03.03.2014 Views

Numerical Methods Course Notes Version 0.1 (UCSD Math 174, Fall ...

Numerical Methods Course Notes Version 0.1 (UCSD Math 174, Fall ...

Numerical Methods Course Notes Version 0.1 (UCSD Math 174, Fall ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

46 CHAPTER 3. SOLVING LINEAR SYSTEMS<br />

(e) Show that when 0 < α, this quantity is always smaller than 1.<br />

(f) Prove that if A is positive definite, then there is an ω such that Richardson’s Iteration<br />

with this ω will converge for any choice of x (0) .<br />

(g) For which matrix do you expect faster convergence of Richardson’s Iteration: A 1 with<br />

eigenvalues in [10, 20] or A 2 with eigenvalues in [1010, 1020]? Why?<br />

(3.11) Implement Richardson’s Iteration to solve the system Ax = b. Your m-file should have<br />

header line like:<br />

function xk = richardsons(A,b,x0,w,k)<br />

Your code should return x (k) based on the iteration<br />

( )<br />

x (j+1) = x (j) − ω Ax (j) − b .<br />

Let w take the place of ω, and let x0 be the initial iterate x (0) . Test your code for A, b for<br />

which you know the actual solution to the problem. (Hint: Start with A and the solution x<br />

and generate b.) Test your code on the following matrices:<br />

• Let A be the Hilbert Matrix. This is generated by the octave command A = hilb(10),<br />

or whatever (smallish) integer. Try different values of ω, including ω = 1.<br />

• Let A be a Toeplitz matrix of the form:<br />

⎡<br />

A =<br />

⎢<br />

⎣<br />

−2 1 0 · · · 0<br />

1 −2 1 · · · 0<br />

0 1 −2 · · · 0<br />

. . .<br />

. .. .<br />

0 0 0 · · · −2<br />

These can be generated by the octave command A = toeplitz([-2 1 0 0 0 0 0 0]).<br />

Try different values of ω, including ω = −1/2.<br />

(3.12) Let A be a nonsingular n × n matrix. We wish to solve Ax = b. Let x (0) be some starting<br />

vector, let D k be span { r (0) , Ar (0) , . . . , A k r (0)} , and let P k be the set of polynomials, p(x) of<br />

degree k with p(0) = 1.<br />

Consider the following iterative method: Let x (k+1) be the x that solves<br />

min<br />

x∈x (0) +D k<br />

‖b − Ax‖ 2<br />

.<br />

Let r (k) = b − Ax (k) .<br />

(a) Show that if x ∈ x (0) + D k , then b − Ax = p(A)r (0) for some p ∈ P k .<br />

(b) Prove that, conversely, for any p ∈ P k there is some x ∈ x (0) + D k , such that b − Ax =<br />

p(A)r (0) .<br />

(c) Argue that ∥ ∥ ∥∥r (k+1)∥<br />

∥∥p(A)r (0)<br />

∥ = min ∥ .<br />

2 p∈P k 2<br />

(d) Prove that this iteration converges in at most n steps. (Hint: Argue for the existence<br />

of a polynomial in P n that vanishes at all the eigenvalues of A. Use this polynomial to<br />

show that ∥ ∥ r<br />

(n) ∥ ∥<br />

2<br />

≤ 0.)<br />

⎤<br />

⎥<br />

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!