Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
30 views

Problem 5

This document contains 6 questions about numerical linear algebra and numerical methods. The questions cover topics like LU-factorization, iterative methods for solving systems of equations, analyzing convergence of Jacobi and Gauss-Seidel methods, matrix norms, and finite difference schemes for solving Poisson's equation.

Uploaded by

Djaafer Mezhoud
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
30 views

Problem 5

This document contains 6 questions about numerical linear algebra and numerical methods. The questions cover topics like LU-factorization, iterative methods for solving systems of equations, analyzing convergence of Jacobi and Gauss-Seidel methods, matrix norms, and finite difference schemes for solving Poisson's equation.

Uploaded by

Djaafer Mezhoud
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

MATH3476 Numerical Methods

Problem Set 5
Numerical Linear Algebra

The following questions are based upon the material covered in Chapter 3 of the notes.
Q1 Use LU-factorisation to show that the solution of the tridiagonal system of equations
   
4 1 0 0 1
 1 2 3 0   0 
 0 3 1 1 x =  1 
   

0 0 2 2 0
1 1 5 5 T

is x = 6 , 3 , − 18 , 18 . Retain the use of exact fractions throughout your calculations.

Q2 Consider the system of equations


   
2 1 4
x= .
1 2 5

(i) Determine HJ , HGS and Hω .


(ii) Evaluate ρ(HJ ) and ρ(HGS ).
(iii) Show that the eigenvalues µ of Hω satisfy the quadratic equation
 2 
2 ω
µ − − 2ω + 2 µ + (ω − 1)2 = 0,
4
and, along the lines of argument in §3.2.5, determine ω ∗ and ρ(Hω∗ ). Confirm this value of ω ∗
using (3.21).
(iv) Using the spectral radius of each matrix, evaluate the theoretical number of iterations, NJ , NGS
1
and Nω∗ ,which ensure that e(N ) ≤ 10 e(0) , i.e. to reduce the initial error by a factor of 10.
(v) If now x(0) = 0, determine the exact solution and evaluate e(0) 2 . Hence calculate the actual
value of N required to reduce e(0) 2 by a factor of 10 for each of the three iterative schemes
(using ω ∗ in SOR).

Q3 Example 3.2 in the notes demonstrates that Gauss-Seidel is not always superior to Jacobi. This is also
the case for the system of equations with coefficient matrix
 
1 0 1
A =  −1 1 0 .
1 2 −3

(i) Determine both HJ and HGS in terms of an ILU splitting of A.


(ii) Determine the eigenvalues of HGS , and so its spectral radius ρ(HGS ). What can you say about
the convergence of Gauss-Seidel?
(iii) Show that the characteristic equation for the eigenvalues of HJ is
1 2
C(λ) ≡ λ3 + λ + = 0.
3 3
(iv) By considering the turning points and sign properties of C(λ), deduce that there is only one real
root (λ1 ) and two complex (conjugate) roots (λ2 and λ̄2 ) of C(λ). Using Python, or whatever
other method you choose, determine the approximate value of the real root.

1
(v) By comparing the above characteristic equation with the expanded form of C(λ) ≡ (λ−λ1 )(λ−
λ2 )(λ− λ̄2 ) = 0, deduce the magnitude of the complex roots. Hence show that Jacobi converges
in this case, even though you have not yet derived the exact form of all the eigenvalues.
(vi) Use the coefficient of λ2 in C(λ) to deduce both the real and imaginary parts of λ2 .
(vii) Hence plot two diagrams showing the eigenvalues and the Gerschgorin row and column disks
of HJ .

Q4 By expressing the n-dimensional initial error vector e(0) in equation (3.17) as a linear combination
of n linearly independent eigenvectors of H, deduce the necessary condition for convergence, i.e.
ρ(H) < 1.

Q5 (a) For the matrix  


2 1
A= ,
3 6
evaluate ||A||F , ||A||1 , ||A||2 and ||A||∞ using the norm definitions of §3.1.1.

(b) Consider the general 2 × 2 matrix  


a b
A= .
c d

(i) Show that


1
  1 
||A||22 = ||A||2F + ||A||4F − 4|A|2
2
,
2
where |A| is the determinant of A.
(ii) Show that
||A||2F ≥ 2|A|
for any 2 × 2 matrix, and hence that ||A||2 is real.
(iii) Hence deduce that
1
√ ||A||F ≤ ||A||2 ≤ ||A||F .
2

Q6 A function satisfies the Poisson equation ∆u = −2 in the square region −1 < x, y < 1, and is zero
everywhere on the boundary of the square.

(i) Use the (5,+) molecule on a square mesh with h = k = 1 to approximate the value of u(0, 0).
(ii) On a square mesh with h = k = 21 , use symmetry arguments to show that there are just three
points (of the nine possible) at which u takes different values. Using the (5,+) molecule, estab-
lish three equations in the three unknown values of u and show that they have solution 0.5625,
0.34375 and 0.4375.
(iii) Solve the problem with h = k = 1 using the 9-point Mehrstellenverfahren scheme in §2.2.2.
1
(iv) Solve the problem with h = k = 2 using the 9-point Mehrstellenverfahren scheme.
(v) Solve (3.43) in the notes for p. Given that the exact answers to 5 decimal places are 0.58937,
0.36229 and 0.45868, use the above numerical results at (0, 0) to determine the error order p for
both the 5-point and 9-point schemes.

You might also like