Parallelization of a cluster algorithm

Abstract

We present two algorithms to parallelize the identification of clusters on a lattice. Such algorithms are necessary for a variety of problems such as percolation and non-local spin update algorithms. The algorithms were tested for the Swendsen-Wang method for the simulation of the Ising model. The tests were run on a multi-transputer system using up to 128 processors. A scaling law for the performance of geometric parallel algorithms is proposed and tested. © 1989.