Re: [Help-gsl] GSL Error code in multimin optimization

[Brian Gough]

At Tue, 16 Feb 2010 14:30:02 -0500,
Sushmita Roy wrote:
> I am using the GSL multimin package to find the minima of
> a high (~1000) dimensional function. I am using the 
> "gsl_multimin_fdfminimizer_vector_bfgs2"
> optimizer and I am following the example in the GSL multimin manual. 
> However, every time I
> execute an iteration using "gsl_multimin_fdfminimizer_iterate" I always 
> get the error code GSL_ENOPROG,
> which stands  for "iteration is not making progress..".
> 
> I know that there are several calls to my likelihood and its derivative 
> functions, so the optimizer
> does search for some time. But I am just concerned about the severity of 
> this error code, and
> whether it indicates that my results are wrong. I have checked the 
> likelihood and derivative
> functions and I am quite certain that I have them correctly.
> 

If you still have the problem, an example program might help people to
see what is wrong.

In the code there are only 2 places where ENOPROG is returned

  if (pnorm == 0.0 || g0norm == 0.0 || state->fp0 == 0)
    {
      gsl_vector_set_zero (dx);
      return GSL_ENOPROG;
    }


   if ((a-alpha)*fpa <= GSL_DBL_EPSILON) {
        /* roundoff prevents progress */
        return GSL_ENOPROG;
   };



-- 
Brian Gough

GNU Scientific Library -
http://www.gnu.org/software/gsl/