Regensburg 2025 – scientific programme

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

HL: Fachverband Halbleiterphysik

HL 3: Focus Session: Machine Learning of semiconductor properties and spectra

HL 3.7: Talk

Monday, March 17, 2025, 12:30–12:45, H17

Exploring Strongly Anharmonic Thermal Insulators with Machine-Learned Interatomic Potential using an Active Learning Scheme — •Shuo Zhao, Kisung Kang, and Matthias Scheffler — The NOMAD Laboratory at the FHI of the Max Planck Society

Thermal insulating semiconductors often exhibit significant anharmonicity, particularly associated with rare events such as defect creation and phase-transition precursors [1]. These phenomena disrupt the conventional phonon picture and render perturbative methods ineffective or even incorrect for describing heat transport, leading to a substantial challenge for reliable prediction of thermal conductivity. This work presents a framework that combines the Green-Kubo formalism with machine-learned interatomic potentials, enhanced by a sequential active learning scheme [2]. Equivariant neural networks NequIP [3] and So3krates [4] are employed and systematically compared for this purpose. Based on this framework, we examine 15 materials that possibly have ultra-low thermal conductivity previously, predicted by a symbolic regression machine-learning model [5]. Our demonstrations and results not only provide precise thermal conductivity predictions for strongly anharmonic systems but also pave the way for accelerated exploration and design of novel thermal insulators.

[1] F. Knoop, et al., Phys. Rev. Lett. 130, 236301 (2023). [2] K. Kang, et al., arXiv:2409.11808 (2024). [3] S. Batzner, et al., Nat. Commun. 13, 2453 (2022). [4] J.T. Frank, et al., Nat. Commun. 15, 6539 (2024). [5] T.A.R. Purcell, et al., Npj Comput. Mater. 9, 112 (2023).

Keywords: Thermal Transport; Machine Learning Interatomic Potential; Active Learning