Chapter 8.

Numerical Solution of Ordinary Differential

Equations

8. 1. Initial- and boundary-value problems

For an n-th order ordinary differential equation, n conditions are required to solve the
equation.

Example-1: Consider the vibration of a spring with a mass m

The governing equation is

d 2x
m 2 = − Kx
dt K
2
0
d x x
2
= −K m
dt m
x
with the initial off balance displacement:

x(t = 0) = x0
and initial speed:
dx(t = 0)
v(t = 0) = = v0
dt
The problem is called an initial-value problem (which is associated with time t ).

99
Example-2: Now consider the deflection of the sea boat mast
The governing equation is

f EI

d 2 y M ( z) f
2
= = ( L − z )2 L
dz EI 2 EI
z

y
f: wind force, EI: the bending stiffness
The conditions (boundary conditions) at the built-in end are:

y ( z = 0) = 0

dy
( z = 0) = 0
dz

The problem is called a boundary-value problem (which is associated with the

edge of the space domain).

100
8.2. First-order ordinary differential equations

Euler’s method
Consider a first-order equation
d
= f ( x,  ) (1)
dx
with
 ( x0 ) =  0


 (b) = ?
d
( x = a)
dx
 (a)

x

a b
x
Taylor series

d (a ) d 2 (a) x 2
b = a + x + +
dx dx 2
ignoring the second and higher order terms, we have
d (a)
b = a + x = a + f [a,  (a )]x
dx
We use the tangent at point a to approximate the function/curve.
Now we divide the solution domain into a number of small grid cells.

101
 3 4
1
2
numerical solution
0
true solution
x x x x
x0 x1 x2 x3 x4 x
0 1 2 3 4
x1 = x0 + x, 1 = 0 + f ( x0 , 0 )x

x2 = x1 + x, 2 = 1 + f ( x1 , 1 )x

x3 = x2 + x, 3 = 2 + f ( x2 , 2 )x

x j +1 = x j + x,  j +1 =  j + f ( x j ,  j )x (2)

This is called Euler’s method. The accuracy of the Euler’s method is first -order.

Example: to solve the following equation

d
= x +  = f ( x,  )  (0) = 0
dx ,
we choose x = 0.2

X=0 0.2 0.4 0.6 0.8

x
j=0 1 2 3 4

From equation (2) we have

d
 j +1 =  j + f ( x j ,  j )  x =  j + 0.2 (x = x j )
dx
=  j + ( x j +  j )  0.2 = 0.2 x j + 1.2 j

102
The exact solution is  ( x) = e x − x − 1

 j +1 = 0.2 x j + 1.2 j

j xj j  j +1 exact  j error

0 0 0 0 0 0

1 0.2 0 0.04 0.021 0.021

2 0.4 0.04 0.128 0.092 0.052

3 0.6 0.1280 0.274 0.222 0.094


j x analytical error

0 0 0 0 0
1 0.2 0 0.021 0.021
2 0.4 0.04 0.092 0.052
3 0.6 0.1280 0.222 0.094
4 0.8 0.2736 0.4255 0.1519
5 1.0 0.4883 0.7183 0.2300

103
Improved Euler’s Method

d
to solve = f ( x,  )
dx
This is a two-step method:

(i) to compute the ‘predicted’ value of  j +1 using an intermediate step

 j +1 =  j + f ( x j ,  j )x (3)


f ( x j +1 ,  j +1 )

1
[ f ( x j ,  j ) + f ( x j +1 ,  j +1 )]
2
f ( x j , j )

xj x j +1 x

(ii) To compute the ‘corrected’ value of  j +1 using

1
 j +1 =  j + [ f ( x j ,  j ) + f ( x j +1 ,  j +1 )]x (4)
2
It can be shown that the improved Euler’s method is a second-order method
(error  O ( x 2 ) ).

104
Example: to solve the same problem:
d
= x + with  (0) = 0
dx
using the improved Euler’s method

from eq. (3)  j +1 =  j + f ( x j , j )x =  j + ( x j +  j )  0.2 = 0.2 x j + 1.2 j

1
from eq. (4)  j +1 =  j + [ f ( x j , j ) + f ( x j +1 ,  j +1 )]x
2

1
=  j + [( x j +  j ) + ( x j +1 +  j +1 )]  0.2
2
1
=  j + [( x j +  j ) + ( x j + 0.2 + 0.2 x j + 1.2 j )]  0.2
2
= 0.02 + 0.22x j + 1.22 j

 j = 0.02 + 0.22 x j −1 + 1.22 j −1

j x j exact error

0 0 0 0 0
1 0.2 0.020 0.0214 0.0014
2 0.4 0.0884 0.0918 0.0034
3 0.6 0.2158 0.2221 0.0063
4 0.8 0.4153 0.4255 0.0102
5 1.0 0.7027 0.7183 0.0156

The errors are smaller than those derived from the original Euler’s
method.

105
8. 3. Linked (simultaneous) First-order Ordinary Differential Equations
We have studied methods for the numerical solution of equations of the form
dx
= f (t , x) with x(t = t0 ) = x0
dt
It is also possible to have a set of linked first-order equations
dx
= f (t , x, y ) , x(t = t0 ) = x0 (5, a)
dt

dy
= g (t , x, y ), y (t = t0 ) = y0 (5, b)
dt
For example,

dx
= x − y 2 + xt
dt
dy
= 2 x 2 + xy − t
dt

These two equations must be solved simultaneously, or side by side.

Euler’s method for the solution of equations (5.a) and (5.b) takes the form:

x j +1 = x j + tf (t , x j , y j ) (6. a)

y j +1 = y j + tg (t , x j , y j ) (6. b)

The improved Euler’s method then becomes:

xj +1 = x j + tf (t , x j , y j ) (7. a)

y j +1 = y j + tg (t , x j , y j ) (7. b)

t (8. a)
x j +1 = x j + [ f (t j , x j , y j ) + f (t j +1 , xj +1 , y j +1 )]
2
t (8. b)
y j +1 = y j + [ g (t j , x j , y j ) + g (t j +1 , xj +1 , y j +1 )]
2

106
Example: to find the value of x(1.4) which satisfies the following problem:
dx
= x − y 2 + xt , x(t = 1) = 0.5
dt

dy
= 2 x 2 + xy − t , y (t = 1) = 1.2
dt

Solution:

using Euler’s method with t = 0.1

j=0 1 2 3 4

0.1 0.1 0.1 0.1

t
t=1.0 1.1 1.2 1.3 1.4

From Eqs (6.a) and (6.b)

x j +1 = x j + tf (t j , x j , y j ) = x j + 0.1 ( x j − y 2j + x j t j )

y j +1 = y j + tg (t j , x j , y j ) = y + 0.1 (2 x 2 + x y − t )
j j j j j

For j=0 1 (i.e., t=1.0 1.1)

x1 = x0 + 0.1 ( x0 − y02 + x0t0 )

= 0.5 + 0.1 (0.5 − 1.2 2 + 0.5  1.0) = 0.4560

y1 = y0 + 0.1 (2 x02 + x0 y0 − t0 )

= 1.2 + 0.1 (2  0.52 + 0.5  1.2 − 1.0) = 1.210

107
 For j=1 2 (i.e., t=1.1 to 1.2)

x2 = x1 + 0.1 ( x1 − y12 + x1t1 )

= 0.4560 + 0.1 (0.4560 − 1.212 + 0.4560 1.1) = 0.4054

y2 = y1 + 0.1 (2 x12 + x1 y1 − t1 )

= 1.21 + 0.1 (2  0.4560 2 + 0.4560  1.21 − 1.1) = 1.1968

……
x4 = 0.2980

Please complete the calculations !!

Also please use the improved Euler’s method to solve this problem !

The method can be used to solve a set of n simultaneous first-order

ordinary differential equations:

dy1
= f1 ( x, y1 , y2 ,, yn ) y1 ( x = x0 ) = y1,0
dx

dy2 y2 ( x = x0 ) = y2,0
= f 2 ( x, y1 , y2 ,, yn )
dx

…………… …………..

dyn
= f n ( x, y1 , y2 ,, yn ) yn ( x = x0 ) = yn,0
dx

108
8. 4. Numerical Solution of Higher-order Equations
consider
d 2x dx 1
2
+ x 2t − xt 2 = t 2 (9)
dt dt 2

dx
x(0) = 1.2 , (t = 0) = 0.8
dt
If we introduce an additional variable

dx d 2 x d dx dy
y= 2
= ( )=
dt dt dt dt dt

Then we have

dx
=y , x(0) = 1.2
dt

dy 1
+ x 2ty − xt 2 = t 2 , y (0) = 0.8
dt 2
Thus the original single second-order differential equation is replaced by a pair of
linked first-order differential equations. (one 2nd order equation = two 1st order
equations).

This process can be extended to higher-order equations.

(i.e. one nth-order equation = n 1st-order equations)

109
Example:
Find the value of x(0.2) satisfying the initial-value problem

d 3x d 2x dx
3
+ xt 2 + t − t 2 x = 0
dt dt dt
dx d 2x
x(t = 0) = 1 , (t = 0) = 0.5 (t = 0) = −0.2
dt , dt 2

using Euler’s method with step size t = 0.1

Solution
Since this is a third-order equation, we need to introduce two new variables:
dx
y=
dt
dy d 2 x
z= =
dt dt 2
Then the equation is transformed into:

dx
= y = f1 (t , x, y, z ) , x(0) = 1
dt
dy
= z = f 2 (t , x, y, z ) , y (0) = 0.5
dt
dz
= − xtz − ty + t 2 x = f 3 (t , x, y, z ), z (0) = −0.2
dt
Applying Euler’s method

x1 = x0 + tf1 (t0 , x0 , y0 , z0 ) = 1 + 0.1 0.5 = 1.05

y1 = y0 + tf 2 (t0 , x0 , y0 , z0 ) = 0.5 + 0.1 (−0.2) = 0.48
z1 = z0 + tf 3 (t0 , x0 , y0 , z0 ) = −0.2 + 0.1 (0) = −0.2

x2 = x(0.2) = x1 + tf1 (t1 , x1 , y1 , z1 ) = x1 + t  y1

= 1.05 + 0.1 0.48 = 1.098

110
8.5. Boundary-value Problems

The ‘shooting method’

Suppose we wish to solve

d 2x t  dy t
− (1 − ) x = t  dt = (1 − )x + t
2
 5
dt 5  dx
 =y
 dt

with x(1) = 2 , x(3) = −1

If dx / dt (t = 1) = x ' (1) is known, it is an initial value problem. The shooting

method is a form of trial and error in which the boundary-value problem is replaced
by an initial-value problem.

x
3
2
1
0 t
1 2 3
-1

111
Main steps

To assume an initial value of x ' (0) = dx / dt(t = t0 ) (slope)

To solve this initial value problem

To compare the solution at the other boundary

To adjust the value of x’ 0

To repeat the process until agreement is achieved

at the other boundary

112
Finite difference method

Consider

d 2x t
2
− (1 − ) x = t (1)
dt 5

with x(1) = 2 , x(3) = −1

Main steps:

(1) To divide the physical region

t =1 t =3
t t
t
j=0 1 2 j −1 j j +1 N

(2) To replace the derivative terms by finite differences

d 2x d dx = ( x j +1 − x j − x j − x j −1 ) / t x j −1 − 2 x j + x j +1
= ( ) = (2)
dt 2
dt dt t t t 2

Substituting (2) into (1) we get

x j −1 − 2 x j + x j +1 tj (3)
− (1 − )x j = t j
t 2 5
or
'
x j −1 − [2 + t 2 (1 − t j / 5)]x j + x j +1 = t j t 2 (3 )

If we set t = 0.5 , we have 3 internal points

113
 1.5 2.0 2.5 3.0
t =1

j=0 1 2 3 4

From eq. (3) x j −1 − [2 + t 2 (1 − t j / 5)]x j + x j +1 = t j t 2 we have

at j=1, t=1.5

x0 − [2 + 0.52 (1 − 1.5 / 5)]x1 + x2 = 1.5  0.52

j=2, t=2.0

x1 − [2 + 0.52 (1 − 2.0 / 5)]x2 + x3 = 2.0  0.52

j=3, t=2.5

x2 − [2 + 0.52 (1 − 2.5 / 5)]x3 + x4 = 2.5  0.52

Rewriting the above equations

− 2.175 x1 + x2 = 0.375 − x0

x1 − 2.15 x2 + x3 = 0.5

x2 − 2.125 x3 = 0.625 − x 4

Applying the boundary conditions:

x0 = 2 and x4 = −1

114
 − 2.175 x1 + x2 = −1.625 (a)

x1 − 2.15 x2 + x3 = 0.5 (b)

(c)
x2 − 2.125 x3 = 1.625

Elimination: (a) + (b)  2.175

0 + (1 − 4.676) x2 + 2.175 x3 = −1.625 + 1.0875

− 3.676 x2 + 2.175 x3 = −0.5375

(b)’

(b) , + (c)  3.676

0 + (2.175 − 7.8115) x3 = −0.5375 + 5.9735

− 5.6365 x3 = 5.436 (c)’

Now, we have:

− 2.175 x1 + x2 = −1.625 (a)

− 3.676 x2 + 2.175 x3 = −0.5375 (b)’

− 5.6365 x3 = 5.436 (c)’

115
Backward substitution:

From (c)’ x3 = 5.436 /( −5.6365) = −0.964

From (b)’ − 3.676 x2 + 2.175  (−0.964) = −0.5375

x2 = −0.424

From (a) − 2.175 x1 + (−0.424) = −1.625

x1 = 0.552

116