hw1p3sol

# hw1p3sol - We sse that the factorization errors are all...

This preview shows page 1. Sign up to view the full content.

% Solution of HW1 Problem 3 % % We examine the accuracy of the Cholesky factorization and the solutions it produces % Cholerr=zeros(200,1); Solerr=zeros(200,1); for k=1:200 S=randn(100,100); A=S'*S; R=chol(A); x=randn(100,1); b=A*x; y=R\(R'\b); Cholerr(k)=norm(R'*R-A)/norm(A); Solerr(k)=norm(x-y)/norm(x); end figure hist(Cholerr) title(' Relative Errors in the Cholesky Factorization ') figure hist(Solerr) % To see a more interesting distribution lets take the log of the solution errors hist(log(Solerr)) hist(log10(Solerr)) title(' Relative Errors in the Solutions ')
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: % % We sse that the factorization errors are all near machine epsilon - clustered % between 1 and 2 times 10^(-16). The solution errors are significantly larger % They are clustered around 10^(-12) with at least one larger than 10^(-9). % This can be explained by the fact that the Cholesky factorization algorithm % is always stable, but positive definite symmetric systems themselves can be % sensitive (ill-conditioned) % % The graphs are found in Cholerr.pdf and Solerr.pdf...
View Full Document

{[ snackBarMessage ]}

Ask a homework question - tutors are online