postmaster@ames.arc.nasa.gov (Bernard Aboba) (08/22/88)
My field is Petroleum Reservoir Simulation, and I've become interested lately in approaching things from a completely new angle. Most simulators are currently finite difference based, with second generation codes being vectorized. The problem with this is that this approach cannot benefit from massive parallelization, in addition to the fact that it cannot take into account the statistical variation of geologic media very well; it can be shown that the classical diffusion equation approach cannot be matched to field data without allowing the diffusion coefficient to be a function of time, for example. The monte carlo approach has the potential for better dealing with the physics AND the ability to be amenable to massive parallelization. [ Monte Carlo is heavily used by the physicists already. Contact folks at Argonne and OakRidge. For OakRidge, try Chris Bottcher oakridge cby%ornlstc@relay.cs.net or Mike Strayer mys%ornlstc@relay.cs.net. They're working on this directly. Try John Kogut at Univ of Ill (don't have id) as well. ---- Steve ]
lkaplan@BBN.COM (Larry Kaplan) (08/23/88)
Some work on parallel Monte-Carlo simulations has been done at the NYU Ultracomputer project. The issue of good random number generation is one of the topics that has been addressed. Contact Mal Kalos kalos@nyu.edu for more information on the state of NYU's monte-carlo work. ________________________________________________________________________________ Laurence S. Kaplan |-\|/-| BBN Advanced Computers lkaplan@bbn.com \ * / 10 Fawcett St. (617) 873-2431 |/ \| Cambridge, MA 02238
wagner@june.cs.washington.edu (Bullwinkle J. Moose) (08/29/88)
There is a paper on exactly this topic in the '88 SIGMETRICS proceedings.
It is by Boris Lubachevsky of AT&T Bell labs. He has an implementation
running on a connection machine.
Dave Wagner
University of Washington Comp Sci Department
wagner@cs.washington.edu
{ihnp4,harvard,tektronix}!uw-beaver!wagner