A scalable molecular-dynamics algorithm suite for materials simulations: design-space diagram on 1024 Cray T3E processors

Abstract

A suite of scalable molecular-dynamics (MD) algorithms has been developed for materials simulations. The linear scaling MD algorithms encompass a wide spectrum of physical reality: (i) classical MD based on a many-body interatomic potential model; (ii) environment-dependent, variable-charge MD; (iii) quantum mechanical MD based on the tight-binding method; and (iv) self-consistent quantum MD based on the density functional theory. Benchmark tests on 1024 Cray T3E processors including 1.02-billion-atom many-body and 22 500-atom density functional MD simulations demonstrate that these algorithms are highly scalable. A design-space diagram spanning seven decades of system size and computational time is constructed for materials scientists to design an optimal MD simulation incorporating maximal physical realism within a given computational budget.