[alt.sources.wanted] C Singular Value Decomposition package?

skh@nl.cs.cmu.edu (Steve Handerson) (09/18/90)

Folks,

I'm in dire need of a singular-value-decomposition package
for arbitrary size (large) matrices.

If anyone has one already implemented and working in C,
I'd really appreciate it.
I have algorithms with non-C code,
but have been unable to write code from them that works.

Alternatively, if anyone can tell me in simple terms
what to do with the eigenvalues and eigenvectors of
A * transpose(A) and transpose(A) * A 
so that the three actually multiply together to form the original
matrix, I'd appreciate that too.

Please mail me at skh@nl.cs.cmu.edu internet, and I guess
...!nl.cs.cmu.edu!skh from a backbone sight on the internet
otherwise, and I'll post a summary on sci.math.num-analysis.

-- Steve

reeves@dvinci (Malcolm Reeves) (09/20/90)

From article <10489@pt.cs.cmu.edu>, by skh@nl.cs.cmu.edu (Steve Handerson):
> 
> I'm in dire need of a singular-value-decomposition package
> for arbitrary size (large) matrices.
> 
> If anyone has one already implemented and working in C,
> I'd really appreciate it.
>
There is C source (or CORTRAN :-) in Numerical Recipes in C by Press et al.
Cambridge University Press. Can't remember the ISBN.
There are algorithms in John Nash's book "Compact Numerical Methods for
Computers". Publisher is Pitman I think. Both are good and I've written
svd code using them - but in Pascal. Hope this helps. I could look up
the references properly if you really need them send me mail.

pmm@acsu.buffalo.edu (patrick m mullhaupt) (09/27/90)

In article <1990Sep20.011833.22595@herald.usask.ca> reeves@dvinci writes:
>From article <10489@pt.cs.cmu.edu>, by skh@nl.cs.cmu.edu (Steve Handerson):
>> 
>> I'm in dire need of a singular-value-decomposition package
>> for arbitrary size (large) matrices.
>
>There is C source (or CORTRAN :-) in Numerical Recipes in C by Press et al.

	I'm not absolutely sure, but I believe there was recently a
discussion in this newsgroup, (maybe it was in sci.math), about a bug
in the Numerical Recipes singular value decomposition algorithm.
	If you're not adverse to fortran, you might want to try the
numerical sofware library available at netlib@ornl.gov.  Just send
mail to netlib@ornl.gov, with the message,
	send index
and you will get a reply with more information about how to get
software (fortran source code), for free.

	Hope this helps,
		Patrick Mullhaupt

PS Numerical Recipes is still an excellent text worth getting.