On-the-fly machine learning force field generation: Application to melting points

Abstract

An on-the-fly force field generation method is developed and applied to liquid-solid phase transitions. The method allows the machine to automatically self-learn interatomic potentials during molecular dynamics simulations and to generate force fields with the distinctive chemical precision of first-principles methods. Applications show that more than 99% of the expensive first-principles calculations are bypassed, and molecular dynamics simulations are accelerated by more than two orders of magnitude already during learning, with many more orders during production runs.