University of Yangon

Department of Mathematics
Second Year Honours Project

Candidate : Mg Satt Paing Soe

Roll No. : 2H Math-4

Title of Project : Numerical Methods for Nonlinear Differential Equations

Supervisor : Dr Aung Moe Khing

Professor
Department of Mathematics
University of Yangon

Abstract

In this project, we first study mumeric methods for ordinary differential equations such
as Euler’s Method, Improved Euler Method and classical Runge-Kutta Method. Differential
equations are used in mathematical modeling problems. Linear differential equations can be
solved analytically but most real life applications are approximate solution. Euler and Runge-
Kutta Methods are explained and illustrated as useful numerical methods for solving nonlinear
differential equations. Accuracy of a numerical method depends on the step size used and
degree of nonlinearity of the equations. Although better accuracy can be obtained with
smaller step size, this takes more computational effort and time. Authors use the MATLAB
software to overcome this challenge and to avoid computational error. The Runge-Kutta
Method is more applicable and accurate for diverse classes of differential equations.

Second Year Honours 21.3.2024

Chapter 1

Methods of First Order ODEs

1.1 Introdrction
An ordinary differential equation (ODE) of the first order is of the form F (x, y, y ′ ) = 0 and
can be written in the explict form y ′ = f (x, y). An initial value problem for the equation is
of the form
(1) y ′ = f (x, y) , y(x0 ) = y0

where x0 and y0 are given, y ′ is a function of the independent variable x and the dependent
variable y. We assume that the problem has a unique solution on some open interval a < x < b
containing x0 . From calculus we know that the derivative y ′ (x) of y(x) is the slope of y(x).
Hence a solution curve of (1) that passes through a point (x0 , y0 ) have, at that point, the
slope y ′ (x0 ) equal to the value of f at that point;

Figure 1.1: Approximating y(x1 ) using a tangent line

1
 y ′ (x0 ) = f (x0 , y0 )

Using this fast, we can develop numeric methods for obtaining approximate solutions of ODEs
(1).

1.2 Euler’s Method

Figure 1.2: Euler’s Method for approximating the solution to the initial value problem

Suppose we wish to approximate the solution (1) IVP problem at x = x1 = x0 + h where

h is small. The idea behind Euler’s Method is to use tanget line equation through (x0 , y0 ) to
obtain such an appromation. (See Figure 1.2)
The equation of the tangent line through (x0 , y0 ) is

y(x) − y0 = m(x − x0 )

y(x) = y0 + m(x − x0 )

where m is the slope of the curve at (x0 , y0 ). From (1), m = f (x0 , y0 ), So

y(x) = y0 + f (x0 , y0 )(x − x0 )

2
Setting x = x0 yields approximate solution value at x1 namely

y1 = y0 + f (x0 , y0 )(x1 , x0 )

y1 = y0 + hf (x0 , y0 )

We can use the same idea, at x2 , tangent line to the solution curve through (x1 , y1 )

y2 = y1 + hf (x1 , y1 )

Continuing in this manner, we obtain

yn+1 = yn + hf (xn , yn ), n = 0, 1

to the solution to the initial value problem at the point xn+1 = xn + h.

In Summary, Euler’s Method for approximating the solution to IVP

y ′ = f (x, y), y(x0 ) = y0

at the points xn+1 = x0 + nh, (n = 0, 1, ...) is

yn+1 = yn + hf (xn , yn ), n = 0, 1, ... (2)

Example 1.2.1. Consider the initial value problem of non-linear equation

y′ = 1 + y2, y(0) = 0

Use Euler’s Method with (a) h = 0.1 and h = 0.2 to obtain an approximation to y(1). Given
that the exact solution is y(x) = tanx at (0, 0). Compare the error.
Solution:

y′ = 1 + y2, y(0) = 0, x0 = 0, y0 = 0

f (x, y) = 1 + y 2 , f (xn , yn ) = 1 + yn 2

3
(a) h = 0.1
xn = xn−1 + h

1 1
Number of iterlation, n = = = 10
h 0.1

yn+1 = yn + 0.1(1 + yn 2 )

Hence,
x = 0.1,
y1 = y0 + 0.1(1 + y0 2 ) = 0 + 0.1(1 + 02 ) = 0.1

x = 0.2,
y2 = y1 + 0.1(1 + y1 2 ) = 0.1 + 0.1(1 + 0.12 ) = 0.201

Continuing in this manner, we generate the approximations in Table 1.1

(b) h = 0.2
1
yn+1 = yn + 0.2(1 + yn 2 ), n= =5
0.2

Hence,
x = 0.2,
y1 = y0 + 0.2(1 + y0 2 ) = 0 + 0.2(1 + 02 ) = 0.2

x = 0.4,
y2 = y1 + 0.2(1 + y1 2 ) = 0.1 + 0.2(1 + (0.2)2 ) = 0.0.408

Continue to y5 , we generate the approximation given in Table 1.1

4
 x h = 0.1, yn h = 0.2, yn Extract(6D) h=0.2, h= 0.4,
Error Error
0.0 0.000000 0.000000 0.000000 0.000000 0.000000
0.1 0.100000 0.100335 0.000335
0.2 0.201000 0.200000 0.202710 0.001710 0.002710
0.3 0.305040 0.309336 0.004296
0.4 0.414345 0.408000 0.422793 0.008448 0.0147930
0.5 0.531513 0.546302 0.014789
0.6 0.659764 0.641293 0.684137 0.024373 0.042844
0.7 0.803293 0.842288 0.038996
0.8 0.967821 0.923544 1.029639 0.061818 0.106095
0.9 1.161488 1.260158 0.098670
1.0 1.396394 1.294131 1.557408 0.161014 0.263277

Table 1.1: The results of applying Euler’s Method

with h = 0.1 to the initial value Problem in Example 1.1

Figure 1.3: The two appromation curves obtained using Euler’s Method and exact solution
of example 1.1

In this figure 1.3, we see that the smaller step has led to a better approximation.

5
1.3 Drivation for Euler’s Method
We briefy introduced Euler’s Method with an example. In this section, we shall develop
Euler’s method by paying close attention to the derivation that uses Taylor’s formula from
calculus to approximate the solution to a first-order ODE at points that are a distance h
apart.
The Taylor series

h2
(2) y(x + h) = y(x) + hy ′ (x) + y”(x) + ...
2
This formula is the key idea that let us develop Euler’s method. Let us start by deriving
Euler’s method. For small h the higher powers h2 , h3 , ... in (2) are very small. Dropping all
of the gives the crude approximation.

y(x + h) ≈ y(x) + hy ′ (x)

= y(x) + hf (x, y)

and the corresponding Euler Method

yn+1 = yn + hf (xn , yn ), (n = 0, 1, ...)

Geometrically, this is an approximation of the curve of y(x) by a polygon whose first side is
tangent to this curve at x0 . (See Figure 1.2)

1.4 Error of the Euler Method

Taylor’s formula with remainder has the form

1
y(x + h) = y(x) + hy ′ (x) + h2 y”(η), x≥η ≥x+h
2

It shows that the error in each step or local error in the Euler Method is proportional to h2 ,
written O(h2 ). Now, over a fixed x interval, the number of steps is proportional to h1 . Hence
global error is proportional to h2 ( h1 ) = h. Hence, the Euler Method is called a first-order
method.

6
1.5 Improved Euler Method(Heun’s Method)

Figure 1.4: Heun’s Method for approximating the solution to the initial value problem

The method that we consider here is also called a predictor-corrector method. The idea
is to use the formula from Euler’s method to obtain a first approximation to the solution
y(xn+1 ). We denote this approximation by yn+1 , so that first the predictor
(3a)
∗
yn+1 = yn + hf (xn , yn )

which is an auxiliary value, and then we now improve (or correct) this approximation by one
more applying Euler’s Method.
∗ ). This gives the
We use the average of the solution curves through (xn , yn ) and (xn+1 , yn+1
corrector
(3b)
1 ∗
yn+1 = yn + h[f (xn , yn ) + f (xn+1 , yn+1 )]
2
In each step, we predict a value (3a) and then we correct it by (3b)
In summmary, the Improved Euler Method for approximating the solution to the initial value
problem
y ′ = f (x, y), y(x0 ) = y0

at the points xn+1 = x0 + nh (n = 0, 1, ...) is

1
yn+1 = yn + [k1 + k2 ]
2

using the notations,

k1 = hf (xn , yn ) (3a)
∗
k2 = hf (xn+1 , yn+1 ) (3b)

where
∗
yn+1 = yn + hf (xn , yn ), n = 0, 1, ...

7
1.6 Derivation for Heun’s Method
Recall First Order derivative of the given function f (x, y)
Using Chain rule,
δf δf dy δf δf
f ′ (xn , yn ) = + = + f
δx δy dx δx δy
Let

yn+1 = yn + ϕ(x, y, h)h

ϕ = a1 k1 + a2 k2 + ... + an kn
k1 = f (xn , yn )
k2 = f (xn + p1 , yn + q11 k1 h)
.
.
.
kn = f (xn + pn−1 h, yn + q(n−1) k1 h + q(n−2) k2 h + ... + q(n−2)(n−1) k(n−1) h)

For First Order Runge Kutta, it is actually the Euler’s Method.

For Second Order Runge Kutta,

yn+1 = yn + (a1 k1 + a2 k2 )h
k1 = f (xn , yn )
k2 = f (xn + p1 , yn + q11 k1 h)

A Taylor Series analysis can be used to determine a1 , a2 , p1 , q11 ,

f ′ (xn , yn ) 2
yn+1 = yn + f (xn , yn )h + h + O(h3 )
2
δf δf h2
= yn + f (xn , yn )h + + f + O(h3 ) ........(ii)
δx δy 2

Taylor series of k2 around (xn , yn )

δf δf
k2 = f (xn + p1 , yn + q11 k1 h) = f (xn , yn ) + p1 h + q11 k1 h + O(h2 )
δx δy

8
Using this in eq(i), we generate

δf δf
yn+1 = yn + [a1 f (xn , yn ) + a2 (f (xn , yn ) + p1 h + q11 k1 h + O(h2 ))]h
δx δy
δf δf
= yn + [a1 f (xn , yn ) + a2 f (xn , yn )]h + [a2 p1 + a2 q11 f (xn , yn ) ]h2 + O(h3 ) ........(iii)
δx δy

Compare eq(ii) and eq(iii)

a1 + a2 = 1
1
a2 p 1 =
2
1
a2 q11 =
2

4 unknowns and 3 equations. We choose one of them arbitrary and calculate the others. This
means there are many different 2nd Order Runge-Kutta Methods.
Select a2 = 12 ⇒ a1 = 12 , p1 = 1, q11 = 1

1 1
yn+1 = yn + ( k1 + k2 )h, k1 = f (xn , yn ), k2 = f (xn + h, yn + hk1 )
2 2

This is Heun’s Method.

Example 1.6.1. Apply the improved Euler’s method with h = 0.2 to determine an approxi-
mation to the solution to the initial value problem

y′ = 1 + y2, y(0) = 0, at x=1

For h = 0.1,
n = 0, 1, ..., N − 1, N = 10

xn+1 = xn + h

1
yn+1 = yn + (k1 + k2 )
2

9
where

k1 = hf (xn , yn )

k2 = hf (xn+1 , yn + k1 )

k1 = 0.2(1 + yn2 )

k2 = 0.2(1 + (yn + k1 )2 )

When n = 0,

k1 = 0.2(1) = 0.2

k2 = 0.2(1 + 0.04) = 0.208

1
y1 = y0 + (0.2 + 0.208)
2
= 0 + 0.204

= 0.204

When n = 1,

k1 = 0.2(1 + (0.204)2 ) = 0.208323

k2 = 0.2(1 + (0.204 + 0.208323)2 ) = 0.234002

1
y2 = y1 + (0.208323 + 0.234002)
2
= 0.204 + 0.221163

= 0.425163

Continue to y5 , we generate the approximate in Table 1.2.

10
 n xn yn Exact(6D) Absolute Error
0 0.0 0.000000 0.000000 0.000000
1 0.2 0.204000 0.202710 0.001290
2 0.4 0.425163 0.422793 0.002369
3 0.6 0.686973 0.684137 0.002836
4 0.8 1.030472 1.029639 0.000834
5 1.0 1.544841 1.557408 0.012567

Table 1.2: The results of applying Improved Euler’s Method with h=0.2

Figure 1.5: Hung’s Method and Exact Solution of example 1.6.1

1.7 Runge-Kutta 4th Order Method

The final method that we consider is somewhat harder to use in hand calculations, but is
very easily programmed into a computer. It is Runge-Kutta 4th Order Method. We see that
in each step we first compute four auxiliary quantities k1 , k2 , k3 , k4 and then the new value
yn+1 .
The 4th Order Runge-Kutta method for approximating the solution to the inital value problem

y ′ = f (x, y), y(x0 ) = y0

11
Figure 1.6: Runge-Kutta Method for approximating the solution to the initial value problem

at the points xn+1 = x0 + nh, (n = 0, 1, 2, ...) is

12
 1
yn+1 = yn + (k1 + 2k2 + 2k3 + k4 )
6
where

k1 = hf (xn , yn )
1 1
k2 = hf (xn + h, yn + k1 )
2 2
1 1
k3 = hf (xn + h, yn + k2 )
2 2
k4 = hf (xn+1 , yn + k3 ) n = 0, 1, 2, ...

1.8 Drivation for Kunge-Kutta 4th Order Method

The Runge-Kutta 4th order method is based on the following.

yn+1 = yn + (a1 k1 + a2 k2 + a3 k3 + a4 k4 )h ..........(i)

where knowing the value of y = yn at xn , we can find the value of y = yn at xn+1 and
h = xn+1 − xn .
Equation (i) is equated to finish five terms of Taylor Series

dy 1 d2 y 2 1 d3 y
yn+1 = yn + (xn+1 − xn ) + (x n+1 − x n ) + (xn+1 − xn )3 .....(ii)
dx 2! dx2 3! dx3

dy
We have, dx = f (x, y) and xn+1 − xn = h

1 ′ 1 1
yn+1 = yn + f (xn , yn )h + f (xn , yn )h2 + f ′′ (xn , yn )h3 + f ′′′ (xn , yn )h4 .....(iii)
2! 3! 4!

Based on equating equ(ii) and (iii), one of the popular solutions used is

1
yn+1 = yn + (k1 + 2k2 + 2k3 + k4 )h
6

13
where

k1 = f (xn , yn )
h 1
k2 = f (xn + , yn + k1 h)
2 2
h 1
k3 = f (xn + , yn + k2 h)
2 2
k4 = f (xn+1 + h, yn + k3 h)

Example 1.8.1. Apply the fourth-order Runge-Kutta method with h = 0.2 to determine an
approximation to the solution to the initial value problem below x = 1.

y′ = 1 + y2, y(0) = 0

f (xn , yn ) = 1 + yn2

k1 = 0.2(1 + yn2 )
k1 2
k2 = 0.2(1 + (yn + ) )
2
k2
k3 = 0.2(1 + (yn + )2 )
2
k3 2
k4 = 0.2(1 + (yn + ) )
2

14
when n=0,

k1 = 0.2(1) = 0.2

k2 = 0.2(1 + 0.01) = 0.202

k3 = 0.2(1 + 0.010201) = 0.2020402

k4 = 0.2(1 + 0.040820) = 0.208164

1
y1 = y0 + (0.2 + 2(0.202) + 2(0.2020402) + 0.208164)
6
= 0 + 0.202707

= 0.202707

Continuing in this manner, we obtain the results displayed in Table 1.3

15
 n xn yn Exact(6D) Absolute Error
0 0.0 0.000000 0.000000 0.000000
1 0.2 0.202707 0.202710 0.000003
2 0.4 0.422789 0.422793 0.000004
3 0.6 0.684133 0.684137 0.000003
4 0.8 1.029637 1.029639 0.000002
5 1.0 1.557352 1.557408 0.000056

Table 1.3: The Results of applying Runge-Kutta Fourth Order Method with h=0.2

Figure 1.7: Runge-Kutta Method for approximating the solution to the initial value problem

16
Table 1.4: Comparsion of the Accuracy of the three Methods under Condideration
in the Case of IVP with h=0.2

Error
x y = tan(x) Euler Improved Euler Runge-Kutta
0.2 0.202710 0.002710 0.001290 0.000003
0.4 0.422793 0.014793 0.002360 0.000004
0.6 0.684137 0.042844 0.002836 0.000003
0.8 1.029639 0.106095 0.000834 0.000002
1.0 1.557408 0.263277 0.012567 0.000056

Figure 1.8: Comparing Exact Solution of y(x) with numerical method for h=0.2

1.9 CAS EXPERIMENT. Euler-Caushy Vs. RK

Consider the initial value problem
(4)
y ′ = (y − 0.01x2 )2 sin(x2 ) + 0.02x, y(0) = 0.4

17
(Solution: y = 1/[2.5 − S(x)] + 0.01x2 where S(x) is the Fresnel Integral.)

Solve (4) by Euler, improved Euler and RK methods for 0 ≤ x ≤ 5 with step h = 0.2. Com-
pare the errors for (x = 1, 3, 5) and graph solution curve of the ODE in (4).

Solution: Let f (x, y) = y ′ = (y − 0.01x2 )2 sin(x2 ) + 0.2x, y(0) = 0.4, x0 = 0 and y0 = 0

Let we divide the interval [0, 5] into 25 sub-intervals of equal length h = 0.2 number of steps,
N = 5−0
0.2 = 25
For n = 0, 1, ..., N − 1
xn+1 = xn + h

x1 = 0.2, x2 = 0.4, x3 = 0.6 and son on.

By using Euler’s method,
yn+1 = yn + hf (xn , yn )

whenn = 0,
f (x0 , y0 ) = f (0, 0.4) = 0

y1 = y0 + hf (x0 , y0 ) = 0.4 + 0.2(0) = 0.4

when n = 1,
f (x − 1, y1 ) = f (0.2, 0.4) = 0.01038

y2 = y1 + hf (x1 , y1 ) = 0.4 + 0.2(0.01038) = 0.40208

and so on.

By using Improved Euler’s method,

k1 = hf (xn , yn )
k2 = hf (xn+1 , yn + k)
1
yn+1 = yn + (k1 + k2 )
2

18
when n=0,

k1 = hf (x0 , y0 ) = 0.2(0) = 0
k2 = hf (x1 , y0 + k) = 0.2f (0.2, 0.4)
= 0.2(0.01038)
1
y1 = y0 + (k1 + k2 )
2
1
= 0.4 + (0.2)(0.01038)
2
= 0.40104

and so on.
By using Runge-Kuffa 4th order method,

k1 = hf (xn , yn )
1 1
k2 = hf (xn + h, yn + k1 )
2 2
k3 = hf (xn + h, yn + k3 )
1
yn+1 = yn + (k1 + k2 + k3 + k4 )
6

19
when n=0,

k1 = hf (x0 , y0 ) = 0.2(0) = 0.2

h k1
k2 = hf (x0 + , y0 + ) = hf (0.1, 0.4)
2 2
= 0.2(0.00359) = 0.000718

h k2
k3 = hf (x0 + , y0 + ) = hf (0.1, 0.4004)
2 2
= 0.2(0.003595) = 0.00719

k4 = hf (x0 + h, y0 + k3 ) = hf (0.2, 0.40007)

= 0.00206

1
y1 = y0 + (k1 + 2k2 + 2k3 + k4 )
6
= 0.4 + 0.00083 = 0.40083

and so on.

Table 1.5: Comparsion of the Accuracy of the three Methods under Condideration
in the Case of IVP with h=0.2 at x = 1, 3, 5

Error
x Exact(5D) Euler Improved Euler Runge-Kutta
1.0 0.46668 0.02002 0.00045 0.00000
3.0 0.66896 0.06259 0.01182 0.00025
5.0 0.75788 0.05154 0.01040 0.00134

20
 xn Exact (5D) Euler method Improved Euler Runge-kutta
0 0.4 0.4 0.4 0.4
0.2 0.40083 0.40000 0.40104 0.40083
0.4 0.40504 0.40208 0.40545 0.40504
0.6 0.41534 0.40879 0.41593 0.41535
0.8 0.43481 0.42275 0.43545 0.4348
1 0.46668 0.44666 0.46713 0.46668
1.2 0.51340 0.48275 0.51311 0.51343
1.4 0.57286 0.53104 0.57071 0.57281
1.6 0.63014 0.58505 0.62505 0.62993
1.8 0.65469 0.62583 0.64758 0.65487
2 0.62987 0.62611 0.62355 0.6299
2.2 0.57982 0.58212 0.57815 0.58009
2.4 0.54717 0.53441 0.5483 0.54725
2.6 0.55595 0.52129 0.55625 0.55587
2.8 0.60858 0.55058 0.60403 0.60841
3 0.66896 0.60637 0.65714 0.66921
3.2 0.66929 0.64035 0.65718 0.67007
3.4 0.63101 0.61102 0.62604 0.62929
3.6 0.62947 0.58315 0.62635 0.6287
3.8 0.68869 0.61333 0.67596 0.68821
4 0.73027 0.6705 0.70868 0.73044
4.2 0.70065 0.6715 0.69004 0.69914
4.4 0.70109 0.64244 0.69307 0.69996
4.6 0.76681 0.67972 0.7435 0.76627
4.8 0.78013 0.7305 0.75471 0.77975
5 0.75788 0.70634 0.74748 0.75654

Table 1.6

All this can be done by using MatLab.

21
Figure 1.9: Comparing

22
Reference

[1] Erwin Kreyzig, ADVANCED ENGINEERING MATHEMATICS, Ninth Edition, United

State, (1976) 886-898.

23