Re: [Help-gsl] Question about simulated annealing

[Benjamin Winkel]

Hi,

months ago I had the same problems. It was enlightening for me to have a 
look into the sources. There are two functions, the solver and 
solver_many. The latter does not do iters_fixed_t calculations but tries 
n_tries points in the neighborhood, choosing the best step (and lowering 
the temperature immediately). So if you choose the "normal" solver only 
the iters_fixed_t variable is considered, while for solver_many only the 
n_tries variable is taken into account.

Hope that helps
Benjamin

Sak wrote:
> I'm trying to use the simulated annealing module of GSL.  I understand the
> algorithm pretty well and the settings parameters passed to the main SA
> function
> in the gsl_siman_params_t data structure.
>
> I believe iters_fixed_t is the length of the random walk at each
> temperature.  Is
> this correct?
>
> And, I don't understand the meaning of n_tries from the GSL documentation:
> "The number of points to try for each step."  From what I understand of SA,
> if the e_i+1 <= e_i, then we take the step for sure (p = 1.0).  If e_i+1 >
> e+i, then
> the probability of taking the step is based on the boltzmann distribution.
> I
> don't understand why multiple points(neighbors?) are sampled before taking
> a step in our random walk.
>
> I thought maybe it means, if n_tries = 20, then we try 20 points and take
> the smallest e_i+1 if all 20 e_i+1 <= e_i.  But then I don't understand how
> GSL would handle cases where multiple e_i+1 > e_i, which leads me to
> think I'm wrong about what n_tries means.  Can anyone clear up for me
> what n_tries function is?
>
> Thanks in advance,
> Sak S.
> _______________________________________________
> Help-gsl mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/help-gsl

--
Dipl.-Phys. Benjamin Winkel

Argelander Institute for Astronomy (AIfA)
University of Bonn
Auf dem Hügel 71, D-53121 Bonn, Germany

tel: +49 (0)228 73-3662
address@hidden
http://www.astro.uni-bonn.de/~bwinkel