Parallel Molecular Dynamics with the Embedded Atom Method
- PDF / 450,096 Bytes
- 6 Pages / 420.48 x 639 pts Page_size
- 80 Downloads / 265 Views
PARALLEL MOLECULAR DYNAMICS WITH THE EMBEDDED ATOM METHOD STEVEN J. PLIMPTON and BRUCE A. HENDRICKSON Sandia National Laboratories, Albuquerque, NM 87185 ABSTRACT Parallel computing offers new capabilities for using molecular dynamics (MD) to simulate larger numbers of atoms and longer time scales. In this paper we discuss two methods we have used to implement the embedded atom method (EAM) formalism for molecular dynamics on multiple-instruction/multiple-data (MIMD) parallel computers. The first method (atomdecomposition) is simple and suitable for small numbers of atoms. The second method (forcedecomposition) is new and is particularly appropriate for the EAM because all the computations are between pairs of atoms. Both methods have the advantage of not requiring any geometric information about the physical domain being simulated. We present timing results for the two parallel methods on a benchmark EAM problem and briefly indicate how the methods can be used in other kinds of materials MD simulations. INTRODUCTION Molecular dynamics (MD) simulation is a commonly used tool in the solid-state physics and materials community for modeling solids and liquids at the atomic level. Each atom in the simulation is treated as a point mass and Newton's equations of motion are integrated to track the motion of each atom. All the physics of the model is contained in an energy functional for the system from which the forces on each atom can be computed. Thermodynamic and transport properties as well as general structural and high-temperature properties of materials can be computed from time averaging various quantities over the ensemble of atoms. Although the length scale (A) and time scale (femtosecond timesteps) of MD simulations make them computationally intensive, they have the desirable property that for models of most materials, the computational work scales as N, where N is the number of atoms simulated. This is due to charge-screening effects which limit the range of Coulombic forces meaning each atom interacts only with a small (roughly constant) number of surrounding atoms. Thus in practice, quite large simulations of tens of thousands of atoms can be performed. For fcc and other close-packed metals, the embedded atom method (EAM) [ 1,2] is a popular choice for the energy functional in MD simulations. It overcomes the volumedependent limitation of pair-potentials by adding a term for the energy to embed an atom in the background electron density of its neighbors. The EAM has proven particularly good at modeling bulk and defect properties (energy, structure) of metals and metal alloys. For example, many MD and Monte Carlo simulations of surfaces and grain boundaries have been performed using the EAM to model such phenomena as crack growth [1,3], surface reconstruction [4], and grain boundary structure [5] and diffusivity [6]. Computationally the EAM is attractive because it is short-range and, as discussed in the next section, can be computed by summing over pairwise interactions. In this paper we describe two methods
Data Loading...
