1

Numerical Solutions of Fredholm Integral Equation of Second Kind Using

Piecewise Bernoulli Polynomials

Afroza Shirin
1
, Md. Shafiqul Islam*
2

1
Institute of Natural Sciences, United International University, Dhaka-1209, Bangladesh.
Email: [email protected].
2
Department of Mathematics, University of Dhaka, Dhaka 1000, Bangladesh
*Corresponding author, Email: [email protected]


Abstract:
The aim of this paper is to solve the integral equations numerically using piecewise Bernoulli
polynomials. The Bernoulli polynomials are derived explicitly over the unit interval. A
matrix formulation for a non-singular linear Fredholm integral equation of the second kind is
derived by the technique of Galerkin method. In the Galerkin method, the Bernoulli
polynomials are exploited as the linear combination in the approximation as basis functions.
Numerical examples are considered to verify the effectiveness of the proposed derivations.

Keywords: Fredholm integral equation, Galerkin method, Bernoulli polynomials,
Numerical solutions.

I. Introduction
In the survey of solutions of integral equations, a large number of analytical but a few
approximate methods are available for solving numerically various classes of integral
equations [1, 2, 7, 8 ]. Since the piecewise polynomials are differentiable and integrable, the
Bernstein polynomials [5 8] have been used for solving differential and integral equations
numerically. Recently, integral equations have been solved by the well known variational
iteration method [9]. In the literature [7], Mandal and Bhattacharya have attempted to solve
integral equations numerically using Bernstein polynomials, but they obtained the results in
terms of finite series solutions In contrast to this, we solve the linear Fredholm integral
equation of the second kind by exploiting very well known Galerkin method, and Bernoulli
polynomials [4] are used as trial functions. For this, we give a short introduction of Bernoulli
polynomials first. Then we derive a matrix formulation by the technique of Galerkin method.
To verify our formulation we consider three examples, in which we obtain exact solutions for
two examples even using a few and lower order polynomials. On the other hand, the last
example shows an excellent agreement of accuracy compared to exact solution, which
confirms the convergence. All the computations are performed using MATHEMATICA.

2
II. Bernoulli Polynomials: The Bernoulli polynomials [Atkinson, 4] upto degree n can
be defined over the interval ] 1 , 0 [ implicitly by
k n
k
n
k
n
x b
k
n
x B

|
|
.
|

\
|
=
0
) (
where
k
b are Bernoulli numbers given by
1
0
= b and
}
=
1
0
) ( dx x B b
k k
1 > k .
These Bernoulli polynomials may be defined explicitly as

= = = =
>
|
|
.
|

\
|

+
+
|
|
.
|

\
|

+
=
=
m
n
n
k
m k
n
k
m k
m
n
m
m k
k
n
n
k x
k
n
n
x B
x B
0 0 0 0
0
1 , ) 1 (
1
1
) ( ) 1 (
1
1
) (
1 ) (
(1)

The first 6 Bernoulli polynomials ) 5 ( = n are given bellow for using in this paper:
5
4 3
5
3
2
3 1
4 3 2
4
2
2 0
2
5
3
5
6
) (
2
3
2
) ( ) (
2 ) ( ) ( 1 ) (
x
x x x
x B x
x x
x B x x B
x x x x B x x x B x B
+ + = + = =
+ = + = =

Note that Bernoulli polynomials have a special property at 0 = x and 1 = x , respectively,
, 0 ) 0 ( =
n
Br 1 > n and , 0 ) 1 ( =
n
B . 2 > n

III. Formulation of Integral Equation in Matrix Form
Consider a general linear Fredholm integral equation (FIE) of second kind [1, 2] is given by
) ( ) ( ) , ( ) ( ) ( x f dt t x t k x x a
b
a
= +
}
| | , b x a s s (2)
where ) (x a and ) (x f are given functions, ) , ( x t k is the kernel, and ) (x | is the unknown
function or exact solution of (2), which is to be determined.
Now we use the technique of Galerkin method [Lewis, 3] to find an approximate solution
) (
~
x | of (2). For this, we assume that

=
=
n
i
i i
x B a x
0
) ( ) (
~
| (3)
where ) (x B
i
are Bernoulli polynomials (basis) of degree i defined in eqn. (1), and
i
a are
unknown parameters, to be determined. Substituting (3) into (2), we obtain

3
) ( ) ( ) , ( ) ( ) (
0 0
x f dt t B a x t k x B a x a
n
i
b
a
n
i
i i i i
=
(

= =

or, | | ) ( ) ( ) , ( ) ( ) (
0 0
x f dt t B x t k a x B x a a
n
i
n
i
b
a
i i i i
= +

}
= =

or, ) ( ) ( ) , ( ) ( ) (
0
x f dt t B x t k x B x a a
b
a
i i
n
i
i
=
(

+
}

=
(4)
Then the Galerkin equations [Lewis, 3] are obtained by multiplying both sides of (3) by
) (x B
j
and then integrating with respect to x from a to b, we have
} } }

=
(
(

+
=
b
a
j
b
a
j
b
a
i i
n
i
i
dx x f x B dx x B dt t B x t k x B x a a ) ( ) ( ) ( ) ( ) , ( ) ( ) (
0

or,
, ) ( ) ( ) ( ) ( ) , ( ) ( ) (
0

} } }
=
=
(
(

+
n
i
b
a
j j
b
a
b
a
i i i
dx x f x B dx x B dt t B x t k x B x a a n j , , 1 , 0 = (5a)

In each equation, there are three integrals. The inner integrand of the left side is a function of
x and t and is integrated with respect to t from a to b. As a result the outer integrand
becomes a function of x only and integration with respect to x yields a constant. Thus for
each j ( n j , , 1 , 0 = ) we have a linear equation with 1 + n unknowns
i
a ( n i , , 1 , 0 = ) .
Finally (5a) represents the system of 1 + n linear equations in 1 + n unknowns.
Equivalently,

=
=
n
i
j j i i
F C a
0
,
, , , , 2 , 1 , 0 n j = (5b)

where
} }
(

+ =
b
a
j
b
a
i i j i
dx x B dt t B x t k x B x a C ) ( ) ( ) , ( ) ( ) (
,
. , , 2 , 1 , 0 , n j i = (5c)
, ) ( ) (
}
=
b
a
i j
dx x f x B F n j , , 2 , 1 , 0 = (5d)
Now the unknown parameters
i
a are determined by solving the system of equations (5), and
substituting these values of parameters in (3), we get the approximate solution ) (
~
x | of the
integral equation (2). The absolute error E for this formulation is defined by

) (
) (
~
) (
x
x x
E

= .


4
IV. Numerical Examples
In this section, we explain three integral equations which are available in the existing
literatures [2, 7]. For each example we find the approximate solutions using Bernoulli
polynomials.

Example 1: We consider the FIE of 2nd kind given by [7]
, 1 ) ( ) ( ) (
1
1
2 2
= +
}

dt t t x xt x | | , 1 1 s s x (6)
having the exact solution .
9
10
1 ) (
2
x x + = |
Using the formulation described in the previous section, the equations (5) lead us,
respectively,
} } }

(

+ =
1
1
1
1
2 2
1
1
,
) ( ) ( ) ( ) ( ) ( dx x B dt t B t x xt dx x B x B C
j i j i j i
, n j i , , 2 , 1 , 0 , = (7a)
}

=
1
1
) ( dx x B F
j j
, n j , , 2 , 1 , 0 = (7b)

Solving the system (7) for , 3 = n the values of the parameters are:
0 ,
9
10
,
9
10
, 1
3 2 1 0
= = = = a a a a
and the approximate solution is
2
9
10
1 ) (
~
x x + = |
which is the exact solution.

Example 2: Now we consider another FIE of 2nd kind given by [Mandal, 7]
x dt t t x x =
}

1
1
4 4
) ( ) ( ) ( | | , 1 1 s s x (8)
having the exact solution x x = ) ( |
Proceeding as the example 1, the system of equations becomes as
where,
} } }

(

=
1
1
1
1
4 4
1
1
,
) ( ) ( ) ( ) ( ) ( dx x B dt t B t x dx x B x B C
j i j i j i
, , , 2 , 1 , 0 , n j i = (9a)

5
}

=
1
1
) ( dx x B x F
j j
, , , 2 , 1 , 0 n j = (9b)
For , 3 = n solving system (9), the values of the parameters (
i
a ) are:
0 , 0 , 1 , 0
3 2 1 0
= = = = a a a a ,
and the approximate solution is x x = ) (
~
| which is the exact solution.
Example 3: Consider another FIE of 2nd kind given by [pp 213 (1), with 1 = , Jerry, 1]
x dt t xt tx x = +
}
1
0
2 2
) ( ) ( ) ( | | , 1 0 s s x (10)
having the exact solution
2
119
80
119
180
) ( x x x + = |
Proceeding as the previous examples, the system of equations becomes as

=
=
n
i
j j i i
F C a
0
,
, , , , 2 , 1 , 0 n j = (11a)
where,
} } }
(

+ =
1
0
1
0
2 2
1
0
,
) ( ) ( ) ( ) ( ) ( dx x B dt t B xt tx dx x B x B C
j i j i j i
, , , 2 , 1 , 0 , n j i = (11b)
}
=
1
0
) ( dx x B x F
j j
, , , 2 , 1 , 0 n j = (11c)
For , 3 = n solving system (11), the values of the parameters (
i
a ) are:
0 ,
119
80
,
119
260
, 0
3 2 1 0
= = = = a a a a ,
and the approximate solution is
2
119
80
119
180
) (
~
x x x + = |
which is the exact solution.

Example 4: Consider another FIE of 2nd kind given by [pp 124 (iv), with 1 = , Shanti, 2]
, ) ( 2 ) (
1
0
x t x
e dt t e e x =
}
| | , 1 0 s s x (12)
having the exact solution
2
2
) (
e
e
x
x

= | .
Since the equations (5b) and (5c) are of the form

6
} } }
(

=
1
0
1
0
1
0
,
) ( ) ( 2 ) ( ) ( dx x B dt t B e e dx x B x B C
j i
t x
j i j i
, n j i , , 2 , 1 , 0 , = (13a)
}
=
1
0
) ( dx x B e F
j
x
j
, , , , 2 , 1 , 0 n j = (13b)
Solving the system (5a) using (13a) and (13b) instead of (5b) and (5c) respectively, we have
the following results:

For 3 = n , the approximate solution is
3 2
051702 . 0 078167 . 0 188957 . 0 185387 . 0 ) (
~
x x x x = |

For 4 = n , the approximate solution is
4 3 2
012893 . 0 025916 . 0 0947437 . 0 185273 . 0 185571 . 0 ) (
~
x x x x x = |

For 5 = n , the approximate solution is
5
4 3 2
00257408 . 0
00645779 . 0 0316362 . 0 0925986 . 0 18558 . 0 185561 . 0 ) (
~
x
x x x x x

= |


For 6 = n , the approximate solution is
6 5
4 3 2
000427923 . 0 0012903 . 0
00791665 . 0 0308581 . 0 0925932 . 0 18556 . 0 185561 . 0 ) (
~
x x
x x x x x

= |

Plot of relative error, E is depicted in Fig. for various values of n.
Now the approximate solutions, exact solutions, and the relative error E at various points of
the domain are displayed in Table 1. The accuracy is obvious.

0.2 0.4 0.6 0.8 1
0.0002
0.0004
0.0006
0.0008

0.2 0.4 0.6 0.8 1
5 10
-6
0.00001
0.000015
0.00002
0.000025
0.00003
0.000035


Fig.2b. Relative error E using 5 polynomials
E
E
x
8.010
-4
6.010
-4

4.010
-4
2.010
-4

3.510
-5

3.010
-5

2.510
-5

2.010
-5

1.510
-5

1.010
-5
5.010
-6

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
x
Fig.2a. Relative error E using 4 polynomials

7


0.2 0.4 0.6 0.8 1
2 10
-7
4 10
-7
6 10
-7
8 10
-7
1 10
-6
1.2 10
-6
1.4 10
-6

0.2 0.4 0.6 0.8 1
1 10
-8
2 10
-8
3 10
-8
4 10
-8
5 10
-8








E
x
x
1.410
-6

1.210
-6

1.010
-6
8.010
-7

6.010
-7

4.010
-7
2.010
-7

E
510
-8

410
-8
310
-8

210
-8

110
-8

Fig.2c. Relative error E using 6 polynomials
0.2 0.4 0.6 0.8 1
0.2 0.4 0.6 0.8 1
Fig.2d. Relative error E using 7 polynomials

8
Table-1: Numerical solutions at various points and corresponding absolute errors of the
example 4.

x
Exact
Solutions
Approximate
Solutions
Absolute
Relative Error, E
Approximate
Solutions
Absolute
Relative Error, E
Polynomials used 3 Polynomials used 4
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
-0.1855612526
-0.2050768999
-0.2266450257
-0.2504814912
-0.2768248595
-0.3059387842
-0.3381146470
-0.3736744748
-0.4129741624
-0.4564070342
-0.5044077810
-0.1853868426
-0.2051159200
-0.2267185494
-0.2505049431
-0.2767853131
-0.3058698717
-0.3380688310
-0.3736924032
-0.4130508005
-0.4564542350
-0.5042129189
0.000940
0.000190
0.000324
0.000094
0.000143
0.000225
0.000136
0.000048
0.000186
0.000103
0.000386
-0.1855710208
-0.2050729963
-0.2266433924
-0.2504841199
-0.2768280330
-0.3059389289
-0.3381115484
-0.3736715751
-0.4129756359
-0.4564113012
-0.5043970842
5.26416910
-5
1.90345010
-5

7.20656310
-6
1.04947110
-5
1.14636310
-5

4.73228710
-7

9.16428710
-6
7.76009010
-6

3.56811510
-6

9.34907510
-6
2.12066410
-5


Polynomials used 5 Polynomials used 6
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
-0.1855612526
-0.2050768999
-0.2266450257
-0.2504814912
-0.2768248595
-0.3059387842
-0.3381146470
-0.3736744748
-0.4129741624
-0.4564070342
-0.5044077810
-0.1855610006
-0.2050770088
-0.2266449063
-0.2504813833
-0.2768249425
-0.3059389458
-0.3381146594
-0.3736743009
-0.4129740843
-0.4564072670
-0.5044071950
2.40558710
-6

8.64986010
-7

3.27360610
-7

5.42075410
-7

1.16277810
-7
4.73125910
-7

1.25952210
-7

3.62080610
-7
2.07428510
-7

4.07511810
-7

9.56024510
-7

-0.1855612694
-0.2050768958
-0.2266450312
-0.2504814909
-0.2768248544
-0.3059387842
-0.3381146522
-0.3736744750
-0.4129741564
-0.4564070387
-0.5044077618
9.04919810
-8
1.99028710
-8
2.42540910
-8

9.60047310
-10

1.84625110
-8
7.75201410
-11

1.55402210
-8

5.63510410
-10

1.44192810
-8

9.99779310
-9

3.79778410
-8



V. Conclusion
We have solved numerically Fredholm integral equations of second kind. We have obtained
the approximate solution of the unknown function by the well known Galerkin method using
Bernoulli polynomials as trial functions. The authors concluding remark is that the
numerical solutions coincide with the exact solutions even a few of the polynomials are used
in the approximation.


9
References
1. Abdul J. Jerri, 1999 Introduction to Integral Equations with Applications, John Wiley &
Sons Inc.
2. Shanti Swarup, 2007 Integral Equations, Krishna Prakashan Media (P) Ltd, Fifteenth
Edition.
3. Lewis, P. E., J. P. Ward,1991 The Finite Element Method, Principles and Applications,
Addison-Wesley.
4. Kendal E Atkinson,1989 An Introduction to Numerical Analysis, John Wiley and Sons,
2nd edition, (1989), pp-284.
5. Reinkenhof, J., 1977 Differentiation and integration using Bernsteins polynomials, Int.
J. Numer. Methods Engrg, 11, 1627 1630.
6. Kreyszig, E., 1979 Bernstein polynomials and numerical integration, Int. J. Numer.
Methods Engrg, 14, 292 295.
7. Mandal, B. N., S. Bhattacharya, 2007 Numerical solution of some classes of integral
equations using Bernstein polynomials, Applied Mathematics and Computation, 190,
1707 1716.
8. Bhatti, M.I., P. Bracken, 2007 Solutions of differential equations in a Bernstein
polynomial basis, J. Computational and Applied Mathematics, 205, 272 280
9. Wang, S.Q., J.H. He, 2007 Variational method for solving integro-differential
equations, Physics Letters A, 367, 188 191.