FitSNAP: Atomistic machine learning with LAMMPS
Andrew Rohskopf, C. Sievers, Nicholas Lubbers, Mary Alice Cusentino, James Goff, Jan Janßen, Megan McCarthy, David Montes de Oca Zapiain, Svetoslav Nikolov, Khachik Sargsyan, Dionysios Sema, Ember Sikorski, Logan Williams, Aidan P. Thompson, Mitchell Wood
Abstract
Chemical and physical properties of complex materials emerge from the collective motions of the constituent atoms.These motions are in turn determined by a variety of interatomic interactions mediated by the local redistribution of valence electrons about the fixed core electrons and nuclear charges.Scientific and engineering advances in materials science, chemistry, and many related fields benefit from our ability to directly sample the equilibrium and kinetic probability distributions of large collections of atoms and molecules.Classical molecular dynamics (MD) is a widely used simulation method in which Newton's equations of motion are numerically integrated forward in time to generate representative atomic trajectories of the atoms, from which insight into a wide range of material behaviors can be obtained.This simulation technique is not restricted to the definition of particles as atoms, but is generalizable to other types of interacting particles, such as coarse-grained beads in polymers, discrete elements representing granular materials, and fluid mass elements in dissipative particle dynamics.While all of these simulation methods use the same core algorithms as MD, we restrict our discussion here to the treatment of interactions between atoms.Rohskopf et al. ( 2023).