Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 28: Liquid and Amorphous Metals
MM 28.8: Vortrag
Mittwoch, 29. März 2023, 12:15–12:30, SCH A 118
Properties of Sodium Borosilicate Glasses via Dynamics Simulations with Ultra-Fast Machine-Learning Potentials — •Hendrik Krass1, Benedikt Ziebarth2, Wolfgang Mannstadt2, and Matthias Rupp3 — 1University of Konstanz, Konstanz, Germany — 2Schott AG, Mainz, Germany — 3Luxembourg Institute of Science and Technology (LIST), Belvaux, Luxembourg
Glasses are of scientific interest and have many industrial applications. However, their investigation and development are limited by the duration and costs of experiments. The computational study of glasses can in principle overcome these limits, but current atomistic glass models are either fast but not accurate enough (classical empirical potentials) or accurate but too slow (ab initio potentials). Machine-learning potentials (MLPs) trained on ab initio reference calculations promise to be both fast and accurate enough.
We investigate the suitability of “ultra-fast potentials” (UFPs) [1]—a class of MLPs that are data-efficient, physically interpretable, sufficiently accurate for applications, can be parametrized automatically, and are as fast as the fastest traditional empirical potentials—to study glasses. For this, we compute structure and properties of interest via dynamics simulations with UFPs and compare them against state-of-the-art models and experimental values for sodium borosilicate glasses, a prototypical glass system.
[1] Stephen R. Xie, Matthias Rupp, Richard G. Hennig: Ultra-Fast Interpretable Machine-Learning Potentials, arXiv 2110.00624, 2021.