Large-Scale Atomistic Simulations of Materials using SNAP Potentials*

Molecular dynamics (MD) is a powerful materials simulation method whose accuracy is limited by the interatomic potential (IAP). SNAP is an automated quantum data-driven approach to IAP generation that balances accuracy and computational cost. The energy is formulated in terms of a very general set of geometric invariants that characterize the local neighborhood of each atom. The SNAP approach has been used to develop potentials for studying plasticity in tantalum, intrinsic defects in indium phosphide, and plasma surface interactions in tungsten and beryllium. In each case, large quantum-mechanical data sets of energy, force, and stress are accurately reproduced and cross-validation on additional test data is performed. The resultant potentials enable high-fidelity MD simulations with thousands to millions of atoms. The relatively large computational cost of SNAP is offset by the LAMMPS implementation, enabling the efficient use of large CPU and GPU clusters.