SKM 2023 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 12: Poster I
MM 12.10: Poster
Monday, March 27, 2023, 18:15–20:00, P2/OG1+2
Machine learning force fields in VASP — •Ferenc Karsai1, Andreas Singraber1, Jonathan Lahnsteiner1, Ryosuke Jinnouchi2, and Georg Kresse2 — 1VASP Software GmbH, Sensengasse 8, Vienna, Austria — 2University of Vienna, Department of Physics, Kolingasse 14-16, Vienna, Austria
An efficient and robust method for on-the-fly machine learning force fields implemented into the Vienna Ab-initio Simulation Package (VASP) is presented. This method realizes the automatic generation of machine learning force fields on the basis of Bayesian inference during molecular-dynamics simulations, where the first-principles calculations are only executed when new configurations out of already sampled data sets appear. The power of the method is demonstrated in several applications such as e.g. melting points of ionic and covalent compounds, thermal transport in Zirconia, delta-learning of Carbon monoxide adsorbed on transition-metal surfaces and solid-solid phase transitions in perovskites. The applications show that during learning 99% of the ab-initio calculations are skipped. The implementation of our on-the-fly learning scheme is fully automatized and is mainly controlled by a few parameters. Hence one can optimally sample through a large phase space and the amount of human intervention for the usually laborious task of training is drastically reduced. Finally, the calculations are accelerated by more than 4 orders of magnitude compared to ab initio, while the accuracy remains the same.