Crank Nicholson Method

ENGR90024
COMPUTATIONAL FLUID DYNAMICS
Lecture O10
Crank-Nicolson method &!
Stiff ODEs
ACCURACY OF CRANK-NICOLSON
Going to show that the Crank-Nicolson time-marching method is
2nd-order accurate.
l
l+1
1 d
d
l+1
l
3
= + t
+
+ t (. . . )
2 dt
dt
That is, that the local truncation error is proportional to t3, or
that the global truncation error is proportional to t2.
Taylor-series expand about tl:
l
2 l
d
l+1
l
21 d
= + t
+ t
+
2
dt
2 dt
t (. . . )
(10.1)
Then Taylor-series expand about tl+1:

l
l+1
d
t
dt
l+1
21
d
t
2 dt2
l+1
t3 (. . . )
(10.2)
Take Eq. (10.1)-(10.2)

l+1
d
l+1+ t
dt
2 l
1
d
d
2
+ t
t
2
2
dt
dt
2 l+1
1
d
2
+ t
2 dt2
3
+ t (...)
l+1
l+1
l+1
l+1
d
d
+
dt
dt
d2
dt2
21
t
2
d2
dt2
l+1
Divide the above equation by 2 to give

l+1
1 d
= + t
+
2 dt
2 l
21 d
+ t
4 dt2
l
d
dt
2
l+1
d
dt2
Crank-Nicolson
l+1
t (. . . ) Error
t3 (. . . )
l
l+1
1 d
d
l+1
l
= + t
+
Lets further analyse !
2 dt
dt
this term
2 l
2 l+1
d
21 d
3
+ t
+
t
(.
.
.
)
(10.3)
4 dt2
dt2
Taylor-series expand d2/dt2 about tl:
2
l+1
d
d
d
2
= 2 + t 3 + t (. . . )
2
dt
dt
dt
Substitute back into Eq. (10.3) to get
l
l+1
1 d
d
l+1
l
3
= + t
+
+ t (. . . )
2 dt
dt
l+1
d
1 d
+
t
2 dt
dt
l+1
3
+ t (. . . )
Small
STABILITY OF CRANK-NICOLSON
Consider the model problem
d
=
dt
Apply the Crank-Nicolson method to obtain

1 l
l+1
l
= + t
+
2
l+1
l
(1
t/2)
= (1 +
t/2)
l+1
1+
=
1
=
t/2
t/2
l+1
1+
l+1
=
1
t/2
t/2
= l
For stability (solution not to blow up) we need ||1.
1+
t/2
Crank-Nicolson:
=
1
t/2
1+
t/2
That is,
1
1
t/2
1 + Re t/2 + i Im t/2
1
1
i Im t/2
Re t/2
(1 + Re t/2)2 + ( Im t/2)2
1
2+(
2
(1
t/2)
t/2)
Re
Im
(1 +
Re
t/2) (1
Re
t0
Re
t/2)
Im
12
t)3
t)5
Im
80
+ ...
which is of order O( Im t)3 . So compared to the Euler methods the phase error
here
of the same(solution
order, but the
term has
in the
denominator
instead
Foris stability
notleading
to blow
up)12 we
need
||1.
of 3 and will be smaller.
That is,
Im
Stable
Unstable
Re
The region of stability (||1)

for the Crank-Nicolson method
Figure 1.13: The stability diagram for the is
implicit
Crank-Nicolson
when
Re t1method.
(left halfplane).
RK-2
Stability region
of implicit
Euler and
CrankNicolson
scheme
RK-4
Euler
System of Nonlinear ODEs
So far, we have learnt to solve single ODE using implicit Crank-Nicolson method. What if you are
required a system of M ODEs?
d 1
= f1 ( 1 , 2 , 3 , . . . . . . ,
dt
d 2
= f2 ( 1 , 2 , 3 , . . . . . . ,
dt
d 3
= f3 ( 1 , 2 , 3 , . . . . . . ,
dt
.. ..
.=.
d M
= fM ( 1 , 2 , 3 , . . . . . . ,
dt
M , t)
M , t)
M , t)
M , t)
The Crank-Nicolson method can easily be extended to solve a system of M equations
d 1
= f1 (
dt
1 , t)
Crank-Nicolson
l+1
1
t
l
= 1+
f1 (
2
l l
1, t )
+ f1 (
l+1 l+1
)
1 ,t
d 1
= f1 (
dt
1 , t)
Crank-Nicolson
l+1
1
l
1
t
+
f1 (
2
l l
1, t )
+ f1 (
l+1 l+1
)
1 ,t
d 1
= f1 (
dt
d 2
= f2 (
dt
1,
2 , t)
1,
2 , t)
Crank-Nicolson
l+1
1
l+1
2
t
=
+
f1 (
2
t
l
= 2+
f2 (
2
l
1
l
1,
l l
2, t )
l
1,
l l
2, t )
+ f1 (
l+1
1 ,
l+1 l+1
)
2 ,t
+ f2 (
l+1
1 ,
l+1 l+1
)
2 ,t
d 1
= f1 (
dt
1 , t)
Crank-Nicolson
l+1
1
l
1
t
+
f1 (
2
d 1
= f1 (
dt
d 2
= f2 (
dt
l l
1, t )
1,
2 , t)
1,
2 , t)
+ f1 (
l+1 l+1
)
1 ,t
Crank-Nicolson
l+1
1
l+1
2
t
f1 (
2
t
l
= 2+
f2 (
2
=
l
1
l
1,
l l
2, t )
l
1,
l l
2, t )
+ f1 (
l+1
1 ,
l+1 l+1
)
2 ,t
+ f2 (
l+1
1 ,
l+1 l+1
)
2 ,t
l+1
1
l+1
2
l+1
3
t
=
+
f1 (
2
t
l
= 2+
f2 (
2
t
l
= 3+
f3 (
2
l
1
d 1
= f1 (
dt
d 2
= f2 (
dt
d 3
= f3 (
dt
1,
2,
3 , t)
1,
2,
3 , t)
1,
2,
3 , t)
Crank-Nicolson
l
1,
l
2,
l l
3, t )
l
1,
l
2,
l l
3, t )
l
1,
l
2,
l l
3, t )
+ f1 (
l+1
1 ,
l+1
2 ,
l+1 l+1
)
3 ,t
+ f2 (
l+1
1 ,
l+1
2 ,
l+1 l+1
)
3 ,t
+ f3 (
l+1
1 ,
l+1
2 ,
l+1 l+1
)
3 ,t
Example O10.1: Write a MATLAB program that uses the Crank-Nicolson

method to solve!
!
2
d
!
3
+
+
=
0
!
2
dt
!
!
in the domain 0 < t < 50 given (0)=1, (0)=0.
To do this case study, need to know:
how to write a 2nd-order scalar ODE as a 1st-order vector

ODE, and
how to solve an implicit vector equation using NewtonRaphson iterations (using Jacobians).
d2
+
2
dt
Set:
=0
= d /dt
Rewrite as a set of 1st-order ODEs:

d 1
dt
d 2
dt
=
=
3
1
d 1
= f1 (
dt
d 2
= f2 (
dt
1,
1,
d 1
dt
d 2
dt
2 , t)
2 , t)
=
=
l+1
2
t
=
+
f1 (
2
t
l
= 2+
f2 (
2
l
1
l
1,
l l
2, t )
l
1,
l l
2, t )
3
1
Crank-Nicolson
Crank-Nicolson
l+1
1
+ f1 (
l+1
1 ,
l+1 l+1
)
2 ,t
l+1
1
+ f2 (
l+1
1 ,
l+1 l+1
)
2 ,t
l+1
2
t
+
2
t
l
2+
2
l
1
l
2
l+1
2
l
1
l 3
1)
l+1
1
l+1 3
1 )
l+1
1
t
+
2
t
l
2+
2
l
1
l+1
2
l
2
+
l
1
l+1
2
l+1
1
l 3
1)
l+1 3
1 )
At every time step you need to solve a system of 2 equations for the 2 unknowns, 1l+1 and 2l+1.
Note that the values of 1l and 2l are usually known.
f=
l+1
1
l+1
2
l
1
l
2
t/2
t/2
l
2
l
1
+
+
l+1
2
( l1 )3
l+1
1
+(
l+1 3
1 )
0
0
This equation is nonlinear and we need some iterative method to solve it. In Lecture O07, we saw how
the Newton-Raphson method can be used to solve a system of equations. The generic formula is
2
6
6
6
6
6
6
6
4
g1 /
g2 /
g3 /
g4 /
..
.
p1
p1
p1
p1
gM / p1
g1 /
g2 /
g3 /
g4 /
..
.
p2
p2
p2
p2
...
...
...
...
..
.
g1 /
g2 /
g3 /
g4 /
..
.
pM
pM
pM
pM
gM / p2
...
gM / pM
3 8
>
>
>
7 >
>
7 >
<
7 >
7
7
7 >
>
7 >
>
5 >
>
>
:
p1,new
p2,new
p3,new
p4,new
pM,new
9
>
>
>
>
>
>
>
=
p1,old
p2,old
p3,old
p4,old > =
>
>
..
>
>
.
>
>
;
pM,old
g1 (p1,old , p2,old , . . . , pM,old )
g2 (p1,old , p2,old , . . . , pM,old )
g3 (p1,old , p2,old , . . . , pM,old )

g4 (p1,old , p2,old , . . . , pM,old )
..
gM (p1,old , p2,old , . . . , pM,old )
2
6
6
6
6
6
6
6
4
g1 /
g2 /
g3 /
g4 /
..
.
p1
p1
p1
p1
gM / p1
g1 /
g2 /
g3 /
g4 /
..
.
p2
p2
p2
p2
...
...
...
...
..
.
g1 /
g2 /
g3 /
g4 /
..
.
pM
pM
pM
pM
gM / p2
...
gM / pM
3 8
>
>
>
7 >
>
7 >
<
7 >
7
7
7 >
>
7 >
>
5 >
>
>
:
p1,new
p2,new
p3,new
p4,new
pM,new
9
>
>
>
>
>
>
>
=
p1,old
p2,old
p3,old
p4,old > =
>
>
..
>
>
.
>
>
;
pM,old
g1 (p1,old , p2,old , . . . , pM,old )

g2 (p1,old , p2,old , . . . , pM,old )
g3 (p1,old , p2,old , . . . , pM,old )
g4 (p1,old , p2,old , . . . , pM,old )
..
.
gM (p1,old , p2,old , . . . , pM,old )
For a system consisting of 2 equations, the above simplifies to
@g1 /@p1 @g1 /@p2

p1,new p1,old
=
@g2 /@p1 @g2 /@p2
p2,new p2,old
g1 (p1,old , p2,old )
g2 (p1,old , p2,old )
We want to use the above equation to solve
f=
l+1
1
l+1
2
l
1
l
2
t/2
t/2
l
2
l
1
+
+
l+1
2
( l1 )3
l+1
1
+(
l+1 3
1 )
Using the notation in previous lectures, we let pn= nl+1 to get
f=
p1
p2
l
1
l
2
t/2
t/2
l
2
l
1
+ p2
+ ( l1 )3 + p1 + p31
0
0
0
0
f=
l
1
l
2
p1
p2
l
2
l
1
t/2
t/2
+ p2
+ ( l1 )3 + p1 + p31
0
0
The Jacobian for this system of equation is given by
J=
@f1 /@p1
@f2 /@p1
@f1 /@p2
@f2 /@p2
For the system of equation we are required to solve, this becomes
Jmn
@fm
=
=
@pn
1
2
t/2(1 + 3p1 )
t/2
1
We need to iterate using the equation below at every time step until pnew converges.
1
t/2(1 + 3p21 )
t/2
1
p1,new
p2,new
p1,old
p2,old
p1
p2
l
1
l
2
t/2
t/2
l
2
l
1
+ p2
+ ( l1 )3 + p1 + p31
function MPO10p1()!
clear all;!
close all;!
!
Delta_t=0.1;!
t=0:Delta_t:50; !
%!
%Preallocating Memory!
%!
phi=zeros(length(t),2);!
phi(1,1)=1.0;phi(1,2)=0.0;!
for n=1:length(t)-1!
phi(n+1,:)=FindPhilplus1(Delta_t,phi(n,:));!
end!
%Check with Matlab solution!
[tmat,phimat] = ode23(@ExampleO10p1,[0 50],[1 0]);!
!
hold off!
plot(t,phi(:,1),'ko',tmat,phimat(:,1),'b-');hold on!
legend('Crank-Nicolson','MATLAB ode23()');!
plot(t,phi(:,2),'ko',tmat,phimat(:,2),'b-');!
!
function p=FindPhilplus1(Delta_t,Phil)!
eps=1.0; %Just making sure you go through the loop at least once!
p=Phil; %Guess value for p!
while eps>1.0e-13!
[gfunc,jacobian]=ExampleFunctionWithJacobian(p,Delta_t,Phil);!
eps=max(abs(gfunc));!
temp=jacobian\(-gfunc); !
p=temp'+p;!
end!
!
function [gfunc,jac]=ExampleFunctionWithJacobian(p, Delta_t,Phil)!
gfunc=[p(1)-Phil(1)-(Delta_t/2.)*(Phil(2)+p(2)); p(2)-Phil(2)+(Delta_t/
2.)*(Phil(1)+Phil(1).^3+ p(1)+p(1).^3)];!
jac=[1 -Delta_t/2.0; (Delta_t/2.0)*(1+3*p(1).^2)
1];!
!
function dphidt = ExampleO10p1(t,phi)!
dphidt = zeros(2,1);
% a column vector!
dphidt(1) = phi(2);!
dphidt(2) = -phi(1)-phi(1).^3;
1
t/2(1 + 3p21 )
t/2
1
p1,new
p2,new
p1,old
p2,old
1.5
CrankNicolson
MATLAB ode23()
1
0.5
0.5
1.5
p1
p2
l
1
l
2
10
15
t/2
t/2
20
25
l
2
l
1
30
35
40
+ p2
+ ( l1 )3 + p1 + p31
45
50
Stiff Systems
A stiff system has rapidly changing components as
well as slowly changing ones. An example of a stiff
system is given by!
!
d
=
dt
1000 + 3000
2000e
with initial condition

(0) = 0.0
The exact solution is

(t) = 3
0.998e
1000t
2.002e
(t) = 3
0.998e
1000t
2.002e
Note that this system has a rapidly varying term

0.998e
1000t
which decays very quickly.

There is also another term
2.002e
which decays a lot slower.

Systems whose solutions are made up of slow terms
together with fast terms are called stiff systems. For
MATLAB, the functions that
are specially for stiff systems have an s in them. For
example
ode23s()
ode15s()
(t) = 3
3
0.998e
2.002e
1000t
2.002e
2.5
(t)
1.5
Slowly varying part of !

the solution
0.998e
0.5
0.5
1000t
1.5
Fast varying part of the solution!

in the vicinity of t=0.
2.5
3.5
Example 10.2:
solution to!
!
!
!
!
Use ode23(), ode23s(), explicit Euler and 4th order Runge-Kutta to obtain numerical
d
=
dt
1000 + 3000
2000e
Use initial conditions x(0)=0 and solve in the domain 0<t<4.
dx
= x
dt
2.83
1+ h+
2 2
h
2
eI = 0
From stability analysis, we can predict that for the computed

solution to be stable, we need t<0.002=2/1000 for the
Eulers method and t<0.00279=2.79/1000 for the RK4
method
-2.79
-2.0
h+
2 2
h
+
2
3 3
4 4
h
h
+
+1
6
24
eI = 0
-2.83
Output
3
Exact solution
ode23()
ode23s()
MyEuler()
MyRK4
2.5
function MPO10p2()!
!
%Solving ODE using different Matlab functions!
% and compare with exact solution!
clear all!
close all!
Delta_t=0.002;!
!
%Exact solution!
texact=linspace(0,4,500);!
xexact=3-0.998*exp(-1000*texact)-2.002*exp(-texact);!
!
%Solving solution using different ODE!
[t1,x1]=ode23(@dxdt,[0, 4],0.0);!
[t2,x2]=ode23s(@dxdt,[0, 4],0.0);!
[t3,x3]=MyEuler(@dxdt,[0, 4],0.0,Delta_t);!
[t4,x4]=MyRK4(@dxdt,[0, 4],0.0,Delta_t);!
!
%Plotting solution!
plot(texact,xexact,'k-',t1,x1,'r-', t2,x2,'g-', t3,x3,'b-',t4,x4,'mo-')!
axis([0 4 0 3]);!
legend('Exact solution','ode23()','ode23s()','MyEuler()','MyRK4')!
!
xlabel('t','Fontsize',14);!
ylabel('x','Fontsize',14);!
!
function [t,x]=MyEuler(ODEfunc,tspan,x0,Delta_t)!
%This function solves only one ODE.!
t=tspan(1):Delta_t:tspan(2);!
x(1)=x0;!
x(n+1)=x(n)+ODEfunc(t(n),x(n))*Delta_t;!
end!
!
function [t,x]=MyRK4(ODEfunc,tspan,x0,Delta_t)!
%This function assumes you are only solving!
%one ODE.!
t=tspan(1):Delta_t:tspan(2);!
x(1)=x0;!
k1=ODEfunc(t(n),x(n));!
k2=ODEfunc(t(n)+(Delta_t/2),x(n)+(Delta_t/2)*k1);!
k3=ODEfunc(t(n)+(Delta_t/2),x(n)+(Delta_t/2)*k2);!
k4=ODEfunc(t(n)+Delta_t,x(n)+Delta_t*k3);!
x(n+1)=x(n)+((1/6)*k1+(1/3)*(k2+k3)+(1/6)*k4)*Delta_t;!
end!
!
function xp=dxdt(tn,xn) !
xp=-1000*xn+3000-2000*exp(-tn);
1.5
t=0.002
0.5
0.5
1.5
2.5
3.5
3
Exact solution
ode23()
ode23s()
MyEuler()
MyRK4
2.5
1.5
0.5
0.5
1.5
2.5
3.5
At first glance, it looks like all MATLAB functions give the same
solution. But upon close inspection, it is clear that ode23() gives

noisy oscillatory solution whereas the solution computed using
ode23s() agree very well with the analytical solution. So if you have a
stiff system, please use ode23s(), instead of ode23().
The MyEuler() solution computed using t=0.002 is unstable (as
expected) whereas the computed solution using MyRK4() with the

same value of t is stable and accurate.
Implicit methods
Implicit methods are methods where the solution at!
time level (l+1) is defined implicitly. These methods are
useful for solving ODEs that are stiff. In this subject, you
have been introduced to two implicit methods
d
(t) = f ( , t)
dt
Implicit Euler
l+1
Crank-Nicolson
l+1
t f(
l+1
l+1
,t
t
l l
+
f( , t ) + f(
2
l+1
l+1
,t
Example 10.3: Use Implicit Eulers method to solve the equation!

!
! d
t
!
= 1000 + 3000 2000e
!
dt
!
!
with initial conditions x(0)=0.0
d
=
dt
1000 + 3000
2000e
Applying the implicit Eulers formula gives

l
l
1000
1000
l+1
l+1
+ 3000
+ 3000
2000e
This is an linear equation and can easily be solved for l+1
(1 + 1000 t)
l+1
l+1
=
=
t 3000
l
+ t 3000
2000e
l+1
l+1
2000e
2000e
(1 + 1000 t)
l+1
l+1
l+1
=
=
l+1
function MPO10p3()!
clear all;!
close all;!
!
!
Delta_t=0.1;!
t=0:Delta_t:4.0;!
!
%Preallocating memory!
phi=zeros(size(t));!
!
phi(1)=0.0;!
phi(n+1)=(phi(n)+Delta_t*(3000-2000*exp(-t(n+1))))/(1+1000*Delta_t);!
end!
plot(t,phi);!
l
l+1
3
2.5
1.5
0.5
0.5
1.5
2.5
3.5
t 3000
2000e
(1 + 1000 t)
l+1
2.5
1.5
0.5
0.5
1.5
2.5
3.5
t
The solution above was calculated with implicit Eulers method with t=0.1. This is a about 2
orders of magnitude greater than the t value used for RK4 and explicit Eulers methods (see
example O10.2). Also, note, no oscillations in the numerical solution!! In general, for stiff problems,
it is usually better to use implicit methods.
Example 10.4: In ODE workshop 3, you were asked to solve

the following set of ordinary differential equations!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
d 1
dt
d 2
dt
d 3
dt
=
=
=
1 3
1 2
with initial conditions f1(0)=f2(0)=f3(0)=5.0. s=10, r=28 and

b=2.6666667 in the domain 0<t<20. Solve the above set of 3
equations using the Crank-Nicolson method and compare with the
solution obtained using MATLAB function ode45()
Becareful, sometimes you cannot blindly trust the !

default solutions given by MATLAB functions
Applying the Crank-Nicolson scheme will require you to solve the following algebraic!
sytem of equations every time step.
2
6
6
6
6
6
4
1/2 t
1/2 t r 1 l+1
l+1
1/2 t
b
3
1
l+1
l+1
1
2
3
l+1
l+1
l+1
2
1
l+1
l+1
l+1
3
2
l+1
l+1
We have 3 unknowns, f1l+1, f2l+1, f3l+1,

and we have 3 equations.
Use Newton Raphson every time step.
1/2
2
1/2
1/2
t r
t
b
2
2
0
7
7 6
7
7 6
l
l
7=4 0 7
1 3
5
7
5
0
l
function MPO10p4()!
clear all;!
close all;!
!
Delta_t=0.01;!
t=0:Delta_t:20; !
%!
%Preallocating Memory!
%!
phiCN=zeros(length(t),3);!
InitialConditions=[5 5 5];!
phiCN(1,:)=InitialConditions;!
for l=1:length(t)-1!
phiCN(l+1,:)=FindPhilplus1CrankNicolson(Delta_t,phiCN(l,:));!
end!
figure(1);!
plot(t,phiCN(:,1),'b-');!
ylabel('\phi_1(t)','Fontsize',18);!
%Check with Matlab solution!

options=odeset('RelTol',1.0e-3);!
[tmat,phimat] = ode45(@LorenzEquations,[0 20],InitialConditions,options);!
figure(2)!
plot(tmat,phimat(:,1),'k-');!
ylabel('\phi_1(t)','Fontsize',18);!
!!
!!
function p=FindPhilplus1CrankNicolson(Delta_t,Phil)!
eps=1.0; %Just making sure you go through the loop at least once!
p=Phil; %Guess value for p!
while eps>1.0e-13!
[gfunc,jacobian]=ExampleFunctionWithJacobianCN(p,Delta_t,Phil);!
eps=max(abs(gfunc));!
temp=jacobian\(-gfunc); %this is more efficient than inv(jacobian)*(-gfunc)!
p=temp'+p;!
end!
function [gfunc,jac]=ExampleFunctionWithJacobianCN(p, Delta_t,Phil)!

%Functions for Crank-Nicolson method!
gfunc=[p(1)-Phil(1)-(-10*p(1)+10*p(2))*Delta_t/2-(-10*Phil(1)+10*Phil(2))*Delta_t/2; p(2)Phil(2)-(28*p(1)-p(2)-p(1)*p(3))*Delta_t/2-(28*Phil(1)-Phil(2)-Phil(1)*Phil(3))*Delta_t/2;
p(3)-Phil(3)-(-2.666667*p(3)+p(1)*p(2))*Delta_t/2-(-2.666667*Phil(3)+Phil(1)*Phil(2))*Delta_t/
2];!
!
jac=[1+10*Delta_t/2 -10*Delta_t/2 0; (-28+p(3))*Delta_t/2 1+Delta_t/2 p(1)*Delta_t/2;p(2)*Delta_t/2 -p(1)*Delta_t/2 1+2.666667*Delta_t/2];!
function dphidt = LorenzEquations(t,phi)!

dphidt = zeros(3,1);
% a column vector!
dphidt(1) = -10*phi(1)+10*phi(2);!
dphidt(2) = 28*phi(1)-phi(2)-phi(1)*phi(3);!
dphidt(3) = -2.666667*phi(3)+phi(1)*phi(2);
2
6
6
6
6
6
4
1/2 t
1/2 t r 1 l+1
l+1
1/2 t
b
3
1
l+1
l+1
1
2
3
l+1
l+1
l+1
2
1
l+1
l+1
l+1
3
2
l+1
l+1
1/2
2
1/2
1/2
t r
t
b
1
1
2
2
2
3
0
7
7 6
7
7 6
l
l
7=4 0 7
1 3
5
7
5
0
l
20
20
Crank Nicolson Dt=1.0e-2
10
10
1(t)
15
1(t)
15
10
10
15
15
20
10
12
14
16
18
20
20
15
10
10
10
10
15
15
10
12
10
12
14
16
18
20
10
12
14
16
18
20
MATLAB ode45()
15
20
20
1(t)
1(t)
20
14
16
18
20
20

20
15
10
1(t)
10
15
20
10
12
14
16
18
20
Solutions do not match. So need to decrease Dt
20
15
10
1(t)
10
15
20
10
12
14
16
18
20
Solutions match up. So you can be confident that your solution is

converged at Dt=1.0e-5 (approximately)
20
15
10
1(t)
10
15
20
10
12
14
16
18
20
MATLAB ode45()
Solutions do not match!! Is your Crank-Nicolson correct? Is there

a problem with the solution given by MATLAB ode45()?
Check the solution given by MATLAB ode45().

Compute with different values of RelTol. Default is RelTol=1.0e-3.
..but is this good enough for this problem?
Solution using MATLAB function ode45()

20
RelTol=1.0e-3
15
15
10
10
1(t)
1(t)
20
10
10
15
15
20
10
12
14
16
18
20
20
RelTol=1.0e-5
14
16
18
20
12
14
16
18
20
20
RelTol=1.0e-8
15
10
10
10
10
15
15
RelTol=1.0e-12
15
1(t)
1(t)
12
20
20
10
10
12
14
16
18
20
20
10
RelTol=1.0e-3
20
15
10
1(t)
10
15
20
10
12
14
16
18
20
RelTol=1.0e-5
RelTol=1.0e-5
20
15
10
1(t)
10
RelTol=1.0e-8
15
20
0
10
12
14
16
18
20
RelTol=1.0e-8
20
15
10
1(t)
10
RelTol=1.0e-12
15
20
10
12
14
16
18
20
20
15
10
10
15
20
10
12
14
16
18
20
t
20
RelTol=1.0e-12
15
10
1(t)
1(t)
10
15
20
10
12
14
16
18
20

20
15
10
1(t)
10
15
20
10
12
14
16
18
20
RelTol=1.0e-12

Crank Nicholson Method

Uploaded by

Copyright:

Available Formats

Crank Nicholson Method

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Crank Nicholson Method

Uploaded by

Copyright:

Available Formats

What numerical method is being analyzed for accuracy and stability?

What numerical method is being analyzed for accuracy and stability?

What conditions must be satisfied for the Crank-Nicolson method to be stable?

What conditions must be satisfied for the Crank-Nicolson method to be stable?

ENGR90024

COMPUTATIONAL FLUID DYNAMICS

Then Taylor-series expand about tl+1:

Take Eq. (10.1)-(10.2)

Divide the above equation by 2 to give

Apply the Crank-Nicolson method to obtain

The region of stability (||1)

System of Nonlinear ODEs

The Crank-Nicolson method can easily be extended to solve a system of M equations

Example O10.1: Write a MATLAB program that uses the Crank-Nicolson

To do this case study, need to know:

how to write a 2nd-order scalar ODE as a 1st-order vector

Rewrite as a set of 1st-order ODEs:

g1 (p1,old , p2,old , . . . , pM,old )

g2 (p1,old , p2,old , . . . , pM,old )

g3 (p1,old , p2,old , . . . , pM,old )

gM (p1,old , p2,old , . . . , pM,old )

g1 (p1,old , p2,old , . . . , pM,old )

gM (p1,old , p2,old , . . . , pM,old )

For a system consisting of 2 equations, the above simplifies to

@g1 /@p1 @g1 /@p2

We want to use the above equation to solve

Using the notation in previous lectures, we let pn= nl+1 to get

The Jacobian for this system of equation is given by

For the system of equation we are required to solve, this becomes

with initial condition

The exact solution is

Note that this system has a rapidly varying term

which decays very quickly.

which decays a lot slower.

Slowly varying part of !

Fast varying part of the solution!

Use initial conditions x(0)=0 and solve in the domain 0<t<4.

From stability analysis, we can predict that for the computed

solution. But upon close inspection, it is clear that ode23() gives

The MyEuler() solution computed using t=0.002 is unstable (as

expected) whereas the computed solution using MyRK4() with the

Example 10.3: Use Implicit Eulers method to solve the equation!

Applying the implicit Eulers formula gives

This is an linear equation and can easily be solved for l+1

Example 10.4: In ODE workshop 3, you were asked to solve

with initial conditions f1(0)=f2(0)=f3(0)=5.0. s=10, r=28 and

Becareful, sometimes you cannot blindly trust the !

We have 3 unknowns, f1l+1, f2l+1, f3l+1,

%Check with Matlab solution!

function [gfunc,jac]=ExampleFunctionWithJacobianCN(p, Delta_t,Phil)!

function dphidt = LorenzEquations(t,phi)!

Crank Nicolson Dt=1.0e-2

Crank Nicolson Dt=1.0e-6

Crank Nicolson Dt=1.0e-5

Crank Nicolson Dt=1.0e-5

Crank Nicolson Dt=1.0e-2

Solutions do not match. So need to decrease Dt

Crank Nicolson Dt=1.0e-6

Crank Nicolson Dt=1.0e-5