Re: [Bug-gsl] gsl_eigen_symmv

[Katrin Wolff]

I sent the wrong version of the program before, which will not reproduce 
the error. I attach the correctly faulty one to this mail.

Benedicte Jourdain tried the program as well and answered the following 
to my mail:

without the diagonal at 1.0, I 've the same problem on the IMAC  
(infinite loop) but not on a machine with cygwin.
After analysis I think the problem is in function  
gsl_linalg_householder_transform called from gsl_linalg_symmtd_decomp  
called from gsl_eigen_symmv.
This function (gsl_linalg_householder_transform) called the function  
hypot which is either the standard library function or gsl_hypot  
depending of your configuration.
On my IMAC this is the standard library function which is called. you  
can verify this in the file config.h with #define HAVE_DECL_HYPOT 1.

I'am not an official maintener of GSL, so i can only give advices.

After ./configure edit config.h to change #define HAVE_DECL_HYPOT 0  
before make, make check and make install


Katrin

On Sunday 08 November 2009, Brian Gough wrote:
> At Wed, 4 Nov 2009 15:50:22 +0100,
>
> Katrin Wolff wrote:
> > I have a problem computing the eigensystem of one particular matrix
> > that I chanced upon. gsl_eigen_symmv() appears to be caught in an
> > infinite loop. I attach a minimal program and the matrix producing
> > the error.
> >
> > Version and architecture are:
> > Architecture: amd64
> > Source: gsl
> > Version: 1.11+dfsg-1
>
> Thanks for the bug report.  I have tried the test program on x86 and
> was not able to reproduce the error.  I am busy at a conference this
> week, but I will try it again on a 64 bit machine when I get back.

matrix_test_main.cc
Description: Text Data