Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!udel!princeton!handel!jtoth
From: jtoth@handel.Princeton.EDU (Gabor J.Toth)
Subject: Re: Physics and GAs ?
Message-ID: <1994Nov24.001206.10559@Princeton.EDU>
Originator: news@hedgehog.Princeton.EDU
Keywords: physics solitons genetic GA
Sender: news@Princeton.EDU (USENET News System)
Nntp-Posting-Host: handel.princeton.edu
Organization: Princeton University
References: <3b097d$lt@btr0x1.hrz.uni-bayreuth.de>
Date: Thu, 24 Nov 1994 00:12:06 GMT
Lines: 29

In article <3b097d$lt@btr0x1.hrz.uni-bayreuth.de>,
Roland Huss <btp313@btp4x9.phy.uni-bayreuth.de> wrote:

> Especially I am  interested in the  application of GA's in the realm
> of  physics, since I  was    thrown in a  university  education   as
> physicist  by some  strange destiny.  So  my  first (your  fears are
> right, there will be several) question is, whether there are already
> some applications in the field of physics.

Don't try to find a field GA's have not been applied to...

> [...]
> So one   evaluation of the  fitness function  takes   ca.  50 -- 250
> minutes.   It's a  lot, I   know.  So my  question is,   could it be
> possible to tackle this problem ?  What do you think, how large must
> be a reasonable population,  and how many  generations it would take
> for the algorithm to converge  ?  (it would  be nice, if I could get
> some results in _this_ life ;-)

I appreciate the  strength of GA, but it  is absolutely not  the right
method for  your  problem, unless  you'll have  an exceptionally  long
life-span.  Seriously, GA's  need  far too many function  evaluations.
I'd suggest you try a  greedy algorithm.  I know,  they tend to end up
in local minima,  but they  at least  do  it quickly. If that  doesn't
work, try  simulated annealing, there you  have a fairly direct way of
setting the    greediness of your  algorithm.   If   that doesn't work
either, than you may try either GA or rather another problem.

Gabor
