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

MM: Fachverband Metall- und Materialphysik

MM 31: Data-driven Materials Science: Big Data and Worksflows

MM 31.2: Vortrag

Donnerstag, 20. März 2025, 15:15–15:30, H10

high-throughput computation and machine learning modeling of magnetic moments and Mössbauer spectroscopy for Fe-based intermetallics — •Bo Zhao, Xiankang Tang, and Hongbin Zhang — Institute of Materials Science, Technische Universität Darmstadt, Otto-Berndt-Str. 3, 64287 Darmstadt, Germany

Understanding the relationship between the local crystalline environment and magnetic properties is a fundamental challenge in condensed matter physics and materials science. This study explores this relationship in Fe-based intermetallic compounds, focusing on the magnetic moments and Mössbauer parameters of iron atoms, including the isomer shift, electric field gradient, and magnetic hyperfine field. High-throughput calculations and machine learning techniques are employed to predict magnetic properties based on local atomic structures, using smooth overlap of atomic positions (SOAP) as local descriptors. The results first reveal the sparsity of relevant materials in the Materials Project database. Leveraging high-throughput, system-specific data, the study demonstrates strong correlations between local atomic environments and magnetic properties, achieved through machine learning models. Furthermore, the limitations of symmetry-invariant descriptors in predicting tensor-like properties, such as the electric field gradient, are highlighted. By incorporating a graph-based equivariant autoencoder, the model achieves improved predictions by effectively capturing the symmetry of local environments.

Keywords: Mössbauer spectroscopy; Machine learning