Newsgroups: comp.parallel.pvm,comp.parallel.mpi
From: ez043844@rocky.ucdavis.edu (Carey Huscroft)
Subject: Re: Montecarlo Simulations
Organization: University of California, Davis
Date: 18 Nov 1995 16:51:28 GMT
Message-ID: <48l2ug$10v@mark.ucdavis.edu>

Aaron Jackson (shadow@apricot.com) wrote:
: Has anyone seen, and/or heard of a package which performs
: Montecarlo simulations in parallel.

: Send replies to ebell@egs.com.

What type of Monte Carlo simulations are you trying to do?  The field is
pretty broad.  Are you trying to calculate a thermal expectation value
for some classical or quantum physical system?  Are you doing an integral?
Something else?

It is pretty straightforward to parallelize the Metropolis Algorithm, at
least if you want to average over random number sequences.  (I discuss
this briefly on my web page http://onsager.ucdavis.edu/~huscroft -- I
do simulations of quantum physical systems, so my page is focused in this
direction.)  Further parallelizations are harder to achieve.  I.e., if
you want to parallelize the calculations that you use to find the
metropolis weights, that is another kettle of fish.

Some of Jim Demmel's students have a discussion of a parallel monte carlo
at: http://dnclab.berkeley.edu/codef/may/prj.html

I guess you can tell that I'm biased toward writing your own code!

Good luck.

Carey
huscroft@solid.ucdavis.edu


