[Fwd: Re: Possible bug/problem in Jama package]

[Ron Boisvert <boisvert@nist.gov>]

Sender: "G. W. (Pete) Stewart" <stewart@cs.umd.edu>
Subject: Re: Possible bug/problem in Jama package



Rather structured example.  The matrix is orthogonal with two complex
eigenvalues.  It's possible, though unlikely, that such an example
could send a QR code into a loop.  But LAPACK, courtesy of Matlab,
handles it just fine.

Pete Stewart



On Fri, 27 Mar 2009, Alexej Abyzov wrote:

> Date: Fri, 27 Mar 2009 13:44:43 -0400
> From: Alexej Abyzov <aabyzov@gmail.com>
> Reply-To: jama@nist.gov
> To: Multiple recipients of list <jama@nist.gov>
> Subject: Possible bug/problem in Jama package
>
> 		Dear developers,
> 	thank you for developing and maintaining Jama package. It is very 
> useful for my research.
> 	However, it appears that I found bug or potential problem with 
> eigenvalue decomposition. Namely, the decomposition can never be completed 
> (calculations continue for days without any result) for the following matrix
>
> 0.07115342717017722 -0.9847493690987374 0.15876356591352137 
> -0.09827567440311949 0.15147152347373427 0.9835640647142199 
> -0.9926122513965614 -0.08558655055664227 -0.0859991903523808
>
> 	Currently, I'm resolving the problem by lowering the precision in 
> hqr2 function from
>     double eps = Math.pow(2.0,-52.0);
> to
>     double eps = Math.pow(2.0,-20.0);
>
> 	I hope you'll fix/address this issue in future releases of Jama.
> 	Thank you.
>
>
> 		Best regards.
>
>                              Alex Abyzov
> ----------------------------------------------------
> Department of Molecular Biophysics and Biochemistry,
> Yale University, 260 Whitney ave., P.O. Box 208114,
> New Haven, CT, 06520, USA
> Phone: 1-(203)-432-5405
> e-mail: abyzov@gersteinlab.org
> URL: http://homes.gersteinlab.org/people/aabyzov
> ----------------------------------------------------
>
>
>
>