Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 19: Poster Session
MM 19.3: Poster
Dienstag, 10. März 2026, 18:00–20:00, P5
Experimentally supported machine-learning interaction potential for Pd-Si — •Przemyslaw Dziegielewski1, Jerzy Antonowicz1, Zuzanna Kostera1, Oleksii Liubchenko2, and Ryszard Sobierajski2 — 1Faculty of Physics, Warsaw University of Technology, Warsaw, Poland — 2Institute of Physics Polish Academy of Sciences, Warsaw, Poland
Classical Molecular Dynamics (MD) simulations were utilised to reproduce the X-ray Free-Electron Laser (XFEL) experiment. Using the pump-probe technique, a Pd-Si alloy with various Si contents was melted and subsequently solidified with a high cooling rate. The EAM (Embedded Atom Method) potentials available for Classical MD allowed us to obtain simulation results showing high consistency with the experiment for pure Pd and the good glass-forming Pd83Si17 alloy.
However, at low Si contents, they become unreliable, generating a structure that is a combination of the hcp and fcc phases, whereas the experimental data clearly indicate the occurrence of the fcc phase only. The two phases differ only slightly in energy, and the occurrence of the hcp phase for alloys with nearly 100% Pd content takes place, for example, under high-pressure conditions. In our presentation, we introduce an alternative approach to simulating metallic alloys using a Machine Learning (ML)-generated potential within the VASP code. Our proposed procedure may constitute a universal approach for the effective and rapid creation of interaction potentials for the purpose of analysing experimental data.
Keywords: Pd-Si alloy; molecular dynamics; machine learning; solidification; free electron lasers