Re: test release gsl-1.8.90.tar.gz

[Tim Fenn <fenn at stanford dot edu>]

On Fri, Feb 16, 2007 at 09:25:30PM +0000, Brian Gough wrote:
> At Fri, 16 Feb 2007 00:09:34 -0800,
> Tim Fenn wrote:
> > I've been testing the bfgs2 minimizer, and it works well until/if it
> > nears a point in which the interpolator stops progressing - I've
> > noticed there is a check against this in the linear_minimize code near
> > line 209:
> > 
> >       if ((a-alpha)*fpa <= GSL_DBL_EPSILON) {
> >         /* roundoff prevents progress */
> >       };
> > 
> > perhaps there should be a break statement to prevent futile
> > iterations?  Or use this as a stopping criteria test?
> 
> Thanks, there should be a 
> 
>         return GSL_ENOPROG;
> 
> in there.  You might want to try that and see if it stops at an
> acceptable place.

That helps, yes.  It also might be worthwhile to note the bfgs2
minimizer requires positive step_size values (which makes intuitive
sense, but is not true of the bfgs/cgpr/cgfr minimizers).

Regards,
Tim

-- 
---------------------------------------------------------

        Tim Fenn
        fenn@stanford.edu
        Stanford University, School of Medicine
        James H. Clark Center
        318 Campus Drive, Room E300
        Stanford, CA  94305-5432
        Phone:  (650) 736-1714
        FAX:  (650) 736-1961

---------------------------------------------------------