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

MM: Fachverband Metall- und Materialphysik

MM 17: Development of Calculation Methods

MM 17.4: Vortrag

Mittwoch, 19. März 2025, 11:00–11:15, H22

DECAF: An Open Source Local Atomic Environment Classifier — •King Chun Lai, Sebastian Matera, Christoph Scheurer, and Karsten Reuter — Fritz-Haber-Institut der MPG, Berlin

Classification of local atomic environments (LAEs) is an inevitable task in most atomic-scale modeling and simulation. The reason is trivial, atoms’ characteristics are predominantly determined by neighbors within a limited radius. The task itself, however, is abstract and error-prone due to the diversity of structures and the often ambiguous relationship between geometry and atomic behaviors. To address these issues, we have developed the open-source package DECAF during the last years [1]. DECAF automatically identifies equivalence groups within atomic structure datasets on the basis of the LAEs. We showcase the usage of the DECAF package on a set of nanotructures. We explain the theoretical background as well as the influence of different options to control the outcome. A particular feature is DECAF’s ability for out-of-sample classification, identifying LAEs that differ from any groups in the training set. We provide examples where this has been exploited such as automatic process exploration [2] or active learning.

[1] Lai et al., J. Chem. Phys 159, 024129 (2023). DOI: 10.1063/5.0160369 Available: https://gitlab.mpcdf.mpg.de/klai/decaf

[2] Lai et al., ChemRxiv (2024). DOI: 10.26434/chemrxiv-2024-jbzr7

Keywords: Surface Science; Clustering; Nanomaterials; Classification