International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No.

3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

Crank Nicolson Method for Solving

Parabolic Partial Differential Equations
Fadugba S. E. 1 , Edogbanya O. H 2 , Zelibe S. C 3
1
Department of Mathematical Sciences, Ekiti State University, Ado Ekiti, Nigeria
[email protected]
2
Department of Mathematics, Federal University, Lokoja, Kogi State, Nigeria
[email protected]
3
Department of Mathematics, Delta State College of Physical Education, Mosogar, Nigeria
[email protected]
Abstract
This paper presents Crank Nicolson method for solving parabolic partial differential equations. Crank
Nicolson method is a finite difference method used for solving heat equation and similar partial
differential equations. This method is of order two in space, implicit in time, unconditionally stable
and has higher order of accuracy.
Key words: Crank Nicolson Method, Finite Difference Method, Exact Solution, Parabolic Equation,
Stability
Mathematics Subject Classification: 35A20, 35A35, 35B35, 35K05

Corresponding Author: [email protected], +2348067032044, +2347052741523

I. INTRODUCTION
The finite difference approximations are one of the simplest and of the oldest methods to solve partial
differential equations. It was already known by L. Euler (1707-1783) ca. 1768, in one dimension of
space and was probably extended to two dimensions by C. Runge (1856-1927) ca. 1908. The advent
of finite difference techniques in numerical applications began in the early 1950s and their
development was stimulated by the emergence of computers that offered a convenient framework for
dealing with complex problems of science and technology. Theoretical results have been obtained
during the last five decades regarding the accuracy, stability and convergence of the finite difference
method for partial differential equations.

The mathematical formulation of most problems in science involving rates of change with respect to
two or more independent variables, usually representing time, often leads to a partial differential
equation.

Problems involving time as one independent variable sometimes lead to parabolic partial differential
equations, the simplest of which is the diffusion equation, derived from the theory of heat conduction
[11]. The diffusion equation plays an important role in a broad range of practical applications such as
fluid mechanics. Only a limited number of special types of parabolic equation have been solved
analytically and the usefulness of these solutions is further restricted to problems involving shapes for
which the boundary conditions can be satisfied. In such cases numerical methods are some of the very
few means of solution.

Crank Nicolson Method for solving parabolic partial differential equations was developed by John
Crank and Phyllis Nicolson in the mid-20th century. A practical method for numerical evaluation of
partial differential equations of the heat conduction type was considered by [4]. [7] modified the

8
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

simple explicit scheme and proved that it is much more stable than the simple explicit case, enabling
larger time steps to be used. [8] considered the stability and accuracy of finite difference method for
option pricing. However, according to [2], the accuracy of the simple explicit method is barely
improved upon.

There are many exhaustive texts on this subject such as ([1], [3], [5], [6], [9], [10], [12], [13]), just to
mention few.

In this paper we shall only consider the accuracy of Crank Nicolson method for solving parabolic
partial differential equations.

II. CRANK NICOLSON METHOD

This section presents Crank Nicolson method for solving parabolic partial differential equations as
follows:

2.1 PARABOLIC PARTIAL DIFFERENTIAL EQUATION

Partial differential equations occur frequently in mathematics, natural sciences and engineering.

 Advection Equation
These are problems involving rates of change of functions of several variables. For examples

f f
v 0
t x
 Heat Equation
f 2 f
D 2 0
t x
 Poisson Equation
2 f 2 f
   u ( x, y )
x 2 y 2
 Wave Equation
2 f 2  f
 0
2

t 2 y 2
c

In these equations above x, y are the space coordinates, v, D, c are real positive constants and t, x are
often viewed as time and space coordinates respectively.

The general second order linear partial differential equation with two independent variables and one

 2u  2u  2u
dependent variable is given by

A 2 B C 2  D  0
x xy y
(1)

where A, B, C are functions of the independent variables, x , y and D can be a function of

u u
. If B 2  4 AC  0 , (1) is called a parabolic partial differential equation. One of
x y
x, y , u , and

the simple examples of a parabolic PDE is the heat-conduction equation for a metal rod shown in
Figure I below;

9
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

FIGURE I: A METAL ROD

1 f  2 f

 t x 2
(2)

Where f  Temperature as a function of location, x and time, t in which the thermal diffusivity
and  is given by


C
where   Thermal conductivity of rod material,   Density of rod material, C  Specific heat of
the rod material

2.2 FINITE DIFFERENCE METHODS

The finite difference methods attempt to solve partial differential equations by approximating the
differential equation over the area of integration by a system of algebraic equations. They are a means
of obtaining numerical solutions to partial differential equations.
The most common finite difference methods for the solution of partial differential equations are:
 Explicit method
 Implicit method
 Crank Nicolson Method

These methods are closely related but differ in stability, accuracy and execution speed. In the
formulation of a partial differential equation problem, there are three components to be considered:
• The partial differential equation.
• The region of space-time on which the partial differential equation is required to be satisfied.
• The ancillary boundary and initial conditions to be met.
Let us consider the diagram below:

Grid point involved in time difference

Grid point involved in space difference

i-1 i i+1 i-1 i i+1 i-1 i i+1
n n n

n+1 n+1 n+1

Explicit Fully implicit Crank-Nicolson

FIGURE II: A COMPUTATIONAL DIAGRAM FOR EXPLICIT AND IMPLICIT
METHODS

In finite difference methods we replace the partial derivative occurring in the partial differential
equations by approximations based on Taylor series expansions of function near the point or points of
interest. The derivative we seek is expressed with many desired order of accuracy.

evaluated from a combination of the derivatives at time steps n and n  1

The finite difference form of the heat equation (2) is now given with the spatial second derivative

10
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

1  f i n1  f i n   f n  2 f i n  f i n1   f i n11  2 f i n1  f i n11 

   m i 1     
  t 


 (x) 2 


 (x) 2 

(1 m) (3)

From the above formula, we will have an explicit scheme when m  1 and an implicit scheme
when m  0 as shown below in equations (4) and (5) respectively:
1  f i n1  f i n   f i n1  2 f i n  f i n1 
  
  t  
  (x) 2 

(4)

1  f i n1  f i n   f i n11  2 f i n1  f i n11 

  
  t  
   

2
(5)
( x )
In fact m can be any value between 0 and 1, however a common choice for m  0.5 . (3) becomes;
1  f i n1  f i n  1  f i n1  2 f i n  f i n1  1  f i n11  2 f i n1  f i n11 
     
  t  2
    2
   

2 2
(6)
( x ) ( x )

The above equation (6) can be arranged so that the temperatures at the present time step (n  1) are on
(6) is called the Crank-Nicolson scheme which is the average of the explicit and implicit schemes.

the left hand side. By applying these equations to all the nodes, we shall obtain a system with

1   t  n 1   t  n 1 1   t  n 1 1   t  n
tridiagonal coefficient matrix.

   f i 1  1  2  f i    f i 1    f i 1
2  (x) 2   (x) 2  2  (x) 2  2  (x) 2 
  t  n   t  n
(7)
 1  2  f i    f i 1
 (x) 2   (x)
2

 t  t
is known as the diffusion number and will be denoted by r , i. e. r 
(x) (x) 2
The expression 2
,

f i 1  1  2r  f i n1  f i n11  f i n1  1  2r  f i n  rf i n1

then (7) becomes

r n1 r r
(8)

   
2 2 2

 r f i n11  f i n11  2(1  r ) f i n1  r f i n1  f i n1  2(1  r ) f i n

Therefore,
(9)
(9) is called the Crank Nicolson approximation of the problem and its associated boundary and initial
conditions are as given.

If we divide the x  interval 0  x  1 into k equal interval, we have (k  1) internal mesh points per
time row. Then for n  1 and i  1, 2,..., k 1 , (9) gives a system of (k  1) linear equations for the (k  1)
unknown values f11 , f 21 , f31 ,... f k11 in the first time row in terms of the initial values f00 , f10 , f 20 ,... f k0 and
the boundary values f01  0, f k1  0 . Similarly for n  1, n  2 , and so on; that is for each time row we
have to solve such a system of (k  1) linear equations resulting from (9). Equation (9) can be written

Tf  tb
in a matrix equation as
(10)
where the unknown f  f n 1 , the known concentrations b  f n and T , t are tri-diagonal matrices of
coefficients defined as

11
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

(2  2r ) r 0   f1n1  2(1  r ) 0   f1n 

    
0   f 2n1   0   f 2n 
0 r 0
 r 2(1  r ) r  r 2(1  r ) r
 0 r 2(1  r ) 0   f3n1    0 2(1  r ) 0   f3n 
      
r (11)

 r     r  
 0 0 0 r 2(1  r )   f kn11   0 0 0 r 2(1  r )   f kn1 

 t
Crucial to the convergence of this method is the condition (Turner, 1994)
r  , this implies that
(x)
1
2
2
(x)
t 
2

2
(12)

choose x small, which makes t very small by (12). This will make the computation exceptionally
Condition (12) is a drawback in practice. Indeed, in order to attain sufficient accuracy, we have to

 t
lengthy, as more time levels will be required to cover the region. A method that imposes no such
restriction as r 
(x) 2
was proposed by Crank and Nicolson in [11]

2.3 THE STABILITY OF CRANK NICOLSON METHOD

   
 r f i n11  f i n11  2(1  r ) f i n1  r f i n1  f i n1  2(1  r ) f i n
Consider the equation (9) given below:

f i n   n (1) i
Worse case solution is given below as:

(1)  (1) 
(13)
Substituting (13) into (9), we have that
 r n1 (1) (1) i 1  2(1  r )  n1 (1) i  r n (1) (1) i 1  2(1  r )  n (1) i
i 1 i 1 i 1 i 1

[r (1)  1  2(1  r )  r (1)  1]  r (1)  1  2(1  r )  r (1)  1

1  2r
The last equation above can also be written as

1  2r
, then

1  2r
    1,    0
1  2r
The Crank Nicolson Method is unconditionally stable and has higher order of accuracy. The price of
solving a tridiagonal system at each step is worth paying since this method allows large step sizes.
This is the most popular numerical method for heat equation.

III. NUMERICAL EXAMPLES AND RESULTS

This section presents some numerical examples and results as follows:

3.1 NUMERICAL EXAMPLES

EXAMPLE 1
We shall use the Crank-Nicolson method to solve the partial differential equation

12
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

f  2 f

t x 2
8 (14)

Initial conditions: f ( x,0)  2000, 0  x  4

with the following initial and boundary conditions:
(15)
f 0.36 f  25.2, x4

x 0, x0
Boundary conditions: (16)

   
SOLUTION:

 r f i n11  f i n11  2(1  r ) f i n1  r f i n1  f i n1  2(1  r ) f i n

Using the Crank-Nicolson method, from (9), the heat equation becomes

 t
Where r 
(x) 2
Let x  1, t  1, then we have that
 t
r   
(x)
0.125 1 1
2
1 8 2

   1
 
 1
Substituting r into the above equation (9), we have

 f i n11  f i n11  21   f i n1  f i n1  f i n1  21   f i n

1 1
 8  8

   
8 8

 0.125 f i n11  f i n11  2.25 f i n1  0.125 f i n1  f i n1  1.75 f i n

Therefore,
(17)

x=0 x=1 x=2 x=3 x=4

i=0 i=1 i=2 i=3 i=4 i=5 i=6

FIGURE III: THE NODES AT ANY GIVEN TIME STEP

There are five unknown temperatures at any given time as shown in Figure 2. The nodes
corresponding to i = 0 and i = 5 are required to accommodate the derivative boundary conditions at

For node i  1, (17) yields

   
the end points.

 0.125 f 0n1  f 2n1  2.25 f1n1  0.125 f 0n  f 2n  1.75 f1n (18)

n 1 n 1 n 1

 
f can be solved in terms of f and f using the boundary condition
f
0 1 2

 0.36 f  25.2  f 2n1  f 0n1  0.36 f1n1  25.2

2x
1
x x 0

f 2n1  f 0n1  2(1)(0.36 f1n1  25.2)

Therefore,

f 2n1  f 0n1  0.72 f1n1  50.4

f 0n1  f 2n1  0.72 f1n1  50.4

Then,
(19)
n n n
Similarly, f can be solved in terms of f and f using the same boundary condition
f 0n  f 2n  0.72 f1n  50.4
0 1 2

(20)
Substituting (19) and (20) into (18), we have

13
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

  
 0.125 f 2n1  0.72 f1n1  50.4  f 2n1  2.25 f1n1  0.125 f 2n  0.72 f1n  50.4  f 2n  1.75 f1n 
2.34 f1n1  0.25 f 2n1  1.66 f1n  0.25 f 2n  12.6
Simplifying the last equation further, we obtain

For 2  i  4 , we have (17), which is given by

(21)

 0.125 f i n11  f i n11   2.25 f i n1  0.125 f i n1  f i n1   1.75 f i n

At node (i  5),
 0.125 f 4n1  f 6n1   2.25 f 5n1  0.125 f 4n  f 6n   1.75 f 5n (22)
f
 0, therefore f 4  f 6
x
Using the boundary condition
x4

 0.25 f 4n1  2.25 f 5n1  0.125 f  1.75 f 5n

4
n
(23)
The results generated are shown in the Table 1 below using MATLAB codes.

TABLE I: THE RESULTS GENERATED FROM THE FORMULATION OF THE FINITE

DIFFERENCE USING CRANK-NICOLSON METHOD FOR A PARABOLIC EQUATION
WITH DERIVATIVE BOUNDARY CONDITIONS
t
x  0.0 x  1.0 x  3.0
f1 f2 f3 f4 f5
x  2.0 x  4.0
0.0 2000.00 2000.00 2000.00 2000.00 2000.00
1.0 1850.65 1991.68 1999.54 1999.97 2000.00
2.0 1741.57 1970.71 1997.54 1999.82 1999.97
3.0 1659.03 1943.41 1993.27 1999.34 1999.89
4.0 1594.36 1913.39 1986.68 1998.35 1999.65
5.0 1541.99 1882.70 1978.03 1996.69 1999.18
6.0 1498.33 1852.43 1967.70 1994.27 1998.36
7.0 1460.99 1823.17 1956.04 1991.06 1997.09
8.0 1428.39 1795.18 1943.39 1987.03 1995.30
9.0 1399.43 1768.54 1930.02 1982.23 1992.93
10.0 1373.33 1743.25 1916.17 1976.68 1989.94
11.0 1349.56 1719.25 1902.00 1970.44 1986.30
12.0 1327.69 1696.47 1887.65 1963.56 1982.01
13.0 1307.43 1674.81 1873.23 1956.10 1977.08
14.0 1288.54 1654.18 1858.80 1948.12 1971.53
15.0 1270.83 1634.51 1844.42 1939.65 1965.39
16.0 1254.16 1615.71 1830.14 1930.77 1958.68
17.0 1238.40 1597.70 1815.98 1921.50 1951.45
18.0 1223.45 1580.42 1801.96 1911.89 1943.73
19.0 1209.22 1563.81 1788.09 1901.99 1935.55
20.0 1195.65 1547.82 1774.37 1891.82 1926.96

EXAMPLE 2
Use the Crank-Nicolson method to solve the partial differential equation
f  2 f

t x 2
50 (24)

f ( x,0)  100 x, 0  x  1;
with the following initial and boundary conditions:


f ( x,0)  200  100 x, 0  x  2
Initial conditions: (25)

Boundary conditions: f (0, t )  f (2, t )  0 (26)

14
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

Let x  0.2, t  0.5

SOLUTION

x = 0.2, t = 0.5, we have

 t
r   
(x)
(0.02)(0.5) 1 1
2
(27)
(0.2) 2 4 2

   
 0.25 f i n11  f i n11  2(1  0.25) f i n1  0.25 f i n1  f i n1  2(1  0.25) f i n
Substituting (27) into (9), we obtain
(28)
 0.25 f n 1
i 1  2.5 f i n 1
 0.25 f n 1
i 1  0.25 f i 1
n
 1.5 f i  0.25 f
n
i 1
n
(29)

X=0.0 X=0.2 X=0.4 X=0.6 X=0.8 X=1.0

i=0 i=1 i=2 i=3 i=4 i=5 i=6

FIGURE IV: THE NODES AT ANY GIVEN TIME STEP

corresponding to i  0 and i  6 are required to accommodate the derivative boundary conditions at

There are six unknown temperatures at any given time as shown in Figure III. The nodes

the end points.

The initial conditions for the problem are given in the Table II below;

TABLE II
i 1 2 3 4 5 6
t\x 0 0.2 0.4 0.6 0.8 1.0
0 0 20 40 60 80 100
0.5 0 V W X Y Z

Due to the symmetry with respect to node (i  6) , we have

From the above Table II, V ,W , X , Y , Z are to be calculated.

 0.5 f 5n1  2.5 f 6n1  0.5 f 5n  1.5 f 6n

The temperatures at the time t  0.5 are the solutions of the following equations in the Table III
below;

i2  2.5 f 2  0.25 f 3

TABLE III
= 40
i3  0.25 f 2  2.5 f 3  0.25 f 4 = 80
i4  0.25 f 3  2.5 f 4  0.25 f 5 = 120
i5  0.25 f 4  2.5 f 5  0.25 f 6 = 160
i6  0.25 f 5  2.5 f 6 = 190

The results for the above set of equations in Table III above are listed in the Table IV below;

TABLE IV: THE RESULTS GENERATED AT t  0, t  0.5

i 1 2 3 4 5 6
t\x 0 0.2 0.4 0.6 0.8 1.0

15
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

0 0 20 40 60 80 100
0.5 0 20 39.99 59.92 79.18 91.84

From the above Table IV, we can see that V  20, W  39.99, X  59.92, Y  79.18, Z  91.84
The results obtained for other values of t using MATLAB code are shown in the Table V below

TABLE V: THE RESULTS GENERATED FROM CRANK NICOLSON METHOD FOR THE
SOLUTION OF PARTIAL DIFFERENTIAL EQUATION WITH INITIAL AND BOUNDARY
CONDITIONS
t
x  0.0 x  0.2 x  0.6
f1 f2 f3 f4 f5 f6
x  0.4 x  0.8 x  1.0
0.0 0.0 20.00 40.00 60.00 80.00 100.00
0.5 0.0 20.00 39.99 59.92 79.18 91.84
1.0 0.0 19.99 39.94 59.59 77.18 86.39
1.5 0.0 19.97 39.82 58.97 75.09 82.31
2.0 0.0 19.92 39.59 58.12 72.89 78.98
2.5 0.0 19.84 39.25 57.12 70.77 76.12
3.0 0.0 19.71 38.82 56.03 68.75 73.58
3.5 0.0 19.54 38.31 54.89 66.83 71.26
4.0 0.0 19.33 37.73 53.72 64.99 69.12
4.5 0.0 19.08 37.11 52.54 63.25 67.12
5.0 0.0 18.80 36.44 51.36 61.57 65.24
5.5 0.0 18.50 35.75 50.19 59.97 63.45
6.0 0.0 18.18 35.04 49.03 58.42 61.75
6.5 0.0 17.84 34.32 47.89 56.93 60.12
7.0 0.0 17.50 33.59 46.77 55.49 58.56
7.5 0.0 17.14 32.86 45.67 54.10 57.05
8.0 0.0 16.79 32.14 44.59 52.75 55.60
8.5 0.0 16.43 31.41 43.53 51.44 54.20
9.0 0.0 16.07 30.70 42.49 50.17 52.84
9.5 0.0 15.71 29.99 41.47 48.94 51.53
10.0 0.0 15.36 29.30 40.48 47.73 50.25

EXAMPLE 3 [11]

 f 2 f
Let us consider the following linear equation.

 0  x  1, t  0
t  x2
(30)

f ( x,0)  sin  x ; f (0, t )  f (1, t )  0

Subject to the following initial and boundary conditions:
(31)

f ( x, t )  e  t sin x
The exact solution of the above equation is given by:
2

   
(32)

 r f i n11  f i n11  2(1  r ) f i n1  r f i n1  f i n1  2(1  r ) f i n

From (9), we have that

 t
Where r 
(x) 2
The Crank-Nicolson approximation (9) of the problem (30) and its associated boundary and initial

2(1  r ) fi n1  r ( fi n11  fi n11 )  2(1  r ) fi n  r ( fi n1  fi n1 )

conditions (31) and (32) respectively are given below as follows:
(33)

16
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

f0n  0, f10  sin  i x, f10n  0 



for i  1,2,3,4,5, n  0,1,2,...,10 

(34)

Taking x  0.1 and  t  0.001, we obtain the results to this problem with Matlab codes.

At x  0.5 and t  0.005, 0.006, 0.007 , we have the temperature values for the Crank-Nicolson
solution ( f ), as well as their corresponding exact solution values ( f  ), and their corresponding
errors, (e) in Table VI below;

x  0.5,  t  0.001
f 
TABLE VI: CRANK NICOLSON METHOD AND EXACT SOLUTION AT

e f f
f
t Crank Nicolson 
Exact Solution, Error,
Method,
0.005 0.9519 0.9519 0.0000
0.006 0.9423 0.9425 0.0002
0.007 0.9328 0.9333 0.0005

A rod of steel is subjected to a temperature of 100C on the left end and 25C on the right end. If
EXAMPLE 4

from t  0 to t  9 seconds. Use x  0.01m and t  3s .

the rod is of length 0.05m , use Crank-Nicolson method to find the temperature distribution in the rod

Given
  54 ,   7800 3 , C  490
W kg J
mK kg  K
. (35)

The initial temperature of the rod is 20C .

m

SOLUTION:

   1.4129 10 5 m 2 / s
C 7800  490
54
(36)

t
Then
r  1.412  10 5  0.4239
x  0.012
3
2 (37)

i0 1 2 3 4 5

0.01m

FIGURE V: SCHEMATIC DIAGRAM SHOWING THE NODE REPRESENTATION IN THE


MODEL

f 0n  100C, f 5n  25C for n  0,1,2,3

The boundary conditions are

The initial temperature of the rod is 20C , that is, all the temperatures of the nodes inside the rod are
(38)

at 20C , t  0 except for the boundary nodes given by (38). This could be represented as

17
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

f i 0  20C, for i  1,2,3,4 . (39)

The initial temperature at the nodes inside the rod (when t  0 second) is given by
f 00  100C

from (38)
f10  20C, f 20  20C, f 30  20C, f 40  20C from (39)
f  25C
0
from (38)
The temperature at the nodes inside the rod when t  3 seconds, from (38), we have the boundary
5


condition of the form:
f 01  100C , f 51  25C
For all the interior nodes, setting n  0 and i 1, 2, 3, 4 in (9) gives the following equations
i 1
 f 01  2(1   ) f11  f 21  f 00  2(1   ) f10  f 20
(0.4239  100)  2(1  0.4239) f11  0.4239 f 21  (0.4239)100  2(1  0.4239)20  (0.4239)20
 42.39  2.8478 f11  0.4239 f 21  42.39  23.044  8.478
2.8478 f11  0.4239 f 21  116.30
i 2
(40)

 f11  2(1   ) f 21  f 31  f10  2(1   ) f 20  f 30

 0.4239 f11  2(1  0.4239) f 21  0.4239 f 31  (0.4239)20  2(1  0.4239)20  (0.4239)20
 0.4239 f11  2.8478 f 21  0.4239 f 31  40.000
i 3
(41)

 f 21  2(1   ) f 31  f 41  f 20  2(1   ) f 30  f 40
 0.4239 f 21  2(1  0.4239) f 31  0.4239 f 41  (0.4239)20  2(1  0.4239)20  (0.4239)20
 0.4239 f 21  2.8478 f 31  0.4239 f 41  40.000
i4
(42)

 f 31  2(1   ) f 41  f 51  f 30  2(1   ) f 40  f 50
 0.4239 f 31  2(1  0.4239) f 41  (0.4239)25  (0.4239)20  2(1  0.4239)20  (0.4239)25
 0.4239 f 31  2.8478 f 41  10.598  8.478  23.044  10.598
 0.4239 f 31  2.8478 f 41  52.718 (43)
The coefficient matrix in the above set of equations is tridiagonal. Special algorithms such as
Thomas’ algorithm are used to solve equation with tridiagonal coefficient matrices
 2.8478  0.4239   f11  116.30 
 0.4239 2.8478  0.4239   1  40.000
0 0
  f 2    
 0.4239 2.8478  0.4239  f 31  40.000
0
 0
    
 0 0  0.4239 2.8478   f 41  52.718
The above matrix is tridiagonal. Solving the above matrix we get

18
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

 f11  44.3720
 1  
 f 2   23.7460
 f 31  20.7970
 1  
 f 4  21.6070
Hence, the temperature at all the nodes at time, t  3 seconds is
 f 01   100 
 1  
 f1  44.3720
 f 21  23.7460
 1   
 f 3  20.7970
 f 1  21.6070
 41   
 f 5   25 
The temperature at the nodes inside the rod when t  6 seconds:

f 02  100C , f 52  25C Boundary Condition (38)
For all the interior nodes, putting n  1 and i 1, 2, 3, 4 in (9) gives the following equations
i 1
 f 02  2(1   ) f12  f 22  f 01  2(1   ) f11  f 21
(0.4239  100)  2(1  0.4239) f12  0.4239 f 22 
(0.4239)100  2(1  0.4239)44.372  (0.4239)23.746
 42.39  2.8478 f12  0.4239 f 22  42.39  51.125  10.066
2.8478 f12  0.4239 f 22  145.971
i 2
(44)

 f 12  2(1   ) f 22  f 32  f 11  2(1   ) f 21  f 31
 0.4239 f 12  2(1  0.4239) f 22  0.4239 f 32 
(0.4239)44.372  2(1  0.4239)23.746  (0.4239)20.797
 0.4239 f  2.8478 f 22  0.4239 f 32  18.809  27.360  8.8158
2

 0.4239 f  2.8478 f 22  0.4239 f 32  54.985

1
2

i 3
1 (45)

 f 22  2(1   ) f 32  f 42  f 21  2(1   ) f 31  f 41
 0.4239 f 22  2(1  0.4239) f 32  0.4239 f 42 
(0.4239)23.746  2(1  0.4239)20.797  (0.4239)21.607
 0.4239 f 22  2.8478 f 32  0.4239 f 42  10.066  23.962  9.1592
 0.4239 f 22  2.8478 f 32  0.4239 f 42  43.187
i 4
(46)

19
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

 f 32  2(1   ) f 42  f 52  f 31  2(1   ) f 41  f 51
 0.4239 f 32  2(1  0.4239) f 42  (0.4239)25 
(0.4239)20.797  2(1  0.4239)21.607  (0.4239)25
 0.4239 f 32  2.8478 f 42  10.598  8.8158  24.896  10.598
 0.4239 f 32  2.8478 f 42  54.908 (47)
The simultaneous linear equations (44) – (47) can be written in matrix form as
 2.8478  0.4239   f12  145.971
 0.4239 2.8478  0.4239   2   54.985 
0 0
  f 2    
 0.4239 2.8478  0.4239  f 32   43.187 
0
 0
    
 0 0  0.4239 2.8478   f 42   54.908 

 f12  55.8830
Solving the above set of equations, we get

 2  
 f 2   31.0750
 f 32  23.1740
 2  
 f 4  22.7300
Hence, the temperature at all the nodes at time, t  6 seconds is
 f 02   100 
 2  
 f1  55.8830
 f 22  31.0750
 2   
 f 3  23.1740
 f 2  22.7300
 42   
 f 5   25 
The temperature at the nodes inside the rod when t  9 seconds:
f 03  100C , f 53  25C Boundary Condition (38)
For all the interior nodes, setting n  2 and i 1, 2, 3, 4 in (9) gives the following equations
i 1
 f 03  2(1   ) f13  f 23  f 02  2(1   ) f12  f 22
(0.4239  100)  2(1  0.4239) f13  0.4239 f 23 
(0.4239)100  2(1  0.4239)55.883  (0.4239)31.075
 42.39  2.8478 f  0.4239 f 23  42.39  64.388  13.173
3

2.8478 f13  0.4239 f 23  162.34

1

i 2
(48)

 f 13  2(1   ) f 23  f 33  f 12  2(1   ) f 22  f 32
 0.4239 f 13  2(1  0.4239) f 23  0.4239 f 33 
(0.4239)55.883  2(1  0.4239)31.075  (0.4239)23.174
 0.4239 f 13  2.8478 f 23  0.4239 f 33  23.689  35.805  9.8235
 0.4239 f13  2.8478 f 23  0.4239 f 33  69.318 (49)

20
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

i 3
 f 23  2(1   ) f 33  f 43  f 22  2(1   ) f 32  f 42
 0.4239 f 23  2(1  0.4239) f 33  0.4239 f 43 
(0.4239)31.075  2(1  0.4239)23.174  (0.4239)22.730
 0.4239 f 23  2.8478 f 33  0.4239 f 43  13.173  26.701  9.635
 0.4239 f 23  2.8478 f 33  0.4239 f 43  49.509
i 4
(50)

 f 33  2(1   ) f 43  f 53  f 32  2(1   ) f 42  f 52
 0.4239 f 33  2(1  0.4239) f 43  (0.4239)25 
(0.4239)23.174  2(1  0.4239)22.730  (0.4239)25
 0.4239 f 33  2.8478 f 43  10.598  9.8235  26.190  10.598
 0.4239 f 33  2.8478 f 43  57.210 (51)
The simultaneous linear equations (48) – (51) can be written in matrix form as
 2.8478  0.4239   f13  162.34 
 0.4239 2.8478  0.4239   3  69.318
0 0
  f 2    
 0.4239 2.8478  0.4239  f 33  49.509
0
 0
    
 0 0  0.4239 2.8478   f 43  57.210

 f13  62.6040
Solving the above set of equations, we get

 3  
 f 2   37.6130
 f 33  26.5620
 3  
 f 4  24.0420
Hence, the temperature at all the nodes at time, t  9 seconds is
 f 03   100 
 3  
 f1  62.6040
 f 23  37.6130
 3   
 f 3  26.5620
 f 3  24.0420
 43   
 f 5   25 

f ( x, t )  100  1500 x   
10(1) m  160 
 
 exp (1.4129  10 )20m  t sin(20mx) (52)
The Exact solution to the above problem in example 4 is given by:

5

m 1  m 
2

We shall present in the Table VII below, the results obtained at interior nodes n  2 and i 1, 2, 3, 4
only in the context of exact solution by substituting the values of x and t gives the temperature
inside the rod at a particular location and time.

21
International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

n  2 AND i 1, 2, 3, 4 USING CRANK NICOLSON METHOD

TABLE VII: COMPARISON OF TEMPERATURE OBTAINED AT INTERIOR NODES

Temperature Crank Nicolson Exact Error

( C ) ( C ) ( C )
at Nodes Method Solution Generated

f 3
62.6040 62.5100 0.0940
1

f 23 37.6130 37.0840 0.5290

f 33 26.5620 25.8440 0.2820
f 43 24.0420 23.6100 0.4320

The temperature distribution along the length of the rod at different times is plotted in Figure VI.

FIGURE VI: TEMPERATURE DISTRIBUTION IN ROD FROM CRANK-NICOLSON

METHOD

IV. DISCUSSION OF RESULTS

From Tables 1 and 5, we can see that Crank Nicolson method is good for solving parabolic heat
equation. Tables 6 and 7 demonstrate that the Crank Nicolson method performs well, is mutually
consistent and agree with the exact solution. However, Crank Nicolson method provides better
accuracy and it only requires the solution of a very simple system of linear equations (namely, a
tridiagonal system) at every time level.

V. CONCLUSION
Crank Nicolson method has become a very popular finite difference method for solving parabolic

method is robust, unconditionally stable, has higher order of accuracy up to O((x) , (t ) ) and
partial differential equations. This method has its own advantages and disadvantages. Crank Nicolson
2 2

converges faster than other two finite difference methods but the main problem of this method is that
it starts to oscillate when the coefficient of the second derivative (the diffusion term) is very small or
when the coefficient of the first derivative (the convection term) is large (or both).
.
REFERENCES
[1] W. F. Ames, Numerical Methods for Partial Differential Equations, Academic Press, Inc., Third
Edition, 1992
[2] D. Britz, Digital Simulation in Electrochemistry, California, Springer Verlag, 1988.

22
 International Journal of Applied Mathematics and Modeling IJA2M Vol.1, No. 3, 8-23
©KINDI PUBLICATIONS September, 2013
www.kindipublication.com ISSN: 2336-0054

[3] J. Cooper, Introduction to Partial Differential Equations with MATLAB, B0ston, 1998
[4] J. Crank and P. Nicolson, A Practical Method for Numerical Evaluation of Solution of Partial
Differential Equations of the Heat Conduction Type, Proc. Camb. Phil. Soc., 6(1996), 50-67.
[5] D. J. Duffy. Finite Difference Methods in Financial Engineering, A Partial Differential Equations
Approach, Wiley, 2006.
[6] S. E. Fadugba, F. H. Adefolaju and O. H. Akindutire, On the Stability and Accuracy of Finite
Difference Method for Option Pricing, International Institute of Science, Technology and Education,
Mathematical Theory and Modeling, 2(2012), 101-108.
[7] S. P. Frankel and E. C. Du Fort, Conditions in the Numerical Treatment of Parabolic Differential
Equations, Mathematical Tables and Other Aids to Computation, 7(1953), 135-152.
[8] J. D. Hofman, Numerical Methods for Engineers and Scientists, McGraw-Hill, New York, 1992.
[9] A. Kaw and S. H. Garapati, Textbook Notes for the Parabolic Differential Equations,
http://numericalmethods.eng.usf.edu, 2011.
[10] K. W. Morton and D. F. Mayers, Numerical Solution of Solution of Partial Differential
Equations: An Introduction, Cambridge, England, 1994.
[11] J. D. Smith, Numerical Solution of Partial Differential Equations, London, Oxford University
Press, 1965.
[12] D. Tavella and C. Randall, Pricing Financial Instruments: The Finite Difference Method,
Wiley, 2000.
[13] W. T. Vetterling, W. H. Press, S. A. Teukolsky and B. P. Flannery, Numerical Recipes,
Cambridge University Press, 2007.

23

View publication stats