 × 10−7 6.99 × 10−8 0.0093 0.00036
1 5.81 × 10−4 7.12 × 10−7 4.88 × 10−7 8.39 × 10−8 0.01501 0.00035

Table 7
Comparison of the absolute errors of X1 (t ) for different values of n in Example 4.
t BPM HBBFPM Ref. [30]
n = 10 n = 12 n = 5, M = 33 n = 6, M = 20 N = 10 N = 12

−1 9.32 × 10−5 7.44 × 10−6 4.35 × 10−6 5.61 × 10−7 0.01013 0.00010
−0.75 8.12 × 10−5 6.98 × 10−6 5.21 × 10−6 4.67 × 10−7 0.00755 0.00007
−0.5 5.33 × 10−5 4.87 × 10−6 3.97 × 10−6 6.35 × 10−8 0.00501 0.00005
−0.25 3.11 × 10−5 1.84 × 10−6 1.77 × 10−7 4.54 × 10−8 0.00249 0.00002
0 2.60 × 10−5 7.88 × 10−7 2.61 × 10−8 0 0 0
0.25 5.22 × 10−5 2.99 × 10−6 2.28 × 10−7 3.36 × 10−8 0.00246 0.00002
0.5 6.72 × 10−5 7.46 × 10−6 8.56 × 10−7 7.88 × 10−8 0.00489 0.00004
0.75 8.35 × 10−5 4.71 × 10−6 1.22 × 10−6 3.28 × 10−7 0.00730 0.00006
1 1.22 × 10−4 9.59 × 10−6 8.41 × 10−6 9.16 × 10−7 0.00967 0.00008

Table 8
Comparison of the absolute errors of X2 (t ) for different values of n in Example 4.
t BPM HBBFPM Ref. [30]
n = 10 n = 12 n = 5, M = 33 n = 6, M = 20 N = 10 N = 12

−1 2.33 × 10 −3
1.92 × 10 −4
6.12 × 10 −6
9.47 × 10−8
0.01030 0.00057
−0.75 1.39 × 10−4 8.76 × 10−6 5.07 × 10−6 7.25 × 10−8 0.00774 0.00041
−0.5 9.15 × 10−5 6.02 × 10−6 1.33 × 10−6 5.19 × 10−8 0.00518 0.00025
−0.25 7.12 × 10−5 3.99 × 10−6 8.23 × 10−7 1.55 × 10−8 0.00261 0.00011
0 1.06 × 10−5 2.10 × 10−6 5.46 × 10−7 1.02 × 10−8 0.00002 0.00001
0.25 2.34 × 10−5 1.82 × 10−6 7.62 × 10−7 2.87 × 10−8 0.00257 0.00012
0.5 6.90 × 10−5 3.52 × 10−6 2.75 × 10−6 5.42 × 10−8 0.00517 0.00022
0.75 1.03 × 10−4 8.19 × 10−6 4.39 × 10−6 6.99 × 10−8 0.00778 0.00030
1 2.05 × 10−4 1.78 × 10−5 4.88 × 10−6 8.39 × 10−8 0.01040 0.00039

n = 10 and n = 12. In these tables a comparison between these results with the Chebyshev polynomials method in [30] is
also given. It is clear that the results obtained by HBBPFM, with using only a small number of bases, are very promising and
superior to those of BPM and [30]. Tables 9 and 10 exhibit the maximum absolute errors of this example for some different
values of n and M. It is seen that, for a certain value of n, as M increases, the accuracy increases and for a certain value of M, as
n increases, the accuracy increases as well. Therefore, HBBPFM for solving this problem is very effective and more accurate
with respect to BPM and the Chebyshev polynomials method.
192 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194

Table 9
Maximum absolute errors for different values of n and M for X1 (t ) in Example 4.
n M
5 10 15 20 33

3 3.70 × 10−1 1.02 × 10−2 6.44 × 10−2 3.29 × 10−2 1.80 × 10−2
4 7.80 × 10−2 4.14 × 10−2 8.77 × 10−3 5.65 × 10−3 1.80 × 10−3
5 7.92 × 10−3 5.34 × 10−3 1.26 × 10−3 6.48 × 10−4 3.10 × 10−5
6 4.02 × 10−3 2.77 × 10−4 3.29 × 10−4 8.39 × 10−6 9.63 × 10−7
7 4.39 × 10−5 5.54 × 10−6 9.23 × 10−7 7.20 × 10−7 4.70 × 10−8
8 2.81 × 10−6 5.17 × 10−7 7.04 × 10−8 5.92 × 10−8 8.26 × 10−9
9 3.19 × 10−8 6.42 × 10−9 7.02 × 10−10 8.21 × 10−11 4.94 × 10−11
10 3.09 × 10−9 5.88 × 10−10 3.11 × 10−12 7.56 × 10−13 2.99 × 10−13

Table 10
Maximum absolute errors for different values of n and M for X2 (t ) in Example 4.
n M
5 10 15 20 33

3 7.66 × 10 −1
4.54 × 10−2
3.09 × 10 −2
1.45 × 10−2
9.76 × 10−3
4 1.04 × 10−1 8.96 × 10−2 2.56 × 10−2 9.11 × 10−3 2.19 × 10−3
5 6.55 × 10−3 5.79 × 10−4 2.60 × 10−4 1.01 × 10−4 8.47 × 10−5
6 5.89 × 10−4 3.41 × 10−4 7.02 × 10−6 4.00 × 10−6 6.48 × 10−7
7 1.15 × 10−4 6.88 × 10−5 2.11 × 10−5 4.79 × 10−7 2.05 × 10−7
8 6.87 × 10−5 6.63 × 10−6 3.98 × 10−6 8.94 × 10−7 1.38 × 10−8
9 6.26 × 10−7 8.19 × 10−8 2.87 × 10−9 1.38 × 10−10 9.02 × 10−11
10 6.60 × 10−8 1.06 × 10−9 1.04 × 10−10 7.88 × 10−12 3.16 × 10−12

Table 11
Comparison of the maximum absolute errors for X1 (t ) in Example 5.
n BPM Ref. [24] (n , M ) HBBFPM

11 7.46 × 10 −7
1.46 × 10 −6
(5, 25) 2.77 × 10−7
31 8.51 × 10−10 2.09 × 10−8 (7, 20) 6.62 × 10−10
101 9.26 × 10−15 1.77 × 10−10 (10, 15) 7.45 × 10−15

Table 12
Comparison of the maximum absolute errors for X2 (t ) in Example 5.
n BPM Ref. [24] (n , M ) HBBFPM

11 9.16 × 10−7 1.09 × 10−6 (5, 25) 7.39 × 10−7

31 7.66 × 10−10 9.20 × 10−7 (7, 20) 2.08 × 10−10
101 1.22 × 10−13 7.48 × 10−9 (10, 15) 8.70 × 10−14

Example 5. As a final test problem, consider the following linear system of Volterra–Fredholm integral equations
 1  t
2X1 (t ) = f1 (t ) − (cos(t − s)X1 (s) + sin(t − s)X2 (s)) ds + (cos(t − s)X1 (s) + sin(t − s)X2 (s)) ds,
0 0
 1  t
2X2 (t ) = f2 (t ) − (cos(t − s)X1 (s) + sin(t − s)X2 (s)) ds + (cos(t − s)X1 (s) + sin(t − s)X2 (s)) ds, (49)
0 0

where

f1 (t ) = t 4 − 12t 2 + 2t + 24 + 15 cos(t − 1) − 19 sin(t − 1) − 4 sin(t ) − 48 cos(t ),

f2 (t ) = 3t 4 − 12t 2 + 2t + 24 + 15 cos(t − 1) − 19 sin(t − 1) − 4 sin(t ) − 48 cos(t ).

The exact solutions of this system are X1 (t ) = t 2 and X2 (t ) = t 4 [24]. Tables 11–14 show the numerical results for this
example. Tables 11 and 12 compare the maximum absolute errors of HBBFPM and BPM with the results in [24] for different
values of n. These results have been included to demonstrate the validity and capability of HBBPFM with respect to BPM and
the collocation method based on q.i. splines of order of precision p = 4 in [24]. Tables 13 and 14 reveal that, for a certain value
of n, as M increases, the accuracy increases, and also for a certain value of M, as n increases, the accuracy increases as well.
 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194 193

Table 13
Maximum absolute errors for X1 (t ) with different values of n and M for Example 5.
n M
10 15 20 25

5 2.79 × 10−5 7.03 × 10−6 2.44 × 10−6 2.77 × 10−7

7 5.94 × 10−7 3.83 × 10−9 6.62 × 10−10 1.10 × 10−10
10 8.28 × 10−10 7.45 × 10−14 3.09 × 10−15 1.22 × 10−17
16 9.40 × 10−14 6.21 × 10−16 9.33 × 10−17 1.20 × 10−17

Table 14
Maximum absolute errors for X2 (t ) with different values of n and M for Example 5.
n M
10 15 20 25

5 7.08 × 10−4 8.50 × 10−5 6.08 × 10−6 2.39 × 10−7

7 4.02 × 10−6 1.01 × 10−8 2.08 × 10−10 6.99 × 10−12
10 3.05 × 10−11 8.70 × 10−14 1.73 × 10−15 9.24 × 10−16
16 4.81 × 10−15 5.17 × 10−15 7.04 × 10−16 5.92 × 10−17

6. Conclusion

In this paper, the Bernstein polynomials and the hybrid Bernstein Block-Pulse functions methods were applied to obtain
the numerical solution for system of Volterra–Fredholm integral equations. Using these methods, the system of integral
equations has been reduced to a system of algebraic equations. In both methods, the system of integral equations was
approximated by the Gauss quadrature formula with respect to the Legendre weight function and then the convergence of
the proposed methods was discussed. It was evident that, the hybrid Bernstein Block-Pulse functions method for a certain
value of n, as M increased, the accuracy was increased, and also for a certain value of M, as n increased, the accuracy was
increased as well. Several numerical experiments were presented to show the superiority and efficiency of HBBPFM in
comparison with BPM and some other well known methods.

References

[1] M.A. Jaswon, G.T. Symm, Integral Equation Methods in Potential Theory and Elastostatics, Academic Press, London, 1977.
[2] L.M. Delves, J.L. Mohamed, Computational Methods for Integral Equations, Cambridge University Press, Cambridge, 1985.
[3] P. Schiavane, C. Constanda, A. Mioduchowski, Integral Methods in Science and Engineering, Birkhäuser, Boston, 2002.
[4] S. Yalcinbas, Taylor polynomial solution of nonlinear Volterra–Fredholm integral equations, Appl. Math. Comput. 127 (2002) 195–206.
[5] M.A. Abdou, Fredholm–Volterra integral equation of the first kind and contact problem, Appl. Math. Comput. 125 (2002) 177–193.
[6] S. Yalcinbas, Taylor polynomial solution of nonlinear Volterra–Fredholm integral equations, Appl. Math. Comput. 127 (2002) 195–206.
[7] A.A. Badr, Numerical solution of Fredholm–Volterra integral equation in one dimension with time dependent, Appl. Math. Comput. 167 (2005)
1156–1161.
[8] K. Parand, J.A. Rad, Numerical solution of nonlinear Volterra–Fredholm–Hammerstein integral equations via collocation method based on radial basis
functions, Appl. Math. Comput. 218 (2012) 5292–5309.
[9] S. Nemati, Numerical solution of Volterra–Fredholm integral equations using Legendre col location method, J. Comput. Appl. Math. 278 (2015) 29–36.
[10] A. Akyuz-Dascioglu, A Chebyshev polynomial approach for linear Fredholm–Volterra integro-differential equations in the most general form, Appl.
Math. Comput. 181 (2006) 103–112.
[11] M. Rabbani, S.H. Kiasoltani, Solving of nonlinear system of Fredholm–Volterra integro-differential equations by using discrete collocation method,
J. Math. Comput. Sci. 3 (2011) 382–389.
[12] K. Maleknejad, N. Aghazadeh, M. Rabbani, Numerical solution of second kind Fredholm integral equations system by using a Taylor-series expansion
method, Appl. Math. Comput. 175 (2006) 1229–1234.
[13] E. Babolian, M. Mordad, A numerical method for solving systems of linear and nonlinear integral equations of the second kind by hat basis functions,
Comput. Math. Appl. 62 (2011) 187–198.
[14] E. Yusufoglu, A homotopy perturbation algorithm to solve a system of Fredholm–Volterra type integral equations, Math. Comput. Modelling 47 (2008)
1099–1107.
[15] J. Rashidinia, M. Zarebnia, Convergence of approximate solution of system of Fredholm integral equations, J. Math. Anal. Appl. 333 (2007) 1216–1227.
[16] M.I. Bhatti, P. Bracken, Solutions of differential equations in a Bernstein polynomial basis, J. Comput. Appl. Math. 205 (2007) 272–280.
[17] D.D. Bhatta, M.I. Bhatti, Numerical solution of KdV equation using modified Bernstein polynomials, Appl. Math. Comput. 174 (2006) 1255–1268.
[18] F. Hosseini Shekarabi, K. Maleknejad, R. Ezzati, Application of two-dimensional Bernstein polynomials for solving mixed Volterra–Fredholm integral
equations, Afr. Mat. (2014) 1–15.
[19] B.N. Mandal, S. Bhattacharya, Numerical solution of some classes of integral equations using Bernstein polynomials, Appl. Math. Comput. 190 (2007)
1707–1716.
[20] K. Maleknejad, M.T. Kajani, Solving second kind integral equations by Galerkin methods with hybrid Legendre and Block–Pulse functions, Appl. Math.
Comput. 145 (2003) 623–629.
[21] X.T. Wang, Y.M. Li, Numerical solutions of integro-differential systems by hybrid of general Block–Pulse functions and the second Chebyshev
polynomials, Appl. Math. Comput. 209 (2009) 266–272.
[22] M. Razzaghi, Y. Ordokhani, N. Haddadi, Direct method for variational problems by using hybrid of Block–Pulse and Bernoulli polynomials, Rom. J.
Math. Comput. Sci. 2 (2012) 1–17.
[23] B. Asady, M.T. Kajani, A.H. Vencheh, A. Heydari, Direct method for solving integro-differential equations using hybrid Fourier and Block–Pulse
functions, Int. J. Comput. Math. 82 (7) (2005) 889–895.
[24] F. Calio, A.I. Garralda-Guillem, E. Marchetti, M. Ruiz Galan, Numerical approach for systems of Volterra–Fredholm integral equations, Appl. Math.
Comput. 225 (2013) 811–821.
[25] T.J. Rivlin, An Introduction to the Approximation of Functions, Dover Publications, New York, 1969.
194 E. Hesameddini, M. Shahbazi / Journal of Computational and Applied Mathematics 315 (2017) 182–194

[26] E. Hesameddini, M. Khorramizadeh, M. Shahbazi, Bernstein polynomials method for solving Volterra–Fredholm integral equations, Bull. Math. Soc.
Sci. Math. Roumanie (N.S.) (2016) in press.
[27] K. Maleknejad, E. Hashemizadeh, R. Ezzati, A new approach to the numerical solution of Volterra integral equations by using Bernsteins approximation,
Commun. Nonlinear Sci. Numer. Simul. 16 (2011) 647–655.
[28] Y. Ordokhani, S. Davaei far, Application of the Bernstein polynomials for solving the nonlinear Fredholm integro-differential equations, J. Appl. Math.
Bioinform. 1 (2) (2011) 13–31.
[29] G. Tachev, Pointwise approximation by Bernstein polynomials, Bull. Aust. Math. Soc. 85 (3) (2012) 353–358.
[30] R. Ezzati, S. Najafalizadeh, Application of Chebyshev polynomials for solving nonlinear Volterra–Fredholm integral equations system and convergence
analysis, Indian J. Sci. Technol. 5 (2) (2012) 2060–2064.