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

MM: Fachverband Metall- und Materialphysik

MM 9: Topical Session: Physics-driven Artificial Intelligence for Materials II

MM 9.3: Vortrag

Montag, 9. März 2026, 16:30–16:45, SCH/A251

Physics-informed Hamiltonian-learning for large-scale electronic-structure calculations — •Martin Schwade, Shaoming Zhang, Frederik Vonhoff, Frederico P. Delgado, and David A. Egger — Physics Department, TUM School of Natural Sciences, Technical University of Munich, 85748 Garching, Germany

Exploring the optoelectronic properties of large-scale materials systems across a wide temperature range using conventional density functional theory (DFT) is often prohibitively computationally expensive. Recent advances in deep neural network approaches offer a promising route to efficiently predict accurate effective Hamiltonians, yet incorporating temperature dependence remains challenging, largely due to the substantial volume of training data typically required. In this work, we introduce HAMSTER [1], a physics-informed Hamiltonian-learning framework that achieves high accuracy with exceptional data efficiency, requiring only a small fraction of the training data demanded by alternative machine-learning models. We demonstrate the capabilities of Hamster on several halide perovskite systems, known for their soft lattices and strong electron-phonon coupling, and show that it reliably reproduces their optoelectronic properties across a broad range of temperatures.

[1] M. Schwade, S. Zhang, F. Vonhoff, F. P. Delgado, D. A. Egger, Physics-informed Hamiltonian learning for large-scale optoelectronic property prediction, arXiv:2508.20536 (2025)

Keywords: Hamiltonian-learning; Machine learning; Electronic structure; Large scale; tight binding