This is the mail archive of the
gsl-discuss@sources.redhat.com
mailing list for the GSL project.
Re: Problem with Singular Value Decomposition Algorithm
- To: "Jim Love" <Jim dot Love at asml dot com>
- Subject: Re: Problem with Singular Value Decomposition Algorithm
- From: Brian Gough <bjg at network-theory dot co dot uk>
- Date: Fri, 14 Sep 2001 18:10:50 +0100 (BST)
- Cc: <gsl-discuss at sources dot redhat dot com>
- References: <sba1cb3a.019@wiltonhub.svg.com>
Jim Love writes:
> I still think their is a problem with your API. I have done this problem using a pencil and paper, using Matlab, using Mathematica, and using the LAPACK DGESVD routine. These 4 implementations all produce the same exact solution for the U S and Q matrix - only yours is different. How do you explain this?
>
Matlab uses LAPACK. I believe Mathematica might use that too now. So
that's probably why they are giving the same results.
Provided the results returned by GSL satisfy u'*u = I, v'*v = I and
u*diag(s)*v' = A then it's a valid singular value decomposition. From
my own testing this checks out ok, and also with the numbers you sent
in your emails. If you have an example that doesn't then that would
be a bug.