Optimizing the performance of engineering materials requires an ability to control their microstructure and predict their fundamental properties on engineering length and time scales. To this end, ever-increasing computational power and the rise of artificial intelligence have given rise to atomistic simulation capabilities for virtual experimentation with unprecedented accuracy, based on machine learned interatomic potentials (MLPs).

MLPs are uniquely enabling for high fidelity mesoscale material modeling by virtue of their basis in arbitrarily large training sets of ab initioresults. Machine-learned potentials (MLPs) thus provide a unique means to extend the proven fidelity, reliability, and maturity of ab initio methods to the mesoscale where users may study the collective behavior of millions or more of atoms and sample millions of atomic configurations.