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

MA: Fachverband Magnetismus

MA 34: Molecular Magnetism

MA 34.1: Vortrag

Donnerstag, 20. März 2025, 09:30–09:45, H18

Handling higher order ligand field parameters of single molecule magnets using deep learning — •Zayan Ahsan Ali, Julius Mutschler, Preeti Tewatia, and Oliver Waldmann — Physikalisches Institut, Universität Freiburg, D-79104 Freiburg, Germany

In recent decades, Single Molecule Magnets (SMMs) have sparked an interest not only due to their applications in quantum computing and spintronics, but also as an ideal platform for exploring fundamental principles of quantum magnetism. While substantial progress has been made towards the characterization of magnetic properties of 3d SMMs, the study of 4f SMMs remains challenging. This difficulty arises from the involvement of up to 27 ligand field parameters and the typically featureless nature of experimental magnetic data, leading to severe overparameterization. Moreover, the physically relevant regions in this parameter space are mostly unknown a priori. Although deep learning based inverse models, such as Conditional Variational Autoencoders and Invertible Neural Networks, have shown promise in addressing overparameterization, their performance degrades significantly when trained on uninformative parameter spaces, which dominate especially in high dimensional settings. This work investigates the use of Monte Carlo based parameter sampling for the higher order ligand field parameter space as a crucial precursor towards improving the deep learning inverse models. The resulting dataset represents a more informative prior, enabling insights into the effects of higher order ligand field parameters and the correlations between them.

Keywords: Quantum Magnetism; Markov Chain; Deep Learning; Inverse Problems