Sequential multiscale simulations with potfit: First principles data for classical molecular dynamics
Peter Brommer, University of Warwick, Coventry, UK
Large-scale molecular dynamics (MD) simulations are only feasible with classical interaction potentials or force fields, where all electronic degrees of freedom are summarised in an effective interaction between nuclei. This now only depends on the positions of the atoms and can be evaluated rapidly and efficiently. Unfortunately, determining the parameters of such an effective potential is far from trivial. Traditionally, they were chosen to best reproduce a number of experimentally determined quantities of interest, like elastic constants. However, there is a disconcerting lack of usable information, particularly for more complex materials, where not even the atomic positions are sufficiently well known. Force matching, developed by Ercolessi and Adams about 20 years ago [1], circumvents this by directly using forces on individual atoms determined with first-principles methods as reference data, implying that if a potential gets the forces correct, it will reproduce the atomic dynamics correctly and thus all quantities that are derived from that. The open-source tool potfit is an implementation of the force matching method [2]. Originally developed to determine interaction potentials for quasicrystals and other structurally complex metallic alloys, it has since been extended to cover various potential models for metals and ionic solids.
The program is available from https://www.potfit.net/. In this talk, I'll discuss the main features of the program as well as recent advances in using <em>potfit</em> for electron-temperature dependent potentials.
[1] Ercolessi and Adams, Europhys. Lett., 1994, 26, 583.
[2] Brommer and Gähler, Modell. Simul. Mater. Sci. Eng., 2007, 15, 295-304.