You are on page 1of 13

Applied Mathematics and Computation 273 (2016) 142–154

Contents lists available at ScienceDirect

Applied Mathematics and Computation


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

A collocation method based on Bernstein polynomials to solve


nonlinear Fredholm–Volterra integro-differential equations
Şuayip Yüzbaşı∗
Department of Mathematics, Faculty of Science, Akdeniz University, TR 07058 Antalya, Turkey

a r t i c l e i n f o a b s t r a c t

Keywords: In this study, a collocation method that based on Bernstein polynomials is presented for
Nonlinear integro-differential equations nonlinear Fredholm–Volterra integro-differential equations (NFVIDEs). By means of the
Bernstein polynomials
collocation method and the matrix operations, the problem is reduced into a system of the
Collocation method
nonlinear algebraic equations. The approximate solutions are obtained by solving this non-
Bernstein series method
Approximate solution linear system. Error analysis is presented for the Bernstein series solutions of the nonlinear
Fredholm–Volterra integro-differential equations. Several examples are given to illustrate the
efficiency and implementation of the proposed method for solving the NFVIDEs. Comparisons
are made to confirm the reliability of the method. Also error analysis is applied for the
numerical examples.
© 2015 Elsevier Inc. All rights reserved.

1. Introduction

Many scientific phenomena in science and engineering can be modeled by nonlinear integro-differential equations. Nonlinear
integro-differential equations are usually difficult to solve analytically. So, it is important that the approximate solutions of the
nonlinear integro-differential equations can be computed by numerical methods. In recent years, several numerical methods for
solving linear and nonlinear integro-differential equations have been presented by many authors. For example, some linear and
nonlinear integro-differential equations have been solved using the Taylor expansion approach [1–3], the Tau method [4–6], the
Chebyshev method [7], the Taylor method [8,9], the homotopy perturbation method [10,11], the Taylor polynomial method [12],
the Shannon wavelets approximation [13], the He’s homotopy perturbation method [14], the He’s variational iteration method
[15], the direct method [16,17], the sinc-collocation method [18,19], the rationalized Haar functions [20], the homotopy analysis
method [21], the Cas wavelet method [22], the Bessel collocation method [23,24], the Chebyshev matrix method [25], the multi-
parametric homotopy approach [26] and the B-spline collocation method [27].
In this study, we present a Bernstein series method for the solutions of the mth order nonlinear Fredholm–Volterra integro-
differential equation

m 
n  b  x
Pk,r (x)yr (x)y(k) (x) = g(x) + λ1 KF (x, t )[y(t )] dt + λ2 KV (x, t )[y(t )] dt,
p q
0 ≤ a ≤ x, t ≤ b (1)
a a
k=0 r=0
with the mixed conditions

m−1
[a jk y(k) (a) + b jk y(k) (b)] = γ j , j = 0, 1, . . . , m − 1 (2)
k=0


Corresponding author. Tel.: +90 242 310 23 83; fax: +90 242 227 89 11.
E-mail address: syuzbasi@akdeniz.edu.tr, suayipyuzbasi@gmail.com

http://dx.doi.org/10.1016/j.amc.2015.09.091
0096-3003/© 2015 Elsevier Inc. All rights reserved.
Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154 143

where y0 (x) = 1 and y(0) (x) = y(x) are unknown functions, p and q are nonnegative integers, λ1 , λ2 , a jk , b jk and γ j are constants.
Moreover, Pk,r (x), g(x), KF (x, t ) and KV (x, t ) are continuous functions on interval a ≤ x, t ≤ b.
We note that some researchers have considered the various studies regarding existence of solutions of integral and integro-
differential equations in [28–33]
This paper is organized as follows: some properties of Bernstein polynomials are given in Section 2. In Section 3, we sum-
marize the method. The Bernstein series method is applied for nonlinear Fedholm–Volterra integro-differential equations in
Section 4. In Section 5, we give the error analysis for the method. We present five numerical examples to clarify the method in
Section 6. Section 7 concludes the article with a brief summary.

2. Some properties of Bernstein polynomials

The Bernstein basis polynomials of degree n (see, [34], e.g., [35], p. 66) are defined by
 
n k
Bk,n (x) = x (1 − x)n−k , x ∈ [0, 1].
k

By using the binomial expansion


 

n−k
n−k i
(1 − x) n−k
= (−1) i
x,
i
i=0

it can be written as
  

n−k
n n − k k+i
Bk,n (x) = (−1)i x , x ∈ [0, 1].
k i
i=0

Also, the Bernstein basis polynomials of degree n in [0, R] are given by the formula [34]
 
n xk (R − x)
n−k
Bk,n (x) = . (3)
k Rn
By substituting the binomial expansion
 

n−k
n − k n−k−i i
(R − x) n−k
= (−1) i
R x
i
i=0

here, we have the formula


  

n−k
n n − k xk+i
Bk,n (x) = (−1) i
x ∈ [0, R].
k i Rk+i
i=0

The Bernstein basis polynomials given by Eq. (3) can be written in the matrix form [36–38]
Bn (x) = [B0,n (x) B1,n (x) · · · Bn,n (x)] = X(x)DT (4)
where
⎡d d12 ... d1n

11
⎢d21 d22 ... d2n ⎥
⎢ ⎥
X(x) = [1 x x2 ··· xn ], D = ⎢. .. .. .. ⎥,
⎣.. . . . ⎦
dn1 dn2 ... dnn
and
⎧   
⎨ (−1) j−i n n − i
, i≥ j
di j = Rj i j−i

0, i < j.

3. Description of the Bernstein series method

In this study, by using the Bernstein polynomial approximation [36–38], we obtain an approximate solution of the problem
given by Eqs. (1) and (2) in the form

N
yN (x) = ak Bk,N (x − c), x ∈ [0, b]. (5)
k=0
144 Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154

Here, 0 ≤ c ≤ b, ak , k = 0, 1, 2, . . . , N are the unknown Bernstein coefficients; N are chosen any positive integers such that N ≥ m
and Bk,N (x), k = 0, 1, 2, . . . , N are the Bernstein polynomials [25] defined by
 
N xk (b − x)
N−k
Bk,N (x) = , k = 0, 1, 2, . . . , Nx ∈ [0, b]
k bN
We convert the desired solution (5) to the matrix form

yc,N (x) = BN (x − c)A (6)

where

BN (x − c) = [B0,N (x − c) B1,N (x − c) ··· BN,N (x − c)]


and

A = [a0 a1 ··· aN ]T .
By means of the matrix form (6), the collocation points and the matrix operations, we construct the matrix forms for each
term of Eq. (1) and the conditions (2). So, we convert the problem into a matrix equation which is a system of nonlinear algebraic
equations. This system corresponds to a system of the (N + 1) nonlinear algebraic equations with the unknown coefficients an ,
n = 0, 1, 2, . . . , N. Finally, the approximate solution will be obtained by solving this system.

4. Application of the method for nonlinear Fedholm–Volterra integro-differential equations

Firstly, let us write Eq. (1) in the form

H (x) = g(x) + λ1 I(x) + λ2V (x) (7)

where

m 
n  b  x
H (x) = Pk,r (x)yr (x)y(k) (x), I(x) = KF (x, t )[y(t )] dt, V (x) = KV (x, t )[y(t )] dt.
p q

a a
k=0 r=0

By substituting Eq. (4) in the matrix form (6), the matrix form of the approximate solution becomes

yN (x) = X(x − c)DT A (8)

where

X(x − c) = [1 x−c (x − c)2 ··· (x − c)N ],


⎡ ⎤
d11 d12 ... d1n ⎧   
⎢d21 d22 ... d2n ⎥ ⎨ (−1) j−i n n − i
, i≤ j
D = ⎢. .. ⎥
.. ⎦ ; di j = ⎩ R i j−i
j
⎣.. .
..
. . 0, i > j.
dn1 dn2 ... dnn

The relation between the matrix X(x − c) and its k-order derivative X(k) (x − c) is given by

X(k) (x − c) = X(x − c)(BT )k (9)

where
⎡ ⎤
0 1 0 ··· 0
⎢0 0 2 ··· 0⎥
⎢. .. ⎥
BT = ⎢
⎢..
..
.
..
.
..
.

. ⎥.
⎣0 0 0 ··· N⎦
0 0 0 ··· 0
By substituting Eq. (9) into k-order derivative of Eq. (8), we have the matrix form

y(Nk) (x) = X(x − c)(BT )k DT A, k = 0, 1, 2, . . . , m (10)

By placing the collocation points defined by


b−a
xi = a + i, i = 0, 1, . . . , N, (11)
N
into Eq. (10), we have the matrix equation system

y(Nk) (xi ) = X(xi − c)(BT )k DT A, i = 0, 1, . . . , N.


Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154 145

We can write the above system in compact form


Y(k) = X(BT )k DT A (12)
where
⎡ ⎤ ⎡ ⎤
X(x0 − c) y(Nk) (x0 )
⎢ X(x1 − c) ⎥ ⎢ (k) ⎥
⎢ ⎥ ⎢ yN (x1 ) ⎥
X=⎢ .. ⎥ and Y = ⎢ .. ⎥
(k) ⎢

⎣ . ⎦ ⎣ . ⎦
X(xN − c) y(k) (xN ) N

In the similar way, by substituting the collocation points (11) into the yr (x)y(k) (x), we obtain the matrix representation
⎡ ⎤ ⎡ ⎤⎡ (k) ⎤
yrN (x0 )y(Nk) (x0 ) yrN (x0 ) 0 ··· 0 yN (x0 )
⎢ r ⎥ ⎢ (k) ⎥
⎢ yN (x1 )y(Nk) (x1 ) ⎥ ⎢ 0 yrN (x1 ) ··· 0 ⎥ ⎥⎢ yN (x1 ) ⎥  r
⎢ ⎥=⎢ ⎢ ⎥ = Ȳ Y(k)
⎢ .. ⎥ ⎢ .. .. .. .. ⎥⎢ .. ⎥ (13)
⎣ . ⎦ ⎣ . . . . ⎦⎣ . ⎦
yrN (xN )y(Nk) (xN ) 0 0 ··· yrN (xN ) (k)
yN (xN )
where
Ȳ = X̄D̄Ā, (14)

⎡ ⎤ ⎡ ⎤
X(x0 ) 0 ··· 0 BT 0 ··· 0
⎢ 0 X(x1 ) ··· 0 ⎥ ⎢0 BT ··· 0⎥
X̄ = ⎢ . .. ⎥ B̄ = ⎢ . .. ⎥
⎣ .. ..
.
..
. .
⎦, ⎣ .. ..
.
..
. .

0 0 ··· X(xN ) 0 0 ··· BT
⎡ ⎤ ⎡ ⎤
DT 0 ··· 0 A 0 ··· 0
⎢0 DT ··· 0⎥ ⎢0 A ··· 0⎥
D̄ = ⎢ . .. .. ⎥ and Ā = ⎢ . .. .. ⎥
⎣ .. .
..
. .
⎦ ⎣ .. .
..
. .

0 0 ··· D T 0 0 ··· A
b
Let us constitute the matrix representation of the part a KF (x, t )[y(t )] dt. Since KF (x, t ) is continuous function, ai t i KF (x, t )
p

is Riemann integrable with respect to t. Thus, by substituting the matrix form (8) into integral part I(x) of Eq. (4), we have the
matrix relation
 b
 b

p−1
I(x) = KF (x, t )[y(t )] dt = KF (x, t )[X(t − c)D A] X(t − c)dt DT A = KF (x)DT A
p T
(15)
a a

where
 b
p−1
KF (x) = KF (x, t )[X(t − c)DT A] X(t − c)dt
 a

= k f,1 (x)k f,2 (x)k f,3 (x) . . . k f,N (x) .
By putting the collocation points (11) into Eq. (15), the matrix form for I(x) is obtained as
⎡ ⎤ ⎡ ⎤
I(x0 ) KF (x0 )
⎢ I(x1 ) ⎥ ⎢ KF (x1 ) ⎥
I =⎢ . ⎥ = KF DT A; KF = ⎢ . ⎥ , (16)
⎣ .. ⎦ ⎣ .. ⎦
I(xN ) KF (xN )
Similarly, we substitute the matrix form (8) into integral part V (x) of Eq. (7) and thus we gain the matrix relation
 x
 x

q−1
V (x) = KV (x, t )[y(t )] dt = KV (x, t )[X(t − c)D A] X(t − c)dt DT A = KV (x)DT A
q T
(17)
a a

where
 x
q−1
KV (x) = KV (x, t )[X(t − c)DT A] X(t − c)dt
 a

= kv,1 (x)kv,2 (x)kv,3 (x) . . . kv,N (x) .
146 Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154

By placing the collocation points (11) into Eq. (17), we have the matrix form
⎡ ⎤ ⎡ ⎤
V (x0 ) KV (x0 )
⎢V (x1 ) ⎥ ⎢ KV (x1 ) ⎥
V =⎢ . ⎥ = KV DT A; KV = ⎢ . ⎥ (18)
⎣ .. ⎦ ⎣ .. ⎦
V (xN ) KV (xN )

Now, let us substitute the collocation points into Eq. (1) as follows:


m 
n  b  xi
Pk,r (xi )yr (xi )y(k) (xi ) = g(xi ) + λ1 KF (xi , t )[y(t )] dt + λ2 KV (xi , t )[y(t )] dt,
p q
i = 0, 1, . . . , N
a a
k=0 r=0

and then this system is written in the matrix form


m 
n
r
Pk,r (Ȳ) Y(k) = G + λ1 I + λ2 V (19)
k=0 r=0

where
⎡ ⎤ ⎡ ⎤
Pk,r (x0 ) 0 ··· 0 yr (x0 ) 0 ··· 0
⎢ 0 Pk,r (x1 ) ··· 0 ⎥  r ⎢ 0 y (x1 )
r
··· 0 ⎥
Pk,r =⎢ ⎥, =⎢ . .. ⎥
⎣ ... ..
.
..
.
..
.
⎦ Ȳ
⎣ .. ..
.
..
. .
⎦,
0 0 ··· Pk,r (xN ) 0 0 ··· yr (xN )
⎡ ⎤ ⎡ ⎤
y(k) (x0 ) g(x0 )
⎢ y(k) (x1 ) ⎥ ⎢ g(x1 ) ⎥
Y(k) = ⎢ ⎥ and G =⎢ . ⎥
⎣ ... ⎦ ⎣ .. ⎦
y(k) (xN ) g(xN )

After the substitution of the relations (12)–(14), (16) and (18) into Eq. (19), we obtain the matrix equation
 

m 
n
r T k
Pk,r (X̄D̄Ā) X(B ) D − λ1 KF D − λ2 KV D
T T T
A = G. (20)
k=0 r=0

Briefly, we can write Eq. (20) in the form

WA = G or [W; G] (21)

where

m 
n
r k
W= Pk,r (X̄D̄Ā) X(BT ) DT − λ1 KF DT − λ2 KV DT .
k=0 r=0

By using the relation (8) at points a and b, the matrix representation of the mixed conditions (2) is obtained as
 

m−1
T k
[a jk X(a) + b jk X(b)](B ) DT A =γ j , j = 0, 1, 2, . . . , m − 1,
k=0

or briefly

U j A = [γ j ] or [U j ; γ j ]; j = 0, 1, 2, . . . , m − 1 (22)

where


m−1
Uj = [a jk X(a) + b jk X(b)](BT )k DT
k=0
= [u j0 u j1 u j2 . . . u jN ]

Finally, we replace the row matrices (22) by the m rows of the augmented matrix (21) and thus we have the new augmented
matrix

[W̃; G̃] or W̃A = G̃


Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154 147

which is a system of nonlinear algebraic equations. For convenience, if last rows of the augmented matrix (21) are replaced, the
new augmented matrix becomes as follows [23–25,36–38]:
⎡ w w0,1 w0,2 ... w0,N ; g(x0 )

0,0
⎢ w1,0 w1,1 w1,2 ... w1,N ; g(x1 ) ⎥
⎢ w2,0 w2,1 w2,2 ... w2,N ; g(x2 ) ⎥
⎢ ⎥
⎢ .. .. .. .. .. .. .. ⎥
⎢ . . . . . . . ⎥
⎢ ⎥
⎢w wN−m,1 wN−m,2 ... wN−m,N ; g(xN−m )⎥
[W̃; G̃] = ⎢ N−m,0 ⎥. (23)
⎢ u0,0 u0,1 u0,2 ... u0,N ; γ0 ⎥
⎢ u1,0 u1,1 u1,2 ... u1,N ; γ1 ⎥
⎢ ⎥
⎢ u2,0 u2,1 u2,2 ... u2,N ; γ2 ⎥
⎢ . ⎥
⎣ . .. .. .. .. .. .. ⎦
. . . . . . .
um−1,0 um−1,1 um−1,2 ... um−1,N ; γm−1
By solving nonlinear system (23), the unknown Bernstein coefficients an , n = 0, 1, 2, . . . , N are determined and substituted in
(6). Thus we obtain the Bernstein polynomial solution


N
yN (x) = ak Bk,N (x − c), 0 ≤ c ≤ b, (24)
k=0

5. Error analysis

In this section, we give an error bound for the presented method. This error analysis was presented for the rational solutions,
based on Bernstein polynomials, of the boundary value problems of the linear differential equations in [39].

Theorem 5.1 [40]. Let f be a function in C n+1 [a, b], and let P be a polynomial of degree ≤ n that interpolates the function f at n + 1
distinct points x0 , x1 , . . . , xn ∈ [a, b]. Then to each x ∈ [a, b] there exists a point ξx ∈ [a, b] such that

1 n
f (x) − pn (x) = f (n+1) (ξx ) (x − xi ). 
(n + 1)! i=0

Let f be an exact solution of Eq. (1) and pN be the interpolation polynomial of f . If f is sufficiently smooth, we can write f as
f = pN + KN where KN is the error function [36]

(x − x0 )(x − x1 ) · · · (x − xN ) (N+1)
KN (x) = f (ξx ), ξx ∈ (0, R).
(N + 1)!
If yN is the Bernstein series solution of (1) given by Eq. (24), thenyN satisfies Eqs. (1) and (2) on the nodes. So, yN and pN are the
solutions of W̃A = G̃ and W̃A∗ = G̃+G, respectively, where
 

m 
n  b  xi
(k)
Pk,r (xi )KN (xi )KN (xi ) + λ1 KF (xi , t )[KN (t )] dt + λ2 KV (xi , t )[KN (t )] dt
p q
[G]i1 = − r
.
a a
k=0 r=0 i1

An upper bound of the absolute errors for the method is given by the following theorem.

Theorem 5.2 ([39] (Convergence of collocation methods)). Let y and f be the Bernstein series solution and the exact solution of (1),
respectively, and pn be the interpolation polynomial of f . Let W̃, A, A∗ , G̃ and G be defined above. If f is sufficiently smooth, then

| f (x) − yN (x)| ≤ |KN (x)| + X(x − c)DT (A − A∗ )∞ .


Proof. Adding and subtracting the term pN yields

| f (x) − yN (x)| ≤ | f (x) − pN (x)| + |yN (x) − pN (x)|.


Since f is sufficiently smooth, the first term in the right hand side becomes |KN (x)| from Theorem 5.1. For the second term,
Eqs. (4) and (8) and the properties of the norm give the following:

|yN (x) − pN (x)| = |BN (x − c)A − BN (x − c)A∗ |


= |BN (x − c)(A − A∗ )|
≤ X(x − c)DT (A − A∗ )∞ .

Summing of these two terms gives the desired result.


148 Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154

6. Illustrative examples

In this section, we explain the method with five examples. It is also shown that the upper bounds of absolute errors obtained
by the Theorem 5.2 and the ∞-norm  f − yN ∞ = max{| f (x) − yN (x)|, 0 ≤ x ≤ b} are consistent. Numerical results are compared
with the results of the other methods. All numerical computations have been done on the computer using a program written in
Maple. While the program was running for the examples, digits were selected as twenty.

Example 1. Let us first consider the nonlinear Fredholm–Volterra integro-differential equation


643 13 4 2 5 1 6 2 7 1 8
y (x) + y(x)y (x) = 2 − x + 3x2 − 3x3 + x − x + x − x + x
560 6 9 6 15 18
 1  x
1 1
xt[y(t )] dt − x2 t[y(t )] dt
3 2
+
2 0 3 0
with the boundary conditions
y(0) = 1 and y(1) = 1.
For this problem, a = 0, b = 1, p = 3, q = 2, λ1 = 12 , λ2 = − 31 , P1,1 (x) = 1, P2,0 (x) = 1, KF (x, t ) = xt, KV (x, t ) = x2 t, g(x) = 2 −
643
560 x + 3x
2 − 3x3 + 13 4 2 5 1 6 2 7 1 8
6 x − 9 x + 6 x − 15 x + 18 x .
Let us seek the approximate the solution in the form

N
yN (x) = ak Bk,N (x − c)
k=0

for c = 0 and N = 2. The set of collocation points (11) for N = 2 is computed as


 1

x0 = 0, x1 = , x3 = 1
2
and from Eq. (20), the fundamental matrix equation of the problem is written as
 2 
P1,1 X̄D̄ĀXBT DT +P2,0 X(BT ) DT − λ1 KF DT − λ2 KV DT A = G
where
         
1 0 0 1 0 0 X(0) 1 0 0 1 0 0
P0,0 = 0 1 0 , P1,0 = 0 1 0 , X= X(1/2) = 1 1/2 1/4 , DT = −2 2 0 ,
0 0 1 0 0 1 X(1) 1 1 1 1 −2 1
       
0 1 0 0 0 X(0) DT 0 0 A 0 0
T
B = 0 0 2 ,
X̄ = X(1/2) 0 0 , D̄ = 0 DT 0 , Ā = 0 A 0
0 0 0 0 X(1)
0 0 0 DT 0 0 A
 b  x
2
KF (x) = xt[X(t )DT A] X(t )dt, KV (x) = x2 tX(t )DT AX(t )dt,
a    0
 
KF (0) KV (0) 2
KF = KF (1/2) , KV = KV (1/2) and G= 62, 333/32, 256
KF (1) KV (1) 4987/1680
The matrix forms for the boundary conditions from Eq. (22) are obtained as follows
[U0 ; λ0 ] = [1 0 0; 1]
and
[U1 ; λ1 ] = [0 0 1; 1].

Hence, we obtain the new augmented matrix [W̃; G̃] for the problem. Solving this system, the Bernstein coefficients matrix is
calculated as
A = [1 1/2 1]T .
The elements of the Bernstein coefficients matrix are substituted into Eq. (6). Thus, the approximate solution for c = 0 and
N = 2 becomes: y0,2 (x) = x2 − x + 1 which is an exact solution of Eq. (1). We note that CPU time of the method for N = 2 is 0.42 s.

Example 2. As a second example, we consider the nonlinear Fredholm–Volterra integro-differential equation


 1  x
1 3 1 3 3 1 1 1 1
y (x) + xy2 (x)y (x) − y(x) = xe3x − x3 t[y(t )] dt + 2xt[y(t )] dt
3 3
x − x e − x − x2 e3x + (25)
36 18 9 3 4 0 2 0

with the initial conditions y(0) = 1, y(1) (0) = 1, y(2) (0) = 1.


Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154 149

Table 1
Comparison of the maximum absolute errors obtained by the
∞-norm and the Theorem 5.2 of Eq. (25).

N By ∞-norm By Theorem 5.2

3 5.1600e−002 1.6720e−001
4 4.4000e−003 8.1000e−003
5 3.8519e−004 5.5016e−003
6 2.2692e−005 6.2686e−004
7 1.3764e−006 4.4165e−005
8 6.0888e−008 3.0887e−006
9 2.8321e−009 3.5002e−009
10 1.0023e−010 7.8420e−009

CPU times of our method for some values of N for Eq. (25)

N 3 5 7 10
CPU time (s) 0.28 0.66 0.89 2.16

Fig. 1. (a) Comparison of the exact solution and the approximate solutions for N = 3, 5, 7, 10 of Eq. (25) and (b) Comparison of the absolute error functions for
N = 3, 5, 7, 10 of Eq. (25).
150 Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154

Table 2
Comparison of the maximum absolute errors obtained by the
∞-norm and the Theorem 4.2 of Eq. (26).

N By ∞-norm By Theorem 5.2

3 7.8000e−003 2.1200e−002
4 4.3534e−004 2.3000e−003
5 3.6607e−005 1.6574e−004
6 1.5551e−006 1.0404e−005
7 9.1675e−008 5.1426e−007
8 3.1820e−009 2.3157e−008
9 1.4238e−010 8.9254e−010
10 4.1851e−012 3.1997e−011
11 1.4611e−013 1.0175e−012
12 1.0658e−014 3.0437e−014

CPU times of our method for some values of N for Eq. (26)

N 3 5 10 12
CPU time (s) 0.24 1.19 3.73 5.91

Table 3
Comparison of the numerical solutions and the absolute errors of Eq. (26).

Exact solution Taylor method [2] for N = 7 Present method for N = 7

xi y(xi ) = exi y7 (xi ) Absolute error y7 (xi ) Absolute error

0 1 1 0 1 0
0.2 1.22140275816 1.2214031 3.4184e−007 1.22140275634 1.8156e−009
0.4 1.49182469764 1.4918253 6.0236e−007 1.49182469631 1.3355e−009
0.6 1.82211880039 1.8221197 8.9961e−007 1.82211879934 1.0530e−009
0.8 2.22554092849 2.2255420 1.0715e−006 2.22554092747 1.0261e−009
1 2.71828182846 2.7182814 4.2846e−007 2.71828173678 9.1675e−008

For this example, a = 0, b = 1, p = 3, q = 3, λ1 = 14 , λ2 = 12 , P0,0 (x) = −1, P2,2 (x) = x, P3,0 (x) = 1, KF (x, t ) = x3 t, KV (x, t ) =
2xt, g(x) = xe3x − 36
1 3 1 3 3
x − 18 x e − 19 x − 13 x2 e3x .
The fundamental matrix equation of the problem becomes
 2 2 3 
P0,0 XDT +P2,2 (X̄D̄Ā) X(BT ) DT +P3,0 X(BT ) DT − λ1 KF DT − λ2 KV DT A = G.
As in Example 1, following the method given in Section 3, we find the approximate solutions of the problem for N = 5, 7, 10,
and digits = 20, respectively,
y5 (x) = 1 + x + 0.5x2 + 0.166666666667x3 + (0.410299935858e − 1)x4 + (0.101999795341e − 1)x5 ,
y7 (x) = 1 + x + 0.5x2 + 0.166666666667x3 + (0.416623224873e − 1)x4 + (0.835851684467e − 2)x5
+ (0.132832098038e − 2)x6 + (0.264625106161e − 3)x7 ,
and
y10 (x) = 1 + x + 0.5x2 + 0.166666666667x3 + (0.416666673690e − 1)x4 + (0.833332606332e − 2)x5
+ (0.138892524040e − 2)x6 + (0.198306504637e − 3)x7 + (0.249910544335e − 4)x8
+ (0.255357392090e − 5)x9 + (0.391886472495e − 6)x10 .
The exact solution of the problem is y(x) = ex . In Table 1, we tabulate the maximum absolute errors computed by the ∞-norm
 f − yN ∞ = max{| f (x) − yN (x)|, 0 ≤ x ≤ 1} and the Theorem 5.2 for various values of N. We compare the exact solution and the
approximate solutions for N = 3, 5, 7, 10 (Fig. 1(a)). Fig. 1(b) shows a plot of the error functions for N = 3, 5, 7, 10. Also, CPU times
of our method for some values of N for Eq. (25) are given in Table 1.

Example 3 [2]. Now, let us solve the nonlinear Fredholm–Volterra integro-differential equation
2
 1 3x
 4  8
 1  x
y (x) + 2xy(x) = − x + e + (2x + 1)ex + −e x+ + (x − t )y(t )dt + (x + t )[y(t )]3 dt (26)
3 9 3 9 0 0
with the initial condition y(0) = 1. Here, a = 0, b = 1, p = 1, q = 3, λ1 = 1, λ2 = 1, P0,0 (x) = 2x,
2
 1 3x
 4  8
P1,0 (x) = 1, KF (x, t ) = x − t, KV (x, t ) = x + t, g(x) = − x + e + (2x + 1)ex + −e x+ .
3 9 3 9
The exact solution is given by y(x) = ex . From Eq. (20), the fundamental matrix equation is written as
 
P0,0 XDT +P1,0 XBT DT − λ1 KF DT − λ2 KV DT A = G.
Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154 151

Fig. 2. (a) Comparison of the absolute error functions for N = 3, 5, 7, 10, 12 of Eq. (26) and (b) Comparison of the absolute error functions obtained by our method
and the Taylor method for N = 7 of Eq. (26).

Table 4
Comparison of the solutions for the xi value of Eq. (27).

xi Exact solution Present method Direct method BPFs method Adomian’s


y(xi ) = cos (xi ) for N = 7 [16] for m = 8 [35] for m = 16 method [36]

0 1 1 1 1 1
0.2 0.9800665778 0.980064907018 0.978496 0.975784 0.980303
0.4 0.9210609940 0.921060345026 0.920841 0.918443 0.921165
0.6 0.8253356149 0.825335722687 0.826070 0.828963 0.822596
0.8 0.6967067093 0.696706121129 0.698009 0.710418 0.667739
1 0.5403017838 0.540314307450 0.543491 0.566917 0.364798

By following the procedure in Section 3, we obtain the approximate solutions for various values of N. The maximum absolute
errors obtained by the ∞-norm and the Theorem 5.2 for various values of N are given in Table 2. Table 3 denotes the numerical
results of the absolute error functions, the exact solutions and the approximate solutions obtained by our method and the Tay-
lor method [2] for N = 7. The absolute error functions are compared for N = 3, 5, 7, 10, 12 in Fig. 2(a). Also, the absolute error
function is compared with one of the Taylor method for N = 7. From Table 2 and Fig. 2, we see that the results obtained by the
152 Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154

Table 5
Comparison of the absolute errors for the xi value of Eq. (27).

xi Present method for N = 7 Direct method [16] for m = 8 BPFs method [35] for m = 16 Adomian’s method [36]

0 0 0 0 0
0.2 1.6708e−006 1.5706e−3 4.2825e−3 2.3642e−4
0.4 6.4898e−007 2.1999e−4 2.6179e−3 1.0400e−4
0.6 1.0778e−007 7.3438e−4 3.6278e−3 2.7396e−3
0.8 5.8822e−007 1.3022e−3 1.3711e−2 2.8967e−2
1 1.2002e−005 3.1892e−3 2.6615e−2 1.7550e−1

Fig. 3. (a) Comparison of the exact and the approximate solutions of Eq. (27) and (b) Comparison of the absolute error functions of Eq. (27).

present method are better than that obtained by the Taylor method. In addition, it is seen from Table 2 and Fig. 2 that the errors
decrease when the integer N is increased.
We tabulate the CPU times of our method for some values ofN for Eq. (26) in Table 2.

Example 4 ([16]). Consider the nonlinear Volterra integro-differential equation


 x
y (x) = 2 sin (x) cos (x) − 3 cos (x − t )[y(t )] dt, 0 ≤ x, t ≤ 1,
2
(27)
0
Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154 153

with the initial condition y(0) = 1. For this problem a = 0, b = 1, q = 2, λ1 = 0, λ2 = −1, P1,0 (x) = 1, KV (x, t ) = 3 cos (x −
t ), g(x) = 2 sin (x) cos (x).

The exact solution of the problem isy(x) = cos (x).


From Eq. (20), we write the fundamental matrix equation of problem as

{P1,0 XBT DT − λ2 KV DT }A = G.
This problem is solved by our method for N = 7. In Tables 4 and 5, our results are compared with the results of the direct
method based on the triangular functions [16], the Block-pulse functions method (BPFs) [41] and the Adomian’s decomposition
method (AM) [42]. Fig. 3(a) displays a comparison of the exact solution, our solution and the solutions of the other methods. The
absolute error functions are compared with the other methods in Fig. 3(b). It is observed from Tables 4 and 5 and Fig. 3 that our
results are better than those obtained by other methods. We note that CPU time of the method for N = 7 is 21.17 s.

Example 5. Finally, we consider the nonlinear Volterra integro-differential equation


⎧ 
⎨y(5) (x) = ex − 1 e3x + 1 + x [y(t )]3 dt, 0 ≤ t ≤ 1
3 3 0

y(0) = y(1) (0) = y(2) (0) = y(3) (0) = y(4) (0) = 1.

By using our method for N = 3, 6, we have the approximate solutions

y3 (x) = 1 + x + 0.5x2 + 0.166666666667x3


and

y6 (x) = 1 + x + 0.5x2 + 0.166666666667x3 + (0.416666666667e − 1)x4


+ (0.833333333333e − 2)x5 + (0.151133677547e − 2)x6 .
If y = f (x)is an exact solution of Example 5, then it is of C 4 [0, 1]. So, by using Theorem 5.2 and property of norm, the upper
bound of the absolute error is computed as

 f (x) − y6 (x)∞ ≤  f (x) − y3 (x)∞ + y3 (x) − y6 (x)∞ = O(10−4 ) + 5.150 × 10−2 .
We note that the approximate solution can be calculated with an acceptable error bound when the exact solution is unknown.
This problem is an example of this case.
We note that CPU time of the method for N = 6 is 0.67.

7. Conclusions

In this study, we have presented the Bernstein series method with error analysis to obtain solutions of the nonlinear
Fredholm–Volterra integro-differential equations. It is observed from the examples that the proposed method for nonlinear
integro-differential equations is a good approximation. We see from Examples 2 and 3 that the upper bounds of the ab-
solute errors obtained by using Theorem 5.2 are consistent with the upper bounds obtained by the ∞-norm  f − yN ∞ =
max{| f (x) − yN (x)|, a ≤ x ≤ b}. Also, the comparisons of the exact solutions and the numerical solutions of our method and
the other methods show that the present method is very effective. If the exact solution of the problem is a polynomial function,
then the exact solution can be obtained by this method. This feature can be seen from Example 1. The upper bound of the abso-
lute errors can be computed by using Theorem 5.2 when the exact solution of the problem is not known. This situation can be
seen from Example 5. One advantage of the method that the approximate solutions can be easily computed by aid of computer
programs such as Maple, Matlab and Mathematica.

Acknowledgments

The author is supported by the Scientific Research Project Administration of Akdeniz University. I would like to thank reviewers
for their constructive comments and suggestions to improve the paper.

References

[1] S. Yalçinbaş, Taylor polynomial solutions of nonlinear Volterra–Fredholm integral equations, Appl. Math. Comput. 127 (2002) 195–206.
[2] K. Maleknejad, Y. Mahmoudi, Taylor polynomial solution of high-order nonlinear Volterra–Fredholm integro-differential equations, Appl. Math. Comput.
145 (2003) 641–653.
[3] P. Darania, K. Ivaz, Numerical solution of nonlinear Volterra–Fredholm integro-differential, Comput. Math. Appl. 56 (2008) 2197–2209.
[4] S. Abbasbandy, A. Taati, Numerical solution of the system of nonlinear Volterra integro-differential equations with nonlinear differential part by the opera-
tional Tau method and error estimation, J. Comput. Appl. Math. 231 (2009) 106–113.
[5] S. Shahmorad, Numerical solution of the general form linear Fredholm–Volterra integrodifferential equations by the Tau method with an error estimation,
Appl. Math. Comput. 167 (2005) 1418–1429.
[6] S.M. Hosseini, S. Shahmorad, Numerical solution of a class of integro-differential equations by the Tau method with an error estimation, Appl. Math. Comput.
136 (2003) 559–570.
154 Ş. Yüzbaşı / Applied Mathematics and Computation 273 (2016) 142–154

[7] A. Akyüz-Daşcıoğlu, M. Sezer, Chebyshev polynomial solutions of systems of higher-order linear Fredholm–Volterra integro-differential equations, J.
Franklin Inst 342 (2005) 688–701.
[8] A. Karamete, M. Sezer, A Taylor collocation method for the solution of linear integro-differential equations, Int. J. Comput. Math 79 (2002) 987–1000.
[9] B. Bülbül, M. Gulsu, M. Sezer, A new Taylor collocation method for nonlinear Fredholm-Volterra integro-differential equations, Numer. Methods Part. Differ.
Equ 26 (2010) 1006–1020.
[10] E. Yusufoglu, Numerical expansion methods for solving systems of linear integral equations using interpolation and quadrature rules, Int. J. Comput. Math
84 (2007) 133–149.
[11] M. Ghasemi, M.T. Kajani, E. Babolian, Numerical solutions of the nonlinear Volterra-Fredholm integral equations by using homotopy perturbation method,
Appl. Math. Comput. 188 (2007) 446–449.
[12] W. Wang, An algorithm for solving the high-order nonlinear Volterra–Fredholm integro-differential equation with mechanization, Appl. Math. Comput. 172
(2006) 1–23.
[13] K. Maleknejad, M. Attary, An efficient numerical approximation for the linear class of Fredholm integro-differential equations based on Cattani’s method,
Commun. Nonlinear Sci. Numer. Simul 16 (2011) 2672–2679.
[14] J. Saberi-Nadjafi, A. Ghorbani, He’s homotopy perturbation method: An effective tool for solving nonlinear integral and integro-differential equations,
Comput. Math. Appl. 58 (2009) 2379–2390.
[15] S.A. Yousefi, A. Lotfi, Mehdi Dehghan, He’s variational iteration method for solving nonlinear mixed Volterra-Fredholm integral equations, Comput. Math.
Appl. 58 (2009) 2172–2176.
[16] 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.
[17] K. Maleknejad, K. Mahdiani, Solving nonlinear mixed Volterra–Fredholm integral equations with two dimensional block-pulse functions using direct
method, Commun. Nonlinear Sci. Numer. Simul 16 (2011) 3512–3519.
[18] M. Zarebnia, M.G. Ali Abadi, Numerical solution of system of nonlinear second-order integro-differential equations, Comput. Math. Appl. 60 (2010) 591–601.
[19] A. Mohsen, M. El-Gamel, On the numerical solution of linear and nonlinear volterra integral and integro-differential equations, Appl. Math. Comput. 217
(2010) 3330–3337.
[20] Y. Ordokhani, M. Razzaghi, Solution of nonlinear Volterra–Fredholm–Hammerstein integral equations via a collocation method and rationalized Haar func-
tions, Appl. Math. Lett. 21 (2008) 4–9.
[21] A. Shidfar, A. Molabahrami, A. Babaei, A. Yazdanian, A series solution of the nonlinear Volterra and Fredholm integro-differential equations, Commun.
Nonlinear Sci. Numer. Simul 15 (2010) 205–215.
[22] H. Danfu, S. Xufeng, Numerical solution of integro-differential equations by using CAS wavelet operational matrix of integration, Appl. Math. Comput 194
(2007) 460–466.
[23] Ş. Yüzbaşı, N. Şahin, M. Sezer, Bessel polynomial solutions of high-order linear Volterra integro-differential equations, Comput. Math. Appl. 62 (2011) 1940–
1956.
[24] Ş. Yüzbaşı, N. Şahin, M. Sezer, Numerical solutions of systems of linear Fredholm integro-differential equations with Bessel polynomial bases, Comput.
Math. Appl 61 (2011) 3079–3096.
[25] G. Yüksel, Ş; Yüzbaşı, M. Sezer, A Chebyshev method for a class of high-order linear Fredholm integro-differential equations, J. Adv. Res. Appl. Math 4 (2012)
49–67.
[26] Y. Khan, K. Sayevand, M. Fardi, M. Ghasemi, A novel computing multi-parametric homotopy approach for system of linear and nonlinear Fredholm integral
equation, Appl. Math. Comput. 249 (2014) 229–236.
[27] Z. Mahmoodi, J. Rashidinia, E. Babolian, B-Spline collocation method for linear and nonlinear Fredholm and Volterra integro-differential equations, Appl.
Anal. 92 (9) (2012) 1787–1802.
[28] S. Luan, W. Yan, Q. Sun, Existence and uniqueness of solution for second order nonlinear integrodifferential equations of mixed type in Banach spaces, Int.
J. Nonlinear Sci 9 (2010) 55–60.
[29] H.L. Tidke, C.T. Aage, J.N. Salunke, Existence, Uniqueness of continuous solution of mixed type inregral equations in cone metric space, Kathmandu Univ. J.
Sci. Eng. Technol. 7 (2011) 48–55.
[30] M.B. Dhanke, H.L. Tidke, Existence and uniqueness of solutions of nonlinear mixed integrodifferential equations with nonlocal condition in Banach spaces,
Elect. J. Differ. Equ 2011 (31) (2011) 1–10.
[31] K. Balachandran, K. Uchıyama, Existence of solutions of nonlinear integrodifferential equations of Sobolev type with nonlocal condition in Banach spaces,
Proc. Indian Acad. Sci. (Math. Sci.) 110 (2) (2000) 225–232.
[32] T.E. Gilsdorf, M. Khavanin, Existence and uniqueness for nonlinear integro-differential equations in real locally complete spaces, Sci. Math. Jpn. 76 (3) (2013)
395–400.
[33] H.L. Tidke, R.T. More, Existence and uniqueness of solution of integrodifferential equation in cone metric space, SOP Trans. Appl. Math 2 (1) (2015) 13–19.
[34] M.I. Bhatti, P. Bracken, Solutions of differential equations in a Bernstein polynomial basis, J. Comput. Appl. Math. 205 (2007) 272–280.
[35] E.W. Cheney, Introduction to Approximation Theory, Second Edition, AMS Chelsea Publishing, 1982.
[36] O.R. Işık, Z. Güney, M. Sezer, Bernstein series solutions of pantograph equations using polynomial interpolation, J. Differ. Equ. Appl. 18 (2012) 357–374.
[37] O.R. Işık, M. Sezer, Z. Güney, Bernstein series solution of a class of linear integro-differential equations with weakly singular kernel, Appl. Math. Comput.
217 (2011) 7009–7020.
[38] Ş. Yüzbaşı, Numerical solutions of fractional Riccati type differential equations by means of the Bernstein polynomials, Appl. Math. Comput. 219 (2013)
6328–6343.
[39] O.R. Işık, M. Sezer, Z. Güney, A rational approximation based on Bernstein polynomials for high order initial and boundary values problems, Appl. Math.
Comput. 217 (2011) 9438–9450.
[40] G.M. Phillips, Interpolation and Approximation by Polynomials, Springer, New York, 2003.
[41] E. Babolian, Z. Masouri, S. Hatamzadeh-Varmazyar, New direct method to solve nonlinear Volterra-Fredholm integral and integro differential equation using
operational matrix with Block-pulse functions, Prog. Electromagn. Res. B 8 (2008) 59–76.
[42] K. Maleknejad, M. Hadizadeh, The numerical analysis of Adomian’s decomposition method for nonlinear Volterra integral and integro-differential equations,
International Journal of Engineering Science, Iran University of Science & Technology 8 (2a) (1997) 33–48.

You might also like