A general strategy to create machine learning (ML) based atomistic force fields is outlined. In contrast to conventional force fields, ML force fields can be systematically improved by adding progressively new training data (representing new atomic environments). Our ML scheme involves (1) preparing a big reference dataset of atomic environments and forces with sufficiently low noise, e.g., using density functional theory or higher-level methods, (2) utilizing a generalizable class of structural fingerprints for representing atomic environments, (3) optimally selecting diverse and non-redundant training datasets from the reference data, and (4) suitable learning approaches to predict atomic forces directly (and rapidly) from atomic configurations. From the atomistic forces, accurate potential energies can then be obtained by appropriate integration along a reaction coordinate or along a molecular dynamics trajectory. The concept is demonstrated for the case of Al, including extensive tests and molecular dynamics studies of diverse phenomena.

Machine learning based atomistic force fields

255th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.