Home »
Public Forums »
archive »
SVDC routine. Is it o.k.?
| SVDC routine. Is it o.k.? [message #9462] |
Thu, 03 July 1997 00:00 |
Paul van Delst
Messages: 364
Registered: March 1997
|
Senior Member |
|
|
Hello,
I use the SVDC routine in IDL to perform matrix inversion on rather
large square symmetric matrices (100s x 100s). Does anyone know if the
IDL version of SVDC (uses the Numerical Recipes algorithm) is bug free?
I'm not in any way, form, or manner suggested that it's busted but I'm
told there are bugs in the FORTRAN version of this code. A colleague of
mine perfomed some comparisons of NR's SVD decomposition with that from
LAPACK. He decomposed a rectangular matrix (MxN; M ne N), reconstructed
it and differenced the reconstructed matrix with the original. He found
that the LAPACK and NR SVDs gave peak residuals of 1.0e-3 - 1.0e-4 and
1.0e-1 - 1.0e-4 respectively. He also noticed that the NR SVD did not
always produce singular values in descending order.
Has anyone experienced any problems/worrying results with the IDL
version of SVDC? I am also wondering what the dependence on possible
problems is if you are decomposing a rectangular matrix as opposed to a
square symmetric matrix (that's a question for the sci.math types).
Any information would be appreciated.
Paul van Delst
|
|
|
|
Current Time: Wed Oct 08 18:07:46 PDT 2025
Total time taken to generate the page: 0.00426 seconds
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
