Re: [Bug-gsl] gsl BFGS multimin bug

[Carl Glen Henshaw]

On Feb 2, 2005, at 1:11 PM, Brian Gough wrote:

> Carl Glen Henshaw writes:
> > All:
> >
> > I believe I have located a bug in the bfgs multivariate minimization
> > routine. Specifically, in some circumstances it is possible for the
> > vector returned by the bfgs minimization routine to contain NaN
> > elements. This is because in the file vector_bfgs.c, in the routine
> > vector_bfgs_iterate, starting on or near line 289, it is possible for
> > dxdg to be uniformly equal to zero. When the BFGS update is 
> > calculated,
> > dxg/dxdg and dgg/dxdg are then undefined.
>
> Thanks for the bug report.  From you description I've got a reasonable
> idea of how this happens.
>
> Do you have a simple test program which demonstrates the problem?---so
> that I could use it for testing purposes and make sure the bug is
> really fixed.

Hi Brian,

Unfortunately I don't. I ran into the bug in the course of writing a 
fairly complex robotic path planner; basically you can think of it as 
part of a continuation scheme. Unfortunately it relies on several other 
pieces of code to run correctly. The bug is fairly intermittent, as you 
might expect, so it's not easy for me to figure out exactly what kind 
of system and what initial conditions replicate it. I have a conference 
coming up next week, but if it would be really helpful I might be able 
to take a few hours and try to replicate it in a simple piece of code 
after I get back.

Thanks,
Glen

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

Dr. Glen Henshaw
Attitude Control Systems Branch
Naval Center for Space Technology
U.S. Naval Research Laboratory