Regensburg 2022 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 33: Computational Materials Modelling: Process Schemes / Oxides

MM 33.2: Vortrag

Donnerstag, 8. September 2022, 16:00–16:15, H44

A machine-learned interatomic potential for crystalline and amorphous silica — •Linus Erhard¹, Jochen Rohrer¹, Karsten Albe¹, and Volker Deringer² — ¹Institute of Materials Science, Technische Universität Darmstadt, Otto-Berndt-Strasse 3, 64287 Darmstadt, Germany — ²Department of Chemistry, Inorganic Chemistry Laboratory, University of Oxford, Oxford OX1 3QR, United Kingdom

Fitting an interatomic potential for silicon oxide that can be used for both the amorphous and numerous crystalline phases has proven to be difficult. This is already shown by the large number of interatomic potentials published in the last decades. Here, we present a machine-learned interatomic potential for silica, which is highly transferable between different crystalline polymorphs and the amorphous phase. It predicts the thermodynamics of the system accurately and is able to generate low-defect amorphous models by melt and quench simulations. We also discuss the importance of choosing an appropriated exchange-correlation functional for density-functional data input, which is particularly important for silica. Since the generation of realistic amorphous structure models by melt-quench simulations is highly dependent on the quench rate, we show new ways via hybrid simulations that combine the speed of classical interatomic potentials with the accuracy of machine-learning potentials. We also investigate the extrapolation behavior of our machine-learning potential using high-pressure simulations. Finally, we show first steps towards an interatomic potential for mixed Si-SiO2 systems.