The Crank-Nicolson Method Insulated Boundaries: Douglas Wilhelm Harder, M.Math. LEL

The Crank-Nicolson Method
and
Insulated Boundaries
Douglas Wilhelm Harder, M.Math. LEL

Department of Electrical and Computer Engineering
University of Waterloo
Waterloo, Ontario, Canada
ece.uwaterloo.ca
dwharder@alumni.uwaterloo.ca
© 2012 by Douglas Wilhelm Harder. Some rights reserved.

Outline
This topic discusses numerical approximations to

solutions to the heat-conduction/diffusion equation:
– Consider the Crank-Nicolson method for approximating the heat-
conduction/diffusion equation
– This is an implicit method
• Uses Matlab from Laboratories 1 and 2
• Unconditionally stable
– Defines the characteristics of insulated boundaries
– Implement insulated boundaries into the Crank-Nicolson method
2
Outcomes Based Learning Objectives
By the end of this laboratory, you will:

– Understand the Crank-Nicolson method
– Understand the definition and approximations of insulated
boundaries
3
Review
In Laboratory 1, you solved a boundary-value problem:

– Given the finite-difference equation
d+uk − 1 + duk + d−uk + 1 = g(xk),
there are three unknowns:
uk – 1 uk uk + 1
– This requires us to set up a system

of n – 2 linear equations which must be
solved
– The boundary values give us u1 and un
4
Review
In Laboratory 2, we used an explicit method:

– Given the finite-difference equation
t
ui ,k 1  ui ,k  2  ui 1,k  2ui ,k  ui 1,k 
h
– All the values on the right-hand side

are known, we need only evaluate this
for u2, k + 1 through un x − 1 , k + 1
5
Review
We found the finite-difference equation

t
ui ,k 1  ui ,k 
h 2 u
i 1, k  2ui ,k  ui 1,k 
by substituting
 ui ,k 1  ui ,k
u  xi , tk  
t t
2 ui 1,k  2ui ,k  ui 1,k
u  xi , tk  
 x
2
h2
u  2u
into  2
t x
Both focus on (xi, tk)
6
What happens if we focus on the point (xi, tk + 1)?

 ui ,k 1  ui ,k
u  xi , tk 1  
t h
2 ui 1,k 1  2ui ,k 1  ui 1,k 1
u  i k 1 
x , t 
2 x h2
These focus on (xi, tk + 1)
7
This gives us the finite-difference equation

t
ui ,k 1  ui ,k 
h 2 u
i 1, k 1  2ui , k 1  ui 1, k 1 
The linear equation now has:

– One known ui, k
– Three unknowns ui – 1,k + 1, ui,k + 1, ui + 1, k + 1
8
Compare the two: t

ui ,k 1  ui ,k 
h 2 u
i 1, k 1  2ui , k 1  ui 1, k 1 
t
ui ,k 1  ui ,k 
h 2 u
i 1, k  2ui ,k  ui 1,k 
9
Given two equations, we can add them:
t
ui ,k 1  ui ,k 
h 2 u
i 1, k  2ui , k  ui 1, k 
t
+ ui , k 1  ui , k  2  ui 1, k 1  2ui , k 1  ui 1, k 1 
h
t
2ui ,k 1  2ui ,k  2  ui 1,k  2ui ,k  ui 1,k 
h
t
 2  ui 1,k 1  2ui ,k 1  ui 1, k 1 
h
10

t
First, substitute r 
h2
2ui ,k 1  2ui ,k  r  ui 1,k  2ui ,k  ui 1,k 
 r  ui 1,k 1  2ui ,k 1  ui 1,k 1 
11
Next, collect similar terms and bring

– All unknowns to the left, and
– All knowns to the right
rui 1,k 1  2  r  1 ui ,k 1  rui 1,k 1

 2ui ,k  r  ui 1,k  2ui ,k  ui 1,k 
12
We now have a new finite-difference equation:

rui 1,k 1  2  r  1 ui ,k 1  rui 1,k 1  2ui ,k  r  ui 1,k  2ui ,k  ui 1,k 
Unknowns Knowns
13
As we did in Laboratory 1, we could then write nx – 2

equations
ru1,k 1  2 1  r  u2,k 1  ru3,k 1  2u2,k  r  u1, k  2u2, k  u3, k 
ru2,k 1  2 1  r  u3,k 1  ru4,k 1  2u3,k  r  u2,k  2u3,k  u4,k 

... nx – 2

runx 3,k 1  2 1  r  unx 2,k 1  runx 1,k 1  2unx 2,k  r unx 3,k  2unx 2,k  unx 1,k 
runx 2,k 1  2 1  r  unx 1,k 1  runx ,k 1  2unx 1,k  r u nx  2, k  2unx 1,k  unx ,k 
Unknowns Knowns
14
Again, there appear to be nx unknowns; however, the

boundary conditions provide two of those values:

...

Unknowns Knowns
15
That is: u1, k + 1 = abndry(tk + 1)

un , k + 1 = bbndry(tk + 1)
x

...

Unknowns Knowns
16
We now have nx – 2 linear equations and nx – 2 unknowns
2 1  r  u2,k 1  ru3,k 1  2u2,k  r  u1,k  2u2,k  u3,k   rabndry  tk 1 


...

runx 2,k 1  2 1  r  unx 1,k 1  2unx 1,k  r u nx  2, k 
 2unx 1,k  unx ,k  rbbndry  tk 1 
Unknowns Knowns
17
This can be written in the form Mx = b:

 2 1  r  r  u
   2,k 1 
 r 2 1  r  r   u3,k 1 
  
 r 2 1  r     u4,k 1 
  
   r  
   unx  2,k 1 
 r 2 1  r  r   
 un 1,k 1 
 r 2 1  r    x 

 2u2,k  r  u1,k  2u2,k  u3, k   rabndry  tk 1  
 
 2u3,k  r  u2,k  2u3, k  u4, k  
 
 4,k
2u  r  u3, k  2u 4, k  u5, k  Knowns 
 
 
 2u

 
nx  2, k  r unx 3, k  2unx  2, k  unx 1, k


 2u
 n x 1, k  
 r unx  2,k  2unx 1,k  unx ,k  rbbndry  tk 1   18

Note the structure of b:

 2u2,k  r  u1,k  2u2,k  u3, k   rabndry  t 1  
 
 2u3,k  r  u2,k  2u3, k  u4, k  
 
 4,k
2u  r  u3, k  2u 4, k  u5, k  
 
 
 2u

 
nx  2, k  r unx 3, k  2unx  2, k  unx 1, k

 diff
 2u
 nx 1,k  
 r unx  2,k  2unx 1,k  unx ,k  rbbndry  t 1  

 u2,k   u1,k  2u2,k  u3,k   rabndry  tk 1  
     
u
 3,k   2,k u  2u  u  
3, k 4, k 
 u4,k   u3,k  2u4, k  u5, k   
2 r  
     
 un  2,k   un 3,k  2un 2,k  un 1,k   
 x
  x x x
  
 un 1,k   un  2,k  2un 1,k  un ,k   rbbndry  tk 1   19
 x   x x x   
Approximating the Solution
Thus, given the initial state at time t1, we create a system

of equations with k = 1 to solve for u2, 2 through unx – 1, 2
20
We will simultaneously solve for these values and assign

them to our solution matrix U
21
Given the initial state at time t2, we will create the system
of equations with k = 2 to solve for u2, 3 through unx – 1, 3
22
Again, having solved for the values u2, 3 through unx – 1, 3,

we assign those entries to our matrix U
23
In general, at time tk, we will create the system of

equations with k and then solve for u2, k + 1 through unx – 1, k + 1
24
Doing this, we will fill in the balance of the matrix
25
The diffusion1d Function
The signature will be

function [x_out, t_out, U_out] = ...
crank_nicolson1d( kappa, x_rng, nx, t_rng, nt, u_init, u_bndry )
where
kappa the diffusivity coefficient
x_rng the space range [a, b]
nx the number of points into which we will divide [a, b]
t_rng the time interval [t0, tfinal]
nt the number of points into which we will divide [t0, tfinal]
u_init a function handle giving the initial state uinit:[a, b] → R
u_bndry a function handle giving the two boundary conditions
 abndry  t  
where ubndry  t   
 bbndry  t  
ubndry:[t0, tfinal] → R2
 
26
Step 1: Error Checking
Unlike the previous method we used which was subject

to a catastrophic error if
t
2
 0.5
h
the Crank-Nicolson method is unconditionally stable
– There are no values which will cause the divergence we saw
using the previous technique
t
Never-the-less, if 2
 0.5 , there may be decaying
h
osillations
27
Step 1: Error Checking
Rather than throwing an exception, issue a warning:
warning( 'MATLAB:questionable_argument', ...

'the arguments of %d and %d are sub-optimal', ...
a, b )
This warning will be seen by the user; however,

it will not terminate the execution of the function
28
Step 2: Initialization
It would still be useful to initialize the matrix U and then

use the values as appropriate
nt
uinit  x1  abndry  t2  abndry  t3  abndry  t4  abndry  t5  abndry  t6  abndry  t7  abndry  t8  abndry  t9  abndry  t10  abndry  t11  abndry t12 
uinit  x2  ? ? ? ? ? ? ? ? ? ? ?
uinit  x3  ? ? ? ? ? ? ? ? ? ? ?
uinit  x4  ? ? ? ? ? ? ? ? ? ? ?
nx uinit  x5  ? ? ? ? ? ? ? ? ? ? ?
uinit  x6  ? ? ? ? ? ? ? ? ? ? ?
uinit  x7  ? ? ? ? ? ? ? ? ? ? ?
uinit  x8  ? ? ? ? ? ? ? ? ? ? ?
uinit  x9  bbndry  t2  bbndry  t3  bbndry  t4  bbndry  t5  bbndry  t6  bbndry  t7  bbndry  t8  bbndry  t9  bbndry  t10  bbndry  t11  bbndry t12 
29
Step 3: Solving
As with the previous case, we will find solutions for the

interior points
t for t through t
2 nt
nt
uinit  x1  abndry  t2  abndry  t3  abndry  t4  abndry  t5  abndry  t6  abndry  t7  abndry  t8  abndry  t9  abndry  t10  abndry  t11  abndry t12 
uinit  x2  ? ? ? ? ? ? ? ? ? ? ?
uinit  x3  ? ? ? ? ? ? ? ? ? ? ?
uinit  x4  ? ? ? ? ? ? ? ? ? ? ?
nx uinit  x5  ? ? ? ? ? ? ? ? ? ? ?
uinit  x6  ? ? ? ? ? ? ? ? ? ? ?
uinit  x7  ? ? ? ? ? ? ? ? ? ? ?
uinit  x8  ? ? ? ? ? ? ? ? ? ? ?
uinit  x9  bbndry  t2  bbndry  t3  bbndry  t4  bbndry  t5  bbndry  t6  bbndry  t7  bbndry  t8  bbndry  t9  bbndry  t10  bbndry  t11  bbndry t12 
30
Step 3: Solving
For each time step from 1 to nt – 1, we will, however,

have to perform the following:
3a. Set up the system of linear equations
3b. Solve the system of linear equations, and
3c. Assign the values to the next column of the solution matrix
31
Useful Matlab Commands
There are no new additional Matlab commands for the

Crank-Nicolson method that you have not already been
introduced to in Laboratories 1 and 2
32
Examples
Consider the following example:

– The initial temperature of the bar is 1 oC
– The bar is placed in contact with two barriers at –1 oC and 2 oC
function [u] = u3a_init( x )

u = x*0 + 1;
end
function [u] = u3a_bndry( t )

u = [t*0 - 1; t*0 + 2];
end
33
Examples
[xs, ts, Us] = crank_nicolson1d( 1.5, [0 1], 6, [0 1], 21, @u3a_init, @u3a_bndry );
mesh( ts, xs, Us )
t
 1.875
h2
34
Examples
mesh( ts, xs, Us )
t
 240
h2
Note the transient oscillations
35
Examples
mesh( ts, xs, Us )
t
 60
h2
Note the transient oscillations
36
Examples
mesh( ts, xs, Us )
t
2
h2
37
Examples
Consider an alternate example:

– The initial temperature of the bar is 0 oC
– The bar is placed in contact with two barriers at 1 oC and 4 oC
function [u] = u3b_init( x )

u = x*0;
end
function [u] = u3b_bndry( t )

u = [t*0 + 1; t*0 + 4];
end
38
Examples
[x3b, t3b, U3b] = crank_nicolson1d( 0.25, [0 1], 11, [0 1], 11, @u3b_init, @u3b_bndry );
mesh( t3b, x3b, U3b );
frames3b = animate( U3b );
frames2gif( frames3b, 'plot3b.i.gif' ); t
 2.5
h2
39
Examples
[x3b, t3b, U3b] = crank_nicolson1d( 0.25, [0 1], 41, [0 1], 161, @u3b_init, @u3b_bndry );
mesh( t3b, x3b, U3b );
frames3b = animate( U3b );
t
frames2gif( frames3b, 'plot3b.ii.gif' );
 2.5
h2
40
We have looked at situations where we have known

temperatures or concentrations at each end of the bar;
however, what happens if one end of the bar is
insulated?
41
Terminology
Up to this point, we have discussed boundary values

where the value of the function is specified
– These are termed Dirichlet boundary condition
Alternatively, one can specify the value of the derivative

at the boundary points
– These are termed Neumann boundary conditions
Specifically, we will focus on insulated boundary

conditions where the derivative at the boundaries are
zero
42
Suppose you have a metal bar in contact with a body at

100 oC
– If the bar is insulated, over time, the entire length of the bar will
be at 100 oC
0 oC 100 oC
43
At time t = 0, one end of the bar is brought in contact

with a heat sink at 0 oC; the other end is insulated
0 oC 100 oC
44
Over time, the bar continues to cool
0 oC 100 oC
45
The cooling process will be much slower
0 oC 100 oC
46
Even the furthest end, however, will begin to cool after

some time
0 oC 42 oC
47
Until, ultimately, the entire bar is at 0 oC
0 oC 0 oC
48
The mathematical property of an insulated boundary is

that there is no transfer w.r.t. space of the property
(temperature or concentration) across that boundary:

u  a, tk 1   0
x
or

u  b, tk 1   0
x
49
We could use the O(h) approximations:

 u  a  h, t   u  a, t   u  b, t   u  b  h, t 
u  a, t   u  b, t  
x h x h
However, as other approximations of the 2nd derivative

are O(h2), we will use:
 3u  a, t   4u  a  h, t   u  a  2h, t 
u  a, t  
x 2h
 u  b  2h, t   4u  b  h, t   3u  b, t 
u  b, t  
x 2h
50
Thus, our approximations of the insulated boundary

conditions
 
u  a, tk 1   0 u  b, tk 1   0
x x
are
u  a  2h, tk 1   4u  a  h, tk 1   3u  a, tk 1 
0
2h
3u  b, tk 1   4u  b  h, tk 1   u  b  2h, tk 1 
0
2h
51
First multiply by 2h:

u  a  2h, tk 1   4u  a  h, tk 1   3u  a, tk 1   0
3u  b, tk 1   4u  b  h, tk 1   u b  2h, tk 1   0
Next, substitute
u3,k 1  4u2,k 1  3u1, k 1  0
3unx ,k 1  4unx 1,k 1  unx  2,k 1  0
Finally, solve for u1,k + 1 and unx ,k + 1 :
4 1 4 1
u1,k 1  u2,k 1  u3,k 1 unx ,k 1  unx 1,k 1  unx  2,k 1
3 3 3 3
52
The first and last linear equations we prepared use both

u1, k + 1 and unx , k + 1, respectively
runx 2, 1 
 2 1  r  unx 1,k 1  runx ,k 1  2unx 1,k  r unx 2,k  2unx 1,k  unx ,k 
53
Applying our substitutions:

4 1
u1,k 1  u2,k 1  u3,k 1
3 3
4 
r  u2,k 1  u3,k 1   2 1  r  u2, k 1  ru3, k 1  2u2, k  r  u1, k  2u2, k  u3, k 
1
3 3 
4 
runx  2, 1  2 1  r  u nx 1, k 1  r  nx 1,k 1
3
u 
1
3
u


nx  2, k 1   2unx 1, k  r unx  2, k  2unx 1, k  unx , k 
4 1
unx ,k 1  unx 1,k 1  unx  2,k 1
3 3
54
Thus, the first and last equations simplify to:
 2 
ru3, k 1  2u2, k  r  u1, k  2u2, k  u3, k 
2
 2  r  2,k 1
u 
 3  3
 2 
2
 runx  2,
3
1  

2  
r  unx 1,k 1  2unx 1,k  r unx  2,k  2unx 1,k  unx ,k
3 

55
Recall our original system Mx = b
 2 1  r  r   2u2,k  r  u1,k  2u2,k  u3,k   rabndry  tk 1  

   
 r 2 1  r  r   3,k
2u  r  2,k 3,k 4,k 
u  2u  u 
   
 r 2 1  r     2u 4, k  r  u 3, k  2u 4, k  u 5, k  
  u intr   
   r   

 r 2 1  r  r 
 


2unx  2,k  r unx 3,k  2unx  2, k  unx 1, k  


 r 
2 1  r    2u
 nx 1,k  
 r unx  2,k  2unx 1,k  unx , k  rbbndry  tk 1  

56
Two changes are required if the left boundary is insulated:

– m1,1 and m1,2 are modified
– The original change to b1 is no longer necessary
2 2
2 r  r
3 3
   
  r 2 1  r   r   2u3,k  r  u2,k  2u3,k  u4,k  
   
  r 2 1  r     2u4,k  r  u3,k  2u4,k  u5,k  
  uintr  
   r   

 r 2 1  r  r 
  2u


nx  2, k  r unx 3, k  2unx  2, k  unx 1, k 


 r 2 1  r    2u
 nx 1,k  

57
Two changes are required if the right boundary is insulated:

– mn x – 2, nx – 2 and mnx – 2, nx – 3 are modified
– The original change to bnx – 2 is no longer necessary

   
 r 2 1  r  r   3,k
2u  r  2,k 3,k 4,k 
u  2 u  u 
   
 r 2 1  r     2u 4, k  r  u 3, k  2u 4, k  u 5, k  
  u intr   
   r 2 2  
  r 2  r2u  r u  2u  u   
 r 2 1  r  3 r  3  nx 2,k nx 3, k nx  2, k nx 1, k


 r 
2 1  r    2u
 nx 1,k  

58
How do we indicate an insulated boundary?

– One of the best ideas I have found is to have the boundary
function return NaN
– Recall that NaN is the result of floating-point operations such as
0/0:
>> 0/0
ans =
NaN
– It is also appropriate: an insulated boundary has an undefined

temperature
59
For example, the following would implement an insulated

boundary at the left-hand end point:
function u = u3c_bndry(t)
u = [0*t + NaN;
0*t + 2];
end
60
Problem: IEEE 754 standard requires that NaN ≠ NaN:

>> NaN == NaN
ans =
0
>> NaN ~= NaN

ans =
1
Solution: use the isnan command:

>> isnan( NaN )
ans =
1
61
Suppose that our system begins as follows:

>> [x3c, t3c, U3c] = crank_nicolson1d( 1.5, [0 2], 6, [0 1], 9, @u3c_init, @u3c_bndry );
1 NaN NaN NaN NaN NaN NaN NaN NaN

1 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0
1 2 2 2 2 2 2 2 2
t
2
 1.1719
h
62
Suppose that our system begins as follows:

>> [x3c, t3c, U3c] = crank_nicolson1d( 1.5, [0 2], 6, [0 1], 9, @u3c_init, @u3c_bndry );

1 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0
1 2 2 2 2 2 2 2 2
The first system of equations yields the solution

1.0070
1.0250
1.0858
1.2929
63
These values are assigned to the 2nd column:

1 1.0070 0 0 0 0 0 0 0
1 1.0250 0 0 0 0 0 0 0
1 1.0858 0 0 0 0 0 0 0
1 1.2929 0 0 0 0 0 0 0
1 2 2 2 2 2 2 2 2

1.0070
1.0250
1.0858
1.2929
64
4 1
Now we reuse the formula: u1,2  u2,2  u3,2
3 3
1 1.0010 NaN NaN NaN NaN NaN NaN NaN
1 1.0070 0 0 0 0 0 0 0
1 1.0250 0 0 0 0 0 0 0
1 1.0858 0 0 0 0 0 0 0
1 1.2929 0 0 0 0 0 0 0
1 2 2 2 2 2 2 2 2

1.0070
1.0250
1.0858
1.2929
65
Repeating the process...

1 1.0070 0 0 0 0 0 0 0
1 1.0250 0 0 0 0 0 0 0
1 1.0858 0 0 0 0 0 0 0
1 1.2929 0 0 0 0 0 0 0
1 2 2 2 2 2 2 2 2
The second system of equations yields the solution

1.0404
1.1077
1.2735
1.6133
66
Copy the values:

1 1.0070 1.0404 0 0 0 0 0 0
1 1.0250 1.1077 0 0 0 0 0 0
1 1.0858 1.2735 0 0 0 0 0 0
1 1.2929 1.6133 0 0 0 0 0 0
1 2 2 2 2 2 2 2 2

1.0404
1.1077
1.2735
1.6133
67
4 1
And use the formula: u1,2  u2,2  u3,2
3 3
1 1.0010 1.0179 NaN NaN NaN NaN NaN NaN
1 1.0070 1.0404 0 0 0 0 0 0
1 1.0250 1.1077 0 0 0 0 0 0
1 1.0858 1.2735 0 0 0 0 0 0
1 1.2929 1.6133 0 0 0 0 0 0
1 2 2 2 2 2 2 2 2

1.0404
1.1077
1.2735
1.6133
68
If the insulated boundary condition is at the other end,

we would use the formula:
4 1
unx ,k 1  unx 1,k 1  unx  2,k 1
3 3
to calculate the missing entry
69
Step 3: Solving
We only need two small changes:

– First, in the argument checking, nx ≥ 4
– Second, for each time step from 1 to nt – 1, we will, however,
have to perform the following:
3a. Set up the system of linear equations and modify:
– The vector if it is a Dirichlet boundary condition
– The matrix if it is an insulated boundaries
3b. Solve the system, and
3c. Copy the values back to U in the next column
– If the boundaries are insulated in that column, use the
appropriate formula to find the end points
70
Examples
Ultimately, we get the image:

[x3c, t3c, U3c] = crank_nicolson1d( 1.5, [0 2], 6, [0 1], 9, @u3c_init, @u3c_bndry );
mesh( t3c, x3c, U3c )
71
Examples
If we increase the resolution:

[x3d, t3d, U3d] = crank_nicolson1d( 1.5, [0 2], 20, [0 1], 100, @u3c_init, @u3c_bndry );
mesh( t3d, x3d, U3d )
72
Examples
After three seconds, the temperature near the insulated

boundary has increased significantly
[x3e, t3e, U3e] = crank_nicolson1d( 1.5, [0 2], 20, [0 3], 600, @u3c_init, @u3c_bndry );
mesh( t3e, x3e, U3e )
frames = animate( U3e );
frames2gif( frames, 'U3e.gif' );
73
Summary
We have looked at the heat-conduction/diffusion

equation
– We developed the Crank-Nicolson method
– You will implement the algorithm
– Unlike the previous implementation, there are less
t
restrictions on
h2
• Large values of this ratio may cause transient oscillations
– We defined insulated boundaries
• Considered their approximation using Matlab
74
References
[1] Glyn James, Modern Engineering Mathematics, 4th Ed., Prentice

Hall, 2007, p.782.
[2] Glyn James, Advanced Modern Engineering Mathematics, 4th Ed.,

Prentice Hall, 2011, p.164.
75

The Crank-Nicolson Method Insulated Boundaries: Douglas Wilhelm Harder, M.Math. LEL

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

The Crank-Nicolson Method Insulated Boundaries: Douglas Wilhelm Harder, M.Math. LEL

Uploaded by

Copyright:

Available Formats

The Crank-Nicolson Method

Douglas Wilhelm Harder, M.Math. LEL

© 2012 by Douglas Wilhelm Harder. Some rights reserved.

This topic discusses numerical approximations to

Outcomes Based Learning Objectives

By the end of this laboratory, you will:

In Laboratory 1, you solved a boundary-value problem:

– This requires us to set up a system

In Laboratory 2, we used an explicit method:

– All the values on the right-hand side

We found the finite-difference equation

Both focus on (xi, tk)