Newsgroups: comp.parallel
From: edemaine@afton.cs.dal.ca (Erik Demaine)
Subject: Parallel Iterative Solvers
Organization: Math, Stats & CS, Dalhousie University, Halifax, NS, Canada
Date: 29 May 1995 05:59:47 GMT
Message-ID: <3qbnsj$kva@usenet.srv.cis.pitt.edu>

Hello again,

This parallels a posting requiring parallel Cholesky solver for a CFD
application.  We are using 2-D or 3-D finite element viscoplastic
incompressible fluid codes written for primitive U,P variable and applying
them to the understanding of continent-continent or continent-ocean collisions
(tectonically active margins).

We have tried both direct and iterative linear solvers to solve this type
of system (illconditioned in several respects: 1. incompressibility
(penalization) 2. plasticity (penalization)).  The iterative solvers we have
tried include: dual schur method, standard CG preconditioned by
element-by-element inversion, CG with projection on incompressibility
constraint, and Uzawa.  All of them have been run (for now) on an HP 735/125
workstation (i.e. in serial), and they do not seem to compete easily with
the direct solvers.  We would like to have an idea of which preconditioner
deals best with the pressure problem, for Stokes flows, if such a general
assessment is possible, regardless of the details of the implementation.

We are going to do parallel tests on a four-node SP2.  We can afford a few
days of computation per model.  We are currently running 2-D models with
typical grid size of 40x300.  Here is the stupid question:  can we run
200x400 grid size with a better numerical method?

Any hint or answer would be very much appreciated.

Again, a summary will be posted if we get enough responses.

Thanks in advance,
Erik
--
Erik Demaine        || edemaine@ug.cs.dal.ca  || edemaine@fx2800.dal.ca
edemaine@cs.dal.ca  || 01ERIK@ac.dal.ca       || edemaine@is.dal.ca
URL: http://ug.cs.dal.ca:3400/~edemaine/edemaine.html
--> A glob is a round substance of thick liquid. A blob is an unformed object.

