Parallel embedded-atom method simulations with delayed electron density approximations

Abstract The embedded-atom method (EAM) is a popular technique for the atomic simulation of metals and alloys. The EAM procedure involves two computational phases; the first to evaluate electron densities and the second to evaluate embedding energies and repulsive forces. Substantial computational costs are required for each phase, particularly for the simulation of large particle systems. On distributed-memory architectures each phase also requires communication overhead, reducing parallel efficiency. We introduce a pseudo-EAM (PEAM) technique to improve the performance for particle simulations of metals. The key PEAM procedure is the approximation of electron densities from the previous timestep, allowing all computations to be performed in a single phase. We demonstrate the efficiency of the PEAM procedure and show that it produces identical behavior to EAM systems. On both serial and parallel architectures, PEAM simulations are nearly twice as fast as EAM simulations for the same atomic system.