Low-energy electron diffraction calculations using a parallel supercomputer

Abstract A first implementation of a dynamical low-energy electron diffraction theory on a parallel supercomputer is described. The computational performance of the Duke-Laramore-Beeby LEED formulation is demonstrated using an Intel iPSC TM /860 parallel computer. The algorithm developed is highly parallelizable and scales nearly ideally with the number of nodes used. A performance equivalent to that experienced using a Cray YMP was attained with as few as 16 nodes. However, it was not possible to exceed significantly the YMP performance due to the increasing inter-node communication overhead. A simple mathematical model is described to explain this limitation, which is encountered in all parallel computing applications and hardware designs. The model indicates that an alternate parallelization scheme is required to attain greater performance.