Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 27: Data Driven Material Science: Big Data and Workflows IV
MM 27.2: Vortrag
Dienstag, 19. März 2024, 13:45–14:00, C 243
Machine learning modelling of local magnetic moments in Fe-based intermetallic compounds — •Bo Zhao1, Kun Hu2, and Hongbin Zhang1 — 1Institute of Materials Science, Technische Universität Darmstadt, Peter-Grünberg-Str. 2, 64287 Darmstadt, Germany — 2School of Materials Science and Engineering, Central South University, Changsha, 410083, Hunan, PR China
It is well known that the magnitude of magnetic moments is determined by the local crystalline environments as verified by studying the dimensional crossover behavior of a few representative systems such as Fe. To further generalise and quantify the mapping between the crystal environments and magnetic moments, we carried out machine learning modelling on Fe-based intermetallic compounds, with the corresponding magnitude of magnetic moments for Fe atoms varying between zero and four Bohr magneton (the atomic limit). Using the symmetry-adapted smooth overlap of atomic positions (SOAPs) as descriptors, it is observed that 2374 data for binary and ternary compounds collected from Materials Project are not sufficient for a reliable modelling. We further enriched the dataset by performing high-throughput calculations on Fe-(B, Co, Ti, Rh) binary intermetallic compounds, and the corresponding accuracy of the machine learning modelling is over 90% across all the systems. The results are further understood based on the Stoner model. Our work establishes a valid approach to model the physical properties dependent on the local crystalline environments.