Solving ODEs

in Matlab
BP205
M.Tremont
1.30.2009
 - Outline -

I. Defining an ODE function in an M-file

II. Solving first-order ODEs
III. Solving systems of first-order ODEs
IV. Solving higher order ODEs
 What
Numerical areare
methods weused
doing when
to solve initial value
problemsnumerically
where it is difficult to obtain
solving exact solutions
ODE’s? !
"#"$%&'((&)'*

+,'+-./00(1-.2/3+'+45+67.8
• An ODE is an equation that contains one independent variable (e.g. time) 9

and one or more derivatives with respect to that independent variable.

• In the time domain, ODEs are initial-value problems, so all the conditions
are specified at the initial time t = 0.

:;<
-=,4.2+4'14+<;>',==,0'?--
dy t (.+,+-..+
/0/@A2,B/@ 3. ?
&3714,B/@ = y(0) = 1 9 /

712C+' 712C+' dt y
>/@BC/.
?
;3@14

!
y(t) = t 2 + 1
;8/B2 .+@32,+0
!
!
• Matlab has several different functions (built-ins) for the numerical
solution of ODEs. !
These solvers can be used with the following syntax:
,0'1D10'102 6/4,/(@1

E [outputs] = function_handle(inputs) D,0,0=/4,0/

[t,state] = solver(@dstate,tspan,ICs,options)
2@+-+6+.A 2++2C=,.C

An array. The solution of Matlab algorithm Handle for function Vector that specifiecs the A vector of the
the ODE (the values of (e.g., ode45, containing the interval of the solution initial conditions
the state at every time). ode23) derivatives (e.g., [t0:5:tf]) for the system
(row or column)
 What are we doing when
numerically solving ODE’s?
F 6/,0 71 ?

y(t)
We know t0 and y(t0)
y!
* * y(t0) and we know the slope of
* * y(t), dy/dt =f(t,y).
* * *
t0 t
We don’t know y(t) for any
values of t other than t0.

Integrators compute nearby value of y(t) using

what we already know and repeat.

Higher order numerical methods reduce error at the cost of speed:

• Euler’s Method - 1st order expansion
• Midpoint method - 2nd order expansion
• Runge-Kutta - 4th order expansion
G ! H (4+I 0+7,0/@ J( 0+2+4,+3.
 (3,@29,0 =0 ?

0/-,0/2/

Solver Accuracy Description

K ode45 Medium This should be the
L+I
first solver you try Runge-Kutta
(4,5) formula
ode23 Low Less accurate
than ode45

ode113 Low to high For

4/0-1 .+M+340
computationally
III?70'+.,0+7 intensive
NJ" O problems
K ode15s Low to Use if ode45
)PQR++'1HS medium fails because the
L4+I0,/0 problem is stiff*

*No precise definition of stiffness, but the main idea is that the equation
includes some terms that can lead to rapid variation in the solution.
 Defining an ODE function in an M-file

+32D32 A
=+ = 0

[t,state] = ode45(@dstate,tspan,ICs,options)
III

1. Define tspan, ICs and options in one file (e.g.

call_dstate.m), which sets up ode45

2. Define your constants and derivatives in another file

(e.g. dstate.m) or as a function dstate within the call
file

3. Run call_dstate.m

4. Analyze the results as a plot of state vs. t

 II. Solving first-order ODEs
dy
= y'(t) = "y(t) # $y(t) 2
Example: dt
I,0

$TU T VTW

NTE QT)
-.2 +4'14
T HUST y(0) = 10
T

=@/,4
,0D32

1 fu nct ion [t,y ]X = cal l_d stat e()

2-
!
ts pan = [0 9]; % set tim e inte rval
9

3- y0 = 10; % set ini !tia l condi tio n

4 % ds tat e ev alu ates r. h.s . of th e ode
5- [t ,y] = ode4 5( @d sta te ,ts pan ,y 0);

Y
6- pl ot( t,y)
7- di sp ([ t,y ]) % dis play s t an d y( t)
8 fu nct ion dy dt = ds tat e (t ,y)V?
9- al pha =2; gam ma= 0.00 01;
10- dy dt = alp ha* y- gam ma *y ^2;
11- en d
12- en d

Save as call_dstate.m in some directory, and cd to that directory in the matlab GUI
Matlab ode45’s numerical solution

dy
= y'(t) = "y(t) # $y(t) 2
dt
y(0) = 10

At t = 9, have we reached
! steady state?
! $
limt"># y(t) = = 20,000
% DC0/,2+,/,071.+0./+

EDU>> [t, y] = call_dstate;

From the command line: ! steady_state = y(end)
EDU>>

steady_state =

1.9999e+04 Z /0 7/2C @/(

III. Solving systems of first-order ODEs

van der Pol equations in relaxation oscillation:

dy1 y1 (0) = 0
= y2
dt
dy 2
= 1000(1" y12 )y 2 " y1 y 2 (0) = 1
dt

! O %% [37. \ L/C0 0/C E +,+ C+34 +0 I

• This is a system of ODEs because we have more than one derivative with
respect!to our independent variable,!time.

• This is a stiff system because the limit cycle has portions where the
solution components change slowly alternating with regions of very sharp
change - so we will need ode15s. 9

• This is a example from mathworks, a great resource @ mathworks.com or

the software manual.

• This time we’ll create separate files for the call function (call_osc.m) and
the ode function (osc.m)
 III. Solving systems of first-order ODEs

van der Pol equations in relaxation oscillation:

dy1 y1 (0) = 0
= y2
dt
dy 2
= 1000(1" y12 )y 2 " y1 y 2 (0) = 1
dt

To simulate this system, create a function osc containing the equations.

! !
Method 1: preallocate space in a column vector, and fill with derivative functions
1 function dydt = osc(t,y)
2- dydt = zeros(2,1); % this creates an empty column
3 %vector that you can fill with your two derivatives:
4- dydt(1) = y(2);
5- dydt(2) = 1000*(1 - y(1)^2)*y(2) - y(1);
6 %In this case, y(1) is y1 and y(2) is y2, and dydt(1)
7 %is dy1/dt and dydt(2) is dy2/dt.
8- end
[+,0 = 0 ?

Save as osc.m in the same directory as before.

III. Solving systems of first-order ODEs

van der Pol equations in relaxation oscillation:

dy1 y1 (0) = 0
= y2
dt
dy 2
= 1000(1" y12 )y 2 " y1 y 2 (0) = 1
dt

To simulate this system, create a function osc containing the equations.

! !
Method 2: vectorize the differential functions
] .C,014. ]

1 function dydt = osc(t,y)

2- dydt = [y(2)
3 1000*(1 - y(1)^2)*y(2) - y(1)];
4 %Still y(1) is y1 and y(2) is y2, and dydt(1)
5 %is dy1/dt and dydt(2) is dy2/dt.
6- end

Save as osc.m in the same directory as before.

III. Solving systems of first-order ODEs

van der Pol equations in relaxation oscillation:

dy1 y1 (0) = 2
= y2
dt
dy 2
= 1000(1" y12 )y 2 " y1 y 2 (0) = 0
dt

Now solve on a time interval from 0 to 3000 with the above initial conditions.
! !
Create a scatter plot of y1 with time.

1 function [T,Y] = call_osc()

2- tspan = [0 3000];
3- y1_0 = 2;
4- y2_0 = 0;
5- [T,Y] = ode15s(@osc,tspan,[y1_0 y2_0]);
6- plot(T,Y(:,1),'o')
7-end G
I4+0- D@+2214. / /. 0+I

Save as call_osc.m in the same directory as before.

Plot of numerical solution

van der Pol equations in

relaxation oscillation:
dy1
= y2
dt
dy 2
= 1000(1" y12 )y 2 " y1
dt

y1 (0) = 2
!
y 2 (0) = 0

!
 IV. Solving higher order ODEs

Simple pendulum:
d 2!
ML 2 = " MG sin !
dt B+0.2/02
/
7/
2
d! Z
G
2
= " sin !
dt L G "I

G 4,7

• Second order non-linear ODE

• Convert the 2nd order ODE to standard form: MG
9
^+7+,+3.,/0
_41 z1 = ", z2 = d" /dt
dz1
B+0B1D2 O =30'14.
C,-C14dt
= z2 `<;
+4'14
.A.217
Z
+= U.2 +4'14 `<;

dz2 G
V0' +4'14 2+
V ,2 = #
+4'14 ^+25
sin(z[/7(
1)
dt L
W4' +4'14 Z . ,2 +4'14 ^+2. '37(
 Non-linear pendulum function file

• G = 9.8 m/s z1 = ", z2 = d" /dt

• L=2m
dz1
• Time 0 to 2π = z2
X Oa
.217 +=

• Initial θ = π/3 dt
+4'14 `<;

• Try ode23 dz2 G

= # sin(z1 )
• Plot θ with time dt L

1 function [] = call_pend()
2- tspan=[0 2*pi]; % set time interval
3- z0=[pi/3,0]; % !set initial conditions
4- [t,z]=ode23(@pend,tspan,z0);
5- plot(t,z(:,1))
6 function dzdt = pend(t,z)
7- G=9.8; L=2; % set constants
8- z1=z(1); % get z1
9- z2=z(2); % get z2
10- dzdt = [z2 ; -G/L*sin(z1);];
11- end
12- end