Dresden 2026 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 31: Transport in Materials: Diffusion, Charge, or Heat Conduction II
MM 31.5: Vortrag
Donnerstag, 12. März 2026, 11:15–11:30, SCH/A216
Building Trust in Ab Initio Machine-Learning Potentials for Extreme Materials – Applications to Strongly Anharmonic Ceramics and Thermal Insulators — •Shuo Zhao1, Kisung Kang1,2, and Matthias Scheffler1 — 1The NOMAD Laboratory at FHI, Max Planck Society — 2School of Materials Science and Engineering, Chonnam National University
Thermal insulating ceramics and semiconductors often exhibit significant anharmonicity, particularly associated with rare events such as Frenkel defect creation and rattling phonon modes. These phenomena not only disrupt the phonon picture and the conventional perturbative methods for heat transport, but also pose challenges for the effective and trustful training of machine-learned interatomic potentials (MLIPs). Our contribution describes the implementation of a framework that combines the non-perturbative Green-Kubo formalism with a sequential, uncertainty guided active learning scheme using AlmoMD[1] with equivariant neural networks NequIP[2] and So3krates[3]. The approach is demonstrated by application to possibly ultra-low thermal conductivity materials[4]. Our results not only substantiate reliable predictions of thermal conductivity for strongly anharmonic systems but also pave the way for the accelerated exploration and design of novel thermal insulators.
[1] K. Kang, et al., Phys. Rev. Mater. 9, 063801 (2025). [2] S. Batzner, et al., Nat. Commun. 13, 2453 (2022). [3] J.T. Frank, et al., Nat. Commun. 15, 6539 (2024). [4] T.A.R. Purcell, et al., Npj Comput. Mater. 9, 112 (2023).
Keywords: Thermal Transport; Thermal Insulators; Machine Learning Interatomic Potential; Active Learning