Differential equations using

Scilab

Kannan M. Moudgalya
IIT Bombay
[email protected]

Scilab Workshop
13 September 2010

Kannan Moudgalya Differential equations using Scilab 1/31

Outline

I Weight reduction ODE model - analytical

solution
I Numerical integration
I Functions in Scilab
I Euler’s method
I Predator-prey system
I Modelling
I Euler method - user created integrator
I Backward difference method - built-in
function

Kannan Moudgalya Differential equations using Scilab 2/31

Weight Reduction Model
I Weight of person = x kg
I Tries to reduce weight
I Weight loss per month = 10% of weight
I Starting weight = 100 kg

dx
= −0.1x
dt
Initial conditions:

x = 100 at t=0

Determine x(t) as a function of t.

Kannan Moudgalya Differential equations using Scilab 3/31
Analytical Solution of Simple Model I
Recall the model:
dx
= −0.1x
dt
x(t = 0) = 100
dx
Cross multiplying, = −0.1dt
x
Integrating both sides from 0 to t,

dx
Z Z
= −0.1 dt
x
C + ln x(t) = −0.1t
Kannan Moudgalya Differential equations using Scilab 4/31
Analytical Solution of Simple Model II

Using initial conditions,

C = − ln 100

Thus, the final solution is,

x(t)
ln = −0.1t or x(t) = 100e−0.1t
100

Kannan Moudgalya Differential equations using Scilab 5/31

Solution, Continued
I Weight of person = x kg
I Tries to reduce weight
I Weight loss per month = 10% of weight
I Starting weight = 100 kg
x(t) = 100e−0.1t
Compute, plot for 2 years, i.e. for 24 months:
1 T=0:0.1:24;
2 p l o t 2 d (T, 1 0 0 ∗ e x p ( −0.1∗T ) ) ;
3 x t i t l e ( ’ Weight v s . month ’ , . . .
4 ’ Time i n months ’ , . .
5 ’ Weight ( kg ) ’ )
Kannan Moudgalya Differential equations using Scilab 6/31
 Weight vs. month
Weight (kg)
100

90

80

70

60

50

40

30

20

10 Time, in months

0
0 4 8 12 16 20 24

Kannan Moudgalya Differential equations using Scilab 7/31

Need for Numerical Solution
I Exact solution is ok for simple models
I What if the model is complicated?
I Consider integrating the more difficult
problem:

dx 2 + 18t + 68t2 + 180t3 + 250t4 + 250t5

=
dt x2
with initial condition, x(t = 0) = 1
I Analytical (i.e. exact) solution difficult to
find
I Let us look for a numerical solution
Kannan Moudgalya Differential equations using Scilab 8/31
Simple numerical solution: Explicit Euler
I Suppose that we want to integrate
dx
= g(x, t)
dt
with initial condition: x(t = 0) = x0
I Approximate numerical method - divide
time into equal intervals: t0 , t1 , t2 , etc.
xn − xn−1
= g(xn−1 , tn−1 )
∆t
I Simplifying,
xn − xn−1 = ∆t g(xn−1 , tn−1 )
xn = xn−1 + ∆t g(xn−1 , tn−1 )
Kannan Moudgalya Differential equations using Scilab 9/31
Example revisited
Recall the problem statement for numerical
solution:
dx 2 + 18t + 68t2 + 180t3 + 250t4 + 250t5
=
dt x2
with initial condition,
x(t = 0) = 1
Recall the Euler method:
dx
= g(x, t)
dt
Solution for initial condition, x(t = 0) = x0 is,
Kannan Moudgalya Differential equations using Scilab 10/31
Scilab Code I

1 exec ( ” d i f f 1 . s c i ” ) ;
2 exec ( ” Euler . s c i ” ) ;
3 x0 =1; t 0 =0; T = 0 : 1 : 2 4 ;
4 s o l = E u l e r ( x0 , t0 , T , d i f f 1 ) ;
5 // s o l = o d e ( x0 , t0 , T , d i f f 1 ) ;
6 p l o t 2 d (T , s o l ) , p a u s e
7 p l o t 2 d (T,1+2∗T+5∗Tˆ 2 , 5 )
8 x t i t l e ( ’ x v s . t : P o l y n o m i a l Problem ’ , ’ t ’ , ’ x

Kannan Moudgalya Differential equations using Scilab 11/31

Scilab Code II

1 f u n c t i o n x = E u l e r ( x0 , t0 , t , g )
2 n = l e n g t h ( t ) , x = x0 ;
3 f o r j = 1 : n−1
4 x0 = x0 + ( t ( j +1)− t ( j ) ) ∗ g ( t ( j ) , x0 ) ;
5 x = [ x x0 ] ;
6 end ;
7 endfunction

1 f u n c t i o n xdot = d i f f 1 ( t , x )
2 x d o t = (2+18∗ t +68∗ t ˆ2+180∗ t ˆ3+250∗ t ˆ4+250∗ t
3 endfunction

Kannan Moudgalya Differential equations using Scilab 12/31

Comparison of numerical, exact solution
x vs. t: Polynomial Problem
x
3200

2800

2400

2000

1600

1200

800

400
t

0
0 4 8 12 16 20 24

Kannan Moudgalya Differential equations using Scilab 13/31

Predator-Prey Problem I

I Population dynamics of predator-prey

I Prey can find food, but gets killed on
meeting predator
I Examples: parasites and certain hosts;
wolves and rabbits
I x1 (t) - number of prey; x2 (t) - number of
predator at time t
I Prey, if left alone, grows at a rate
proportional to x1

Kannan Moudgalya Differential equations using Scilab 14/31

Predator-Prey Problem II
I Predator, on meeting prey, kills it ⇒
proportional to x1 x2
dx1
= 0.25x1 − 0.01x1 x2
dt
I Predator, if left alone, decrease by natural
causes
I Predators increase their number on
meeting prey
dx2
= −x2 + 0.01x1 x2
dt
I Determine x1 (t), x2 (t) when x1 (0) = 80,
x2 (0) = 30
Kannan Moudgalya Differential equations using Scilab 15/31
Explicit Euler for a System of Equations I

dx1
= g1 (x1 , . . . , xn , t)
dt
..
.
dxN
= gn (x1 , . . . , xn , t)
dt

Kannan Moudgalya Differential equations using Scilab 16/31

Explicit Euler for a System of Equations II
g1 (x1 , . . . , xn , t − 1)
   
x1
d . ..
 ..  =  . 
dt
xn gn (x1 , . . . , xn , t − 1)

g1 ((x1 , . . . , xn )|t−1 , t − 1)
     
x1 x1
 ...  =  ...  + ∆t  ..
. 
xn t xn t−1 gN ((x1 , . . . , xn )|t−1 , t − 1)

Solution in vector form:

xt = xt−1 + ∆tg(xt−1 )

Kannan Moudgalya Differential equations using Scilab 17/31

Scilab Code for Predator-Prey Problem I

1 exec ( ” pred . s c i ” ) ;
2 exec ( ” Euler . s c i ” ) ;
3 x0 = [ 8 0 , 3 0 ] ’ ; t 0 =0; T = 0 : 0 . 1 : 2 0 ; T=T ’ ;
4 s o l = E u l e r ( x0 , t0 , T , p r e d ) ;
5 // s o l = o d e ( x0 , t0 , T , p r e d ) ;
6 clf ();
7 p l o t 2 d (T , s o l ’ )
8 x s e t ( ’ window ’ , 1 )
9 plot2d ( sol (2 ,:) , sol (1 ,:))

Kannan Moudgalya Differential equations using Scilab 18/31

Scilab Code for Predator-Prey Problem II
1 f u n c t i o n x = E u l e r ( x0 , t0 , t , g )
2 n = l e n g t h ( t ) , x = x0 ;
3 f o r j = 1 : n−1
4 x0 = x0 + ( t ( j +1)− t ( j ) ) ∗ g ( t ( j ) , x0 ) ;
5 x = [ x x0 ] ;
6 end ;
7 endfunction

1 f u n c t i o n xdot = pred ( t , x )
2 xdot ( 1 ) = 0 . 2 5 ∗ x (1) −0.01∗ x ( 1 ) ∗ x ( 2 ) ;
3 x d o t ( 2 ) = −x ( 2 ) + 0 . 0 1 ∗ x ( 1 ) ∗ x ( 2 ) ;
4 endfunction

Kannan Moudgalya Differential equations using Scilab 19/31

Predator-Prey Problem: Solution by Euler
Predator−Prey by Euler for different step sizes, Black = 0.25, Red = 0.1
Prey
140

130

120

110

100

90

80

70
Predator

60
10 14 18 22 26 30 34 38 42 46

As step size increases, the solution diverges

Kannan Moudgalya Differential equations using Scilab 20/31
General method to handle stiff systems
I The predator-prey problem is an example
of a stiff system
I Results because of sudden changes in the
derivative
I Approximation of using previous time
values does not work
I General approach to solve this problem:
xn = xn−1 + ∆t g(xn , tn )
I Requires solution by trial and error, as
g(xn , tn ) is unknown
I Scilab has state of the art methods (ode)
to solve such systems
Kannan Moudgalya Differential equations using Scilab 21/31
General method to handle stiff systems

I Derived from ODEPACK

I FOSS
I In use for thirty years
I Bugs have been removed by millions of users

Kannan Moudgalya Differential equations using Scilab 22/31

Predator-Prey Problem by Scilab
Integrator I
Execute the following code, after commenting
out Euler and uncommenting ode:
1 exec ( ” pred . s c i ” ) ;
2 exec ( ” Euler . s c i ” ) ;
3 x0 = [ 8 0 , 3 0 ] ’ ; t 0 =0; T = 0 : 0 . 1 : 2 0 ; T=T ’ ;
4 s o l = E u l e r ( x0 , t0 , T , p r e d ) ;
5 / / s o l = o d e ( x0 , t0 , T , p r e d ) ;

6 clf ();
7 p l o t 2 d (T , s o l ’ )
8 x s e t ( ’ window ’ , 1 )
Kannan Moudgalya Differential equations using Scilab 23/31
Predator-Prey Problem by Scilab
Integrator II

9 plot2d ( sol (2 ,:) , sol (1 ,:))

1 f u n c t i o n xdot = pred ( t , x )
2 xdot ( 1 ) = 0 . 2 5 ∗ x (1) −0.01∗ x ( 1 ) ∗ x ( 2 ) ;
3 x d o t ( 2 ) = −x ( 2 ) + 0 . 0 1 ∗ x ( 1 ) ∗ x ( 2 ) ;
4 endfunction

Kannan Moudgalya Differential equations using Scilab 24/31

Predator-Prey Problem: Scilab Integrator
Predator−Prey by Scilab integrator
Prey
130

120

110

100

90

80

Predator

70
15 19 23 27 31 35 39

Use the Scilab built-in integrator to get the

correct solution
Kannan Moudgalya Differential equations using Scilab 25/31
Partial Differential Equations

Kannan Moudgalya Differential equations using Scilab 26/31

Parabolic Differential Equations
I Heat conduction equation
I Diffusion equation
∂u(t, x) ∂ 2 u(t, x)
=c
∂t ∂x2
I Initial condition:
u(0, x) = g(x), 0≤x≤1
I Boundary conditions:
u(t, 0) = α, u(t, 1) = β, t≥0
I Let um
j be approximate solution at xj = j∆x,
tm = m∆t
um+1
j − um
j c
= 2
(um m
j−1 − 2uj + uj+1 )
m
∆tKannan Moudgalya
(∆x)Differential equations using Scilab 27/31
Finite Difference Approach

um+1
j − um
j c c∆t
= (um m m
j−1 − 2uj + uj+1 ), µ=
∆t (∆x)2 (∆x)2
um+1
j = um m m m
j + µ(uj−1 − 2uj + uj+1 )
= µum m m
j−1 + (1 − 2µ)uj + µuj+1

Write this equation at every spatial grid:

um+1
1 = µum m m
0 + (1 − 2µ)u1 + µu2
um+1
2 = µum m m
1 + (1 − 2µ)u2 + µu3
..
.
um+1
N = µum m m
N−1 + (1 − 2µ)uN + µuN+1

Kannan Moudgalya Differential equations using Scilab 28/31

Finite Difference Approach - Continued

um+1
1 = µum m m
0 + (1 − 2µ)u1 + µu2
um+1
2 = µum m m
1 + (1 − 2µ)u2 + µu3
..
.
um+1
N = µum m m
N−1 + (1 − 2µ)uN + µuN+1

In matrix form,
m+1   m 
u1 1 − 2µ µ u1 µ
 
 u2   µ 1 − 2µ µ u
  .2  + 
   
 . 
 ..  = .. .. 
 .   
uN µ 1 − 2µ uN µu

Kannan Moudgalya Differential equations using Scilab 29/31

Conclusions

I Scilab is ideal for educational institutions,

including schools
I Built on a sound numerical platform
I It has good integrators for differential
equations
I It is free
I Also suitable for industrial applications

Kannan Moudgalya Differential equations using Scilab 30/31

Thank you

Kannan Moudgalya Differential equations using Scilab 31/31