Services & ResourcesWolfram ForumsMathGroup Archive

[Date Index] [Thread Index] [Author Index]

SingularValueDecomposition bug: fifty examples

• To: mathgroup at smc.vnet.net
• Subject: [mg60753] SingularValueDecomposition bug: fifty examples
• From: John Sidles <sidles at u.washington.edu>
• Date: Tue, 27 Sep 2005 03:45:23 -0400 (EDT)
• References: <906C0B48-1EB7-403C-A2F4-ED8FEA05BE1F@u.washington.edu>
• Sender: owner-wri-mathgroup at wolfram.com

Dear Mathgroup folks


I have documented fifty different examples of matrices for which
SingularValueDecomposition[] fails under Mathematica 5.2
(Mac OS X, Tiger).  Full details are here:

     http://faculty.washington.edu/sidles/mma_svd_failure_example/ 
README.pdf

The matrices themselves (plus a notebook to successively load them and
document the SVD failure) are in this gzipped tarball:

     http://faculty.washington.edu/sidles/mma_svd_failure_example/ 
SVDfailures.tar.gz

A Makefile is included -- run "make unpack" to unpack the tarball:

     http://faculty.washington.edu/sidles/mma_svd_failure_example/ 
Makefile

Just as a point of interest, the above files include:

   -> 25 matrices that generate the following (undocumented) error  
message:

      "SingularValueDecomposition::cfail:  Algorithm failed to  
converge."

   -> 25 matrices for which SingularValueDecomposition[] returns a
         wrong result, with no error message.

Needless to say, a call to SingularValueDecomposition[] should never fail ...
our application (a terrain-following mapping program) surely depends  
on this!

Sincerely
John Sidles

• Follow-Ups:
  • Re: SingularValueDecomposition bug: fifty examples
    • From: Sseziwa Mukasa <mukasa@jeol.com>