Feature activated molecular dynamics: An efficient approach for atomistic simulation of solid-state aggregation phenomena

Abstract

A new approach is presented for performing efficient molecular dynamics simulations of solute aggregation in crystalline solids. The method dynamically divides the total simulation space into "active" regions centered about each minority species, in which regular molecular dynamics is performed. The number, size and shape of these regions is updated periodically based on the distribution of solute atoms within the overall simulation cell. The remainder of the system is essentially static except for periodic rescaling of the entire simulation cell in order to balance the pressure between the isolated molecular dynamics regions. The method is shown to be accurate and robust for the Environment-Dependant Interatomic Potential (EDIP) for silicon and an Embedded Atom Method (EAM) potential for copper. Several tests are performed beginning with the diffusion of a single vacancy all the way to large-scale simulations of vacancy clustering. In both material systems, the predicted evolutions agree closely with the results of standard molecular dynamics simulations.Computationally, the method is demonstrated to scale almost linearly with the concentration of solute atoms, but is essentially independent of the total system size. This scaling behavior allows for the full dynamical simulation of aggregation under conditions that are more experimentally realizable than would be possible with standard molecular dynamics. ABSTRACTA new approach is presented for performing efficient molecular dynamics simulations of solute aggregation in crystalline solids. The method dynamically divides the total simulation space into "active" regions centered about each minority species, in which regular molecular dynamics is performed. The number, size and shape of these regions is updated periodically based on the distribution of solute atoms within the overall simulation cell. The remainder of the system is essentially static except for periodic rescaling of the entire simulation cell in order to balance the pressure between the isolated molecular dynamics regions. The method is shown to be accurate and robust for the Environment-Dependant Interatomic Potential (EDIP) for silicon and an Embedded Atom Method (EAM) potential for copper. Several tests are performed beginning with the diffusion of a single vacancy all the way to large-scale simulations of vacancy clustering. In both material systems, the predicted evolutions agree closely with the results of standard molecular dynamics simulations. Computationally, the method is demonstrated to scale almost linearly with the concentration of solute atoms, but is essentially independent of the total system size. This scaling behavior allows for the full dynamical simulation of aggregation under conditions that are more experimentally realizable than would be possible with standard molecular dynamics.

Feature Activated Molecular Dynamics: Parallelization and Application to Systems with Globally Varying Mechanical Fields

“…Future work will address approaches for computing the static lattice positions on the fly, which is a non-trivial task for static lattices that contain a (varying) strain field. Next, the positions of all feature atoms are communicated to all the processors and feature clusters are created based on the Stillinger criterion [20] with an interaction distance determined by the range of the interatomic potential [1,17]. These clusters do not represent physical clusters but rather are groupings that are used to construct independent (isolated) MD regions.…”

“…Next, thermal activities, T A , are assigned to the atoms in MD regions based on their location relative to feature atoms [1]. The thermal activity of an atom is simply its temperature scaled by the simulation temperature, so that 0 ≤ T A ≤ 1.…”

“…Recently, a molecular dynamics-based method was presented for performing efficient atomistic simulations of certain types of solid-state aggregation phenomena [1]. The approach, called Feature-Activated Molecular Dynamics, or FAMD, was shown to extend the length scales and process conditions accessible with molecular dynamics simulations of important phenomena such as point defect and impurity atom clustering in crystalline bulk materials such as silicon and copper.…”

“…Maintaining a simple boundary between the static and active regions in FAMD allows for relatively easy update of the active region geometries as well as for seamless region coalescence and fragmentation as demonstrated in Refs. [1,17].…”

“…The central concept of FAMD is a multiresolution approach, which retains the basic atomic representation throughout the simulation space [1]. The total space is divided up into "active" and "static" regions.…”