Regensburg 2025 – wissenschaftliches Programm

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

MM: Fachverband Metall- und Materialphysik

MM 9: Poster

MM 9.31: Poster

Montag, 17. März 2025, 18:30–20:30, P1

Phases of AlN by machine learning potentials — •Simon Liebing, Oliver Heymer, and Jens Kortus — Institute of Theoretical Physics, TU Bergakademie Freiberg, Germany

AlN is an important wide-band gap semiconductor with e.g. applications in high power electronics. Under high pressure (about 13 GPa) the wurtzite phase transforms to the rocksalt phase. Here, we attempt to simulate this phase transition as function of temperature and pressure by means of machine learned interatomic potentials trained on accurate density functional theory molecular dynamics data. In particular we will utilize use the open-source library FitSNAP [1] for atomistic machine learning in combination with the molecular dynamics code LAMMPS [2]. FitSNAP is used to provide fast interaction potentials with accuracy inherited from DFT. It already provides interfaces for popular open-source codes such as Quantum ESPRESSO[3], PyTorch and LAMMPS and it supports the state-of-the-art atomic cluster expansion (ACE) descriptors [4]. The ACE descriptors transform structural information into machine learning models. This enables us to carry out large-scale classical MD systems of AlN with thou- sands of atoms with DFT accuracy. The results will be compared to earlier works based on small unit cells using density functional theory. [5] References [1] Rohskopf et al., Journal of Open Source Software, 8 (84), 5118 (2023). [2] A. P. Thompson et al. Comp Phys Comm, 271 10817 (2022). [3] P. Giannozzi et al. J. Chem. Phys. 152, 15 (2020). [4] Drautz, R. Physical Review B, 99 (1), 014104 (2019). [5] S.Schmerler and J. Kortus Physical Review B, 89, 6, (2014).

Keywords: DFT; phase transition; machine learning; interatomic potentials