Berlin 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

MA: Fachverband Magnetismus

MA 17: Computational Magnetism II

MA 17.1: Talk

Tuesday, March 19, 2024, 09:30–09:45, EB 202

Rational design of RE-free magnets from first principles via Bayesian optimisation — •Fabian Eilers, Stephan Erdmann, Halil Ibrahim Sözen, and Thorsten Klüner — Institute of Chemistry, Carl-von-Ossietzky University of Oldenburg, D-26129 Oldenburg, Germany

The design of new rare-earth-free permanent magnetic materials to replace the powerful, but critically limited, rare earth-based magnets remains a a major challenge for the scientific community. Here we report on the rational design of new RE-free magnets using Bayesian optimisation. We attempt to improve the magnetic properties and phase stability of Fe2P-based magnetic materials by optimising the chemical composition and the lattice configuration of the unit cell. This problem involves global optimisation in a high configuration space. Bayesian optimisation can be used as a sequential design strategy to effectively explore this high configuration space. In an active learning approach, new trials are generated by sampling based on the expected improvement of the Gaussian process surrogate model. Trial evaluation is then performed via first-principles calculations using the Korringa-Kohn-Rostoker and Coherent Potential Approximation (KKR-CPA) method.

Keywords: Machine Learning; Density-functional theory; Magnetism