MathGroup Archive 2004

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: parallel NMinimize[]

  • To: mathgroup at smc.vnet.net
  • Subject: [mg50614] Re: parallel NMinimize[]
  • From: "Michal Kvasnicka" <michal.kvasnicka at No_s_Pam.quick.cz>
  • Date: Mon, 13 Sep 2004 02:19:28 -0400 (EDT)
  • Organization: Czech Technical University
  • References: <chmova$a4j$1@smc.vnet.net> <chroam$50c$1@smc.vnet.net> <chulqn$jm7$1@smc.vnet.net>
  • Sender: owner-wri-mathgroup at wolfram.com

Robert,
if your problem has extremely time consuming objective function (like FEM)
you must focus on complete different methods. The standard differential
evolution (via NMinimize) is not suitable for this kind of optimization
problems.

A very powerful method is SNOBFIT, for example. For more details see:
http://www.mat.univie.ac.at/~neum/software/snobfit/

In this time the SNOBFIT is not implemented in Mathematica, but I think that
this task is not very difficult. On the other hand, way not to make
Mathematica package for global optimization of the extremely time consuming
objective functions. Think about it...!!!

Regards,
Michal

>
>
> hi mark, hi michal
>
> thank you for your contributions,
> mark, yes this could be a way, ....
>
> but first of all i dont own a parallel grid version of Mathematica.
> second its not the NMinimize aktion thats time consuming its the
> evaluation of the objective function which is extreme time consuming
> consisting on FEM calculations by an linux based FEM package. so the
> natural way would be try to calculate the function values parallel on
> a linux cluster (which i have access to).
>
> by the time i got some hints from dan lichtblau perhaps its posssible
> for me to modify the NMinimize behavier.
>
> regards robert
>




  • Prev by Date: Some printing questions
  • Next by Date: Re: DSolve Question
  • Previous by thread: Re: parallel NMinimize[]
  • Next by thread: Exact real numbers