next up previous contents
Next: Self-labeling Up: The Swendsen-Wang Cluster Algorithm Previous: Hoshen and Kopelman   Contents

Parallel Cluster Algorithm

A parallel cluster algorithm involves distributing the lattice onto an array of processors using the usual domain decomposition. Clearly a sequential algorithm can be used to label the clusters on each processor, but we also need a procedure for converting these labels to their correct global values. We need to be able to tell many processors, which may be any distance apart, that some of their clusters are actually the same. Thus we need to be able to agree on which of many different local labels for a given cluster should be assigned to be the global cluster label, and to pass this label to all the processors containing a part of that cluster. We will discuss one of the methods called self-labeling [#!Coddington91!#,#!Tamayo!#] for tackling this problem.



Subsections

Bryan Carpenter 2004-06-09