CHAPTER 2

Numerical Solution of Partial Differential Equation: Elliptic Type

Dr. N.J. Wange
Assistant Professor of Department of Applied Mathematics and Humanities,
Yeshwantrao Chavan college of Engineering, Nagpur-441110, India

E-mail address for Correspondence: [email protected]

Introduction

Partial differential equations (PDEs) arise in every field of science and engineering like
hydrodynamics, elasticity, quantum mechanics and electromagnetic theory. A physical problem in
applied mathematics or science and engineering can be formulated in terms of PDE, so the solution
of these PDEs is of great interest in understanding various physical phenomena. Many PDEs
cannot be solved by analytical methods, we go in for sufficiently approximate solution by simple
numerical methods, and the method of finite differences is commonly used. In this method, the
finite difference equation is first obtained by replacing the partial derivatives by the finite
difference approximations. Most of the problems can be formulated as second order partial
differential equations.

Classification

The general second order linear PDE is of form

𝜕2𝑢 𝜕2𝑢 𝜕𝑢 𝜕𝑢
𝜕2𝑢
𝐴 𝜕𝑥2 + 𝐵 𝜕𝑥𝜕𝑦 + 𝐶 𝜕𝑦2 + 𝐷 𝜕𝑥 + 𝐸 𝜕𝑦 + 𝐹𝑢 = 𝐺 2.1
where A, B, C, D, E, F, G are function of x and y only.
An equation of the form
𝜕2𝑢 𝜕2𝑢 𝜕2𝑢 𝜕𝑢 𝜕𝑢 𝜕𝑢 𝜕𝑢
,
𝐴 𝜕𝑥2 + 𝐵 𝜕𝑥𝜕𝑦 + 𝐶 𝜕𝑦2 + 𝐷 𝜕𝑥 + 𝐸 𝜕𝑦 + (𝑥, 𝑦, 𝑢, 𝜕𝑥 𝜕𝑦 ) = 0 2.2

In which the terms involving the second order partial derivatives alone are linear is known as a
quasi-linear PDE of the 2nd order.
The equations 2.1 or 2.2 is said to be of
(i) the elliptic type, if 𝐵2 − 4𝐴𝐶 < 0
(ii) the parabolic type, if 𝐵2 − 4𝐴𝐶 = 0
(iii) the hyperbolic type, if 𝐵2 − 4𝐴𝐶 > 0

For example,
Let us consider the following equations
(a) 𝑢𝑥𝑥 − 2𝑥𝑢𝑥𝑦 + 𝑥2𝑢𝑦𝑦 − 2𝑢𝑦 = 0
Here, 𝐴 = 1, 𝐵 = −2𝑥, 𝐶 = 𝑥2
𝐵2 − 4𝐴𝐶 = (−2𝑥)2 − 4𝑥2 = 0
Therefore, the equation is parabolic at all point.
(b) 𝑦2𝑢𝑥𝑥 + 𝑥2𝑢𝑦𝑦 = 0
Here, 𝐴 = 𝑦2, 𝐵 = 0, 𝐶 = 𝑥2
𝐵2 − 4𝐴𝐶 = 0 − 4𝑥2𝑦2 < 0
Therefore, the equation is elliptic at all point.
(c ) 𝑥2𝑢𝑥𝑥 − 𝑦2𝑢𝑦𝑦 = 0

Here, 𝐴 = 𝑥2, 𝐵 = 0, 𝐶 = −𝑦2

𝐵2 − 4𝐴𝐶 = 0 + 4𝑥2𝑦2 > 0
Therefore, the equation is hyperbolic at all point.
(d ) 𝑢𝑥𝑥 − 𝑥𝑢𝑦𝑦 = 0

Here, 𝐴 = 1, 𝐵 = 0, 𝐶 = −𝑥
𝐵2 − 4𝐴𝐶 = 4𝑥
Therefore, the equation is elliptic, parabolic or hyperbolic, according as
𝑥 < 0, 𝑥 = 0 𝑜𝑟 𝑥 > 0.
(e) 𝑢𝑥𝑥 + 4𝑢𝑥𝑦 + (𝑥2 + 4𝑦2)𝑦𝑦 = sin(𝑥 + 𝑦)
𝐵2 − 4𝐴𝐶 = 16 − 4(𝑥2 + 4𝑦2) = 4[4 − (𝑥2 + 4𝑦2)]
The equation is elliptic if 4 − (𝑥2 + 4𝑦2) < 0
i.e., (𝑥2 + 4𝑦2) > 4
It is hyperbolic if (𝑥2 + 4𝑦2) < 4.It
is parabolic if (𝑥2 + 4𝑦2) = 4.
Some practical examples
(a) 𝜕
2𝑢 + 𝜕2𝑢 = 0 (Laplace equation, elliptic type)
𝜕𝑥2 𝜕𝑦2

(b) 𝜕
2𝑢 + 𝜕2𝑢 = −𝑓(𝑥, 𝑦) (Poisson equation, elliptic type)
𝜕𝑥2 𝜕𝑦2

(1D heat flow equation, parabolic type)

(c) 𝜕𝑢 = 𝛼2 𝜕2𝑢
𝜕𝑡
 𝜕𝑥2

(1D wave equation, hyperbolic type)

(d) 𝜕2𝑢 = 𝑎2 𝜕2𝑢
𝜕𝑡2 𝜕𝑥2

Finite Difference Approximation for Solving Elliptic PDE

Laplace’s Equation

The equation defined by

𝜕2𝑇 + 𝜕2𝑇 = 0 2.3
𝜕𝑥2 𝜕𝑦2

is called Laplace’s equation. It occurs in all problems involving steady state temperature
distribution and is of elliptic type. To derive this equation we consider a heated plate which is
insulated everywhere except at its edges where the temperature is constant. This equation will give
the steady state temperature distribution.
The exact solution of above equation (2.3) using method of separation of variables, can be
(𝑥, 𝑦) = (𝑐1𝑒𝑘𝑥 + 𝑐2𝑒−𝑘𝑥)(𝑐3𝑐𝑜𝑠𝑘𝑦 + 𝑐4𝑠𝑖𝑛𝑘𝑦)
(𝑥, 𝑦) = (𝑐5𝑒𝑘𝑦 + 𝑐6𝑒−𝑘𝑦)(𝑐7𝑐𝑜𝑠𝑘𝑥 + 𝑐8𝑠𝑖𝑛𝑘𝑥)
The proper form of the solution has to be chosen depending upon the physical conditions of the
problem.
The finite difference method is based on the replacement of differential equation with
approximately equivalent finite difference equation. Finite difference solution depends on the
following steps:
(i) Discretize the solution domain using grids and nodes into small rectangles.
(ii) Approximate the given differential equation by equivalent finite difference equations
that give the solutions to the grid points.
(iii) Using the initial boundary condition difference equation produce the system of
simultaneous equation.
 (iv) Solve the system of equation using Gauss Seidel iteration method (Leibmann’s
method).

Consider Laplace equation is defined by

𝜕2𝑇 𝜕2𝑇
𝜕𝑥2 + 𝜕𝑦2 = 0 , 𝑅 = {0 ≤ 𝑥, 𝑦 ≤ 1} 2.4

or 𝑇𝑥𝑥 + 𝑇𝑦𝑦 = 0

where 𝑇(𝑥, 𝑦) is the steady state temperature distribution in the square domain.
Boundary conditions:
Dirchlet’s boundary condition T(𝑥, 𝑦) = 𝑓(𝑥, 𝑦) on S
Neumann’s boundary condition 𝜕𝑇
= ℎ(𝑥, 𝑦) on S
𝜕𝑛

Robin boundary (Mixed boundary) condition 𝜕𝑇

+ (𝑥, 𝑦) = 𝑔(𝑥, 𝑦) on S in square domain R.
𝜕𝑛

In partial differential equations, the form of equation is always associated with a particular type of
boundary conditions. In the case of Dirichlet , the problem is said to be well defined (or well-
posed). If, we associate Laplace equation with Neumann and mixed boundary conditions, the
problem is said to be ill-posed.

To find the solution of the function T(𝑥, 𝑦) on the region R, we divide the region on XY plane
using vertical and horizontal lines into equal rectangles or meshes of sides ∆𝑥 = ℎ and ∆𝑦 = 𝑘
along 𝑥 and 𝑦 direction respectively such that 𝑥 = 𝑖ℎ, 𝑖 = 0, 1, 2 … … and 𝑦 = 𝑗𝑘, 𝑗 = 0,
1, 2 … … . The point (𝑖ℎ, 𝑗𝑘) is called the grid point (𝑖, 𝑗). If 𝑇 is a function of the two

independent variables 𝑥, 𝑦 the value of (𝑥, 𝑦) at the point (i, j) is denoted by 𝑇𝑖,𝑗.
Then we have the following finite difference formulae for partial derivatives using Taylor’s series
expansion

𝜕2𝑇 = 𝑇𝑥𝑥 = 𝑇𝑖−1, − 2𝑇𝑖,𝑗 + 𝑇𝑖+1,𝑗ℎ2 + (ℎ2) 2.5

𝜕𝑥2
 𝜕2𝑇 = 𝑇𝑦𝑦 = 𝑇𝑖,−1 − 2𝑇𝑖,𝑗 + 𝑇𝑖,𝑗+1 + (𝑘2) 2.6
𝜕𝑦2 𝑘2

The terms (ℎ2) and 𝑂(𝑘2) in equations 2.5 and 2.6 denote the order of local truncation errorand
is also known as the order of method. After neglecting the truncation, error and substituting in
equation 2.4 we obtain the following difference equations

𝑇𝑖−1,−2𝑇𝑖,𝑗ℎ+𝑇
2 𝑖+1,𝑗
+ 𝑇𝑖,𝑗−1−2𝑇𝑖,𝑗+𝑇𝑖,𝑗+1
𝑘2 =0
If we consider ℎ = 𝑘, then

 𝑖,𝑗 = 4 1(𝑇𝑖+1,𝑗 + 𝑇𝑖−1,𝑗 + 𝑇𝑖,𝑗+1 + 𝑇𝑖,𝑗−1) 2.7

𝑇
This shows that value of 𝑇𝑖𝑗 at any interior grid points of the region R is the average of values
of 𝑇 at surrounding four nearest grid points shown in Figure 1. This is called standard five
point formula (SFPF).

T(i,j+1)

h h
T(i,j)
T(i+1,j)
T(i-1,j)

T(i,j-1)
Figure 1 Temperature at any point average of surrounding points
 T(i-1,j+1) T(i+1,j+1)

T(i,j)

T(i-1,j-1) T(i+1,j-1)

Figure 2 Temperature at any point average of corner points

Instead of formula (2.7), we may also use the following formula which is called diagonal five
point formula (DFPF); it can be shown as Figure 2,

1
𝑇𝑖, = 4 (𝑇𝑖+1,𝑗−1 + 𝑇𝑖−1,𝑗−1 + 𝑇𝑖+1,𝑗+1 + 𝑇𝑖−1,𝑗+1) 2.8

The temperature on the four sides of a region is given, at all the internal points the temperature is
assumed according to boundary point. Region is divided into a finite number of rectangular
elements. Here the selection of mesh size ℎ = (1/4) in x direction and 𝑘 = (1/4) in y direction will
split the region into 16 rectangular elements. The interior nodes lie inside the region and exterior
nodes lie on the boundary. The nodes of the region are numbered and shown in Figure 3.
 T(0,3) T(1,3) T(2,3) T(3,3)

T(0,2) T(1,2) T(2,2) T(3,2)

T(0,1) T(1,1) T(2,1) T(3,1)

T(0,0) T(1,0) T(2,0) T(3,0)

Figure 3 Rectangular elements with nodes numbers

Solution of Laplace’s Equation:

We wish to solve Laplace’s equation
𝑇𝑥𝑥 + 𝑇𝑦𝑦 = 0 or ∇2𝑇 = 0
Consider a boundary region R with boundary C. As Dirichlet’s problem, let the value of T be
specified everywhere on C. For simplicity, let R be a square region so that it can be divided into a
network of small squares of side h. Let the values of (𝑥, 𝑦) on the boundary C be given by 𝑐𝑖 andlet
the interior mesh points and boundary points be as in Figure 4. To get initial values for the
interior points of the network, we first find a value of 𝑇5 at the centre of the square by taking the
mean of the four boundary values at the ends of the horizontal and vertical lines passing through
it.

We first use the diagonal five point formula (DFPF) and compute 𝑇5, 𝑇7, 𝑇9, 𝑇1, 𝑇3 .
 Thus we obtain,
1 1
𝑇5 = [𝑐1 + 𝑐5 + 𝑐9 + 𝑐13] (DFPF); 𝑇7 = [𝑐15 + T5 + 𝑐11 + 𝑐13] (SFPF);
4 4

𝑇9 = 1
[𝑇5 + 𝑐7 + 𝑐9 + 𝑐11] (DFPF); 𝑇1 = 1
[𝑐1 + 𝑐3 + 𝑐9 + 𝑐15] (SFPF) ;
4 [𝑐3 + 𝑐5 + 𝑐7 + 𝑢5] (DEPF). 4
𝑇3 = 1
4

We then compute the remaining quantities 𝑇8, 𝑇4, 𝑇6, 𝑇2 by the standard formula (2.7).
Now we know all the boundary values of 𝑇 and rough values of 𝑇 at every grid point in the interior
of the region R. We proceed with an iteration to improve their accuracy. Now we improve the
values by using Standard Five Point Formula (SFPF).
The iterative formula is
𝑖,𝑗 (𝑛+1) = 41 [𝑇(𝑛+1) + 𝑇(𝑛)
𝑖+1, + 𝑇(𝑛)
𝑖,−1 + 𝑇(𝑛+1)
𝑖,+1]
𝑖−1,

where the superscript of 𝑇 denotes the iteration number.

This is known as Gauss-Seidel scheme converges twice as fast as the Jacobi scheme. This method
is also referred to as Leibmann’s method.

C13 C12 C11 C10

C9

C14
C8
T7 T8 T9
C15
C7
T4 T5 T6

C16
C6
T1 T2 T3

C3 C4 C5
C1 C2

Figure 5 Interior mesh points and boundary points

Here our problem is to find numerical value of a function (𝑥, 𝑦) at the interior nodes points of the
region provided that the Laplace equation and the boundary condition given in equations 2.3 are
satisfied.
This study used the Gauss-Seidel iterative method for solving the system of equations.
The stopping criterion for the iteration is
= |𝑇𝑖, 𝑝𝑟𝑒𝑠𝑒𝑛𝑡 − 𝑇𝑖,𝑗 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠| < 10−𝑛
𝜀𝑖𝑗 𝑇𝑖,𝑗 𝑝𝑟𝑒𝑠𝑒𝑛𝑡

Example 1:
Laplace Equation with Dirichlet boundary condition in a square (regular) domain defined in
Figure 4 to find the steady state temperature distribution.

1 1

0 0
T4 T3

0 0
T1 T2

0 0

Figure 4 Dirichlet boundary condition on the four sides of a square domain for Laplace
equation

Solution: Using Standard five point finite difference formula we obtain the following

1
𝑇1 = 4 [0 + 0 + 𝑇2 + 𝑇4]
1
𝑇2 = 4 [0 + 0 + 𝑇3 + 𝑇1]
 1
𝑇3 = 4 [0 + 1 + 𝑇2 + 𝑇4]
1
𝑇4 = 4 [0 + 1 + 𝑇1 + 𝑇3]

We will improve the solution by using Leibmann’s iterative method.

(𝑛+1) = 1 [0 + 0 + 𝑇(𝑛) + 𝑇(𝑛)]

1 4 2 4
(𝑛+1) = 1 [0 + 0 + 𝑇(𝑛) + 𝑇(𝑛+1)]
2 4 3 1
(𝑛+1) = 1 [0 + 1 + 𝑇(𝑛+1) + 𝑇(𝑛)]
3 4 2 4
(𝑛+1) = 1 [0 + 1 + 𝑇(𝑛+1) + 𝑇(𝑛+1)]
4 4 1 3
For first iteration ( = 0 )

(1)
= 1 [0 + 0 + 𝑇(0) + 𝑇(0)]
1 4 2 4
(1)
= 1 [0 + 0 + 𝑇(0) + 𝑇(1)]
2 4 3 1
(1)
= 1 [0 + 1 + 𝑇(1) + 𝑇(0)]
3 4 2 4
(1)
= 1 [0 + 1 + 𝑇(1) + 𝑇(1)]
4 4 1 3

Let 𝑇0 = 0, (since it is nearer to the boundary value 𝑇 = 0 ), and 𝑇0 = 𝑇0 = 1 (maximum

2 4 3
value is taken from nearer boundary value).
Hence
(1)
= 1 [0 + 0 + 0 + 1] = 0.25
1 4
(1)
= 1 [0 + 0 + 1 + 0.25] = 0.3125
2 4
(1)
= 1 [0 + 1 + 0.3125 + 1] = 0.5625
3 4
(1)
= 1 [0 + 1 + 0.25 + 0.5625] = 0.46875
4 4

Similarly putting 𝑛 = 1, 2, 3, …. we obtain the second, third, …. iteration for temperature

distribution at 𝑇1, 𝑇2 , 𝑇3, 𝑇4.
 The Leibmann’s iterative process for numerical solution shown in Table 1.

Table1 Leibmann’s iterative process for Laplace Equation with Dirichlet boundary condition
in a square domain

Iterations 𝑇1 𝑇2 𝑇3 𝑇4
1 0.25 0.3125 0.5625 0.46875
2 0.21875 0.17187 0.42187 0.39844
3 0.14844 0.13672 0.38672 0.38086
4 0.13086 0.12793 0.37793 0.37646
5 0.12646 0.12573 0.37573 0.37537
6 0.125275 0.12525 0.375155 0.3751075
7 0.125275 0.125061 0.375015 0.3750327
8 0.1250706 0.125028 0.375015 0.3750214

Here all the values of 𝑇 of eight iteration are same (upto four decimal) as the corresponding
values of the seven iteration. Hence, we stop the iteration procedure. Eight iteration is the
approximate solution of Laplace equation with Dirichlet condition in square domain.
𝑇1 = 0.125, 𝑇2 = 0.125, 𝑇3 = 0.375, 𝑇4 = 0.375 .

Example 2:
Consider a steel plate of size 15𝑐𝑚 × 15𝑐𝑚. If two of the sides are held at 1000𝐶 and the other
two sides are held at 00𝐶, what are the steady- state temperature at interior points. Assuming a grid
size of 5𝑐𝑚 × 5𝑐𝑚.
Solution:
Since all values are known on the boundary, then it is Dirichlet boundary value problem.
Here grid size of 5𝑐𝑚 × 5𝑐𝑚 and domain is square with ℎ = 5. We have to obtain the steady state
temperature distribution at interior points in a square domain shown in figure 5.
Therefore, the steady-state heat conduction equation is
𝜕2𝑇 + 𝜕2𝑇 = 0 , 𝑅 = {0 ≤ 𝑥, 𝑦 ≤ 15}
𝜕𝑥2 𝜕𝑦2

which is Laplace equation.

 100 100

100 0
T4 T3

100 0
T1 T2

0 0

Using five point standard formula, we get

1
𝑇1 = 4 [100 + 0 + 𝑇2 + 𝑇4]
1
𝑇2 = 4 [0 + 0 + 𝑇3 + 𝑇1]
1
𝑇3 = 4 [0 + 100 + 𝑇2 + 𝑇4]
1
𝑇4 = 4 [100 + 100 + 𝑇1 + 𝑇3]

We will improve the solution by using Leibmann’s iterative method.

(𝑛+1)
= 1 [100 + 0 + 𝑇(𝑛) + 𝑇(𝑛)]
1 4 2 4
(𝑛+1) = 1 [0 + 0 + 𝑇(𝑛) + 𝑇(𝑛+1)]
2 4 3 1
(𝑛+1) = 1 [0 + 100 + 𝑇(𝑛+1) + 𝑇(𝑛)]
3 4 2 4
(𝑛+1) = 1 [100 + 100 + 𝑇(𝑛+1) + 𝑇(𝑛+1)]
4 4 1 3
For first iteration ( = 0 ) (1)
= 1 [100 + 0 + 𝑇(0) + 𝑇(0)]

1 4 2 4
(1)
= 1 [0 + 0 + 𝑇(0) + 𝑇(1)]
2 4 3 1
(1)
= 1 [0 + 100 + 𝑇(1) + 𝑇(0)]
3 4 2 4
(1)
= 1 [100 + 100 + 𝑇(1) + 𝑇(1)]
4 4 1 3

Let 𝑇0 = 0, (since it is nearer to the boundary value 𝑇 = 0 ), and 𝑇0 = 𝑇0 = 100 (maximum

2 4 3
value is taken from nearer boundary value).
Hence
(1)
= 1 [100 + 0 + 0 + 100] = 50
1 4
(1)
= 1 [0 + 0 + 100 + 50] = 37.5
2 4
(1)
= 1 [0 + 100 + 37.5 + 100] = 59.375
3 4
(1)
= 1 [100 + 100 + 50 + 59.375] = 77.34
4 4
Similarly putting 𝑛 = 1, 2, 3, …. we obtain the second, third, …. iteration for temperature
distribution at 𝑇1, 𝑇2 , 𝑇3, 𝑇4.
The Leibmann’s iterative process for numerical solution shown in Table 2.

Table2 Leibmann’s iterative process for Laplace Equation with Dirichlet boundary condition
in a square domain

Iterations 𝑇1 𝑇2 𝑇3 𝑇4
1 50 37.5 59.375 77.34
2 53.71 28.27 51.401 76.278
3 51.137 25.634 50.478 75.403
4 50.259 25.184 50.147 75.101
5 50.071 25.0545 50.039 75.0275
6 50.0205 25.0149 50.0106 75.008
7 50.006 25.00415 50.003 75.0022
8 50.001 25.001 50.0008 75.0004
9 50.0004 25.002 50.0015 75.001
10 50.000 25.000 50.000 75.000
11 50.000 25.000 50.000 75.000
 Here all the values of 𝑇 at eleven iteration are same (upto four decimal) as the corresponding
values at the ten iteration. Hence, we stop the iteration procedure. Eleven iteration is the
approximate solution of Laplace equation with Dirichlet condition in square domain.
𝑇1 = 50, 𝑇2 = 25, 𝑇3 = 50, 𝑇4 = 75 .

Example 3:
Solve the elliptic equation 𝑇𝑥𝑥 + 𝑇𝑦𝑦 = 0 in the square mesh with boundary values shown in the
Figure 6.
0 500 1000 500

1000
0
C1000

T7

T
8

T
9

2000 T4 T5 T6 B 2000

A
T1 T2 T3 1000
1000
D
1000 500 0
0 500

Figure 6 Mesh grids with boundary values

Solution:
Figure 6 is symmetry about AB and CD, so
𝑇1 = 𝑇3, 𝑇4 = 𝑇6, 𝑇7 = 𝑇9, and 𝑇1 = 𝑇7, 𝑇2 = 𝑇8, 𝑇3 = 𝑇9,
Therefore, 𝑇1 = 𝑇3 = 𝑇7 = 𝑇9, and 𝑇2 = 𝑇8, 𝑇4 = 𝑇6.So, we
find only 𝑇1, 𝑇2, 𝑇4, 𝑇5.
Using five point standard formula, we get
1 1 1
𝑇5 = 4 [𝑇2 + 𝑇8 + 𝑇6 + 𝑇4] = 4 [2𝑇2 + 2𝑇4] = 2 [𝑇2 + 𝑇4]
 𝑇 = 1 [𝑇 + 𝑇 ]
5 2 2 4
 1
𝑇1 = 4 [1000 + 500 + 𝑇2 + 𝑇4]
1
𝑇2 = 4 [1000 + 𝑇1 + 𝑇3 + 𝑇5]
 𝑇 = 1 [1000 + 2𝑇 + 𝑇 ]
2 4 1 5

1
𝑇4 = 4 [2000
1
+ 𝑇1 + 𝑇5 + 𝑇7]

𝑇4 = [2000 + 2𝑇1 + 𝑇5]
4

We will improve the solution by using Leibmann’s iterative method.

= 1 [𝑇(𝑛) + 𝑇(𝑛)]
(𝑛+1)
5 2 2 4
(𝑛+1)
= 1 [1500 + 𝑇(𝑛) + 𝑇(𝑛)]
1 4 2 4
(𝑛+1) = 1 [1000 + 2𝑇(𝑛+1) + 𝑇(𝑛+1)]
2 4 1 5
(𝑛+1) = 1 [2000 + 2𝑇(𝑛+1) + 𝑇(𝑛+1)]
4 4 1 5
For first iteration ( = 0 )
= 1 [𝑇(0) + 𝑇(0)]
(1)
5 2 2 4
(1)
= 1 [1500 + 𝑇(0) + 𝑇(0)]
1 4 2 4
(1)
= 1 [1000 + 2𝑇(1) + 𝑇(1)]
2 4 1 5
(1)
= 1 [2000 + 2𝑇(1) + 𝑇(1)]
4 4 1 5

Let 𝑇0 = 1000, 𝑇0 = 2000 (since it is nearer to the boundary value 𝑇 = 0 ), and 𝑇0 = 1000
2 4 7
(maximum value is taken from nearer boundary value).
 Hence (1)
= 1 [1000 + 2000] = 1500

5 2
(1)
= 1 [1500 + 1000 + 2000] = 1125
1 4
(1)
= 1 [1000 + 2 × 1125 + 1500] = 1187.5
2 4
(1)
= 1 [2000 + 2 × 1125 + 1500] = 1437.5
4 4
Similarly putting 𝑛 = 1, 2, 3, …. we obtain the second, third, …. iteration for temperature
distribution at 𝑇1, 𝑇2 , 𝑇4, 𝑇5.

After nine iteration 𝑇1 = 𝑇3 = 𝑇7 = 𝑇9 = 938.05, and 𝑇2 = 𝑇8 = 1000.55,

𝑇4 = 𝑇6 = 1250.55 and 𝑇5 = 1125.55

Example 4: (Dirichlet condition in regular domain)

Using the given boundary values, solve the Laplace equation ∇2𝑢 = 0 at the nodal points of the
square grid shown in Figure 7.

60 60

40 50
U1 U2

20 40
U3 u4

10 20
Figure 7 Dirichlet conditions on all boundary points

Ans: After 5th step 𝑢1 = 43.313, 𝑢2 = 46.656, 𝑢3 = 26, 𝑢4 = 33.328.

Example 5: (Mixed boundary condition with regular domain)

Laplace Equation with Robin boundary conditions in a square domain as follow and defined
in Figure 8.

∇2𝑇 = 0, 0 ≤ 𝑥, 𝑦 ≤ 1
𝑇 = 2𝑥, 𝑦 = 0
𝑇 = 2𝑥 − 1, 𝑦=1
𝑇𝑥 + 𝑇 = 2 − 𝑦, 𝑥=0
𝑇 = 2 − 𝑦, 𝑥=1

T=2x-1

Tx
+T T4 T5 T=2-y
T6
=2
-y
T1 T2 T3

T=2x

Figure 8 Robin boundary condition on the left side of a square domain for Laplace equation.

Solution:

At 𝑇1 and 𝑇4 mixed boundary condition is defined. So, at 𝑇1, consider boundary condition
𝑇𝑥 + 𝑇 = 2 − 𝑦
Apply forward difference approximation for derivative in condition, then
=2−𝑦 Assume ℎ = 1/3.
𝑇2−𝑇1 + 𝑇 5 1 4
1
ℎ ) , similarly, T4 =
1 3 2 (3T5 − )3
T1 = 2 (3T2 −

At 𝑇2, 𝑇3, 𝑇5, 𝑇6 , Gauss-Seidel iteration formulae are

T(n+1) 2 (n) (n) 1 (n+1) 1 (n+1) (n)

2 = 5 (T3
 − ) + 3)
T
+ T5 6 =
(T42 3 + T6
T(n+1) 2 (n+1)
(n) (n+1) 1 (n+1)
(n+1) 5

5 + T6 − 1) T6
=
(T52
 +)
= 4 (T3 +T.
35

The Gauss-Seidel iterative process for numerical solution shown in Table 3.

Table 3 Gauss-Seidel iterative process for Laplace Equation with Robin boundary condition
in a square domain

Iterations 𝑇2 𝑇3 𝑇5 𝑇6
1 0.466687 1.19999 0.319871 0.7966405
2 0.5412768 1.084479 0.1351669 0.721586
3 0.421178 1.035693 0.0571056 0.6898739
4 0.37045 1.015081 0.0241312 0.6764775
5 0.349016 1.006373 0.0101975 0.67081764
6 0.32725 0.983240 0.0098431 0.65320425
7 0.30456 0.952452 0.0094061 0.6332156
8 0.30748 0.950471 0.0094043 0.6321457
9 0.30732 0.950456 0.0094033 0.6320852

Nine iterations are necessary step for a solution of Dirichlet’s boundary value problem with
Robin’s condition in a square domain.

Example 6 : (Dirichlet condition with irregular domain)

Laplace Equation with Dirichlet boundary condition in a triangular (irregular) domain with
vertices (0, 0), (7, 0), (0,7) shown in figure 9. Assume mesh size 2.
 T= x-y
T3 T= x-y

T1 T2
T= x-y (7, 0)
(0,0)

Figure 9 Dirichlet boundary condition on the four sides of a tringular domain for Laplace
equation

Solution:
Five point finite difference formula with Gauss seidel iteration is
+ T(n) ) T(n+1) = 1 (T(n+1)
1 4
+ 8)
T(n+1) = 1 (T
4 (n)
1 2 3 2
T(n+1) 1 (n+1)
= 4 (T )
3 1

Table 4 shows the Gauss-Seidel iteration process for numerical solution of Laplace equation
in irregular domain .

Table 4 Gauss-Seidel iterative process for Laplace Equation with Dirichlet boundary condition
in a triangular domain

Iterations 𝑇1 𝑇2 𝑇3
1 0.75000 2.187506 0.1875
2 0.593750 2.1484375 0.1484375
3 0.5742187 2.143555 0.1435547
4 0.571777 2.142944 0.14294425
5 0.571472 2.142868 0.14286806
6 0.571470 2.142839 0.14286800

Six iterations are necessary to reach a solution of Laplace equation with Dirichlet condition in
triangular domain.
Poisson’s Equation:
An equation of the form ∇2𝑢 = −(𝑥, 𝑦)
i.e 2𝑢 + 𝜕2𝑢 = −(𝑥, 𝑦) 2.9
𝜕

𝜕𝑥2 𝜕𝑦2

where 𝑓(𝑥, 𝑦) is a function of x and y only, is called Poisson’s equation.

This is an elliptic equation which can be solved numerically at the interior grid points of a square
network when the boundary values are known.
Replace the derivatives in equation (2.9) by difference formula by taking
𝑥 = 𝑖ℎ, 𝑦 = 𝑗ℎ (∵ ℎ = 𝑘), then equation (2.9) reduces to

𝑦)
𝑢𝑖−1,−2𝑢𝑖,𝑗+𝑢𝑖+1,𝑗 + 𝑢𝑖,𝑗−1−2𝑢𝑖,𝑗+𝑢𝑖,𝑗+1 = −𝑓(𝑥
ℎ2 ℎ2
 𝑖, 𝑗

 4 𝑢𝑖,𝑗 = (𝑢𝑖+1,𝑗 + 𝑢+𝑖−1,𝑗

𝑢 + 𝑢𝑖,𝑗+1
+ 𝑢+ 𝑢𝑖,𝑗−1+) 𝑢+ ℎ2𝑓(𝑥𝑖,+𝑦𝑗ℎ) 2(𝑥 𝑦 ) ) 2.10
 𝑢 1

𝑖, = 4 (𝑢𝑖+1,𝑗
𝑖−1, 𝑖,+1 𝑖,−1 𝑖,
𝑗

By applying equation (2.10) at each interior grid point, we get a system of linear equations at (𝑖, 𝑗)
point. The equations can be solved by Gauss-Seidel / Leibmann’s method.

Example 1:
Solve Poisson equation 𝜕
2𝑇 + 𝜕2𝑇 = −3(𝑥 + 𝑦) , 0 ≤ 𝑥, 𝑦 ≤ 3 with Dirichlet boundary
𝜕𝑥2 𝜕𝑦2
condition in a square domain defined in Figure 10 to find the potential distribution.

T=0

T=0 T4 T3
T=0
T1 T2

T=0
Figure 10 Dirichlet boundary condition on the four sides of a square domain for Poisson equation.

Using equation 2.10 we obtain the approximate solution at each node.

 1 2 𝑓( 1, 1)]
𝑇1 = 4 [0 + 0 + 𝑇2 + 𝑇4 + ℎ
1 2 𝑓( 2, 1)]
𝑇2 = 4 [0 + 0 + 𝑇1 + 𝑇3 + ℎ
1 2
𝑓( 2, 2)]
𝑇3 = 4 [0 + 0 + 𝑇2 + 𝑇4 + ℎ 1 𝑇 =
[0 + 0 + 𝑇
+𝑇 +2
ℎ(
1,
2)]

4 4 1 3

(1, 1) = 3(𝑥 + 𝑦) = 6
(2, 1) = 3(𝑥 + 𝑦) = 9
(2, 2) = 3(𝑥 + 𝑦) = 12
(1 ,2) = 3(𝑥 + 𝑦) = 9
For 1st iteration

1 0 + 𝑇40 + ℎ2 1 1 3
𝑇11 = 1.5
= 4 [0 + 0 + 𝑇2 𝑓 ( ,3 )] = 2
3

𝑇 1 = 1 [0 + 0 + 𝑇 1 + 𝑇 0 + ℎ2(2, 1)] = 2.625

2 4 1 3
𝑇 1 = 1 [0 + 0 + 𝑇 1 + 𝑇 0 + ℎ2(2, 2)] = 3.65625
2 4
3 4
𝑇 1 1 1 1 2 ( )
+ 𝑇3 + ℎ 𝑓 1, 2 ] = 3.539
4 = 4 [0 + 0 + 𝑇1

The Gauss-Seidel iterative process for numerical solution shown in Table 5.

Table 5 Gauss-Seidel iterative process for Poisson Equation with Dirichlet boundary
condition in a square domain

Iterations 𝑇1 𝑇2 𝑇3 𝑇4
1 1.5 2.625 3.65625 3.539
2 3.041 3.92431 4.8658 4.2267
3 3.53775 4.35088 5.14439 4.4205
4 3.6928 4.4593 5.21995 4.4788
5 3.73437 4.48858 5.24169 4.494013
6 3.745648 4.49683 5.24674 4.49809
7 3.74873 4.49886 5.24924 4.49949
8 3.74958 4.499707 5.24992 4.499846
9 3.74988 4.49757 5.24918 4.49302
 Nine iterations are necessary to reach a solution of Poisson’s equation with Dirichlet condition
in square domain.

Example 2:
Solve ∇2𝑢 = −10(𝑥2 + 𝑦2 + 10) over the square mesh with sides 𝑥 = 0, 𝑦 = 0, 𝑥 = 3, 𝑦 = 3
with 𝑢 = 0 on the boundary and mesh length 1 unit.
Ans: 𝒖𝟏 = 𝟕𝟓 = 𝒖𝟒 𝒖𝟐 = 𝟖𝟐. 𝟓, 𝒖𝟑 = 𝟔𝟕. 𝟓

Example 3:
Solve ∇2𝑢 = −4(𝑥2 + 𝑦2) over the square mesh with sides 𝑥 = 0, 𝑦 = 0, 𝑥 = 3, 𝑦 = 3 with 𝑢 = 0 on
the boundary and mesh length 1 unit.
Ans: 𝒖𝟏 = 𝟏𝟎, 𝒖𝟐 = 𝟏𝟑, 𝒖𝟑 = 𝟕, 𝒖𝟒 = 𝟏𝟎

Example 4:
Solve ∇2𝑢 = 8𝑥2𝑦2 for the square mesh shown below in figure 11 (ℎ = 1)
u=0

u=0 u3 u4
u=0
u1 u2

u=0
Figure 11 Square domain with Dirichlet condition
Ans: 𝒖𝟏 = −𝟏𝟑, 𝒖𝟐 = −𝟐𝟐 = 𝒖𝟑, 𝒖𝟒 = −𝟒𝟑

Alternating Direction Implicit (ADI ) Method:

The Alternating Direction Implicit (ADI) method is an iterative method used to solve matrix
equations in control theory. It is a popular method for solving the large matrix equations that arise
in system theory and control theory and can be formulated to construct solutions in a memory-
efficient, factored form. It is also used to numerically solve parabolic and elliptic partial differential
equations, and is a classic method used for modeling heat conduction and solving the diffusion
equation in two or more dimensions.
We consider Laplace’s equation in two dimensions, viz.,
𝜕22𝑇
𝜕𝑥 + 𝜕𝜕𝑦2𝑇2 = 0

And the standard five point formula

𝑇𝑖+1, + 𝑇𝑖−1,𝑗 + 𝑇𝑖,𝑗+1 + 𝑇𝑖,𝑗−1 − 4 𝑇𝑖,𝑗 = 0 2.11

We rearrange Equation (2.11) in either of two ways:

𝑇𝑖−1, − 4 𝑇𝑖,𝑗 + 𝑇𝑖+1,𝑗 = −𝑇𝑖,𝑗−1 − 𝑇𝑖,𝑗+1 2.12
Or 𝑇𝑖,−1 − 4 𝑇𝑖,𝑗 + 𝑇𝑖,𝑗+1 = −𝑇𝑖−1,𝑗 − 𝑇𝑖+1,𝑗 2.13

The ADI is an iterative method and equations (2.12) and (2.13) are used as iteration formulae

𝑇𝑟+1 − 4𝑇𝑟+1 + 𝑇𝑟+1 = −𝑇𝑟 − 𝑇𝑟 2.14

𝑖−1, 𝑖,𝑗 𝑖+1, 𝑖,−1 𝑖,+1

and 𝑇𝑟+2 − 4𝑇𝑟+2 + 𝑇𝑟+2 = −𝑇𝑟+1 − 𝑇𝑟+1 2.15

𝑖,−1 𝑖,𝑗 𝑖,+1 𝑖−1, 𝑖+1,

Equation (2.14) is used to compute function values at all internal mesh points along rows and
equation (2.15) those along columns. For 𝑗 = 1,2,3, … . , 𝑛 − 1. Equation (2.14) yields a tridiagonal
system of equation and can easily be solved. Similarly, for 𝑖 = 1,2,3, … . , 𝑛 − 1. Equation (2.15)
also yields a tridiagonal system of equations.
In the ADI method, formulae (2.14) and (2.15) are used alternately. For example, for the first row
𝑗 = 1, and equation (2.14) gives
𝑇𝑟+1 − 4𝑇𝑟+1 + 𝑇𝑟+1 = −𝑇𝑟 − 𝑇𝑟 (𝑖 = 1, 2, 3, … . . (𝑛 − 1) 2.16
𝑖−1,1 𝑖,1 𝑖+1,1 𝑖,0 𝑖,2

Together with the boundary conditions, equation (2.16) represents a tridiagonal system of
equations and are easily solved for 𝑢𝑟+1. We next put 𝑗 = 2 and obtain the values of 𝑢𝑟+1 on the
𝑖,1 𝑖,2

second row. The process is repeated for all the rows, viz up to 𝑗 = 𝑛 − 1. We next alternate the
direction, i.e we use equation (2.15) to compute 𝑢𝑟+2
𝑖,𝑗
. It is easy to see that at every stage we will
be solving a tridiagonal system of equation.

Example 1:
Solve Laplace’s equation, 𝜕2𝑇 + 𝜕2𝑇 = 0, in the domain of Figure 12

𝜕𝑥2 𝜕𝑦2
T3,3
T0,3 T=1

T3,2
T=0 T1,2 T2,2
T=0
T0,1,
T1,1 T2.1

T0,0 T=0 T2,0 T3,0

Figure 12 Mesh grids in a square domain with Dirichlet condition

Solution:
To apply formulae given in equations (2.14) and (2.15), we relabel the internal mesh points, as in
Figures 12. To start the iterations, we set 𝑟 = 0. For the first row, 𝑗 = 1. Then, equation (2.14)
gives

𝑖−1,1 −
(1) 4𝑇(1)𝑖,1+ 𝑇(1)𝑖+1,1 = −𝑇𝑖,0
(0) − 𝑇(0)
𝑖,2

With 𝑖 = 1 and 𝑖 = 2, this gives two equations

0,1(1) − 4𝑇1,1
(1) + 𝑇2,1
(1) = −𝑇(0)1,0
− 𝑇(0) 1,2

1,1(1) − 4𝑇2,1
(1) + 𝑇3,1
(1) = −𝑇(0)2,0
− 𝑇(0) 2,2

Substituting the boundary values and assuming that (0) = 1 and 𝑇(0) = 1, the above equations
1,2 2,2
−4(1) + 𝑇(1) = −1
1,1 2,1
(1)
− 4𝑇 = −1
(1)

1,1 2,1

Solving above two equations we obtain

𝑇1 = 𝑇1 1
1,1
= 3 = 0.3333
2,1

For computing the function values on the second row, we set 𝑗 = 2 in equation (2.14) to obtain
 𝑖−1,2 −
(1) 4𝑇(1)𝑖,2+ 𝑇(1)𝑖+1,2 = −𝑇𝑖,1
(0) − 𝑇(0)
𝑖,3

With 𝑖 = 1 and 𝑖 = 2, this gives two equations

0,2(1) − 4𝑇1,2
(1) + 𝑇2,2
(1) = −𝑇(0)1,1
− 𝑇(0) 1,3

1,2(1) − 4𝑇2,2
(1) + 𝑇3,2
(1) = −𝑇(0)2,1
− 𝑇(0) 2,3

Substituting the boundary values and assuming (0) = 1, 𝑇(0) = 1 ,

1,3 2,3
−4(1) + 𝑇(1) = −1
1,2 2,2
(1)
− 4𝑇 = −1
(1)

1,2 2,2
Solving these we get, 𝑇1 = 𝑇1 1
1,2
= 3 = 0.3333
2,2

Having completed the computations on two rows, we now alternate the direction and compute the
function values on the columns, starting with the first one. For this, we use equation (2.15) with
𝑟 = 0. Setting 𝑖 = 1 equation (2.15) becomes

(2)1,−1 − 4𝑇1,
(2) + 𝑇1,+1
(2) = −𝑇(1)
0, − 𝑇(1)
2,

Putting 𝑗 = 1 and = 2 in the above, we obtain the equations

1,0(2) − 4𝑇1,1(2) + 𝑇1,2

(2) = −𝑇(1)0,1− 𝑇(1) 2,1

And

1,1(2) − 4𝑇1,2
(2) + 𝑇1,3
(2) = −𝑇(1)0,2
− 𝑇(1) 2,2

Substituting the boundary values we obtain,

−4(2) + 𝑇(2) = − 1
1,1
1,2 3
(2)
− 4𝑇(2) + 1 = − 1
1,1
1,2 3
Solving these equations we get,
(2)
= 0.1778 and 𝑇(2) = 0.3778
2,1 2,2

The iteration are continued to improve the function values obtained first on the rows, then on the
column, and so on.
Exercise:
1. The function 𝑢 (𝑥, 𝑦) satisfies Laplace equation at all points within the squares given below
and has the boundary values as indicated. Compute a solution correct to two decimal places by
finite difference method problems (a) to (b).
(a)
4 8

0 11
u3 u4

0 10
u1 u2

3 6

(b)
60 60

40 50
T4 T3
20 40
T1 T2

10 20
2. Solve 𝜕2𝑢 + 𝜕2𝑢 = 0 for the temperature of the heated plate for the rectangle region shown in
𝜕𝑥2 𝜕𝑦2

figure.
 80 80 80

70 u(1,3) u(2,3) 40

70 u(1,2) u(2,2) 40

70 u(1,1) u(2,1) 40

0 0 0 0

3. Solve Laplace’s equation with ℎ = 1/3 over the boundary of a square of unit length with
(𝑥, 𝑦) = 9𝑥2𝑦2 on the boundary.

4. Solve Poisson’s equation ∇2𝑢 = 8𝑥2𝑦2 for the square grid shown below (ℎ = 1)

1 1

0 0
T4 T3

0 0
T1 T2

0 0
5. Solve 2𝑢 + 𝜕2𝑢 = 0 for the temperature of the heated plate for the square region shown in
𝜕

𝜕𝑥2 𝜕𝑦2

figure by ADI method 60 60

40 50
T4 T3
20 40
T1 T2

10 20

Ans: 1 (a) 𝑢1 = 3.33, 𝑢2 = 6.67 𝑢3 = 3.67, 𝑢4 = 7.33

(b) 𝑢1 = 26.7, 𝑢2 = 33.3 𝑢3 = 46.7, 𝑢4 = 43.3
2. 𝑢1 = 26.25, 𝑢1 = 9.84375, 𝑢1 = 10.691406,
11 21 31
𝑢1 = 36.09375, 𝑢1 = 17.226562, 𝑢1 = 25.469238,
12 22 32
𝑢1 = 69.785156, 𝑢1 = 62.629394, 𝑢1 = 78.036987,
13 23 33
3. 𝑢6 = 0.498, 𝑢6 = 0.9998, 𝑢6 = 0.9998, 𝑢6 = 2.4999
1 2 3 4

4. 𝑢11 = −13, 𝑢21 = 𝑢12 = −22, 𝑢22 = −43

5.

Bibliography
Balagurusamy, E., Numerical Methods, McGraw Hill, New Delhi, 1999.
Hildebrand, F. B., Introduction of Numerical Analysis, McGraw Hill, New York, 1956.
Jain M. K., Iyengar S. R. and Jain R. K., Numerical Methods for Scientific and Engineering
Computation, Wiley Eastern Limited, New Delhi, 1985.
Levy, H. and Baggott, E. A., Numerical Solution of Differential Equations, Dover, New York,
1950.
Sastry, S. S., Engineering Mathematics, 3rd eds., Vols., 1 and 2, Prentice-Hall of India, New Delhi,
2004.
Sastry, S. S., Introductory Methods of Numerical Analysis, PHI, Fifth Edition, New Delhi, 2012.
Scheid, Francis, Theory and Problems of Numerical Analysis, Schaum Series, McGraw Hill, New
York, 1968.
G. Haribaskaran, Numerical methods, Laxmi Publications (P) LTD, New Delhi, Boston, USA,
2006.