Numerical Solution of Ordinary Differential Equations

CHAPTER

10

UMERICAL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

Chapter Objectives

Introduction

Picard’s method

Taylor’s series method

Euler’s method

Modified Euler’s method

Runge’s method

Runge-Kutta method

Predictor-corrector methods.

Milne’s method

Adams-Bashforth method

Simultaneous first order differential equations

Second order differential equations.

Error analysis

Convergence of a method

Stability analysis

Boundary-value problems
Finite-difference method

Shooting method

Objective type of questions

420

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

10.1 Introduction

A number of problems in science and technology can be formulated

into differential equations. The analytical methods of solving differential

equations are applicable only to a limited class of equations. Quite often

differential equations appearing in physical problems do not belong to any

of these familiar types and one is obliged to resort to numerical methods.

These methods are of even greater importance when we realize that com-

puting machines are now readily available which reduce numerical work

considerably.

Solution of a differential equation.

The solution of an ordinary differen-

tial equation means finding an explicit expression for

in terms of a finite

number of elementary functions of

. Such a solution of a differential equa-

tion is known as the

closed

or

finite form of solution

. In the absence of such

a solution, we have recourse to numerical methods of solution.

Let us consider the first order differential equation

 dy

dx

), given

(1)

to study the various numerical methods of solving such equations. In most

of these methods, we replace the differential equation by a difference equa-

tion and then solve it. These methods yield solutions

either

as a power se-

ries in
x

from which the values of

can be found by direct substitution,

or

a set of values of

and

. The methods of Picard and Taylor series belong

to the former class of solutions. In these methods,

in (1) is approximated

by a truncated series, each term of which is a function of

The information

about the curve at one point is utilized and the solution is not iterated. As

such, these are referred to as

single-step methods.

The methods of Euler, Runge-Kutta, Milne, Adams-Bashforth, etc. be-

long to the latter class of solutions.

In these methods, the next point on the

curve is evaluated in short steps ahead, by performing iterations until suf-

ficient accuracy is achieved. As such, these methods are called

step-by-step

methods

Euler and Runga-Kutta methods are used for computing

over a lim-

ited range of

- values whereas Milne and Adams methods may be applied

for finding

over a wider range of

-values. Therefore Milne and Adams

methods require starting values which are found by Picard™s Taylor series

or Runge-Kutta methods.

UMERIC AL

OLUTION

OF

O
RDINARY

IFFERENTIAL

QUATIONS

421

Initial and boundary conditions.

An ordinary differential equation

of the

th order is of the form

22

,,/,/ ,)

(/0

nn

Fxydy dxdydx dydx

(2)

Its general solution contains

arbitrary constants and is of the form

12

,,)

,,

(,0

xyc cc

(3)

To obtain its particular solution,

conditions must be given so that the

constants

can be determined.

If these conditions are prescribed at one point only

say:x

)
, then the dif-

ferential equation together with the conditions constitute an

initial value

problem

of the nth order.

If the conditions are prescribed at two or more points, then the problem

is termed as

boundary value problem.

In this chapter, we shall first describe methods for solving initial value

problems and then explain the

finite difference method

and

shooting

method

for solving boundary value problems.

10.2 Picard’s Method

Consider the first order equation

(,)

dy

fxy

dx

(1)

It is required to find that particular solution of (1) which assumes the

value

0
when

. Integrating (1) between limits, we get

(,) or (,)

dyfxydxyyfxydx

(2)

This is an integral equation equivalent to (1), for it contains the un-

known

under the integral sign.

As a first approximation y

to the solution

, we put

y
0

in

) and

integrate (2), giving

100

(, )

yyfxydx

For a second approximation y

, we put

in

(
x

) and integrate

(2), giving

yyfxydx

422

UMERIC AL

ETHODS

IN

NGINEERING
AND

CIENCE

Similarly,

a third approximation is

302

(, )

yyfxydx

Continuing this process, we obtain

where

0
01

(,)

nn

yyfxydx

Hence this method gives a sequence of approximations

each giving a better result than the preceding one.

Obs.

PicardÕs method is of considerable theoretical value, but

can be applied only to a limited class of equations in which the

successive integrations can be performed easily. The method can

be extended to simultaneous equations and equations of higher

order (See Sections 10.11 and 10.12).

EXAMPLE 10.1

Using Picard™s process of successive approximations, obtain a solution

up to the fifth approximation of the equation

dy

dx

, such that

1 when

0. Check your answer by finding the exact particular solution.

Solution:

) We have

1()

x
yxydx

First approximation

Put

1 in

, giving

1(1)1/2

yxdxxx

Second approximation.

Put

y
1

/2

in

giving

223

1(1/2)1/6

yxxdxxxx
Third approximation.

Put

/6 in

x,

giving

34

232
3

1(1/6)12

324

xx

yxxxdxxx

Fourth approximation

. Put

in

, giving

34

0
345

112

324

312120

xx

yxxdx

xxx

xx

NOTE

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

E
QUATIONS

423

Fifth approximation, Put

in

, giving

345

3456

112

312120

31260720
x

xxx

yxxdx

xxxx

xx

(1)

ii

) Given equation

dy

yx

dx

is a Leibnitzs linear in

Its, I.F. being

the solution is

xx

xxxx

yexedxc
xeedxcxeec

ycex

Since

1, when

0,

2.

Thus the desired particular solution is

21

yex

(2)

Or using the series:

234

2!3!4!

xxx

ex

We get

3456

31260360

xxxx

yxx

(3)

Comparing (1) and (3), it is clear that (1), approximates to the exact

particular solution (3) upto the term in

Obs.

At x

1,

the fourth approximation y

4

3.433

and the fifth

approximation y

3.434

whereas the exact value is

3.44.

EXAMPLE 10.2

Find the value of

for

0.1 by Picard™s method, given that

,(0) 1.

dyyx

dxyx

Solution:

We have
0

yx

ydx

yx

NOTE

424

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

First approximation

. Put
y

1 in the integrand, giving

00

111

12log(1)12log(1)

xx

yx

ydxdx

yxx

xxxx

Second approximation. Put

2 log(1 +
x

) in the integrand,

giving

12log(1)

12log(1)

11

12log(1)

xxx

ydx

xxx

dx

which is very difficult to integrate.

Hence we use the first approximation and taking

x

0.1 in (

) we obtain

(0.1)

1 Π(0.1)

2 log 1.1

0.9828.

10.3 Taylor’s Series Method

Consider the first order equation

(,)

dy

fxy

dx

(1)

Differentiating (1), we have

dydy

ff

xydx

dx

i.e.
xy

yf ff

(2)

Differentiating this successively, we can get

iv

yy

etc. Putting

and

0, the

Values of

000

(),( ),( )

yy y

can be obtained.

Hence the Taylor™s series

23
00

00000

()()

()() () ()...

2!3!

xxxx

yyxxyyy

(3)

gives the values of

for every value of

for which (3) converges.

On finding the value

for

from (3),

,
y

etc. can be evaluated

at

by means of (1), (2) etc. Then

can be expanded about

. In

this way, the solution can be extended beyond the range of convergence of

series (3).

UMERIC AL

OLUTION

OF
O

RDINARY

IFFERENTIAL

QUATIONS

425

Obs.

This is a single step method and works well so long as

the successive derivatives can be calculated easily. If (x, y) is

somewhat complicated and the calculation of higher order

derivatives becomes tedious, then TaylorÕs method cannot

be used gainfully. This is the main drawback of this method

and therefore, has little application for computer programs.

However, it is useful for finding starting values for the

application of powerful methods like Runga-Kutta, Milne and

Adams- Bashforth which will be described in the subsequent

sections.

EXAMPLE 10.3

Solve

y, y

(0)

1 by Taylor™s series method. Hence find the val-

ues of y at

0.1 and

0.2.

Solution:

Differentiating successively, we get

x
y

(0)

1[

(0)

1]

(0)

(0)

2
y

(0)

2, etc.

Taylor™s series is

23

00

00000

()()

()()()()

2!3!

xxxx

yyxxyyy

Here

0,

y
0

234

()()

11(2)(2)(4)

23! 4!

xxx

yx

Thus

34

0.10.1

0.11 0.1 0.1

3!4!

1.1103

and

34

0.20.2
0.21 0.20.2

36

1.2427

NOTE

426

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

EXAMPLE 10.4

Find by Taylor™s series method, the values of

y
at

0.1 and

0.2 to

five places of decimals from

dy

dx

yÐ

1,

(0)

1.

Solution:

Differentiating successively, we get

2
y

Π1, (

Œ1[

(0)

1]

xy

, (

y
2

xy

, (

iv

xy

x
2

, (

iv

Π6, etc.

Putting these values in the Taylor™s series, we have

234

34

()()

11(0)(2)(6)

23! 4!

34

xxx

yx

xx

Hence

y
(0.1)

0.90033 and

(0.21)

0.80227

EXAMPLE 10.5

Employ Taylor™s method to obtain approximate value of

at

0.2 for

the differential equation

dy

dx

,y

(0)

0. Compare the numerical

solution obtained with the exact solution.

Solution:

) We have

(0)

(0)

3.

Differentiating successively and substituting

x
 0,

0 we get

(0)

(0)

3
e

(0)

(0)

21

iv

iv

(0)

(0)
3

45 etc.

Putting these values in the Taylor™s series, we have

234

234

234

()(0)(0)(0)(0)(0)

2!3!4!

92145

03

2624

92115

268

iv

xxx

yxyxyy y y

xxxx

xxxx
N

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

427

Hence

234

(0.2)3(0.2)4.5(0.2)3.5(0.2) 1.875(0.2) 0.8110

b
) Now

23

dy

ye

dx

is a Leibnitz™s linear in

Its I.F. being

Œ2

, the solution is

Œ2 .2 Œ2

3Œ3 or Œ3

xxxxxx

yeeedxcecyece

Since

0 when

0,

c
3.

Thus the exact solution is

3(

When

0.2,

3(

0.4

0.2

0.8112 (

ii

)
Comparing (

) and (

ii

), it is clear that (

) approximates to the exact value

up to three decimal places

EXAMPLE 10.6

Solve by Taylor series method of third order the equation

32

dy xxy

dx

(0)

1 for

at

0.1,

0.2 and
x

0.3

Solution:

We have

32

;00

yxxyey

Differentiating successively and substituting

0,

1.

3222

3222Œ

Œ3.2 .

ŒŒ32;

()()()

() 01

x
x

yxxyexyxyye

xxyxyxyyey

3222Œ

22

2Œ

ŒŒ32 Œ

Œ3Œ.2 .62

()()

{()

[()]}

20Œ2

y xxyxy xyye

xyxyyxyy

yyxy yyey
Substituting these values in the Taylor™s series, we have

23

23

23

()(0)(0)(0)(0)

2!3!

1(0)(1)(2)

26

26

xx

yxyxyy y

xx

xx

428
•

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

Hence

23

23

23

11

0.11 0.1 0.1 1.005

23

11

0.210.2 0.2 1.017

23

11

0.310.3
()()

()()

() 0.3 1.036

23

()

EXAMPLE 10.7

Solve by Taylor™s series method the equation

log()

dy

xy

dx

for

(1.1)

and
y

(1.2), given

(1)

Solution:

We have

log

log

(1)

log 2

Differentiating w.r.t.,

and substituting

1,

2, we get
111

1log2

yyy

xy

22

111

111

11log2log2

224

yyyy

xy

Substituting these values in the Taylor™s series about

s

1, we have

23

'

11

()(1)1(1)(1)(1)

2!3!

11

2( 1)log2( 1)1log2

22

1111

(1)log2log2

6244

xx

yxyxyyy

xx

x
23

(0.1)1(0.1)111

(1.1)2 (0.1)log21log2log2log2

22 6244

2.036

23

(0.2)1(0.2)111

(1.2)2 (0.2)log21log2log2log2

22 6244

2.081

UMERIC AL

OLUTION

OF

RDINARY
D

IFFERENTIAL

QUATIONS

429

Exercises 10.1

1.

Using Picard™s method, solve

dy

dx

xy

with

0,

1 up to the

third approximation.
2.

Employ Picard ™s method to obtain, correct to four places of decimals

the, solution of the differential equation

dy

dx

for

0.4, given

that

0 when

0.

3.

Obtain Picard™s second approximate solution of the initial value problem

y
x

/(

1),

(0)

0.

4.

Find an approximate value of

when

0.1, if

dy

dx

and

y
1

at

0, using

) Picard™s method (

) Taylor™s series.

5.

Solve

given

(1)

0. Find

(1.1) and

(1.2) by Taylor™s meth-

od. Compare the result with its exact value.

6.

Using Taylor™s series method, compute

(0.2) to three places of deci-

mals from

12

dy

xy

dx

given that

(0)

0.

7.

Evaluate

(0.1) correct to six places of decimals by Taylor™s series

method if

) satisfies

xy
 1,

(0)

1.

8.

Solve

(0)

1 using Taylor™s series method and compute

(0.1) and

(0.2).

9.

Evaluate

(0.1) correct to four decimal places using Taylor™s series

methods if
dy

dx

(0)

1.

10.

Using Taylor series method, find

(0.1) correct to three decimal places

given that

dy

dx

y
2

(0)

10.4 Euler’s Method

Consider the equation

dy

dx

(1)

given that

.Its curve of solution through

0
,

)is shown dotted

in Figure.10.1. Now we have to find the ordinate of any other point

on

this curve.

LL

+ hx

hx

+ nh

y
0

True value of

Error

Approx. value of

1
FIGURE 10.1

Let us divide

LM

into

sub-intervals each of width

at

so that

is quite small

In the interval

LL

, we approximate the curve by the tangent at

. If the

ordinate through

1
 meets this tangent in

), then

LP

R
1

PR

tan

0000

()

dy

yhyhfxy

dx

Let

1
 be the curve of solution of (1) through

and let its tangent at

meet the ordinate through

in

). Then

2
y

hf

) (1)

Repeating this process

times, we finally reach on an approximation

MP

of

MQ

given by

Œ1 0 Œ1

()
1,

nnn

yyhfxnhy

This is

EulerÕs method

of finding an approximate solution of (1).

Obs.

In EulerÕs method, we approximate the curve of solution

by the tangent in each interval, i.e., by a sequence of short lines.

Unless h is small, the error is bound to be quite significant. This

sequence of lines may also deviate considerably from the curve

of solution. As such, the method is very slow and hence there is

a modification of this method which is given in the next section.

NOTE

UMERIC AL

OLUTION

OF
O

RDINARY

IFFERENTIAL

QUATIONS

431

EXAMPLE 10.8

Using Euler™s method, find an approximate value of

corresponding to

1, given that

dy

dx

and

1 when
x

0.

Solution:

We take

10 and

0.1 which is sufficiently small. The various cal-

culations are arranged as follows:

xy

dy

dx

Old

dy
/

dx

new y

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.00
1.10

1.22

1.36

1.53

1.72

1.94

2.19

2.48

2.81

3.18

1.00

1.20

1.42

1.66
1.93

2.22

2.54

2.89

3.29

3.71

1.00

0.1 (1.00)

1.10

1.10

0.1 (1.20)

1.22

1.22

0.1 (1.42)

1.36

1.36

0.1 (1.66)

1.53

1.53

0.1 (1.93)

1.72
1.72

0.1 (2.22)

1.94

1.94

0.1 (2.54)

2.19

2.19

0.1 (2.89)

2.48

2.48

0.1 (3.29)

2.81

2.81

0.1 (3.71)

3.18

Thus the required approximate value of

3.18.

Obs.

In Example 10.1(Obs.), we obtained the true values of y

from its exact solution to be 3.44 where as by EulerÕs method

3.18 and by PicardÕs method y

3.434. In the above

solution, had we chosen n

 20, the accuracy would have been

considerably increased but at the expense of double the labor of

computation. EulerÕs method is no doubt very simple but cannot

be considered as one of the best

EXAMPLE 10.9

Given

dyyx

dxyx

with initial condition

at

0; find

for

0
.

by Euler™s method.

Solution:

We divide the interval (0, 0.1) in to five steps,

i.e.,

we take

5 and

0.02. The various calculations are arranged as follows:

NOTE

432

UMERIC AL

ETHODS

IN

E
NGINEERING

AND

CIENCE

xydy

dx

Oldy

02

dy

dx

new y

0.00

0.02

0.04
0.06

0.08

0.10

1.0000

1.0200

1.0392

1.0577

1.0756

1.0928

1.0000

0.9615

0.926

0.893

0.862

1.0000

0.02(1.0000)

1.0200
1.0200

0.02(0.9615)

1.0392

1.0392

0.02(0.926)

1.0577

1.0577

0.02(0.893)

1.0756

1.0756

0.02(0.862)

1.0928

Hence the required approximate value of

1.0928.

10.5 Modified Euler’s Method

In Euler™s method, the curve of solution in the interval

LL

is approxi-

mated by the tangent at

(Figure 10.1) such that at

, we have
 y

hf

) (1)

Then the slope of the curve of solution through

[i.e.,

dy

dx

)
P

)]

is computed and the tangent at

to

is drawn meeting the ordinate

through

L
2

in

).

Now we find a better approximation

(1)

of

0
h

) by taking the slope

of the curve as the mean of the slopes of the tangents at

and

i.e.,

(1)

100001

[(,)(,)]

yyfxyfxhy

As the slope of the tangent at

1 is not known, we take

as found in (1)

by Euler™s method and insert it on R.H.S. of (2) to obtain the first modified

value
y

(1)

Again (2) is applied and we find a still better value

1(2)

corresponding

to

as

(2)(1)

100001

[(,)(,)]

yyfxyfxhy

We repeat this step, until two consecutive values of

agree. This is then

taken as the starting point for the next interval

L
2

Once

is obtained to a desired degree of accuracy,

corresponding to

is found from (1).

hf

y
1

and a better approximation

(1)

is obtained from (2)

(1)

210 10 2

[(,)(2,)]

yy fxhyfxhy

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL
E

QUATIONS

433

We repeat this step until

becomes stationary. Then we proceed to calcu-

late

as above and so on.

This is the

modified EulerÕs method

which gives great improvement in

accuracy over the original method.

EXAMPLE 10.10

Using modified Euler™s method, find an approximate value of

when

0
.

3, given that

dy

dx

and

when

0.

Solution:

The various calculations are arranged as follows taking

0.1:

Mean slope
Old y

mean slope

new y

0.0

1.00

0.1 (1.00)

1.10

0.1

0.1

1.1

()

11.2

1.00
0.1 (1.1)

1.11

0.1

0.1

1.11

11

()

.21

1.00

0.1 (1.105)

1.1105

0.1

0.1

1.1105

11.

()

2105

1.00

0.1 (1.1052)

1.1105

Since the last two values are equal, we take

y
(0.1)

1.1105.

0.1 1.2105Š

1.1105

0.1 (1.2105)

1.2316

0.2

0.2

1.2316

1.121051.4

()

316

1.1105

0.1 (1.3211)

1.2426

0.2

0.2

1.2426

1.21051.4

()

426

1.1105
0.1 (1.3266)

1.2432

0.2

0.2

1.2432

1.21051.4

()

432

1.1105

0.1 (1.3268)

1.2432

Since the last two values are equal, we take

(0.2)

1.2432.

0.2 1.4432Š

1.2432

0.1 (1.4432)

1.3875

0.3

0.3

1.3875

2
1.44321.6

()

875

1.2432

0.1 (1.5654)

1.3997

0.3

0.3

1.3997

1.44321.6

()

997

1.2432

0.1 (1.5715)

1.4003

0.3

0.3

1.4003

1.44321.7

()

003

1.2432
0.1 (1.5718)

1.4004

0.3

0.3

1.4004

1.4432 1.7

()

004

1.2432

0.1 (1.5718)

1.4004

Since the last two values are equal, we take

(0.3)

1.4004.

434

UMERIC AL

ETHODS
IN

NGINEERING

AND

CIENCE

Hence

(0.3)

1.4004 approximately.

Obs.

In Example 10.8, we obtained the approximate value of y

for x

0.3 to be 1.53 whereas by the modified EulerÕs method

the corresponding value is 1.4003 which is nearer its true

value 1.3997, obtained from its exact solution y

2ex Ð x Ð 1 by

putting x

0.3.

EXAMPLE 10.11

Using the modified Euler™s method, find

(0
.

2) and

(0

4)

given

,y

(0)

Solution:

We have

y
ex

);

0,

0 and

0.2

The various calculations are arranged as under:

To calculate

(0

2)

ex

y
Mean slope

Old y

Mean slope

new y

0.01 Š

0.2 (1)

0.2

0.2

0.2

0.2

1.4214

1 1.42141.

)07

(21

0.2 (1.2107)
0.2421

0.2

0.2421

0.2

1.4635

1 1.46351.

)17

(23

0.2 (1.2317)

0.2463

0.2

0.2463

0.2

1.4677

11.467

()
2

71.2338

0.2 (1.2338)

0.2468

0.2

0.2468

0.2

1.4682

1 1.468

()1

2.2341

0.2 (1.2341)

0.2468

Since the last two values of

are equal, we take

(0.2)

0.2468.
To calculate y

(0

4)

ex

Mean slope

Oldy

mean slope

new y

0.2

0.2468

0.2

1.4682
Š

0.2468

0.2 (1.4682)

0.5404

0.4

0.5404

0.4

2.0322

1.46822.0322)

1.7502

0.2468

0.2 (1.7502)

0.5968

NOTE

UMERIC AL

OLUTION
OF

RDINARY

IFFERENTIAL

QUATIONS

435

ex

Mean slope

Oldy

mean slope

new y

0.4
0.5968

0.4

2.0887

1.46822.0

()

887

1.7784

0.2468

0.2 (1.7784)

0.6025

0.4

0.6025

0.4

2.0943

(1.46822.09)

= 1.7 5
43

812

0.2468

0.2 (1.78125)

0.6030

0.4

0.6030

0.4

2.0949

1.46822.0

= 1.

949

781

0.2468

0.2 (1.7815)

0.6031

0.4

0.6031
e

0.4

2.0949

1.46822.0

()

= 1.

49

781

0.2468

0.2 (1.7815)

0.6031

Since the last two value of

are equal, we take

(0.4)

0.6031

Hence

(0.2)
 0.2468 an d

(0.4)

0.6031 approximately.

EXAMPLE 10.12

Solve the following by Euler™s modified method:

log()()

,02

dy

xyy

dx

at

and

4
with

2.

Solution:

The various calculations are arranged as follows:

log

Mean slope

Old y

(
mean slope

new y

0.0

0.2

0.2

log (0

2)

log (0.2

2.0602)

log (0.2

2.0655)

0.310

()

0.3541

0.301

()

0.3552

2
0.2(0.301)

2.0602

0.2 (0.3276)

2.0655

0.2 (0.3281)

2.0656

0.2

0.4

0.4

0.3552

log (0.4

2.1366)

log (0.4

2.1415)

0.35520.4)

(042

0.35520.4

()
051

2.0656

0.2 (0.3552)

2.1366

2.0656

0.2 (0.3797)

2.1415

2.0656

0.2 (0.3801)

2.1416

0.4

0.6

0.6

0.4051

log (0.6

2.2226)

log (0.6

2.2272)

0.4051 0.4

()

506

1
2

0.4051 0.4

()

514

2.1416

0.2 (0.4051)

2.2226

2.1416

0.2 (0.4279)

2.2272

2.1416

0.2 (0.4282)

2.2272

436

UMERIC AL

ETHODS

IN

NGINEERING
AND

CIENCE

log

Mean slope

Old y

mean slope

new y

0.6
0.8

0.8

0.4514

log (0.8

2.3175)

log (0.8

2.3217)

0.45140.4

()

938

0.45140.4

()

943

2.2272

0.2 (0.4514)

2.3175

2.2272

0.2 (0.4726)

2.3217

2.2272
0.2 (0.4727)

2.3217

0.80.4943Š

2.3217

0.2 (0.4943)

2.4206

1.0

log (1

2.4206)

0.4943)

0.5341

2.3217

0.2 (0.5142)

2.4245

1.0

log (1

2.4245)

0.49430.5

()

346

2.3217

0.2 (0.5144)
2.4245

1.00.5346Š

2.4245

0.2 (0.5346)

2.5314

1.2

log (1.2

2.5314)

0.53460.5

()

719

2.4245

0.2 (0.5532)

2.5351

1.2

log (1.2

2.5351)

0.53460.5

()

723

2.4245

0.2 (0.5534)
2.5351

1.20.5723Š

2.5351

0.2 (0.5723)

2.6496

1.4

log (1.4

2.6496)

0.57230.6

()

074

2.5351

0.2 (0.5898)

2.6531

1.4

log (1.4

2.6531)

0.57230.6

()

078

2.5351

0.2 (0.5900)
 2.6531

Hence

(1.2)

2.5351 an d

(1.4)

2.6531 approximately.

EXAMPLE 10.13

Using Euler™s modified method, obtain a solution of the equation

dydxxy

with initial conditions

at

for the range

0

£0

in steps of

2.

Solution:

The various calculations are arranged as follows:

xyy

Mean slope

Old y

mean slope
)

new y

0.0

0.2 (1)

1.2

0.2

0.21.2

= 1.2954

11.2954

()

= 1.1477

0.2 (1.1477)

1.2295

UMERIC AL

S
OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

437

xyy

Mean slope

Old y

mean slope

new y
0.2

0.21.2295

= 1.3088

11.3088

()

= 1.1544

0.2 (1.1544)

1.2309

0.2

0.21.2309

= 1.3094

11.3094

()

= 1.1547

0.2 (1.1547)

1.2309

0.21.3094Š

1.2309

0.2 (1.3094)

1.4927
0.4

0.41.4927

= 1.6218

1.30941.6

()

= 1.

18

465

1.2309

0.2 (1.4654)

1.5240

0.4

0.41.524

= 1.6345

1.30941.6

()

= 1.

45
471

1.2309

0.2 (1.4718)

1.5253

0.4

0.41.5253

= 1.6350

1.3094 1.6

()

= 1.

50

472

1.2309

0.2 (1.4721)

1.5253

0.41.6350Š

1.5253

0.2 (1.635)

1.8523

0.6

0.61.8523
= 1.9610

1.6351.961)

1.798

1.5253

0.2 (1.798)

1.8849

0.6

0.61.8849

= 1.9729

1.6351.97

()

= 1. 0

29

804

1.5253

0.2 (1.804)

1.8861

0.6

0.61.8861

= 1.9734

1
2

1.6351.97

()

= 1. 2

34

804

1.5253

0.2 (1.8042)

1.8861

Hence

(0.6)

1.8861 approximately.

Exercises 10.2

1.

Apply Euler™s method to solve

(0)

0,
choosing the step length

0.2. (Carry out six steps).

2.

Using Euler™s method, find the approximate value of

when

0.6 of

dy

dx

1Œ2

xy

, given that

0 when

0 (take

0.2).

3.

Using the simple Euler™s method solve for

at

0.1 from
dy

dx

xy

(0)

1, taking step size

0.025.

438

UMERIC AL

ETHODS

IN
 E

NGINEERING

AND

CIENCE

4.

Solve

1Œ

(0)

by the modified Euler™s method and obtain

at

0.1, 0.2, 0.3

5.

Given that

dy

dx
x

and

(0)

1. Find an approximate value of

(0.1), taking

0.05 by the modified Euler™s method.

6.

Given

sin

(0)

1. Compute
y

(0.2) and

(0.4) with

0.2

using Euler™s modified method.

7.

Given

dyyx

dxyx

with boundary conditions

1 when

0, find

approximately

for

0.1, by Euler™s modified method (five steps)

8.

Given that

/2

dydxxy

and
y

1 when

1. Find approximate

value of

at

2 in steps of 0.2, using Euler™s modified method.

10.6 Runge’s Method

Consider the differential equation,

00

(,),()

dy

fxyyxy

dx

(1)

Clearly the slope of the curve through

0
) is

) (Figure 10.2).

Integrating both sides of (1) from (

) to (

0
k

), we have

00

00

(,)

ykxh

yx

dyfxydx

(2)

L N M

S
1

FIGURE 10.2

*Called after the German mathematician

Carl Runge

(1856-1927).

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS
•

439

To evaluate the integral on the right, we take

as the mid-point of

LM

and find the values of

)(

i.e., dy

dx

) at the points

h
/2,

. For

this purpose, we first determine the values of

at these points.

Let the ordinate through

cut the curve

PQ

in

and the tangent

PT

in

. The value of

is given by the point

S
1

110

. tan

yNSLPHSyPH

00

00

()

/(,

yhdydxyfxy

(3)

Also

000

.tan( ).

hfxy

yMTLPRTyPRy
Now the value of

at

is given by the point

where the line

through

draw with slope at

y
T

) meets

MQ

Slope at

00000

tan ,,

())]

[(

Tfxhyfxhyhfxy

000000

.tan[,(, )]

yRRTyPRyhfxhyhfxy

(4)

Thus the value of

,
y

) at

),

the value of

) at

x
0

/2, y

and the value of

) at

)
where

and

are given by (3) and (4).

Hence from (2), we obtain

000 0

(,)4

()(/2,)(,

xh

PSQ

SQ

kfxydxfff

fxyfxhyxhy
by Simpson™s rule

Which gives a sufficiently accurate value of k and also

The repeated application of (5) gives the values of

for equi-spaced

points.

Working rule

to solve

(1)

by RungeÕs method

Calculate successively

100

(, ),

khfxy
2001

11

22

khfxhy k

001

khfxhyk

440

UMERIC AL

ETHODS

IN

NGINEERING

AND

S
CIENCE

and

300

khfxhyk

Finally compute,

123

kkkk

which gives the required approximate value as y

k.

(Note that

is the weighted mean of

,
 k

, and

).

EXAMPLE 10.14

Apply Runge™s method to find an approximate value of

when

given that

dy

dx

and

1
when

0.

Solution:

Here we have

0,

1,

0.2,

)
1

100

(, )0.210.200

khfxy

2001

11

0.240

0.20.1,1.1

22

khfxhy kf

001

,0 0.

.20.2,28

20

1.

khfxhyk f

and

300

,0.20.1,1.280.296

khfxhykf
123

11

k4kk0.2000.9600.0.242

66

29

Hence the required approximate value of

is 1.2426.

10.7 Runge-Kutta Method*

The Taylor™s series method of solving differential equations numerical-

ly is restricted by the labor involved in finding the higher order derivatives.

However, there is a class of methods known as Runge-Kutta methods which

do not require the calculations of higher order derivatives and give greater

accuracy. The Runge-Kutta formulae possess the advantage of requiring

only the function values at some selected points. These methods agree with

Taylor™s series solution up to the term in

h

where

differs from method to

method and is called the

order of that method

First order R-K method

. We have seen that Euler™s method (Section

10.4) gives

100000

(, )

yyhfxyyhy

,
y

)]

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

441

Expanding by Taylor™s series

10000

()

h
yyxhyhyy

It follows that the Euler™s method agrees with the Taylor™s series solu-

tion upto the term in

Hence,

EulerÕs method is the Runge-Kutta method of the first order

Second order R-K method

. The modified Euler™s method gives

10001

(, )(, )

yyfxyfxhy

(1)

Substituting

y
0

hf

) on the right-hand side of (1), we obtain

100000

(),

yyf fxhyhf

where

,
y

) (2)

Expanding L.H.S. by Taylor™s series, we get

23

100000

2!3!

hh

yyxhyhyyy

(3)

Expanding

0
hf

) by Taylor™s series for a function of two

variables, (2) gives

2**

1000000

(,)()

ff

yyffxy hhf Oh

xy

23

000

ff

yhfhfhOh

xy
2

000

yhffOh

dfxy

ff

dxxy

000

2!

yhyyOh

(4)

Comparing (3) and (4), it follows that the modified Euler™s method
agrees with the Taylor™s series solution upto the term in

Hence the modified EulerÕs method is the Runge-Kutta method of the

second order.

**

) means fiterms containing second and higher powers of

fl and is read as

order of h

442

UMERIC AL

ETHODS
IN

NGINEERING

AND

CIENCE

The second order Runge-Kutta formula is

1012

yykk

Where

hf

,
y

) and

hf

iii

Third order R-K method.

Similarly, it can be seen that Runge™s meth-

od (Section 10.6) agrees with the Taylor™s series solution upto the term in

As such,

RungeÕs method is the Runge-Kutta method of the third order.

The third order Runge-Kutta formula is

10123

yykkk

Where

1002001

11

(,), ,

22

khfxykhfxhyk

And

300

,,
khfxhyk

where

3001

(,)

kkhfxhyk

iv

Fourth order R-K method.

This method is most commonly used

and is often referred to as the

Runge-Kutta method

only.

Working rule

for finding the increment k of y corresponding to an

increment h of x by Runge-Kutta method from

(,),()

dy

fxyyx

dx

is as follows
:

Calculate successively

hf

),

2001

3002

11

22

11

22

khfxhy k

khfxhyk
and

4003

khfxhyk

Finally compute

1234

22

kkkkk

which gives the required approximate value as

(Note that k is the weighted mean of

k
1

,k

,k

, and

).

Obs.

One of the advantages of these methods is that the

operation is identical whether the differential equation is linear

or non-linear.

NOTE

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL
 E

QUATIONS

443

EXAMPLE 10.15

Apply the Runge-Kutta fourth order method to find an approximate

value of

when

given that

dy

dx

and

y
1

when

Solution:

Here

0,

1,

0.2,

y
0

100

(,)0.21= 0.2000

khfxy

2001

11

0.2400

,0.20.1,1.1

22

khfxhy kf

3002

11

,0.20.1,1.12

22

0.2440

khfxhykf

and

4003
,0.20.2,0.288

1.248

khfxhykf

1234

22

0.20000.48000.48800.2888

1.45680.2428

kkkkk

Hence the required approximate value of

is 1.2428.

EXAMPLE 10.16

Using the Runge-Kutta method of fourth order, solve

22

22

dyy x

dx

yx

with

(0)

at

4.

Solution:

We have

22
22

(,)

yx

fxy

yx

To find

(0.2)

Hence

0,

1,

0.2

100

(,)0.20,10.2000

khfxyf

2001
11

,0.20.1,1.1

22

0.19672

khfxhy kf

444

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

3002

0.1967
11

,0.20.1,1.09836

22

khfxhyk f

4003

,0.20.2,1.0.1

198

6791

khfxhykf

1234

22

kkkkk

0.22 0.196722 0.19670.18 0.19599

91

6
Hence

(0.2)

1.196.

To find

(0

4):

Here

0.2,

1.196,

0.2.
111

(,0.1891

khfxy

2111

0.1795

11

,0.20.3,1.2906

22

khfxhykf

3112

0.1793

11

,0.20.3,1.2858

22

khfxhyk f

4113

,0.20.4,1.0.1

376

5388
khfxhykf

1234

22

kkkkk

0.1891 2 0.17952 0.17930.16880 1 2

.79

Hence

(0.4)

1.196
 0.1792

1.3752.

EXAMPLE 10.17

Apply the Runge-Kutta method to find the approximate value of

for

in steps of

1, if

dy

dx

,y
1

where

0.

Solution:

Given

x, y

Here we take

0.1 and carry out the calculations in two steps.

Step I

.x

0,

y
0

1,

0.1

100

(,)0.10,10.1000

khfxyf

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

445
2001

11

,0.10.05,1.1

22

0.1152

khfxhy kf

3002

0.1168

11

,0.10.05,1.1152

22

khfxhykf

4003

,0.10.1,1.0.1

113

6847

khfxhykf

1234

22
6

kkkkk

0.10000.23040.23360.13470 1

.1 65

giving

(0.1)1.1165

yyk

Step II. x

0.1,

1
1.1165,

0.1

111

(,)0.10.1,1.11650.1347

khfxyf

2111

0.1551

11

,0.10.15,1.1838

22

khfxhykf

3112

0.1576

11

,0.10.15,1.194

22

khfxhykf

4123

( ,)0.10.2,1.0.1823
1576

khfxhykf

1234

0.15

221

kkkkk

Hence

(0.2)1.2736

yyk

EXAMPLE 10.18

Using the Runge-Kutta method of fourth order, solve for

at

2
,

From

dyxye

dx

xxe

given

,y

Solution:
We have

(,)

xye

fxy

xxe

To find

(1

2)

446

UMERIC AL

ETHODS
IN

NGINEERING

AND

CIENCE

Here

,y

,h

100

(,)0.20.1462

khfxy

e
10.1

200

10.1

21 0.100.073

,0.2

22

1 0.11 0.1

0.1402

he

khfxy

1.1

3002

2
1.1

1 121 0.100.07

,0.2

22

1 0.11 0.1

0.1399

khfxhyk

1.2

4003

1.2

21.2 0.1399

,0.2

1.21.2

0.1348

khfxhyk

e
and

1234

11

220.14620.28040.27980.1348

66

0.1402

kkkkk

Hence

(1.2)

0.1402

0.1402.

To find

(1

4)

Here
11

1.2,0.1402,0.2

xyh

111

(,)0.21.2,00.1348

khfxyf

2111

/2,/20.21.3,0.20760.1303

khfxhykf

3112

/2,/20.21.3,0.20530.1301

khfxhyk f

4113

( ,)0.21.3,0.27030.1260

khfxhykf

1234
1

22

0.13480.26060.26020.1260

0.1303

kkkkk

Hence

(1.4)0.14020.13030.2705.

yyk

UMERIC AL

OLUTION

OF

RDINARY

D
IFFERENTIAL

QUATIONS

447

Exercises 10.3

1.

Use Runge™s method to approximate

when

1.1, given that

1.2

when

1 and

dy

dx

y
2

2.

Using the Runge-Kutta method of order 4, find

(0.2) given that

dy

dx

(0)

1 taking

0.1.

3.

Using the Runge-Kutta method of order 4, compute

(0.2) and
y

(0.4)

from 10

22

dy

xy

dx

(0)

1, taking

0.1.

4.

Use the Runge Kutta method to find

when

1.2 in steps of 0.1, given

that

dy

dx

2
y

and

(1)

1.5.

5.

Given

dy

dx

(0)

2. Compute

(0.2),

y
(0.4), and

(0.6) by the

Runge-Kutta method of fourth order.

6.

Find

(0.1) and

(0.2) using the Runge-Kutta fourth order formula,

given that

and

(0)

1.

7.

Using fourth order Runge-Kutta method, solve the following equation,

taking each step of

h
0.1, given

(0)

3.

dy

dx

(4

xy

). Calculate

for

0.1 and 0.2.

8.

Find by the Runge-Kutta method an approximate value of

for

0.6,

given that

y
 0.41 when

0.4 and

dydxxy

9.

Using the Runge-Kutta method of order 4, find

(0.2) for the equation

dyyx

dxyx

(0)

1. Take

0.2.

10.

Using fourth order Runge-Kutta method, integrate

32

Œ212Œ208.5,

yxxx

using a step size of 0.5 and initial condition

of

1 at

0.

11.

Using the fourth order Runge-Kutta method, find

at

0.1 given that

dy

dx

(0)

0 and

0.1.
12.

Given that

dy

dx

Œ2

)/(

) and

1 at

0, find

for

0.1,

0.2, 0.3, 0.4, and 0.5.

448

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

10.8 Predictor-Corrector Methods

If

Œ1

and

i
are two consecutive mesh points, we have

Œ1

. In

Euler™s method (Section 10.4), we have

-10-1

(-1,);1,2,3

iii

yyhfxihyi

(1)

The modified Euler™s method (Section 10.5), gives

-1-1-1

(,)(,)

iiiiii

h
yyfxyfxy

The value of

is first estimated by using (1), then this value is inserted

on the right side of (2), giving a better approximation of

. This value of

is again substituted in (2) to find a still better approximation of

. This step

is repeated until two consecutive values of

agree.

This technique of refin-

ing an initially crude estimate of y

by means of a more accurate formula

is known as

predictor-corrector method.

The equation (1) is therefore

called the

predictor

while (2) serves as a

corrector

of

In the methods so far described to solve a differential equation over an

interval, only the value of

at the beginning of the interval was required. In

the predictor-corrector methods, four prior values are needed for finding

the value of

at

. Though slightly complex, these methods have the ad-

vantage of giving an estimate of error from successive approximations to

i
.

We now describe two such methods, namely: Milne™s method and

Adams-Bashforth method.

10.9 Milne’s Method

Given

dy

dx

x, y

and

,x

to find an approximate value of

for

x
x

nh

by Milne™s method, we proceed as follows

The value y

) being given, we compute

h
),

),

),

by Picard™s or Taylor™s series method.

Next we calculate,

),

y
1

),

),

0
 3

Then to find

), we substitute Newton™s forward interpola-

tion formula

23

0000

(1)(1)(2)

(,)

26

nnnnn
fxyf nfff

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

449

In the relation

40
(,)

xh

yyfxydx

40000

...

xh

nn

yyfnf fdx

[Put

0
nh

dx

hdn

0000

...

nn

yfnffdn

0000

20

48 ...

yhfff
Neglecting fourth and higher order differences and expressing

23

000

,and

ff f

and in terms of the function values, we get

()

40123

22

yyff f

which is called a

predictor

Having found

, we obtain a first approximation to

404
(4,)

ffx hy

Then a better value of

is found by Simpson™s rule as

42234

yyfff

which is called a

corrector

Then an improved value of

is computed and again the corrector is

applied to find a still better value of

. We repeat this step until

y
4

remains

unchanged. Once

and

are obtained to desired degree of accuracy,

) is found from the

predictor

as

()

51234
4

22

yyfff

and

) is calculated. Then a better approximation to the value

of

5
is obtained from the

corrector

as

()

53345

yyfff

We repeat this step until

becomes stationary and, then proceed to

calculate

as before.

450

N
UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

This is

MilneÕs predictor-corrector method

. To insure greater accuracy,

we must first improve the accuracy of the starting values and then sub-

divide the intervals.

EXAMPLE 10.19

Apply Milne™s method, to find a solution of the differential equation

yxy

in the range

0
x

for the boundary condition

at

0.

Solution:

Using Picard™s method, we have

0(,),

yyfxydx

where

(,)

fxyxy

To get the first approximation, we put

y
 0 in

),

Giving

yxdx

To find the second approximation, we put

Giving

425

4220

yxd

xx
x

Similarly, the third approximation is

252581

2202201604400

yxd

xxxxxx

Now let us determine the starting values of the Milne™s method from

), by choosing

0.2.
x

0.0,

0.0000,

0.0000

0.2,

0.020,

0.1996
x

0.4,

0.0795

0.3937

0.5,

0.1762,

0.5689

Using the predictor,

()

40123

22

yyff f

0.8

()

0.3049,

0.7070

and the corrector,

()

42 234

4,
3

yyfff

yields

()

0.3046

0.7072 (

ii

UMERIC AL

OLUTION

OF
 O

RDINARY

IFFERENTIAL

QUATIONS

451

Again using the

corrector

()

0.3046,

, which is the same as in (ii)

Now using the predictor,

()

41234

22
3

yyfff

0.1,

()

0.4554

0.7926

and the corrector

()

53345

yyfff
gives

()

0.4555

0.7925

Again using the corrector,

()

0.4555,

a value which is the same as before.

Hence

(1)

0.4555.
EXAMPLE 10.20

Using Milne™s method find

(4

5)

given

xy

0 given

(4)

(4

.
1)

0049

,y

(4

2)

0097

,y

(4

3)

0143

;y

(4
.

4)

0187

Solution

We have

(2Π)/5()

yyxfx

[say]

Then the starting values of the Milne™s method are

0,

0
1,

212

0.05

54

4.1,

1.0049,

0.0485

4.2,

2
1.0097,

0.0467

4.3,

1.0143,

0.0452

4.4,

4
1.0187,

0.0437

Since

is required, we use the predictor

()

51234

22

yyfff

0.1)

4.5,
()

4 0.1

1.004922.04670.045220.04371.023

452

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE
2

21.023

0.0424

554.5

Now using the corrector

()

53345

yyfff

, we get

()

5
0.1

1.01430.045240.04370.04241.023

Hence

(4.5)

1.023

EXAMPLE 10.21

Given

Ðx

,y

(0)
1

find

at

2,

and

by Taylor™s

series method and compute

(0
.

4)

by Milne™s method.

Solution:

Given

(0)

1 and

0.1

We have

22

()( )

yxxxye

00

32 22

()()()(3(2))

xx

yxxxyexyxyye
3222

32

exxyxy xyy

01

32222

()3262 2'2

02

y xexxyxy xyyxyyxyxyy

Substituting these values in the Taylor™s series,

23

()0000

1!2!3!

xx x

yxyyyy
23

11

(0.1)1(0.1)0(0.1)1 (0.1)2

26

0.005 Π0.0003

1.0047,

i.e.,

1.005

Now taking

0.1,

(0.1)

1.005,

0.1

0.10.092,0.10.849,0.11.247

yy y
Substituting these values in the Taylor™s series about

0.1,

23

0.1 0.10.1

(0.2)0.10.10.10.1

1!2!3!

yyyyy

23

(0.1)(0.1)

1.005(0.1)0.0920.8491247

26

=1.018

UMERIC AL

OLUTION

OF
O

RDINARY

IFFERENTIAL

QUATIONS

453

Now taking

0.2,

(0.2)

1.018,

0.1

0.20.176,0.20.77,0.20.819

yyy

Substituting these values in the Taylor™s series

23
0.10.10.1

(0.2)0.20.2 0.20.2

1!2!3!

yyyyy

1.018

0.0176

0.0039

0.0001

1.04

Thus the starting values of the Milne™s method with

0.1 are

0.0,

0
y

0.1,

1.005

0.092

0.2,

2
1.018

0.176

0.3,

1.04

0.26

Using the predictor,

()

40123

22

h
yyff f

2 0.0920.1

4 0.1

1[76]

2 0.2

1.09.

0.4

()

1.09,

0.40.362

fy

Using the corrector,

()
42 234

4,

yyfff

yields

()

0.1

0.018 (0.176 + 4(0.26) + 0.362) = 1.071

Hence

(0.4)

1.071

EXAMPLE 10.22

Using the Runge-Kutta method of order

find
y

for

given that

dy

dx

xy

,y

(0)
1. Continue the solution at

using

Milne™s method.

Solution:

We have

xy

To find
 y

(0

1)

Here

0,

1,

0.1.

100

(,)0.1 0,0.10

100

khfxyf

454

•
N

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

2001

11

,0.1 0.0.1155

5,1.05

22

khfxhy kf

3002

0.1172

11
,0.1 0.05,1.0577

22

khfxhykf

4003

,0.1 0.1, 0.13598

1.1172

khfxhykf

1234

22

kkkkk

0.1 0.2310.23430.13590.1 6 7

818

Thus

10

(0.1)1.1169
yyyk

To find

(0

2)

Here

,y

1169

,h

1
111

,0.1 0.1,1.1160 13 9

9.5

khfxyf

2111

0.1581

11

,0.1 0.15,1.1848

22

khfxhykf

3112

0.1609

11

,0.1 0.15,1.1959

22

khfxhykf

4113

,0.1 0.20.1888

,1.2778

khfxhykf
1234

0.16

225

kkkkk

Thus

(0.2)

1.2773.

To find

y
(0

3)

Here

0.2,

1.2773,

0.1.

122

(,)0.1 0.2, 0.188

1.27 37

khfxyf

2221

0.2224
11

,0.1 0.25,1.3716

22

khfxhy kf

3222

0.2275

11

,0.10.25,1.3885

22

khfxhykf

4223

,0.10.3,1.0.2

507

4816

khfxhykf

UMERIC AL

OLUTION
OF

RDINARY

IFFERENTIAL

QUATIONS

455

1234

0.22

227

kkkkk

32

(0.3)1.504

yyyk

Now the starting values for the Milne™s method are:

x

0.0

1.0000

1.0000

0.1

1.1169

1.3591

2
0.2

1.2773

1.8869

0.3

1.5049

2.7132

Using the

predictor

()
40123

22

yyff f

()

444

0.41.83444.0988

xyf

and the

corrector

()

42 234

yyfff
()

0.1

1.27731.88694 2.71324.098

1.8397

4.1159.

Again using the

corrector

()

0.1

1.27731.88694 2.71324.1159

c
y

1.8391

4.1182 (i)

Again using the

corrector

()

0.1

1.27731.88694 2.71324.1182

1.8392 which is same as (i)

Hence

(0.4)

1.8392.
Exercises 10.4

1.

Given

dy

xy

dx

(0)

2. The values of

(0.2)

2.073,

(0.4)

2.452,

and

(0.6)

3.023 are gotten by the R.K. method of the order. Find

(0.8) by Milne™s predictor-corrector method taking

0.2
456

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

2.

Given 2

dy

dx

(1

2
)

and

(0)

1,

(0.1)

1.06,

(0.2)

1.12,

(0.3)

1.21, evaluate

(0.4) by Milne™s predictor corrector method.

3.

Solve that initial value problem

1,(0)1

dy

xyy

dx
for

0.4 by using Milne™s method, when it is given that

: 0.1 0.2 0.3

: 1.105 1.223 1.355

4.

From the data given below, find

at

1.4, using Milne™s predictor-

corrector formula:

dy

dx

/2:

x
 1

1.1 1.2 1.3

2.2156 2.4549 2.7514

5.

Using Taylor™s series method, solve

dy

xyx

dx

(0)

1; at

0.1,

0.2, 0.3. Continue the solution at

0.4 by Milne™s predictor-corrector

method.

6.

If

2
e

(0)

2,

(0.1)

2.01,

(0.2)

2.04, and

2.09, find

(0.4) using Milne™s predictor-corrector method.

7.

Using the Runge-Kutta method, calculate

(0.1),

(0.2), and

(0.3)
given that

1.

dyxy

dx

(0)

0. Taking these values as starting val-

ues, find

(0.4) by Milne™s method.

10.10 Adams-Bashforth Method

Given

(,)

dy

fxy

dx

and

0
y

), we compute

102030

(),(2),(3)

yyxhyyxhyyxh

by Taylor™s series or Euler™s method or the Runge-Kutta method.

Next we calculate

10 120230 3

(,), (2,),(3,)

ffxhyffxhyffxhy

Then to find

, we substitute Newton™s backward interpolation formula

23

0000
(1)(1)(2

(,)

26

nnnnn

fxyf nfff

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

457

in

0
0

10

(,)

xh

yy fxy

(1)

1000 0

(1)

nn

yyfnffdx

[Put

0
nh

dx

hdn

0000

(1)

nn

yhfnffdn

23

00 0 00

153

2128

yhffff

Neglecting fourth and higher order differences and expressing

23
00 0

, and

fff

in terms of function values, we get

100123

55 5937 9

24

yyff ff

(2)

This is called the

Adams-Bashforth predictor formula

Having found

, we find

f
(

).

Then to find a better value of

1, we derive a

corrector formula

by sub-

stituting Newton™s backward formula at

1,

i.e.,

23

1111

(1)(1)(2

(,)

26
nnnnn

fxyfnfff

in

(1)

10111

(1)

nn

yy fnffdx

[Put

1
nh

dx

h dn

0111

(1)

nn

yfnffdn

23

01101

111

21224

yhf fff

Neglecting fourth and higher order differences and expressing

23

111
, and

fff

and in terms of function values, we obtain

()

101012

91959

24

yy ffff

which is called the

Adams-Moulton corrector formula

458

UMERIC AL

ETHODS
IN

NGINEERING

AND

CIENCE

Then an improved value of

1 is calculated and again the corrector (3) is

applied to find a still better value

1. This step is repeated until

remains

unchanged and then we proceed to calculate

as above.

Obs.

To apply both Milne and Adams-Bashforth methods, we

require four starting values of y which are calculated by means

of PicardÕs method or TaylorÕs series method or EulerÕs method

or the Runge-Kutta method. In practice, the Adams formulae

(2) and (3) above together with the fourth order Runge-Kutta

formulae have been found to be the most useful.

EXAMPLE 10.23

Given

dy

xy

dx

and

(1)

,y

(1

1)

1
.

233

,y

(1

2)

548

(1

3)

979, evaluate

(1

4) by the Adams-Bashforth method.

Solution:
Here

(1

Starting values of the Adams-Bashforth method with

0.1 are

1.0,

Œ3

1.000,

Œ3
(1.0)

(1

1.000)

2.000

1.1,

Œ2

1.233,

Œ2

2.702

1.2,

Œ1

1.548,

Œ1

3.669

x
1.3,

1.979,

5.035

Using the

predictor,

()

100123

55 5937 9

24

yyffff

411

1.4,2.5737.004

xyf
Using the

corrector

()

101012

9195

24

yy ffff

()

0.1

1.97997.00419 5.03553.6692.7022.575

24

Hence

(1

4)

2
.

575

NOTE

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

459

EXAMPLE 10.24

If

2,

dy

ey
dx

(0)

find

(4) using the Adams predictor corrector

formula by calculating

(1),

(2), and

(3) using Euler™s modified formula.

Solution

We have

x, y

e
x

yMean slope

Oldy

mean slop

new y

04

0.1(4)

2.4

12

0.1

(2.4) = 5.305

2
45.3054.6524

0.1 (4.6524)

2.465

12

0.1

(2.465) = 5.449

45.4654.7244

0.1 (4.7244)

2.472

12

0.1

(2.4724) = 5.465

45.4654.7324

2
 0.1 (4.7324)

2.473

12

0.1

(2.478) = 5.467

45.467 4.7333

0.1 (4.7333)

2.473

15

467

0.1 (5.467)

3.0199

22
e

0.2

(3.0199) = 7.377

5.4677.3776.422

2.473

0.1 (6.422)

3.1155

27

611

5.4677.6116.539

2.473

0.1 (6.539)

3.127

27

639

1
2

5.4677.6396.553

2.473

0.1 (6.553)

3.129

27

643

5.4677.6436.555

2.473

0.1 (6.555)

3.129

27

463

3.129

0.1 (7.643)

3.893

0
.

32

0.3

(3.893) = 10.51

7.64310.519.076

3.129

0.1 (9.076)

4.036

310

897

7.64310.8979.266

3.129

0.1 (9.2696)

4.056

310

.
949

7.64310.9499.296

3.129

0.1 (9.296)

4.058

310

956

7.64310.9569.299

3.129

0.1 (9.299)

4.0586

To find

(0

4)

by AdamÕs method

, the starting values with

h
0.1 are

0.0

Œ3

2.4

Œ3

0.1

Œ2

2.473

Œ2

5.467

0.2

Œ1
3.129

Œ1

7.643

0.3

4.059

10.956

460

UMERIC AL

ETHODS

IN
E

NGINEERING

AND

CIENCE

Using the predictor formula

()

100123

55 5937 9

24

yyffff

0.1

4.0595510.957597.643375.4679 4

24

5.383

Now

0.4

11
0.45.38325.38316.061

xy fe

Using the corrector formula

()

101012

9195

24

yy ffff

0.1

4.0586916.0611910.95657.6435.467

24

5.392

Hence

(0.4)

5.392

EXAMPLE 10.25

Solve the initial value problem

dy
/

dx

xÐy

,y

(0)

to find

(0

4) by

Adam™s method. Starting solutions required are to be obtained using the

Runge-Kutta method of the fourth order using step value

Solution:

We have
f

To find

(0

1)

Here

0,

0
1,

0.1.

100

(,)0.10, Œ0.1000

khfxyf

2001

Œ0.08525

11

,0.10.05,0.95

22

khfxhy kf

3002

11

,0.10.05,0.957Œ0.0867

22

khfxhykf

4003
,0Œ0.

.10

0.1,0.91 3

7741

khfxhykf

1234

22 00.0883

kkkkk

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

E
QUATIONS

461

Thus

10

(0.1)10.0883 0.9117

yyyk

To find

(0

2)

Here x

0.1,

0.9117,

h
0.1

111

,0.1 0.1,0.9117 Œ0.0731

khfxyf

2111

0.0616

11

,0.10.15,0.8751

22

khfxhykf

3112

0.0626

11

,0.10.15,0.8809

22

khfxhykf

4113

,0.1 0.2 0.0521

,0.8491

khfxhykf
1234

0.06

223

kkkkk

Thus

(0.2)

0.8494.

To find

y
(0

3)

Here

0.2,

0.8494,

0.1.

122

(,)0.1 0.25,0.84940.0521

khfxyf

2221

0.0428

11

,0.10.25,0.8233

22
khfxhy kf

3222

0.0436

11

,0.10.25,0.828

22

khfxhykf

4223

,0.10.3,00.0349

.058

khfxhykf

1234

0.04

228

kkkkk

Thus
32

(0.3) 0.8061

yyyk

Now the starting values for the Milne™s method are:

0.0

1.0000

0.0

(0.1)

1.0000

0.1

y
1

0.9117

0.1

(0.9117)

0.7312

0.2

0.8494

0.2

(0.8494)

2
0.5215

0.3

0.8061

0.3

(0.8061)

0.3498

462

UMERIC AL

ETHODS
IN

NGINEERING

AND

CIENCE

Using the predictor,

()

100123

55 5937 9

24

yyffff

0.4

()

1
0.1

0.8061550.3498590.5215370.731291

24

0.77890.2.67

Using the

corrector,

()

101012

9195

24

yy ffff

()

0.1

0.806190.2067 190.349850.5215 0.7312

24

0.7785
c

Hence

(0.4)

0.7785

Exercises 10.5

1.

Using the Adams-Bashforth method, obtain the solution of

dy

dx

at

0.8, given the values

: 0 0.20.40.6

: 00.0200 0.07950.1762

2.
Using the Adams-Bashforth formulae, determine

(0.4) given the dif-

ferential equation

dydxxy

and the data:

: 00.10.2 0.3

: 1 1.002 5 1.0101 1.0228

3.

Given

(0)

1 and the starting values

y

(0.1)

0.90516,

(0.2)

0.82127,

(0.3)

0.74918, evaluate

(0.4) using the Adams-

Bashforth method.

4.

Using the Adams-Bashforth method, find

(4.4) given 5

xy

2,

(4)

1,
y

(4.1)

1.0049,

(4.2)

1.0097 and

(4.3)

1.0143.

5.

Given the differential equation

dy

dx

and the data:

: 1 1.1 1.2 1.3

: 11.2331.548488 1.978921
determine

(1.4) by any numerical method.

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

463

6.

Using the Adams-Bashforth method, evaluate

(1.4); if

satisfies
dy

dx

1/

and

(1)

1,

(1.1)

0.996,

(1.2)

0.986,

(1.3)

0.972.

10.11 Simultaneous First Order Differential Equations

The simultaneous differential equations of the type

(,, )

dy

fxyz

dx

(1)

and

(,, )

dz

xyz

dx

(2)

with initial conditions

and

)
z

can be solved by the meth-

ods discussed in the preceding sections, especially Picard™s or Runge-Kutta

methods.

PicardÕs method gives

10001000

20 112011

30223022

(, ,), (, ,)

(,,),(,,)

(, ,),(, ,)

yyfxyzdxzzxyzdx

yyfxyzdxzzxyzdx

yyfxyzdxzzxyzdx

and so on.

ii
)

TaylorÕs series method

is used as follows:

If

be the step-size,

) and

0
h

). Then Taylor™s al-

gorithm for (1) and (2) gives

23

1000 0

'

2!3!

hh

yyhyyy

(3)

23

1000 0

2! 3!

hh

zzhzzz

(4)

Differentiating (1) and (2) successively, we get

y
,

, etc. So the values

000

,,

yyy

and

000

,,

zzz

are known. Substituting these in (3) and (4), we

obtain

for the next step.

Similarly, we have the algorithms

23

21111

2!3!

hh
yyhyyy

(5)

464

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

23

21111

2!3!

hh
zzhzzz

(6)

Since

and

are known, we can calculate

11

,,

yy

and

11

,,

zz

. Sub-

stituting these in (5) and (6), we get

and

z
2

Proceeding further, we can calculate the other values of

and

step

by step.

iii

Runge-Kutta method

is applied as follows:

Starting at (

) and taking the step-sizes for

y
,

to be

respectively, the Runge-Kutta method gives,

1000

(,,)

khfxyz

1000

(,,)

lhxyz

200101

111

,,

222

khfxhy kzl

200101
111

,,

222

lhxhykzl

300202

111

,,

222

khfxhykzl

300202

111

,,

222

lhxhykzl

400303

111

,,

222

khfxhykzl
400303

111

,,

222

lhxhykzl

Hence

101234

22

yykkkk

and

101234

22

zyllll

To compute

and

z
2

, we simply replace

by

in the

above formulae.

EXAMPLE 10.26

Using Picard™s method, find approximate values of

and

z
correspond-

ing to

1, given that

(0)

,z

(0)

and

dy

dx

,
dz

dx

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

465
Solution:

Here

0,

2,

1,

and

(,, )

dy

fxyzxz

dx

(,, )

dz

xyzxy

dx
0

(,, )

yyfxyzdx

and

(,, )

zz xyzdx

First approximations

00

1000

(, ,)2( 1)2

xx

xx
yyfxyzdxx dxxx

00

10 00

(, ,)1 (4)14

xx

xx

zzxyzdxxdxxx

Second approximations

2011

(,,)2 (14)

xx

yyfxyzdxxx dx
3

26

xx

00

2011

(,,)12

xx

xx

zzxyzdxxxxdx

45

23

14
2420

xx

xxx

Third approximations

23456

3022

3111 1

(, ,)2

2 2 4 20120

yyfxyzdxxxxxxx

3022

(, ,)

zzxyzdx

234 567

357 3111

14
2312 6012 252

xxxxxxx

and so on.

When

0.1

2.105,

2.08517,

2.08447

0.605,

2
0.58397,

0.58672.

466

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

Hence

(0.1)

2.0845,
z

(0.1)

0.5867

correct to four decimal places.

EXAMPLE 10.27

Find an approximate series solution of the simultaneous equations

dx

dt

xy

t, dy

dt

ty

subject to the initial conditions

,y

1
,

0.

Solution:

and

both being functions of

, Taylor™s series gives

and

23

000 0

23

0000

()

2!3!

()

2!3!

tt

xtxtxxx

tt

ytytyyy
(

Differentiating the given equations

xxy t

ii

ytyx

iii

w.r.t.

, we get

xxyxy

xxyxyxyxy
22

222

ytyyx

ytyyyx

(iv)

Putting

1,

Π1,

0
0 in (

ii

), (

iii

), and (

iv

), we obtain

00000

Œ 120Œ1

1.1 Π1 Π124

Œ3Œ 114Œ1Œ1Œ9

xxyxy

0
000

02

2Œ1Œ1Œ3

2248 etc

yyx

Substituting these values in (

), we get

23

23

()14912

2!3!2

tt

xttttt
23

23

34

()1381

2!3!2 3

tt

xttttt

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

467

EXAMPLE 10.28
Solve the differential equations

1,

dy

dz

xzxy

dxdx

for

0.3

using the fourth order Runge-Kuta method. Intial values are

,z

Solution:
Here

(, , )1 , (, , )

fxy zxz xy zxy

0,

0,

1. Let us take

0.3.

hf

(
x

0.3

(0, 0, 1)

0.3 (1

0)

0.3.

1000

,,0.3Œ0 00

()

lhxyz

200101

111

,,

222
0.30.15,0.15,10.31 0.150.345

khfxhy kzl

200101

111

,,

222

0.30.150.150.00675

lhxhykzl

22

3000

,,

222

0.30.15,0.1725,0.996625

0.31 0.9966250.15 0.34485

kl

khfxhyz

f
22

3000

,,

222

0.30.150.17250.007762

kl

lhxhyz

400303

,,

0.30.3,0.34485,0.992240.3893

khfxhykzl

400303

,,

0.30.30.344850.03104
lhxhykzl

Hence

001234

22

yxhykkkk

468

UMERIC AL

ETHODS

IN

NGINEERING

AND

S
CIENCE

i.e.,

(0.3)0 0.32(0.345)2(0.34485)0.38930.34483

and

01234

22

zxhyllll

i.e.

(0.3)1020.006752(0.0077625)(0.03104)

0.98999

10.12 Second Order Differential Equations

Consider the second order differential equation

2

,,

dydy

fxy

dx dx

By writing

dy

dx

, it can be reduced to two first order simultaneous

differential Equations

,f(x, y, z)

dy

dz

dxdx

These equations can be solved as explained above.

EXAMPLE 10.29

Find the value of

(1

.
1)

and

(1

2)

from

;y

(1)

,y

(1)

1
,

using the Taylor series method

Solution:

Let

so that

Then the given equation becomes

z
x

Ðy

such that

(1)

1,

(1)

1,

0.1. (

ii

Now from (i)

,,

yzy zyz

iii
)

and from (ii)

32222

222

223

,32

6(2)2

6(2)22

zxyzzxyzyzyz

zxyzyyzyzyzz

xyzyzzyzz

iv

UMERIC AL

OLUTION
OF

RDINARY

IFFERENTIAL

QUATIONS

469

Taylor™s series for

(1.1) is

23

(1.1) (1)(1)11....

2!3!

hh

yyhyyy

Also

(1)1,(1)1,(1)(1)0,(1)(1)1

yyyz yz
23

0.10.1

(1.1)(1)0.1(1)0 01.1002.

26

Taylor™s series for

(1.1) is

23

(1.1)(1) (1) 11....

2!3!

hh

zzhzzz

Here

(1)1,(1) 0,(1)1,(1) 3

zzzz

23
0.10.1

(1.1)(1)0.1(0)131.0055

26

Hence

(1.1)

1.1002 and

(1.1)

1.0055.

EXAMPLE 10.30

Using the Runge-Kutta method, solve

xy

for

x
0

correct

to 4 decimal places. Initial conditions are

,y

,y

0.

Solution:

Let

dy

dx

(
x

Then

22

Œ(,,)

dy

xzyxyz

dx

We have

0,

1,

0,
h

0.2

Runge-Kutta formulae become

hf

0.2(0)

20101

111

,,

222

0.2Π0.1Π0.02

o

khfxhy kzl

30202

111

,,

222

0.2Œ 0.0999 Œ0.02

khfxhykzl

40303

0.2Π0.1958 Π0 92

0. 3

khfxhykzl

470

N
UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

1234

220.0199

kkkkk

hf

0
,

0)

0.2(Π1)

Π0.2

200101

111

,,

222

0.2(0.999)0.1998

lhxhykzl

300202

111

,,

0.2Œ

222

0.9791Œ0.1958

lhxhykzl
400303

,,

0.2 0.9527Œ0.1905

lhxhykzl

1234

220.1970

lllll

Hence at

0.2,

1 Π0.0199

0.9801

and
y

0 Π0.1970

Π0.1970.

EXAMPLE 10.31

Given

xy

,y

(0)
1

,y

(0)

obtain

for

0(0

1)

3 by

any method. Further, continue the solution by Milne™s method to calculate

(0

4).

Solution:
Putting

, the given equation reduces to the simultaneous equa-

tions

xz

0,

(1)

We employ Taylor™s series method to find

Differentiating the given equation

times, we get
y

n+

ny

At

0, (

0
Œ(

1)(

(0)

1, gives

(0)

Π1,

(0)

3,

(0)

Œ 5 × 3, ......

and
y

(0)

0 yields

(0)

(0)

......

0.

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS
•

471

Expanding

) by Taylor™s series, we have

23

123

()(0)(0)(0)(0)

2!3!

xx

yxyxyyy

46

353

()1

2!4!6!

yxxx
(2)

and

35

11

()(),

28

zxyxxxxxy

(3)

From (2), we have

24

246

0.11

0.110.10.995

28

0.20.2

0.21 0.9802

28

0.30.30.3

0.310.956

2848

y
y

From (3), we have

(0.1)

Π0.0995,

(0.2)

Π0.196,

(0.3)

Π0.2863.

Also from (1),

Œ(

xz

)
z

(0.1)

0.985,

(0.2)

Π0.941,

(0.3)

Π0.87.

Applying Milne™s predictor formula, first to

and then to

, we obtain

0.4

0.400.120.1Π0.220.3

Œ 1.79 0.941Œ1.74Œ0.3692

zzzzz
and

0.4

0.400.120.1Œ0.220.3

0Œ 0.1990.196Œ0.5736 0.9231

yyyyy

Also

(0.4)

Œ{

(0.4)
z

(0.4)

(0.4)}

Π{0.4(Π0.3692)

0.9231}

Π0.7754.

Now applying Milne™s corrector formula, we get

0.40.20.240.30.4

Œ0.196Œ0.941Œ3.48Œ0.7754Œ0.3692

0.1

zzzzz

472

•
N

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

and

0.40.20.240.30.4

0.9802Œ0.196Œ1.1452Œ0.3692 0.923

.1

yyyyy
Hence

(0.4)

0.9232 and

(0.4)

Π0.3692.

Exercises 10.6

1.

Apply Picard™s method to find the third approximations to the values of

and

, given that

dy

dx

dz

dx

3
(

), given

1,

1/2 when

0.

2.

Using Taylor™s series method, find the values of

and

for

0.4,

satisfying the differential equations

dx

dt

x
y

dt

with initial conditions

0,

1,

dy

dt

Π1 at

0.
3.

Solve the following simultaneous differential equations, using Taylor

series method of the fourth order, for

0.1 and 0.2:

1;;01.

dy

dz

xzxyy

dxdy

4.

Find

(0.1),

(0.1),

(0.2), and

(0.2) from the system of equations:

x
z

given

(0)

0,

(0)

1 using Runge-Kutta method of the

fourth order.

5.

Using Picard™s method, obtain the second approximation to the solution

of

33

dydy
xxy

dx

dx

so that

01.0.

yy

6.

Use Picard™s method to approximate

when

0.1, given that

220

dydydy

xxy

dxdx

dx
and

0.5,

0.1

dy

dx

when

0.

7.

Find

(0.2) from the differential equation

xy

Œ6

0 where

(0)

1,

(0)

0.1, using the Taylor series method.

8.

Using the Runge-Kutta method of the fourth, solve

xy

(0)

1,

(0)

0 to find

(0.2) and

(0.2).

UMERIC AL

OLUTION

OF
O

RDINARY

IFFERENTIAL

QUATIONS

473

9.

Consider the second order initial value problem

Œ2

sin

with
y

(0)

Π0.4 and

(0)

Π0.6. Using the fourth order Runga-Kutta

method, find

(0.2).

10.

The angular displacement

of a simple pendulum is given by the equa-

tion

sin0

dt

where

98 cm and

g
980 cm/sec

. If

0 and

dt

4.472 at

0,

use the Runge-Kutta method to find

and

dt

when

0.2 sec.

11.

In a

L-R-C

circuit the voltage

t
) across the capacitor is given by the

equation

dvdv

LCRCv

dt

dt

subject to the conditions

0,

dv

dt

0.

Taking

0.02 sec, use the Runge-Kutta method to calculate

v

and

dv

dt

when

0.02, for the data

10 volts,

0.1 farad,

0.5 henry

and

10 ohms.

10.13 Error Analysis

The numerical solutions of differential equations certainly differ from

their exact solutions.

The difference between the computed value yi and the

true value y

xi
)

at any stage is known as the

total error.

The total error at

any stage is comprised of

truncation error

and

round-off error.

The most important aspect of numerical methods is to minimize the

errors and obtain the solutions with the least errors. It is usually not pos-

sible to follow error development quite closely. We can make only rough

estimates. That is why, our treatment of error analysis at times, has to be

somewhat intuitive.

In any method, the truncation error can be reduced by taking smaller

sub-intervals. The round-off error cannot be controlled easily unless the

computer used has the double precision arithmetic facility. In fact, this er-

ror has proved to be more elusive than the truncation error.

The truncation error in Euler™s method is

,
2

hyn

i.e., of (

) while

that of modified Euler™s method is

hy

.i.e., of (

474

UMERIC AL
M

ETHODS

IN

NGINEERING

AND

CIENCE

Similarly in the fourth order of the Runge-Kutta method, the trunca-

tion error is of

).

In the Milne™s method, the truncation error

due to predictor formula

14

45

yh

and due to corrector formula

5

90

yh

i.e.,

the truncation error in Milne™s method is also of

).

Similarly the error in the Adams-Bashforth method is of the fifth order.

Also the predictor error

and the corrector error

are so related that

19

T
P

Π251

The

relative error

of an approximate solution is the ratio of the total

error to the exact value

. It is of greater importance than the error itself for

if the true value becomes larger, then a larger error may be acceptable. If

the true value diminishes, then the error must also diminish otherwise the

computed results may be absurd.

EXAMPLE 10.32

Does applying Euler™s method to the differential equation

dy

dx

x, y

)
,y

estimate the total error?

When

x, y

Ð y, y

(0)

compute this error neglecting the round-off

error.

Solution:
We know that Euler™s solution of the given differential equation is

n+

hf

) where

0
nh

i.e., y

n+

hy

(1)

Denoting the exact solution of the given equation at

by

) and
expanding

) by Taylor™s series, we obtain

() ()(),,1

2!

nnnnnnn

yxyxhyxyxx

(2)

The truncation error

(
x

)Œ

(1/2)

Thus the truncation error is of

) as

0.

N
UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

475

To include the effect of round-off error

, we introduce a new ap-

proximation

which is defined by the same procedure allowing for the

round-off error.
11

nn nnn

yyhfxyR

(3)

The total error is defined by

111

()Œ

nnn

Eyx

[(2) Π(3)]

11

()()(,)

2!

()()(,)

nnnnnn

nnnnnn

n
h

yxhyxyyhfxyR

yxy hhxfxyTR

(4)

Assuming continuity of

and using Mean-Value theorem, we have

)] Œ

n
,

)Œ

fy

), where

lies between

(
x

and

(4) takes the form

111

()Œ1

[,

nnynnnn

EyxhfxTR

or

n
[1

hf

)]

n+

(5)

This is the

recurrence formula

for finding the total error. The first terms

on the right-hand side is the

inherited error, i.e.,

the propagation of the er-

ror from the previous step

to

) We have

dy

dx

(0)

1.

Taking

0.01 and applying (1) successively, we obtain

y

(0.01)

0.01(Π1)

0.99

(0.02)

0.99

0.01 (Π0.99)

0.9801

(0.03)

0.9703,

(0.04)

0.9606

The truncation error

(1/2)

(
)

0.00005

5 × 10

Œ5

)[

dy

dx

is Πve]

i.e., T

5 × 10

Œ5

(0)

5 × 10
Œ5

5 × 10

Œ5

(0.01)

5 × 10

Œ5

(0.99) < 5 × 10Œ5

5 × 10

Œ5

(0.02)

5 × 10

Œ5

(0.9801) < 5 × 10Œ5

4
5 ×10

Œ5

(0.03)

5 × 10

Œ5

(0.9703) < 5 × 10Œ5 etc.

476

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

Also 1
hf

0.01(Π1)

0.99.

Neglecting the round-off error and using the above results, (5) gives

0,

0(0.99)
T

5 × 10

Œ5

0.00005

(0.99)

< 5 × 10Œ5

5 × 10

Œ5

0.0001

3
E

(0.99)

< 10Œ4

5 × 10

Œ5

0.00015

(0.99)

< 1.5 × 10Œ4

5 × 10

Œ5

0.0002 etc.
Obs.

The exact solution is y

Ðx

Actual error in y(0.03)

Ð0.03

Ð 0.9703

0.00014

and actual error in y(0.04)

Ð0.04

Ð 0.9606

0.00019.

Clearly the total error E

agrees with the actual error in y

(0

04).

10.14 Convergence of a Method

Any

numerical method for solving a differential equation is said to be

convergent if the approximate solution y

n
 approaches the exact solution

as h tends to zero provided the rounding errors arise from the initial

conditions approach zero

. This means that as a method is continually re-

fined by taking smaller and smaller step-sizes, the sequence of approximate

solutions must converge to the exact solution.

Taylor™s series method is convergent provided

) possesses enough

continuous derivatives. The Runge-Kutta methods are also convergent un-

der similar conditions. Predictor corrector methods are convergent if

,
y

satisfies the

Lipschitz condition, i.e.,

(,)Π,

|()|

Œ,

fxyfxy kyy

being a constant, then the sequence of approximations to the numeri-

cal solution converges to the exact solution.

10.15 Stability Analysis

There is a limit to which the step-size

can be reduced for controlling

the truncation error, beyond which a further reduction in

will result in

the increase of round-off error and hence increase in the total error. This

behavior of the error bound is shown in Figure 10.3.

In such situations, we have to use stable methods so that an error intro-

duced at any stage does not get magnified.

NOTE

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

477

A method is said to be

stable

if it produces a bounded solution which

imitates the exact solution. Otherwise it is said to be

unstable.

If a method

is stable for all values of the parameter, it is said to be

absolutely
or

uncon-

ditionally stable

. If it is stable for some values of the parameter, it is said to

be

conditionally stable

The Taylor™s method and Adams-Bashforth method prove to be rela-

tively stable. Euler™s method and the Runge-Kutta method are condition-

ally stable as will be seen from Example 10.23.

The Milne™s method is however, unstable since when the parameter is

negative, each of the errors is magnified while the exact solution decays.

Total error

Truncatio

error

Round-off error

Optimum,

Error

FIGURE 10.3

EXAMPLE 10.33

Does

applying Euler™s method to the equation

dy

dx

y, given y

determine its stability zone? What would be the range of stability when

Π1?

Solution:

We have

x
0

(1)

By Euler™s method,

Œ1

hy

Œ1

Œ1
hy

Œ1

(1

Œ1

[by (1)

478

UMERIC AL

ETHODS

IN

NGINEERING
AND

CIENCE

Œ1

(1

Œ2

.......................................

(1

1
 y

(1

Multiplying all these equations, we obtain

(1

(2)

Integrating (1), we get

ce

Using
y

ce

xÐx
0)

In particular, the exact solution through (

) is

00

xn xnh

yyeye

nh
]

or

00

()1

hn

yyeyh

(3)

Œ2

h)

h)

Unstable

Stable
Œ1

FIGURE 10.4

Clearly the numerical solution (2) agrees with exact solution (3) for

small values of

. The solution (2) increases if |1

| > 1.

Hence |1

|< 1 defines a stable zone.

When

is real

, then the method is stable if |1

|<1

i.e.

Œ2<

<0

When
is complex

ib

), then it is stable if

|1

ib

|<1

i.e.

(1

ah

bh

2
 <1

i.e.,

1)2

< 1, [where

ah

bh

.]

i.e.,

lies within the unit circle shown in Figure 10.4.

When

is imaginary

ib
), |1

1, then we have a

periodic-stability

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

479

Hence Euler™s method is absolutely stable if and only if

(
i

) real

:Œ2<

0.

ii

) complex

lies within the unit circle (Figure 10.4),

i.e.,

Euler™s

method is conditionally convergent.

When

Œ 1, the solution is stable in the range Œ 2 < Œ

<0

i.e.

0<

< 2.

Exercises 10.7

1.

Show that the approximate values

y

, obtained from

with

(0)

by Taylor™s series method, converge to the exact solution for

tending to

zero.

2.

Show that the modified Euler™s method is convergent.

3.

Starting with the equation

, show that the modified Euler™s

method is relatively stable.

4.
Apply the fourth order Runge-Kutta method to the equation

dy

dx

and show that the range of absolute stability is

Π2.78 <

< 0.

5.

Find the range of absolute stability of the equation

10

0,
y

(0)

1, using

) Euler™s method, (

) Runge-Kutta method.

6.

Show that the local truncation errors in the Milne™s predictor and cor-

rector formulae are

14

45

hy

and

90

hy

respectively.

10.16 Boundary Value Problems

Such a problem requires the solution of a differential equation in a

region

subject to the various conditions on the boundary of

. Practical

applications give rise to many such problems. We shall discuss two-point

linear boundary value problems of the following types:

dydy

xxyx

dx

dx

with the conditions

)
a

480

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

b
.

ii

dy

xyx

dx

with the conditions

and
y

There exist two numerical methods for solving such boundary value

problems. The first one is known as the

finite difference method

which

makes use of finite difference equivalents of derivatives. The second one is

called the

shooting method

which makes use of the techniques for solving

initial value problems.

10.17 Finite-Difference Method

In this method, the derivatives appearing in the differential equation

and the boundary conditions are replaced by their finite-difference ap-

proximations and the resulting linear system of equations are solved by any

standard procedure. These roots are the values of the required solution at

the pivotal points.

The

finite-difference approximations to the various derivatives are

derived as under:

If

) and its derivatives are single-valued continuous functions of

then by Taylor™s expansion, we have

23

2!3!

hh

yxhyxhyxyxyx
(1)

and

23

2!3!

hh

yxhyxhyxyxyx

(2)

Equation (1) gives

yxyxhyxy x

i.e.,

()

yxyxhyx Oh

h
which is the

forward difference approximation of y

) with an error of the

order

Similarly (2) gives

()

yxyxyxhOh

UMERIC AL

OLUTION

OF

RDINARY
 D

IFFERENTIAL

QUATIONS

481

which is the

backward difference approximation of y

) with an error of

the order

Subtracting (2) from (1), we obtain

()

ŒŒ()

yyxhy

xx
h

hOh

which is the

central-difference approximation of y

) with an error of the

order

. Clearly this central difference approximation to

) is better than

the forward or backward difference approximations and hence should be

preferred.

Adding (1) and (2), we get

()()

Œ2(Œ) ()
yyxhyyx

hOh

xx

which is the

central difference approximation of y

). Similarly we can de-

rive central difference approximations to higher derivatives.

Hence

the working expressions for the central difference approxima-

tions to the first four derivatives of y

are as under

1Œ1

()

iii

h
yyy

(3)

1Œ1

iiii

yyyy

(4)

21Œ1

Œ2

Œ2Œ

iiiii

yyyyy

(5)
2

1Œ1Œ2

Œ46Œ4

iv

iiiiii

yyyyyy

(6)

Obs.

The accuracy of this method depends on the size of the

sub-interval h and also on the order of approximation. As we

reduce h, the accuracy improves but the number of equations to

be solved also increases.

EXAMPLE 10.34

Solve the equation

with the boundary conditions

y

(0)

(1)

0.

Solution:

We divide the interval (0, 1) into four sub-intervals so that

1/4 and

the pivot points are at

0,

1/4,

2
1/2,

3/4, and

1.

NOTE

482

UMERIC AL

ETHODS

IN

NGINEERING

AND

S
CIENCE

Then the differential equation is approximated as

11

Œ2

iiiii

yyyxy

or 16

Π33

16

Œ1

x
i

1, 2, 3.

Using

0, we get the system of equations

2132132

13

16Œ3316Œ3316 ;Œ 3316

24

yyyyyyy

Their solution gives

1
Π0.03488,

Π0.05632,

Π0.05003.

Obs.

The exact solution being

sinh

(),

sinh1

yxx

the error at

each nodal point is given in the table below:

xComputed value y

Exact value y

)
Error

0.25Π0.03488Π0.035050.00017

0.5Π0.05632Π0.056590.00027

0.75Π0.05003Π0.050280.00025

EXAMPLE 10.35

Using the finite difference method, find

(0

25)

,y

(0

5),

and

(0

75)

sat-

isfying the differential equation

2
,

dy

yx

dx

subject to the boundary condi-

tions

(0)

,y

(1)

Solution:

Dividing the interval (0, 1) into four sub-intervals so that

0.25 and

the pivot points are at

0
0,

0.25,

0.5,

0.75, and

1.

The given equation

x
)

, is approximated as

11

Œ2

iiiii

yyyyx

or 16

Π31

16

Œ1
x

Using

0 and

2, (

) gives the system of equation,

1) 16

Π31

1
0.25; (

ii

2) 16

Π31

16

0.5 (

iii

3) 32 Π31

3
16

0.75,

i.e.,

Π31

16

Π31.25 (

iv

NOTE

UMERIC AL

OLUTION

OF

O
RDINARY

IFFERENTIAL

QUATIONS

483

Solving the equations (

ii

), (

iii

), and (

iv

), we get

0.5443,

1.0701,

3
 1.5604

Hence

(0.25)

0.5443,

(0.5)

1.0701,

(0.75)

1.5604

EXAMPLE 10.36

Determine values of

at the pivotal points of the interval

(0

1)

if

satisfies the boundary value problem

iv
81

81

,y

(0)

(1)

(0)

(1)

(Take

n
3).

Solution:

Here

1/3 and the pivotal points are

0,

1/3,

2/3,

1.

The corresponding

-values are
y

0),

0).

Replacing

iv

by its central difference approximation, the differential

equation becomes

21Œ1Œ2

Œ46

1
Œ48181

iiiiiii

yyyyyyx

or

Œ4

Œ4

Œ1

i
Œ2

1, 2

At

1,

Œ4

Œ4

0
y

Œ1

1/9

At

2,

Œ4

Œ4

4/9

Using

y
0

0, we get Π4

Œ1

1/9 (

2
Œ4

4/9 (

ii

Regarding the conditions

0, we know that

Œ1

(Œ)

iii

yiyyy

h
At

0,

9(

Œ2

Œ1) or

Œ1

0
y

0] (

iii

At

3,

9(

Œ2

) or

y
4

0] (

iv

Using (

iii

), the equation (

) becomes

Π4

2
6

1/9 (

Using (

iv

), the equation (

ii

) reduces to

Œ4

4/9 (

vi

Solving (

) and (

vi
), we obtain

11/90 and

7/45.

Hence

(1/3)

0.1222 and

(2/3)

0.1556.

484

UMERIC AL

ETHODS
IN

NGINEERING

AND

CIENCE

EXAMPLE 10.37

The deflection of a beam is governed by the equation

81(),

dy

yx

dx

where

) is given by the table

1/3 2/3 1

x
)

81 162 243

and boundary condition

(0)

(0)

(1)

(1)

Evaluate the de-

flection at the pivotal points of the beam using three sub-intervals.

Solution:

Here

1/3 and the pivotal points are

x
0

0,

1/3,

2/3,

1.

The corresponding

-values are

0),

2
,

The given differential equation is approximated to

21Œ1Œ

Œ46Œ4

81

iiiiiii

yyyyyyx

At

1,

Œ4

7
y

Œ4

Œ1

1(

At

2,

Œ4

Œ4

y
1

2(

ii

At

3,

Œ4

Œ4

2
y

3(

iii

We have

0(

iv

Since

1Œ1

ii

yi

yy

for

0,
01 1

0 ŒŒ1i.e., Œ1

yyyy

Since

Œ1

Œ2

iii

yiyy

for

i
3,

2432

343

0Œ2,i.e., 2Œ

yyyyyy

vi

Also

21Œ1

Œ2

Œ2Œ

iiiii

yyyyy
for

3,

354

21

ŒŒ

022

yyyyy

i.e., y

Œ2

2
y

vii

Using (

iv

) and (

), the equation (

) reduces to

Œ4

1(

viii

)
N

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

485

Using (

iv

) and (

vi

), the equation (

ii

) becomes

3
3

Œ2

1(

ix

Using (

vi

) and (

vii

), the equation (

iii

) reduces to

Œ4

2
y

3(

Solving (

viii

), (

ix

), and (

), we get

8/13,

22/13,

37/13.

Hence
y

(1/3)

0.6154,

(2/3)

1.6923,

(1)

2.8462.

10.18 Shooting Method

In this method, the given boundary value problem is first transformed

to an initial value problem. Then this initial value problem is solved by Tay-

lor™s series method or Runge-Kutta method, etc. Finally the given bound-

ary value problem is solved. The approach in this method is quite simple.

Consider the boundary value problem

),
y

A, y

(1)

One condition is

and let us assume that

m
which rep-

resents the slope. We start with two initial guesses for

, then find the cor-

responding value of

) using any initial value method.

Let the two guesses be

so that the corresponding values of

are

,
b

) and

). Assuming that the values of

and

) are lin-

early related, we obtain a better approximation

for

from the relation:

10

21

10
,

,,

mm

mm

ybymb

ymbym b

This gives

2110

10

,,

ymbyb

mmmm

ymbym b

(2)

We now solve the initial value problem

),

y
(

A, y

and obtain the solution

).

To obtain a better approximation

for

, we again use the linear rela-

tion (2) with [

m

)] and [

)]. This process is repeated until

the value of

,
b

) agrees with

) to desired accuracy.

486

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

(x)

0
x

=a

=b

(y

(a) = m

= y(m

,b)

2
= y(m

,b)

B=y(b)

=y(m

,b)

o
t

(y

(a) = m

FIGURE 10.5

Obs.

This method resembles an artillery problem and as such

is called the shooting method (Figure 10.5). The speed of

convergence in this method depends on our initial choice of

two guesses for m. However, the shooting method is quite slow

in practice. Also this method is quite tedius to apply to higher

order boundary value problems.

EXAMPLE 10.38

Using the shooting method, solve the boundary value problem:

x
)

,y

(0)

and

(1)

17

Solution:

Let the initial guesses for

(0)
m

be

0.8 and

0.9.

Then

),

(0)

0 gives

(0)
m y

(0)

(0)

(0)

(0)

iv

(0)

(0)

yv

(0)

y
(0)

yvi

(0)

yiv

(0)

and so on.

Putting these values in the Taylor™s series, we have

23

357

()0000

2!3!

61205040

xx

yxyxyyy

xxx

mx
NOTE

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

487

(1)

(1

0.1667
0.0083

0.0002

(1.175)

For

0.8,

, 1)

0.8 × 1.175

0.94

For

0.9,

(
m

, 1)

0.9 × 1.175

1.057

Hence a better approximation for

, i.e.,

is given by

2110

10

,11

,1,1

ym y

mmmm

ym ym

1.0571.175

0.90.1 0.90.100851.00085

1.0570.94
which is closer to the exact value of

(0)

0.996

We now solve the initial value problem

),

(0)

0,

(0)

Taylor™s series solution is given by

(
m

, 1)

(1.175)

1.1759

Hence the solution at

1 is

1.176 which is close to the exact value

of

(1)

1.17.

Exercises 10.8

1.

Solve the boundary value problem for

0.5:

2
10,0 10.

dy

yyy

dx

(Take

4)

2.

Find an approximate solution of the boundary value problem:

8(sin

0, 0

1,

(0)

y
(1)

1. (Take

4)

3.

Solve the boundary value problem:

xy

0,

(1)

1,

(2)

2. (Take

4)

4.

Solve the equation

Œ4

4
y

, with the conditions

(0)

0,

(1)

Π2, taking

4.

5.

Solve the boundary value problem

Π64

10

0 with

(0)

(1)
0

by the finite difference method. Compute the value of

(0.5) and com-

pare with the true value.

488

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

6.

Solve the boundary value problem

y
xy

2,

(0)

0,

(1)

1.

7.

The boundary value problem governing the deflection of a beam of

length three meters is given by

12

24,00330.
93

dy

yx xyyyy

dx

The beam is built-in at the left end (

0) and simply supported at the

right end (

3).

Determine

at the pivotal points

1 and

2.

8.

Solve the boundary value problem,

81729001 10.Us e3
dy

yxyyyyn

dx

9.

Solve the equation

iv

, subject to the boundary conditions

(0)

(0)

0 and

y
(1)

2,

(1)

0. (Take

5).

10.

Apply shooting method to solve the boundary value problem

,00andy11.1752.

dy

yy

dx

11.

Using shooting method, solve the boundary value problem

6,01,0.50.44

2
dy

yyy

dx

10.19 Objective Type of Questions

Exercises 10.9

Select the correct answer or fill up the blanks in the following questions:

1.

Which of the following is a step by step method:

) Taylor™s (

) Adams-Bashforth

) Picard™s (

) None.

2.

The finite difference scheme for the equation 2

5 is ...... .
3.

If

(0)

1 and

(1)

/2, then by Picard™s method, the

value of

(2)

(
x

) is ......

UMERIC AL

OLUTION

OF

RDINARY

IFFERENTIAL

QUATIONS

489

4.

The iterative formula of Euler™s method for solving

(
x

) with

, is ....... .

5.

Taylor™s series for solution of first order ordinary differential equations

is ......... .

6.

The disadvantage of Picard™s method is ...... .

7.

Given

,
y

, Milne™s corrector formula to find

for

dy

dx

), is ...... .

8.

The second order Runge-Kutta formula is ...... .

9.

Adams-Bashforth predictor formula to solve

y
f

x, y

), given

is .... .

10.

The

Runge-Kutta method is better than Taylor™s series method because

...... .

11.

To predict Adam™s method atleast ...... values of

, prior to the desired

value, are required.

12.

Taylor™s series solution of

y

xy

0,

(0)

1 upto

is ...... .

13.

If

dy

dx

is a function of

alone, the fourth order Runge-Kutta method

reduces to .......

14.

Milne™s Predictor formula is ....... .

15.

Adam™s Corrector formula is ....... .

16.

Using Euler™s method,

dy
/

dx

Œ2

)/

(0)

1; gives

(0.1)

..... .

17.

dydy

yy

dx

dx

is equivalent to a set of two first order differential

equations ...... and ...... .

18.

The formula for the fourth order Runge-Kutta method is ...... .

19.

Taylor™s series method will be useful to give some ...... of Milne™s

method.

20.

The names of two self-starting methods to solve

) given

are ...... .
21.

In the derivation of the fourth order Runge-Kutta formula, it is called

fourth order because .....

490

UMERIC AL

ETHODS

IN

NGINEERING

AND

CIENCE

22.

If

x
Œ

(0)

1 then by Picard™s method, the value of

(1)

(1) is ...... .

) 0.915 (

) 0.905 (

) 1.091 (

) none.

23.

The finite difference formulae for

) and

(
x

) are ...... .

24.

If

(0)

1, then by Euler™s method, the value of

(1) is

) 0.99 (

) 0.999 (

) 0.981 (

) none.

25.

Write down the difference between initial value problem and boundary

value problem ..... .

26.

Which of the following methods is the best for solving initial value prob-

lems:

) Taylor™s series method

) Euler™s method

) Runge-Kutta method of the fourth order

) Modified Euler™s method.

27.

The finite difference scheme of the differential equation

0 is

.....

28.

Using the modified Euler™s method, the value of

y

(0.1) for

,01

dy

xyy

dx

is

) 0.809 (

) 0.909 (

) 0.0809 (

) none.

29.

The multi-step methods available for solving ordinary differential equa-

tions are ...... .

30.

Using the Runge Kutta method, the value of

(0.1) for

y
x

Œ2

(0)

1, taking

0.1, is ......

) 0.813 (

) 0.825 (

) 0.0825 (

) none.

31.

In Euler™s method, if

is small the method is too slow, if

is large, it
gives inaccurate value. (True or False)

32.

Runge-Kutta method is a self-starting method. (True or False)

33.

Predictor-corrector methods are self-starting methods. (True or False)