Re: [Help-gsl] gsl_multimin_fminimizer_set fails to return

[Stephen J. Barr]

I think that I see what is happening. From my starting point, some small
deviations quickly lead to inf. So, what it comes down to, I think , is
that I need to be more careful about the size of dx.

Thanks,
Stephn

On Thu, Aug 30, 2012 at 11:47 AM, Brian Hawkins <address@hidden>wrote:

> Stephen,
>
> The set routine does little more than call the objective function at the
> initial guess x0 and at x0+dx for each dx in the initial step sizes.  Have
> you verified that the MPI objective function works with these parameters?
>  Is there any diagnostic information besides a missing print output (e.g.,
> segfault or backtrace).
>
> As an aside, I see lots of vector alloc calls but none to free.  Are you
> using some C++ wrapper for GSL?
>
> Regards,
> Brian
>
> On Thu, Aug 30, 2012 at 9:00 AM, <address@hidden> wrote:
> >
> >
> > Hello,
> >
> > I am working on a project using GSL to minimize a function, where this
> > function calls MPI to do the actual computation. I have both an MPI and a
> > non-MPI version of the objective function. When I use the non-MPI
> version,
> > everything seems to work. However, when I use the MPI version,
> > gsl_multimin_fminimizer_set fails to run ("passed set" never prints)
> >
> >   cout << "GSL: " << "Starting minimizer" << endl;
> >   S = gsl_multimin_fminimizer_alloc(T, MODEL_VALUE_COUNT);
> >   cout << "GSL: passed alloc"  << endl;
> >   gsl_multimin_fminimizer_set(S, &MinexFunc, StartVec, StepVec);
> >   cout << "GSL: passed set" << endl;
> >
> > The full code is available at https://gist.github.com/3522278
> >
> > The non-mpi version code from 596 to 613 runs just as I would want it to.
> > However, the MPI equivalent (lines 622 to 650) do not. I would really
> > appreciate some feedback as to why this isn't working as I would expect.
> I
> > know that the mpi objective function (nll_mpi, which is wrapped by
> > SolveEnergy).
> >
> > Thanks,
> > Stephen
> >
>