GSL-1.2 gsl_eigen_symmv() Problem

GSL-1.2 gsl_eigen_symmv() Problem

[Adam Johansen]

Hello,

I'm rather new to this list, so I hope you'll forgive me if this is
outside it's normal scope or has been answered before (although I can't
find any reference to anything similar in the archive).

When calling gsl_eigen_symmv within a loop in the heart of a C program
(which is still very much in the development stage, but is over 1000 lines
long and hence not easy to include here) the program halts with an
apparently infinite (certainly happy to continue running for 16hrs) loop
within the symmv function (although gdb indicates that it switches between
symmv and symm quite frequently).

The problem is not a simple one, as I tried using the gsl_matrix_fwrite
and gsl_matrix_fread functions to record the matrices being used in this
program and then wrote a quick test program which loaded the matrices from
disk and then ran gsl_eigen_symmv on them. This program ran quite happily
and found the eigenvalues of these matrices.

My question, I suppose, is whether there is any known interaction between
this function and any other elements of the gsl or standard libraries (no
nonstandard libraries are being used, although I am using electricfence
for debuggin) which could lead to this sort of behaviour. I'm reasonably
sure that there are no incorrect memory accesses within the loop (I've
used MALLOC_CHECK_=2 and electric fence to check this...) although
something like that is always possible, I suppose.

Many thanks, in advance,
Adam Johansen

Re: GSL-1.2 gsl_eigen_symmv() Problem

[Adam Johansen]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 829 bytes --]

Hello Again,

Further to my earlier email, I've managed to narrow the problem down
somewhat. It would seem that if the log() function is called before a call
to gsl_eigen_symmmv with a matrix similar to one which has already been
solved (this is a rough summary which may not be entirely accurate, but I
think contains the relevant details) then that routine goes into an
apparently infinite loop.

I've attached a short piece of code which demonstrates this problem and
I'd be interested to know whether it is a general one or is confined to
systems like my own (or indeed, is due to a corrupt library and is only
happening to me...).

Many apologies if this is a known problem, or if there's some obvious
reason why using the C libraries log function in conjunction with gsl
isn't a reasonable thing to do....

Adam Johansen



[-- Attachment #2: The offending code --]
[-- Type: TEXT/PLAIN, Size: 789 bytes --]

#include <stdio.h>

#include <gsl/gsl_eigen.h>
#include <gsl/gsl_matrix.h>
#include <gsl/gsl_vector.h>

int main(int argc, char** argv) {
  double hmmm;
  gsl_eigen_symmv_workspace* wsp;
  gsl_matrix* in, *out;
  gsl_vector* vec;

  in = gsl_matrix_alloc(2,2);
  out = gsl_matrix_alloc(2,2);
  vec = gsl_vector_alloc(2);
  int i;

  int mi = 0;

  for(i = 1; i < 101; i++) { 
     hmmm = log(1.2);
     
     for(mi = 0; mi < 6; mi++) {
	wsp = gsl_eigen_symmv_alloc(2);
       
	gsl_matrix_set(in, 0, 0, i);
	gsl_matrix_set(in, 0, 1, i - 1);
	gsl_matrix_set(in, 1, 0, i - 1);
	gsl_matrix_set(in, 1, 1, i - 0.5);
	
	printf("Solving matrix %i in iteration %i...\n", mi, i);
	gsl_eigen_symmv(in, vec, out, wsp);
	
	gsl_eigen_symmv_free(wsp);
     }
  }
}

Re: GSL-1.2 gsl_eigen_symmv() Problem

[Atakan Gurkan]

On Thu, Nov 07, 2002 at 05:06:04PM +0000, Adam Johansen wrote:
> #include <stdio.h>
> 
> #include <gsl/gsl_eigen.h>
> #include <gsl/gsl_matrix.h>
> #include <gsl/gsl_vector.h>
> 
> int main(int argc, char** argv) {
>   double hmmm;
>   gsl_eigen_symmv_workspace* wsp;
>   gsl_matrix* in, *out;
>   gsl_vector* vec;
> 
>   in = gsl_matrix_alloc(2,2);
>   out = gsl_matrix_alloc(2,2);
>   vec = gsl_vector_alloc(2);
>   int i;
> 
>   int mi = 0;
> 
>   for(i = 1; i < 101; i++) { 
>      hmmm = log(1.2);
>      
>      for(mi = 0; mi < 6; mi++) {
> 	wsp = gsl_eigen_symmv_alloc(2);
>        
> 	gsl_matrix_set(in, 0, 0, i);
> 	gsl_matrix_set(in, 0, 1, i - 1);
> 	gsl_matrix_set(in, 1, 0, i - 1);
> 	gsl_matrix_set(in, 1, 1, i - 0.5);
> 	
> 	printf("Solving matrix %i in iteration %i...\n", mi, i);
> 	gsl_eigen_symmv(in, vec, out, wsp);
> 	
> 	gsl_eigen_symmv_free(wsp);
>      }
>   }
> }
himm, shouldn't you include math.h? since log does not return int this might
cause a problem. BTW, your code doesn't compile with a C compiler, you might
want to move those int statements a few lines earlier.

ato

Re: GSL-1.2 gsl_eigen_symmv() Problem

[Adam Johansen]

On Thu, 7 Nov 2002, Atakan Gurkan wrote:

> himm, shouldn't you include math.h? since log does not return int this might
> cause a problem. BTW, your code doesn't compile with a C compiler, you might
> want to move those int statements a few lines earlier.
>
> ato
>

Thanks. You are, of course, completely right.  I've just spent 4 days
debugging one program and it's amazing how easy it is to miss the obvious.
Many apologies for wasting your time....

AMJ