Numerical Analysis (WI4014TU)

Report : Assignment 1

Ankit Joshi (4617053)

October 31, 2017

Note: MATLAB codes of all the problems are given at the end of the document.

Problem 1

Solving Poisson equation with Dirichlet BC’s and different source functions.

Discretization of equation:

Original equation is given as,

∂ 2u
− =f x ∈ (0 1)
∂x2
with
u(0) = 0, u(1) = 0

Above equation can be discretized with O(h2 ) approximation as,

−ui+1 + 2ui − ui−1

= fi i = 1, 2....n − 1 (1)
h2

where i represents inner grid point of the domain x∈[0 1] and h is the step size of the grid. When
the above equation is written in matrix form for all the inner grid points, it gives
    
2 −1 0 0 0 ... 0 u1 f1 + uh20
−1 2 −1 0 0 ... 0    u2   f2
   
 
1  0 −1 2 −1 0 . . . 0   u3  
     f3 
..   ..  = 

2
 .. .. .. .. .. .. ..
h  .

 . . . . . .  .  
   . 

0 0 0 . . . −1 2 −1 un−2   fn−2 
0 0 0 . . . 0 −1 2 un−1 fn−1 + uhn2

By solving above matrix(Au=f) we can get u vector for all the inner grid points x=ih and u val-
ues for the boundaries(u0 & un ) are already given as BC’s.
Solution: The matrix was solved with n=10 for the following source functions. Below graphs
shows comparison of numerical solution to the analytical(exact) solution.
f1 (x) = −1
f2 (x) = −x + x2 − x3 + x4 − x5 + x6
f3 (x) = −sin(πx)

f(x)=-1
0

-0.02

-0.04

-0.06
u(x)

-0.08

-0.1

-0.12 numerical
exact
-0.14
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

f(x)=-x+x 2-x 3+x 4-x 5+x 6

0

-0.005

-0.01

-0.015
u(x)

-0.02

-0.025

-0.03

-0.035 numerical
exact
-0.04
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

f(x)=sin( x)
0

-0.02

-0.04
u(x)

-0.06

-0.08

-0.1 numerical
exact
-0.12
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

From the graphs of u(x) it is clear that for the source function f1 (x), both analytical(exact) and
numerical solution are same but in the case of f2 (x) & f3 (x) there is some difference between the
two solutions.This is discussed in the next page.
Error in the values of numerical solution can be estimated using the below equation. Here e ma-
trix contains estimated error for each grid point.

Ae = h2 p

where p is the approximation error at each grid point and for the discretization procedure adopted
for this problem,
h2 0000
p ≈ u (x)
12
or
h2 00
p ≈ − f (x)
12
00 00 00
For the source functions given in this problem, it is clear that f1 (x)=0. Whereas, f2 (x) & f3 (x)
are 6= 0. This is the reason why analytical and numerical solution graph coincides only for u1 (x) .

The graphs below show the comparison of estimated error and exact error for u2 (x) & u3 (x).

10-4 f(x)=-x+x 2-x 3+x 4-x 5+x 6

3

2.5

2
error(x)

1.5

0.5
estimated error
exact error
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

10-4 f(x)=sin( x)
9

5
error(x)

1
estimated error
0
exact error
-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
Problem 2

Original equations,
0 0
−(ku ) = f (x), x ∈ (0 1)
u(0) = 0, u(1) = 1
f (x) = 1, x ∈ [0 1]
k(x) = 1 + sin(πx), x ∈ [0 1]

Discretization of equation: Above equation in discretized form is given as,

 
ai+1 (ui+1 − ui ) ai (ui − ui−1 )
− − = fi
h2 h2

To obtain O(h2 ) discretization ai is chosen such that,

ai = ki−1/2 = k(xi − 0.5h)

The discretized equation can also be written in matrix form for i=1,2...n-1 as,
    
a1 + a2 −a2 0 0 0 ... 0 u1 f1 + ah1 u2 0
 −a2 a2 + a3 −a3 0 0 ... 0   u2   f2 
    
1  0 −a 3 a 3 + a 4 −a 4 0 . . . 0   u3   f 3

=
    
. .. .. .. .. .. .. .. ..
h2  ..
     
 . . . . . . 

 . 
 
 . 

 0 0 0 . . . −an−2 an−2 + an−1 −an−1  un−2   fn−2 
an un
0 0 0 ... 0 −an−1 an−1 + an un−1 fn−1 + h2

Here fi is equal to 1 for all the grid points and u0 & un are given as the boundary conditions. Let
us represents above equation in the form of Au = h2 f , here the A matrix of n−1×n−1 dimension
can be formed using A = DT Ac D, where
 
1 0 0 0 ... 0
−1 1 0 0 . . . 0 
1  
Dn×(n−1) =  ... ... ... ... .. .. 

h . . 

 0 0 0 . . . −1 1 
0 0 0 . . . 0 −1
 
a1 0 0 0 ... 0
 0 a2 0 0 ... 0
 
An×n =  ... ... .. .. .. .. 

c
 . . . .
0 0 0 ... an−1 1
0 0 0 ... 0 an

In the code, the above matrices are defined as sparse matrices, which means that only non-zero
values are stored in the matrix and rest of the spaces are empty. This saves a lot of memory ,
especially when the matrices are large.
Results:After solving the matrix equation for n=20, we get the solution of u(x) as shown below.

0.9

0.8

0.7

0.6
u(x)

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1
x

Problem 3

Original equation,

∂ 2u ∂ 2u
 
− + = f (x, y), (x, y) ∈ Ω
∂x2 ∂y 2
f (x, y) = 20y 2 sin(πx) (x, y) ∈ Ω

BC’s are,

u(x, y) = sin(πx/2), (x, y) ∈ Ω1

u(x, y) = 0, (x, y) ∈ Ω2

Discretization of equation: The laplacian was discretized as,

 
ui−1,j − 2ui,j + ui+1,j ui,j−1 − 2ui,j + ui,j+1
− + = fi,j (2)
∆x2 ∆y 2

where i=1,2...m and j=1,2...n. The above equation can also be written in matrix form as Au =
f . Figure in the next page shows how the grid is divided, the blue region shows the domain for
calculating fi,j .
 Grid division

A matrix is calculated using the formula,

A = Iy ⊗ Axx + Ayy ⊗ Ix

where Iy & Ix is identity matrix Im−1 & In−1 and Axx & Ayy is given by,

Axx = DxT Dx , Ayy = DyT Dy

where,
   
1 0 0 0 ... 0 1 0 0 0 ... 0
−1 1 0 0 ...  0 −1 1 0 0 ... 0
1 
 .. .. .. .. ..

.. 1 
 .. .. .. .. ..

..
Dxm×(m−1) = , Dyn×(n−1) =
 
∆x  . . . . . . ∆y  . . . . . .
  
 
0 0 0 ... −1 1  0 0 0 ... −1 1 
0 0 0 ... 0 −1 0 0 0 ... 0 −1

All the matrices above are stored as sparse matrices to save on memory. Matrix u & f are ar-
ranged in lexicographic order. This is given as,
u1,0
f1,1 + ∆y
   
u1,1 2
 u2,1   f2,1 + u2,02 
∆y
..
   
   .. 
 .   . 
   um−1,0 
 um−1,1  fm−1,1 + ∆y2 
   
 u1,2   f1,2 
   
 u2,2   f 2,2

 .
..
 
..

um−1×(n−1) =  , f m−1×(n−1)
=
   
  . 
 um−1,2 
   
 f m−1,2 
 .
..
  . 





 .
.


 u
f1,n−1
  
 1,n−1   
 u
f2,n−1
  
 2,n−1   
 ..   .
.

 .   . 
um−1,n−1 fm−1,n−1
In the f matrix, only for j=1 the values of fi,j will get affected. This is because only the bound-
ary next to j = 1 has some value, all other boundaries are equal to zero.

Results: After solving the above equations for m=20 & n=10 we get u(x,y) as shown below,

0.5
u(x,y)

-0.5
1

0.5
2
1.5
y 1
0 0.5
0 x

Surface plot of u(x,y)

0.8

0.6
y

0.4

0.2

0
0 0.5 1 1.5 2
x

Contour plot of u(x,y)

MATLAB Codes

Problem 1

clc
clear all
close all

n=10; % no. of segments (points starts from 0 to 20)

maindiag=-2*ones(1,n-1); % value in main diagonal of A matrix

offdiag=1*ones(1,n-2); % value in off diagonal of A matrix
A=diag(maindiag)+diag(offdiag,-1)+diag(offdiag,1); % create matrix A

invA=inv(A);

dx=1/n; % Step Size

x=0:dx:1; % x vector

% f matrix for u & error calculation

f1=ones(n-1,1);
f2=ones(n-1,1);
f3=ones(n-1,1);
f2e=ones(n-1,1);
f3e=ones(n-1,1);

for i=1:n-1

B1(i,1)=-dx^2*(-1);

f2(i,1)=x(i+1)^6-x(i+1)^5+x(i+1)^4-x(i+1)^3+x(i+1)^2-x(i+1);
f2e(i,1)=-dx^4/12*(30*x(i+1)^4-20*x(i+1)^3+12*x(i+1)^2-6*x(i+1)+2);
B2(i,1)=-dx^2.*f2(i,1);

f3(i,1)=-sin(pi*x(i+1));
f3e(i,1)=-dx^4/12*(pi^2*sin(pi*(x(i+1))));
B3(i,1)=-dx^2.*f3(i);
end

u1=invA*B1;
u2=invA*B2;
u3=invA*B3;
u2e=invA*f2e;
u3e=invA*f3e;

u1_final=[0; u1 ;0];
u2_final=[0; u2 ;0];
u3_final=[0; u3 ;0];
u2e_final=[0; u2e ;0];
u3e_final=[0; u3e ;0];

func1=-(x)/2+(x).^2/2;
func2=-1/840*(15*x.^8-20*x.^7+28*x.^6-42*x.^5+70*x.^4-140*x.^3+89*x);
func3=-sin(pi*x)/pi^2;
func1=func1’;
func2=func2’;
func3=func3’;

error1=func1-u1_final;
error2=func2-u2_final;
error3=func3-u3_final;

plot(x,u1_final,’o-’)
hold on
plot(x,func1,’o-’)
legend(’numerical’,’exact’,’location’,’southwest’)
xlabel(’x’)
ylabel(’u(x)’)
title(’f(x)=-1’)
set(gca,’fontname’,’times’)

figure
plot(x,u2_final,’o-’)
hold on
plot(x,func2,’o-’)
legend(’numerical’,’exact’,’location’,’southwest’)
xlabel(’x’)
ylabel(’u(x)’)
title(’f(x)=-x+x^2-x^3+x^4-x^5+x^6’)
set(gca,’fontname’,’times’)

figure
plot(x,u3_final,’o-’)
hold on
plot(x,func3,’o-’)
legend(’numerical’,’exact’,’location’,’southwest’)
xlabel(’x’)
ylabel(’u(x)’)
title(’f(x)=sin(\pix)’)
set(gca,’fontname’,’times’)

figure
plot(x,u2e_final,’o-’)
hold on
plot(x,error2,’o-’)
legend(’estimated error’,’exact error’,’location’,’south’)
xlabel(’x’)
ylabel(’error(x)’)
title(’f(x)=-x+x^2-x^3+x^4-x^5+x^6 ’)
set(gca,’fontname’,’times’)

figure
plot(x,u3e_final,’o-’)
hold on
plot(x,error3,’o-’)
legend(’estimated error ’,’exact error’,’location’,’south’)
xlabel(’x’)
ylabel(’error(x)’)
title(’f(x)=sin(\pix) ’)
set(gca,’fontname’,’times’)

Problem 2

clear all

n=20; % divisions in X direction

D=sparse(n,n-1); % creaating sparse matrix D

for i=1:n
for j=1:n-1
if i==j
D(i,j)=1;
elseif i==j+1
D(i,j)=-1;
end
end
end

dx=1/n; % step size X direction

x=[0:dx:1]’; % X vector
xm=[dx/2:dx:1-dx/2]’; % X vector for calcualting a values
a=1+sin(pi*xm);

K=sparse(diag(a));
A=D’*K*D; % A matrix

F=ones(n-1,1); % creating F matrix

u0=0;
u1=1;
F(1)=F(1)+a(1)*u0/dx^2; % Adding boundary conditions
F(n-1)=F(n-1)+a(n)*u1/dx^2;
U=dx^2*inv(A)*F;
Ufinal=[u0 ; U ; u1] % U matrix

plot(x,Ufinal,’*-’)
ylabel(’u(x)’)
xlabel(’x’)
set(gca,’fontname’,’times’,’fontsize’,12)

Problem 3

clear all

m=20; % divisions in X direction

n=10; % divisions in Y direction

dx=2/m; % step size X direction

dy=1/n; % step size Y direction

x=0:dx:2; % X vector
y=0:dy:1; % Y vector

Dx=sparse(eye(m,m-1));
Dy=sparse(eye(n,n-1));

%Creating f matrix
for i=1:m
for j=1:m-1
if i==j
Dx(i,j)=1;
elseif i==j+1
Dx(i,j)=-1;
end
end
end

for i=1:n
for j=1:n-1
if i==j
Dy(i,j)=1;
elseif i==j+1
Dy(i,j)=-1;
end
end
end

Dx=Dx/dx; % Dx matrix
Dy=Dy/dy; % Dy matrix

Axx=Dx’*Dx; % Axx matrix

Ayy=Dy’*Dy; % Ayy matrix

Iy=sparse(eye(n-1));
Ix=sparse(eye(m-1));

A=kron(Iy,Axx)+kron(Ayy,Ix); % A matrix

U_BC=zeros(m+1,n+1); % U matrix with BCs

U_BC(:,1)=sin(pi*x/2); % U matrix with BCs

f=zeros((m-1),(n-1));
for i=1:m-1 % f matrix including BCs
for j=1:n-1
f(i,j)=20*y(j+1)^2*sin(pi*x(i+1));
if j==1
f(i,j)=f(i,j)+ sin(pi/2*x(i+1))/dy^2;
end
end
end

func=reshape(f,(m-1)*(n-1),1); % reshaping f from 2D to 1D matrix

U=inv(A)*func; % calculating U matrix
U=reshape(U,(m-1),(n-1)); % reshaping U from 1D to 2D matrix
U_BC(2:m,2:n)=U; % Adding boundary values , U_BC is the final solution
[xx,yy]=meshgrid(x,y);

surf(x,y,U_BC’)
xlabel(’x’)
ylabel(’y’)
zlabel(’u(x,y)’)
set(gca,’fontname’,’times’,’fontsize’,12)

figure
contour(x,y,U_BC’,15)
xlabel(’x’)
ylabel(’y’)
set(gca,’fontname’,’times’,’fontsize’,12)