Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 92: Electronic Structure Theory
O 92.6: Vortrag
Donnerstag, 20. März 2025, 16:15–16:30, H25
Enhancing structure relaxation with machine-learned interatomic potentials — Sudarshan Vijay1,2, Martijn Marsman1, Georg Kresse1,3, and •Martin Schlipf1 — 1VASP Software GmbH, Berggasse 21, 1090 Vienna, Austria — 2Department of Chemical Engineering, Indian Institute of Technology Bombay, Powai, Mumbai, Maharashtra 400076 India — 3Faculty of Physics and Center for Computational Materials Science, University of Vienna, Kolingasse 14-16, A-1090 Vienna, Austria
Machine learning interatomic potentials (MLIPs) offer a cost-effective way to predict ground-state properties of materials compared to density functional theory (DFT) calculations. MLIPs are often used to replace DFT due to their ability to access large length and longer time scales. In this study, we explore whether MLIPs can also enhance DFT by preconditioning commonly used optimization algorithms for structure relaxation. We demonstrate that applying this preconditioner to methods such as conjugate gradient (CG) and Broyden-Fletcher-Goldfarb-Shanno (BFGS) results in faster convergence but at the expense of stability. For this particular tasks, state-of-the-art MLIPs are not accurate enough to replace DFT; prerelaxing the structures with MLIPs tends to perform worse compared to starting the relaxation directly within DFT. Our analysis of the convergence behavior for both standard and preconditioned methods suggests that advancements in line-search techniques could enhance the effectiveness of this preconditioning approach for structure relaxation.
Keywords: Density functional theory (DFT); Vienna Ab initio Simulation Package (VASP); Machine Learning; Structure Relaxation