Dresden 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 30: Poster Session II
MM 30.35: Poster
Dienstag, 17. März 2020, 18:15–20:00, P4
High-dimensional neural network potential for laser-excited materials — •Pascal Plettenberg, Bernd Bauerhenne, and Martin E. Garcia — Theoretical Physics, University of Kassel, Heinrich-Plett-Straße 40, 34132 Kassel, Germany
In recent years, machine learning techniques have been increasingly used to develop highly accurate representations of ground state or single excited state potential energy surfaces (PES) for large systems. Here, we want to develop such a machine learning potential to electronic temperature (Te) dependent PES of solids excited by intense femtosecond laser pulses. Studying the effects of ultrashort excitation on the long-term (nanoseconds) and large-scale (hundreds of nanometers) dynamics of a material has not been possible so far, because ab-initio simulations are limited to a small number of atoms and short time scales. On the other side, Te-dependent interatomic potentials, which can be efficiently used in large-scale simulations, are difficult to derive and often do not reach the required accuracy. In this work we investigate the possibility to fill this gap of simultaneous efficiency and ab-initio accuracy with a high-dimensional neural network potential including an electronic temperature dependency, which is implemented as an additional input node. We train the network to reference Density-Functional-Theory data of thin-film silicon and demonstrate its performance in molecular dynamics simulations.