Newsgroups: sci.physics.computational.fluid-dynamics,comp.parallel
From: abani@cs.utexas.edu (Abani Kumar Patra)
Subject: Re: How to partition FEMs on parallel computers
Organization: CS Dept, University of Texas at Austin
Date: 21 Dec 1995 04:15:53 GMT
Message-ID: <4ak7b5$l4n@pageboy.cs.utexas.edu>

Current practice appears to be mostly recursive bisection of either
1. coordinates
2. spectral (i.e. 2nd eigen vector of Laplacian of connectivity graph)

Performance of either seems to taper off if the grid is adaptive and or
has bad geometry. I recently wrote a paper reviewing both the above
methods and introducing a simpler algorithm based on bisection of  a 
space filling curve passing through the element centroids. The paper
appeared in Computing Systems in Engg. vol. 6, no. 2, pp 97-109. 
Citations to papers on the other algorithms are in this paper.

abani


