Homework 2

Name: Francisco Quinga

Assume that | | = 1 and | | 1

=
( )1 =

= (D E)
Then for the m-row we have:

= ( )
> <

==> | ( )| | | | |
< > >

==> || ( | | | |) || |( )| ||
< < >

==> || (| | | |) | |
< >

==> || (| || | || | | ) | |
< >

==> || (| | | | ) | |
< >

>| |
==> ||
| | < | |

Since A is strictly diagonally dominant then: | | < | | > >| | , so

( ) < 1
Which means that the Gauss-Seidel iteration converges.
2. Generate a symmetric positive definite matrix.
Using the generateSPDmatrix(n) algorithm for n=5 we get the following matrix:

a) Implement Jacobi method and compute the solution of the linear system.
function y=jacobi(A,b,x0,iter)
n=size(A,1);
for l=1:iter
y=zeros(n,1);
for i=1:n
%-----------sumatorio----------
suma=0;
for j=1:n
if j~=i
suma=suma+A(i,j)*x0(j);
end
end
%--------------------------------
y(i)=(b(i)-suma)/A(i,i);
end
x0=y;
end
end

4 0
If compute the solution for the linear system Ax=b were b=(32) and x0=(00), we get
5 0
3 0
the next solution for 3 iterations:
b) Implement Gauss-Seidel method and compute the solution of the linear system.
function x=seidel(A,b,xo,iter)
n=size(A,1);
for l=1:iter
x=xo;
for i=1:n
%-----------sumatorio----------
suma=0;
for j=1:n
if j~=i
suma=suma+A(i,j)*x(j);
end
end
%--------------------------------
x(i)=(b(i)-suma)/A(i,i);
end
xo=x;
end
end

4 0
If compute the solution for the linear system Ax=b were b=(32) and x0=(00), we get
5 0
3 0
the next solution for 3 iterations:
c) Compare a) and b). How much faster is Gauss-Seidel?
The following graph shows the logarithm in base 10 of the norm 2 of the error in each
iteration. Where the error of the k-th iteration is equal to:
= || ||

Fig 1. Jacobi vs Gauss-Seidel.

( ) ()
0.3590 0.1542
Table 1. Spectral radius of Jacobi and Gauss Seidel.
As we can see in the graph, Gauss-Seidel method converges faster than the Jacobi
method. In addition, the convergence factor of each method indicates that Gauss-Seidel
0.3590
is 0,1542 = 2,32 times faster.
d) Write down the numbers of iterations.
Jacobi iterations:
K=1 K=3 K=5 K=7 K=11 K=15

Gauss-Seidel iterations:
K=1 K=3 K=5 K=7 K=11 K=15

e) Implement Conjugate Gradient method and compute the solution of the linear
system.
function x= conjugrad(A,b,tol)
x=b-b;
r=b;
p=b;
while norm(r,2)>tol
z=A*p;
v=(r'*r)/(p'*A*p);
x=x+v*p;
temp=r;
r=r-v*z;
u=(r'*r)/(temp'*temp);
p=r-u*p;
end
end

4
If compute the solution for the linear system Ax=b were b=(32) , we get the next with a
5
3
tolerance of 0.1 solution:
f) Compare e) with a) and b).
The following graph shows the logarithm in base 10 of the norm 2 of the error in each
iteration. Where the error of the k-th iteration is equal to:
= || ||

Fig 2. Jacobi vs Gauss-Seidel vs Conjugate Gradient.

As we can see in the graph, the Gauss-Seidel method is still being faster than Jacobi and
Conjugate Gradient.

g) Write down the numbers of iterations.