Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 10: Topical Session: Hydrogen in Materials: from Storage to Embrittlement I
MM 10.6: Vortrag
Montag, 18. März 2024, 17:15–17:30, C 130
Accurate representation of hydrogen in metals by machine-learning enhanced modelling of nuclear quantum effects — •Kai Sellschopp1, Thomas Klassen1,2, Paul Jerabek1, and Claudio Pistidda1 — 1Institute of Hydrogen Technology, Helmholtz-Zentrum hereon, Geesthacht, Germany — 2Helmut-Schmidt-University, Hamburg, Germany
In a sustainable economy built on renewable energy, hydrogen plays a key role for storing energy and replacing fossil fuels. An efficient way to store hydrogen is to keep it in the solid state by binding it chemically in a metal hydride, which is particularly useful for seasonal energy storage or for applications where safety is a concern. Despite the fact that hydrogen is known to show nuclear quantum effects (NQE) even at higher temperatures, these have been neglected in computational studies of metal hydrides so far due to the high cost of path-integral molecular dynamics (PIMD) calculations. In this work, a machine-learned potential (MLP) is trained for the Mg-H system, a well-known hydrogen storage material, in order to speed up the simulations and bring down the cost. At the same time, the sample collection is accelerated by training the potential "on-the-fly" during classical molecular dynamics runs, where ab-initio calculations are replaced by the MLP whenever the estimated errors are low enough. In this contribution, I present the training and validation of this MLP and evaluate the speed-up that allows to perform computationally expensive PIMD calculations. First insights obtained from these calculations enhance our understanding of metal-hydrogen systems.
Keywords: hydrogen in metals; ab-initio calculations; machine learning; path-integral molecular dynamics; nuclear quantum effects