Solving Initial Value Problems

Jake Blanchard
University of Wisconsin - Madison
Spring 2008
Example Problem
 Consider an 80 kg paratrooper falling
from 600 meters.
 The trooper is accelerated by gravity, but
decelerated by drag on the parachute
 This problem is from Cleve Moler’s book
called Numerical Computing with Matlab
(my favorite Matlab book)
Governing Equation

 m=paratrooper mass (kg)

 g=acceleration of gravity (m/s2)
 V=trooper velocity (m/s)
 Initial velocity is assumed to be zero

dV 4
m  mg  V * V
dt 15
Solving ODE’s Numerically
 Euler’s method is the simplest approach
 Consider most general first order ODE:
dy/dt=f(t,y)
 Approximate derivative as (yi+1-yi)/dt
 Then:
dy yi 1  yi
  f (t , yi )
dt t
yi 1  yi  t f (t , yi )
A Problem
 Unfortunately, Euler’s method is too good
to be true
 It is unstable, regardless of the time step
chosen
 We must choose a better approach
 The most common is 4th order Runge-
Kutta
Runge-Kutta Techniques
 Runge-Kutta uses
a similar, but more
complicated k1  t * f (t , yi )
stepping algorithm t k1
k 2  t * f (t  , yi  )
2 2
t k2
k3  t * f (t  , yi  )
2 2
k 4  t * f (t  t , yi  k3 )
k1  2(k 2  k3 )  k 4
yi 1  yi 
6
Approach
 Choose a time step
 Set the initial condition
 Run a series of steps
 Adjust time step
 Continue
Preparing to Solve Numerically
 First, we put the equation in the form

dy
 f (t , y )
dt
 For our example, the equation becomes:

dV 4 V *V
 g 
dt 15 m
Solving Numerically
 There are a variety of ODE solvers in
Matlab
 We will use the most common: ode45
 We must provide:
◦ a function that defines the function derived
on previous slide
◦ Initial value for V
◦ Time range over which solution should be
sought
How ode45 works
 ode45 takes two steps, one with a
different error order than the other
 Then it compares results
 If they are different, time step is reduced
and process is repeated
 Otherwise, time step is increased
 The Solution
clear all
timerange=[0 15]; %seconds
initialvelocity=0; %meters/second
[t,y]=ode45(@f,timerange, initialvelocity)
plot(t,y)
ylabel('velocity (m/s)')
xlabel('time(s)')
The Function

function rk=f(t,y)
mass=80;
g=9.81;
rk=-g-4/15*y.*abs(y)/mass;
My Solution
0

-10

-20
v (m/s)

-30

-40

-50

-60
0 5 10 15
time(s)
Practice
 Download the file odeexample.m
 Run it to reproduce my result
 Run again out to t=30 seconds
 Run again for an initial velocity of 10
meters/second
 Change to k=0 and run again (gravity
only)
Practice
 The outbreak of an insect population can
be modeled with the equation below.
 R=growth rate
 C=carrying capacity
 N=# of insects
 Nc=critical population
 Second term is due to bird predation

 
2
dN N rN
 RN 1    2
 C  Nc  N
2
dt
 Parameters
 0<t<50 days  What is steady
 R=0.55 /day state population?
 N(0)=10,000  How long does it
 C=10,000 take to get there?
 Nc=10,000
 
2
dN N rN
 r=10,000 /day
 RN 1    2
 C  Nc  N
2
dt

 Note: this is a first order ode

 Skeleton script is in file: insects.m
Insects.m
function insects
clear all
tr=[0 ??];
initv=??;
[t,y]=ode45(@f, tr, initv);
plot(t,y)
ylabel('Number of Insects')
xlabel('time')
%
function rk=f(t,y)
rk= ??;
Practice
 Let h be the depth of water in a spherical
tank
 If we open a drain at the tank bottom, the
pressure at the bottom will decrease as
the tank empties, so the drain rate
decreases with h
 Find the time to empty the tank
 Parameters

 R=5 ft; Initial height=9 ft

 1 inch hole for drain

dh 0.0334 h

dt 10h  h 2

 How long does it take to drain

the tank?
Rockets

 A rocket’s mass decreases as it burns fuel

 Find the final velocity of a rocket if:
 T=48000 N; m0=2200 kg
 R=0.8; g=9.81 m/s2; b=40 s
dv
m  T  mg
dt
 rt 
m  m0 1  
 b
Options
 Options are available to:
◦ Change relative or absolute error tolerances
◦ Maximum number of steps
◦ Etc.
Some Other Matlab routines
 ode23 – like ode45, but lower order
 ode15s – stiff solver
 ode23s – higher order stiff solver
Advanced IVPs
 Second order equations
 Stiff equations
Second Order Equations
 Consider a falling object with drag

4
y   g  y y
15m
y (0)  h
y (0)  0
Preparing for Solution
 We must break second order equation
into set of first order equations
 We do this by introducing new variable
(z=dy/dt)

z  y 4
z   z z g
z  y 15m
4 y  z
z   z z g
15m y (0)  h; z (0)  0
Solving
 Now we have to send a set of equations
and a set of initial values to the ode45
routine
 We do this via vectors
 Let w be vector of solutions: w(1)=y and
w(2)=z
 Let r be vector of equations: r(1)=dy/dt
and r(2)=dz/dt
Function to Define Equation
dy
 z  w(2)
dt
dz 4
 w(2) * w(2)  g
dt 15m
function r=rkfalling(t,w)
...
r=zeros(2,1);
r(1)=w(2);
r(2)= -k*w(2).*abs(w(2))-g;
The Routines
tr=[0 15]; %seconds
initv=[600 0]; %start 600 m high
[t,y]=ode45(@rkfalling, tr, initv)
plot(t,y(:,1))
ylabel('x (m)')
xlabel('time(s)')
figure
plot(t,y(:,2))
ylabel('velocity (m/s)')
xlabel('time(s)')
Function

function r=rkfalling(t,w)
mass=80;
k=4/15/mass;
g=9.81;
r=zeros(2,1);
r(1)=w(2);
r(2)= -k*w(2).*abs(w(2))-g;
General Second Order Equations

 We can write a general d2y dy

second order equation 2
 f (t , y, )
dt dt
as shown: or
 To solve:
dy
◦ Define f z
dt
◦ Set initial conditions
dz
◦ Set time range  f (t , y, z )
dt
 The Routines
tr=[0 15]; %seconds
initv=[600 0]; %start 600 m high
[t,y]=ode45(@rkfalling, tr, initv)
plot(t,y(:,1))
ylabel('x (m)') function r=rkfalling(t,w)
xlabel('time(s)') mass=80;
figure k=4/15/mass;
plot(t,y(:,2)) g=9.81;
ylabel('velocity (m/s)') r=zeros(2,1);
xlabel('time(s)') r(1)=w(2);
r(2)= -k*w(2).*abs(w(2))-g;
Practice
 Return to paratrooper problem.
 Download ode2ndOrder.m
 Run to duplicate earlier results for
velocity
 Change initial velocity to 10 m/s and run
again
2
d y 4 dy dy
m 2  mg 
dt 15 dt dt
 Practice-nonlinear pendulum
 r=1 m; g=9.81 m/s2
 Initial angle =/8, /2, -0.1

d
2
g
2
  sin( )
dt r
Systems
 For systems of first order ODEs, just
define both equations.
 Practice
 Consider an ecosystem of rabbits r and
foxes f. Rabbits are fox food.
 Start with 300 rabbits and 150 foxes
 =0.01
 r=w(1)
dr  f=w(2)
 2r   rf
dt function z=rkfox(t,w)
df
  f   rf alpha=0.01;
dt r=zeros(2,1);
z(1)=2*w(1)-alpha*w(1)*w(2);
z(2)= -w(2)+alpha*w(1)*w(2);
Approach
 Start with ode2ndOrder.m
 Modify with function from previous slide
 Put in time range (0<t<15) and initial
conditions
 Higher Order Equations

 Suppose we want to model a projectile

x  k xV
y  k y V  g
V  x y
 2
 2
Now we need 4 1st order ODEs

x  s
s   k sV
y  z
z   k zV  g
V  s z2 2
The Code

clear all;
tspan=[0 1.1]
wnot(1)=0; wnot(2)=10;
wnot(3)=0; wnot(4)=10;
[t,y]=ode45('rkprojectile',tspan,wnot);
plot(t,y(:,1),t,y(:,3))
figure
plot(y(:,1),y(:,3))
The Function
function r=rkprojectile(t,w)
g=9.81;
x=w(1); s=w(2); y=w(3); z=w(4);
vel=sqrt(s.^2+z.^2);
r=zeros(4,1);
r(1)=s;
r(2)=-s*vel;
r(3)=z;
r(4)=-z*vel-g;
Questions