Scribd
Upload
93 views

Using MATLAB To Solve Differential Equations Numerically

This document provides an introduction to numerically solving differential equations using MATLAB. It uses the van der Pol equation as an example problem. Key points: - MATLAB's Runge-Kutta solver ode45 can be used to numerically solve non-stiff differential equations. - The stiff van der Pol equation requires the implicit solver ode15s, which takes larger time steps. - Functions defining the right-hand side of the equations must be created as M-files. - Solutions can be plotted versus time or in the phase plane to visualize dynamics. - A procedure is provided to plot multiple phase portraits by integrating from different initial conditions.

Uploaded by

Brij Mohan Singh
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
93 views

Using MATLAB To Solve Differential Equations Numerically

This document provides an introduction to numerically solving differential equations using MATLAB. It uses the van der Pol equation as an example problem. Key points: - MATLAB's Runge-Kutta solver ode45 can be used to numerically solve non-stiff differential equations. - The stiff van der Pol equation requires the implicit solver ode15s, which takes larger time steps. - Functions defining the right-hand side of the equations must be created as M-files. - Solutions can be plotted versus time or in the phase plane to visualize dynamics. - A procedure is provided to plot multiple phase portraits by integrating from different initial conditions.

Uploaded by

Brij Mohan Singh
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Using MATLAB to solve

differential equations numerically


Morten Brns
Department of Mathematics
Technical University of Denmark
September 1998
Unfortunately, the analytical tool-box for understanding nonlinear differential equations which we develop in this course is far from complete. The good news is that with the
present day computer power it is possible to compute numerically approximate solutions
to the equations. The purpose of the present chapter is to give a quick introduction to
this subject in the framework of the programming language MATLAB. We will focus on
practical matters and readers interested in numerical analysis as a mathematical subject
should look elsewhere.
In the G-databar at DTU, MATLAB can be accessed either by typing  at a
terminal prompt or by pressing the middle mouse button when the cursor points at the
terminal background. A small menu pops up, and MATLAB can be chosen under the item
mathematics. In any case, a terminal window with the MATLAB prompt  results, and
if this is your first try you are advised to type  to get a demonstration of the programs
capabilities. Also, the command
  gives an interface to the on-line help system.
You probably want to save all material for this course in a directory, say,  ;
having created this the matlab command    will make it the working directory.
Before anything can be done with a differential equation, a so-called  -file defining
the right hand side must be created. As an example, consider the van der pol equation
x

y y

1 x2y x

(1)

Open your favorite text editor, enter the following

      
    
   
    !"



and save it as   .
The first line defines a function  , a function of time  and the state vector , which
returns the value of the right hand side of the system (1). Since the system is autonomous,
time does not appear explicitly. The next line initialises the function to be a matrix with 2
1

rows and 1 column i.e. a column vector consisting of zeroes. The two final lines then
compute the two coordinates of the right hand side. The lines end with a semi-colon to
prevent the result from being printed when the function is called.
To solve this equation numerically, type in the MATLAB command window

 #$  %&' ' #( ($ # ($


(except for the prompt generated by the computer, of course). This invokes the RungeKutta solver %& with the differential equation defined by the file   . The equation
is solved on the time interval t 0 20 with initial condition x1 x2 1 0. The
solver obtains results at times stored in the one-column vector  with the corresponding
solution in the two-column matrix . The solver has an error control and adjusts the time
step to try to make the error within a certain bound. Hence, the number of steps taken in
not known in advance. Check the actual size by e.g.  . Further information on the
solver can be obtained by
 %&.
You can plot the variables as a function of time using

  )'')' '
A new window opens with the plot. Here ) means everything in the first column in
 and '' gives a red curve. The second column is shown in blue with a dash-dotted line.
Use
   to see how to control the plot.
A plot in the phase plane of the solution may be obtained by

 ))
See figure 1.
Note, that in the MATLAB command window you can scroll through previous commands by using the up and down arrow keys. This is most useful for editing and reusing
previous commands.
The Runge-Kutta method used above is a good choice for a standard solver. However,
for some systems of differential equations the error control will force the solver to take
extremely small steps, and more advanced methods are needed. Such systems are called
stiff, and this occurs typically when the dynamics involve both slow and fast changes.
Methods for stiff problems are generally implicit, that is, in each time step a non-linear
set of algebraic equations must be solved. This is time-consuming, but the advantage is
that much larger steps can be taken and the overall performance is significantly improved.
To illustrate stiffness, consider a modified van der Pol equation
x

y y

10001 x2 y x

Make a file  ((( with the following contents:

      (((
    
   
   (((" !"



and solve as before but with a very short time interval t 0 1:
2

(2)

10

12

14

16

18

20

0.5

1.5

2.5

3
2.5

1.5

0.5

F IGURE 1: MATLAB plots of a solution to the van der Pol equation (1). Top: Time traces.
Bottom: Phase plane.

 #$  %&' (((' #( $ # ($


How many steps where taken?

  
 
*(**

The array  has 3033 rows (and 1 column) so this is the number of steps. MATLAB also
has an implicit solver for stiff systems, &, which may be used as follows:

 #$  &' (((' #( $ # ($


  
3

 
+&

Here, only 165 steps was needed and the computation was performed much faster.
In fact, the time interval considered is much too short to be of interest. These commands give the plots in figure 2:

 #$  &' (((' #( *((($ # ($


  )
  ))
Note the long period of the limit cycle compared to figure 1, the different scales of the two
variables, and the slow-fast behavior which is the source of the stiffness of the system.
2

1.5

0.5

0.5

1.5

2.5

500

1000

1500

2000

2500

3000

1500

1000

500

500

1000

1500
2.5

1.5

0.5

0.5

1.5

F IGURE 2: MATLAB plots of a solution to the van der Pol equation (2). Top: Time trace of x.
Bottom: Phase plane.

Rather than define individual  -files for the different versions of the van der Pol
equation, it is convenient to consider a parametrised family
x

y y

1 x2 y x

(3)

and then afterwards specify the relevant values of the parameter . Parameters may be
transported as follows. Define the file   with the following contents

      , 
    
   
   " !"

(The variable , is a dummy variable which is not important here.) To obtain a solution with 1, type

 #$  %&' ' #( ($ # ($ # $ 


To draw a phase portrait you will need more than just one solution. Here is a primitive
procedure in a file
  that provides this easily:

 
 -     , 

   ,

 
   

    
#     $  ,  
   .  / 
#01$   -     
 1)1)
#01$   -      
 1)1)

In the procedure -  is the name of the  -file defining the equations,   
is the time interval wanted for the solutions,  , is of the form #     
 $ and defines the plotting window in the phase plane, and   is the name of a
MATLAB differential equation solver.
When called, a plotting window opens, and the cursor changes into a cross-hair. Clicking with the left mouse button at a point in the phase space gives the orbit through that
point. First the equations are integrated forwards in time and this part of the orbit is plotted. Then the same is done backwards in time. In either time direction the orbit may
become unbounded, and a warning may be printed in the MATLAB command window.
This is generally of no consequence, but if you have chosen    to large, computations may be unacceptably slow. Then press control-c in the MATLAB command
window to break the execution and adjust the time interval. When an orbit is drawn you
can continue clicking at points, and when you are satisfied click any other mouse button
in the window, and the procedure terminates.
Figure 3 is created by the call
5

3
3

F IGURE 3: Phase portrait of (1) using the procedure  .


 ' '#( *($ # * *

* *$ ' %&'

Again, you may want to include a parameter. This is done by


   :

 
  -     ,   

   ,

 
   

    
#     $  ,  
   .  / 
#01$   -     # $  
 1)1)
#01$   -      # $  
 1)1)

The phase portrait in figure 3 can then be also be obtained by the call


  ' '#( *($ # * *

* *$ ' %&'

You might also like