Vous êtes sur la page 1sur 7

Continental J.

Applied Sciences 5:1 - 7, 2010


Wilolud Journals, 2010

ISSN: 1597 - 9928


http://www.wiloludjournal.com

A MAXIMAL ORDER ALGORITHM FOR SOLVING INITIAL VALUE PROBLEMS OF FIRST ORDER
ORDINARY DIFFERENTIAL EQUATIONS.
Oluwatusin, E. A.
Department of Mathematics, School of Sciences, College of Education, Ikere-Ekiti, Ekiti State, Nigeria,
eaoluwatusin@yahoo.com
ABSTRACT
A maximal order method with continuous variable coefficient is proposed for initial value
problems of first order ordinary differential equations. The main objective of thisstudy is to
develop some continuous multistep collocation methods with all collocation points taken at the
selected grid points. The step number k=4 is considered with order of accuracy equals to 2k
which is the degree of the basis function. Four Predictors n + i i = 1(1) k are also proposed.
The derived scheme belongs to the family of linear multistep methods where collocation and

interpolation of differential system and approximate solution was restricted to grid points

y n+i

i = 1(1)k and i = 1(1)k 1 respectively. Numerical experiment using the new scheme to

solve both linear and non-linear initial value problems reveal high level of consistency and
accuracy.
KEYWORDS and Phrases; Collocation, Interpolation, Grid point, Consistency, Step number,
Initial value problems
INTRODUCTION
In this paper, the numerical solution of the initial value problems in ordinary differential equations of the form

y / ( x) = f ( x, y ), y ( x0 ) = y 0

(1)

is considered where f and y are scalar or vector functions and

( x0 , y 0 )

is the given initial starting point.

The characteristic of discretization methods is that errors are generated when they are adopted for the solution of
Ordinary Differential Equations. Consequently, these errors have to be minimized to ensure the accuracy of the
method. A solution is said to be accurate when the computed value does not deviate significantly from the exact
solution as the iteration progress. The magnitude of these errors referred to as global error, e is in the form of

e = y ( xn ) y n
y( x )

n
and
where
truncating error.

(2)

y n are exact and computed solutions at x n respectively. This global error, e is a function of local
k

[ y ( x), h] = [a; y ( x n +i ) hiy / ( x n +i )]


L

i 0

(3)

And must be negligibly small for high accuracy. Therefore the main objective of this work is to ensure that the local
truncating error (3) is minimized hence the development of maximal order method per even step number k.
Many authors and researchers such as Fatunla (1988), Awoyemi (1992), Oluwatusin (2006), Lambert (1973, 1991),
Onumanyi, Awoyemi, Jator and Sivisena, (1994). Oladele (1991), Kockler (1994), Jacques and Judd (1987),
proposed numerical methods of order of accuracy less than 2k. Awoyemi (2001) points out that some of the
mathematical models resulting to non-linear Ordinary Differential Equations which might not have analytical

Oluwatusin, E. A: Continental J. Applied Sciences 5:1 - 7, 2010

solution could be solved by numerical methods of high order of accuracy. It was also discovered according to Ibijola
and Ogunrinde (2006) that the language of engineering is differential equations.
This implies that the proposed scheme in this paper can also be used to solve problems in engineering, where high
precise accuracy is a must.
DERIVATION OF MAXIMAL ORDER METHOD
The derivation of maximal order method is investigated for k=4.
Consider an approximate solution to problem (1) in the form
2k

y ( x) = a j x j
j =0

(4)

And its derivative is


2k

y / ( x) = ja j x j 1 = f ( x, y ),
j =0

(5)

By collocating the differential system (5) at all grid points and interpolating (4) at
algebraic equations is obtained

x = x n +i , i = 0(1)3

a system of

a 0 + a1 x n + a 2 x 2 n + a3 x 3 n + a 4 x 4 n + a5 x 5 n + a 6 x 6 n + a 7 x 7 n + a8 x 8 n = y n
a 0 + a1 x n +1 + a 2 x 2 n +1 + a3 x 3 n +1 + a 4 x 4 n +1 + a5 x 5 n +1 + a 6 x 6 n +1 + a 7 x 7 n +1 + a8 x 8 n +1 = y n +1
a 0 + a1 x n + 2 + a 2 x 2 n + 2 + a3 x 3 n + 2 + a 4 x 4 n +3 + a5 x 5 n + 2 + a 6 x 6 n + 2 + a 7 x 7 n + 2 + a8 x 8 n + 2 = y n + 2
a 0 + a1 x n +3 + a 2 x 2 n +3 + a3 x 3 n +3 + a 4 x 4 n +3 + a5 x 5 n +3 + a 6 x 6 n +3 + a 7 x 7 n +3 + a8 x 8 n +3 = y n +3
a1 + 2a 2 x n + 3a3 x 2 n + 4a 4 x 3 n + 5a5 x 4 n + 6a 6 x 5 n + 7 a 7 x 6 n + 8a8 x 7 n = f n
a1 + 2a 2 x n +1 + 3a3 x 2 n +1 + 4a 4 x 3 n +1 + 5a5 x 4 n +1 + 6a 6 x 5 n +1 + 7 a 7 x 6 n +1 + 8a8 x 7 n+1 = f n+1
a1 + 2a 2 x n + 2 + 3a3 x 2 n + 2 + 4a 4 x 3 n + 2 + 5a5 x 4 n + 2 + 6a 6 x 5 n + 2 + 7 a 7 x 6 n + 2 + 8a8 x 7 n + 2 = f n + 2
a1 + 2a 2 x n +3 + 3a3 x 2 n +3 + 4a 4 x 3 n +3 + 5a5 x 4 n +3 + 6a 6 x 5 n +3 + 7 a 7 x 6 n +3 + 8a8 x 7 n +3 = f n +3
a1 + 2a 2 x n + 4 + 3a3 x 2 n + 4 + 4a 4 x 3 n + 4 + 5a5 x 4 n + 4 + 6a 6 x 5 n + 4 + 7 a 7 x 6 n + 4 + 8a8 x 7 n + 4 = f n + 4 (6)
Using Gaussian elimination techniques, the values of

aj

s are given as follows

Oluwatusin, E. A: Continental J. Applied Sciences 5:1 - 7, 2010

h 8 1360
550
{
y n+3 + 450 y n+ 2 720 y n +1
y n + 3hf n + 4 152hf n+3 792hf n + 2
7200 3
3
552hf n+1 47hf n }

a8 =

h7
{(281y n +3 + 27 y n + 2 297 y n+1 11y n + 3hf n + 4 105hf n +3 369hf n + 2
1692
764
129hf n ) a8 (8 x n +
h)}
47
h6
a6 =
{(720 y n+1 720 y n 19hf n + 4 106 y n +3 + 264hf n + 2 646 f n+1 251hf n)
9720
986
7888
7514 2
a 7 (7 x n +
h) a8 (28 xn2 +
hxn +
h )}
81
81
81
h6
a5 =
{( f n + 4 4 f n+3 6 f n + 2 4 f n+1 + f n ) a6 (6 x n + 12h) a7 (21xn2 + 84hxn + 91h 2 )
120
a8 (56 xn3 + 336hxn2 + 728h 2 x n + 560h 3 )}
a7 =

h4
15
75
{( f n+3 3 f n+ 2 f n ) a5 (5 x n + h) a6 (15 xn2 + 45h 2 xn + h 2 )
24
2
2
315
525
315
a7 (35 xn3 +
hx 2 n +
h 2 xn +
h 3 ) a8 (701x n4 + 420hxn3 + 1050h 2 xn2 + 1260h 3 x n + 602h 4 )
2
2
2
h3
35
a3 = {( f n+ 2 2 f n +1 + f n ) a 4 (4 xn + 4h) a5 (10 xn2 + 20hxn + h 2 )
6
3
217 4
a6 (20 x n3 + 60hxn2 + 30h 3 ) a 7 (35 x n4 + 140hxn3 + 25h 2 xn2 + 210h 3 xn +
h )
3
1960 2 3
1736 4
a8 (56 x 5 n + 280hxn4 +
h xn + 840h 3 xn2 +
h xn + 168h 5 )}
3
3
h2
a 2 = {( f n +1 f n ) a3 (3x n + 3h) a 4 (6 x n2 + 6hx n + 2h 2 )
2
5
a5 (10 x n3 + 15hx n2 + 10h 2 x n + h 3 ) a6 (15 x n4 + 30hx n3 + 30h 2 x n2 + 15h 3 x n + 3h 4 )
2
105 4
105 3 2
7
a7 ( 21x n5 +
hx n + 70h 2 x n3 +
h x n + 21h 4 x n + h 5 )
2
2
2
6
5
2 4
3 3
4 2
a8 ( 28 x n + 84hx n + 140h x n + 140h x n + 84h x n + 28h 5 x n + 4h 6 )}
a4 =

a1 = f n 2a n x n 3a3 xa 2 n + 4a 4 xa 3 n + 5a5 xa 4 n + 6a 6 xa 5 n + 7 a7 xa 6 n + 8a8 xa 7 n


a 0 = y n a1 x n a 2 x 2 n a 3 x 3 n a 4 x 4 n a 5 x 5 n a 6 x 6 n a 7 x 7 n a8 x 8 n

(7)

Oluwatusin, E. A: Continental J. Applied Sciences 5:1 - 7, 2010

aj

Substitute the values


in the form
k

j =0

j =0

s into the expression (4) and after algebraic manipulation a continuous scheme is obtained

j ( x) y n + j = j ( x) f n + j
Where

and

f n+ j = f ( xn+ j , y n+ j )

t = ( x x n+3 ) / h,

(8)
by taking
(9)

are determined as follows

1
(540 3141t 2 3967t 3 488t 4 + 1886t 5 + 1327t 6 + 3538t 7 + 34t 8 )
540
1
2 = (36t 2 12t 3 47t 4 + 4t 5 + 18t 6 + 8t 7 + t 8 )
16
1
1 = (63t 2 + 141t 3 + 64t 4 58t 5 61t 6 13t 7 t 8 )
20
1
(180t 2 + 452t 3 + 277t 4 148t 5 230t 6 88t 7 11t 8 )
0 =
270
h
(36t 2 + 132t 3 + 193t 4 + 144t 5 + 58t 6 + 12t 7 + t 8 )
4=
2400
h
(900t + 2616t 2 + 2317t 3 67t 4 1286t 5 802t 6 703t 7 19t 8 )
3 =
900
h
(504t 2 + 948t 3 + 302t 4 409t 5 363t 6 107t 7 11t 8 )
2 =
120
h
(600t + 522t 2 + 123t 3 + 661t 4 426t 5 584t 6 201t 7 23t 8 )
1 =
300
h
0 =
(7200t + 708t 2 + 1796t 3 + 1129t 4 586t 5 926t 6 364t 7 47t 8 )
7200

3 =

Put t=1 implies that

(10)

x = x n + 4 and substitute the result in (8) yields a symmetric discrete scheme

5 y n + 4 + 33 y n +3 33 y n +1 5 y n =

6h
( f n + 4 + 16 f n + 3 + 36 f n + 2 + 16 f n +1 + f n )
5

(11)

3. Derivation of predictors for the proposed methods


To Implement the proposed methods, the predictors are required. Hence the derivations of the predictors are
considered in this section. Taking the approximate solution to problem (1) to be

Oluwatusin, E. A: Continental J. Applied Sciences 5:1 - 7, 2010

y ( x) =

2 k 1

a
j =0

xj
(12)

The first derivative of (12) is given as

y / ( x) =

2 k 1

a
j =0

By collocating (13) at

x j 1
(13)

x = x n + j , j = 0(1) k 1

system of non linear equation. By solving for


scheme for k=4 is derived in the form

k 1

k 1

j =0

j =0

and interpolating (12) at

x = x n + j , j = 0(1) k 1

s and substituting is values into the equation 12 then a continuous

y (t ) = j (t ) y n + j + j (t ) f n + j
(14)

Where

1
(36 247t 2 217t 3 283t 4 87t 5 10t 6
36
1
2 (t ) = (12t 2 + 4t 3 9t 4 6t 5 t 6 )
4
1
1 (t ) = (15t 2 + 41t 3 + 39t 4 + 15t 5 + 2t 6 )
4
1
0 (t ) = (4t 2 + 12t 3 + 13t 4 + 6t 5 + t 6 )
36
h
3 (t ) = (12t + 41t 2 + 51t 3 + 31t 4 + 9t 5 + t 6 )
12
h
2 (t ) = (12t 2 + 28t 3 + 23t 4 + 8t 5 + t 6 )
2
h
1 (t ) = (6t 2 + 17t 3 + 17t 4 + 7t 5 + t 6 )
2

3 (t ) =

Evaluating (14) at t=1, that is,

x = x n + 4 , yields a symmetric discrete explicit scheme

y n+ 4 = 28 y n+3 + 28 y n +1 + y n + 12h( f n +3 + 3 f n+ 2 + f n +1 )
Similarly, the predictors for

to obtain a

(16)

y n +3 and y n + 2 are derived to be

y n+3 = 3 y n+ 2 + 9 y n +1 11y n + 2h( f n + 2 5 f n +1 2 f n )


and

(17)

Oluwatusin, E. A: Continental J. Applied Sciences 5:1 - 7, 2010

y n+ 2 = 4 y n+1 + 5 y n + 2h(2 f n +1 + f n )
By adopting Taylor series expansion for

y n+1 = y n + hf n +

y n +1

about

(18)

( xn , yn )

yields

h2
{f / x + f n (f / y)}
2

(19)

4.
Numerical Experiment
The efficiency and accuracy of the method was tested on two sample problems. One linear and one non-linear
Ordinary Differential Equation were solved using the Fortran Computer Program developed for the new methods.

Problem (i)

y / = e10( x y )

y (0) = 0.1

y=
Theoretical solution
Problem (2)

1
ln(e10 x + e 1)
10

y / = (1 x) y 2 y, y (0) = 1
Theoretical solution y = 1 + 1 /(1 + 10 x )

TABLE (1): Results for problem (1) using scheme (11)


X
y( x )

yn

0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0

0.1489880126D+01
0.2209080454D+01
0.3082085127D+01
0.4030986355D+01
0.5011511183D+01
0.6029145431D+01
0.7026527018D+01
0.8025562030D+01
0.9025206796D+01
0.1002507608D+01

0.1489880126D+01
0.2209080454D+01
0.3082085127D+01
0.4030986355D+01
0.5011511183D+01
0.6029145431D+01
0.7026527018D+01
0.8025562030D+01
0.9025206796D+01
0.1002507628D+01

0.7105427358D-14
0.1065814104D-13
0.1465494393D-13
0.1953992523D-13
0.1065814104D-13
0.1154631946D-13
0.1332267630D-13
0.1421085472D-13
0.1776356839D-13
0.1598721155D-13

yn

TABLE (2): Results for problem (2) using scheme (11)


X
y( x )
n

0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0

0.8917402715D+00
0.7740988401D+00
0.6569789426D+00
0.5477828431D+00
0.4507993471D+00
0.3657297906D+00
0.2964696330D+00
0.2395003293D+00
0.1931444741D+00
0.1556760929D+00

0.8917402715D+00
0.7740988401D+00
0.6569789426D+00
0.5477828431D+00
0.4507993471D+00
0.3657297906D+00
0.2964696330D+00
0.2395003293D+00
0.1931444741D+00
0.1556760929D+00

0.3452793607D-13
0.3008704397D-13
0.2542410726D-13
0.2120525977D-13
0.6106226635D-13
0.4940492460D-13
0.4052314040D-13
0.3247402347D-13
0.2609024108D-13
0.2109423747D-13

Oluwatusin, E. A: Continental J. Applied Sciences 5:1 - 7, 2010

5.

CONCLUSION

Maximal order algorithm for Ordinary Differential Equations have been developed, programmed and tested in this
paper. Collocation and interpolation is restricted to grid points only at
respectively.

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

and

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

This new scheme with step number k=4 has order of accuracy p=8. Four predictions were also derived alongside
with the main scheme.
Tables 1 and 2 show that the scheme is highly accurate with small truncation error. It should be noted that the
interval of absolute stability of this scheme is at the origin which is in line with Simpson method with step number
k=2 whose interval of absolute is also at the origin. This maximal scheme of order p=4 can be obtained from (4)
with k=2, also higher methods can be similarly derived for k 5.
REFERENCES:
Awoyemi D. O. (1992): On some continuous linear multistep methods for initial value problems. Ph. D Thesis
(Unpublished). University of Ilorin Nigeria.
Awoyemi D. O. (2001) A family of continuous Linear Multistep Hybrid Methods for a system of first order ordinary
differential equation. NJTE 18, 124-131
Fatunla S. O. (1988): Numerical Methods for 1 VPs in ordinary differential equations Academic press Inc. Harcourt
Brace Jovanorich Publishers, New York.
Ibijola E. A. and Ogunride R. B. (2006). The role of Mathematics in a modern society: International Journal of
Numerical Mathematics 1, 414-426
Jacues I and Judd C. J. (1887): Numerical Analysis: Chapman and Hall, N. York.
Lambert J. D. (1973): Computational methods in ordinary differential systems of initial value problems. John Wiley
and Sons, New York.
Lambert J. D. (1991): Numerical methods for ordinary differential systems of initial value problems. John Wiley and
Sons, New York.
Oladele J. O. (1991): Some New collocation formulae for the continuous numerical solution of initial value problem.
MSc. Degree Dissertation (Unpublished). University of Ilorin
Oluwatusin E. A. (2006): Derivation of collocation methods for initial value problems of first order ordinary
differential equations. Journal of Mathematics Education 1,100-115
Onumanyi P, Awoyemi D. O., Jator J. N. and Sivisena, U. W. (1994). New Linear Multistep methods with
continuous coefficient for first Difference Approximation for solving differential equations, Comp. Maths. 72, 15-27
Received for Publication: 03/02/10
Accepted for Publication: 10/03/10