Dresden 2026 – scientific programme

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

MA: Fachverband Magnetismus

MA 42: Molecular Magnetism and Magnetic Particles / Clusters II

MA 42.1: Talk

Thursday, March 12, 2026, 09:30–09:45, HSZ/0004

Towards a Physics-Informed Deep Learning Model for Parameterizing the Magnetism in 4f based Single-Molecule Magnets — •Zayan Ahsan Ali and Oliver Waldmann — Physikalisches Institut, Universität Freiburg, Germany

Single-molecule magnets (SMMs) have attracted interest in recent decades for their intriguing magnetic behavior and potential applications in quantum computing and spintronics. 4f based SMMs stand out because lanthanide ions can provide large magnetic anisotropies. However, the ligand environments that govern their magnetic properties involve up to 27 ligand-field parameters, while common experimental data such as temperature dependent magnetic susceptibility measured on powdered samples are comparatively featureless. Inferring ligand field parameters from such data is therefore a highly over-parameterized inverse problem. Machine learning approaches have shown promise in addressing this challenge [1]. Two central difficulties are identification of physically relevant regions of the ligand field parameter space and learning of a highly complex one-to-many mapping. This work presents a machine learning framework tailored to overcome both obstacles. It is shown that an informative training data set can be constructed via active learning with uncertainty sampling, and that incorporating the energy spectrum as an intermediate representation significantly improves the learnability of the high dimensional inverse mapping. The resulting architecture is found to be capable of predicting multiple possible ligand field parameter sets.

[1] Z. A. Ali et al., PRB 112, 064403 (2025).

Keywords: single molecule magnets; lanthanides; machine learning; inverse problems