Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 83: Frontiers of Electronic Structure Theory: Focus on Artificial Intelligence Applied to Real Materials 4
O 83.10: Talk
Friday, September 9, 2022, 12:45–13:00, S054
Thermal Transport via Green-Kubo Method and Message-Passing Neural-Network Potentials — Marcel F. Langer1,2, Florian Knoop2,3, Christian Carbogno2, Matthias Scheffler2, and •Matthias Rupp2,4 — 1TU Berlin, Germany — 2The NOMAD Laboratory, FHI-MPG & HU Berlin, Germany — 3Theoretical Physics Division, Linköping U, Sweden — 4Konstanz U, Germany
Accurate, precise, and efficient computational access to thermal conductivities of materials is relevant for scientific understanding and industrial applications. The Green-Kubo method with first-principles calculations enables the determination of thermal conductivities, even for strongly anharmonic materials [1]. However, the high computational cost of long dynamics simulations of large supercells required for convergence limits applicability for large-scale, high-throughput materials discovery. Machine-learning potentials can reduce this cost [2].
Message passing neural networks (MPNNs) are a promising, but for this task yet untested, class of models due to their relational inductive bias, implicit long-range nature, and ability to incorporate directional information. We adapt the heat flux definition for MPNNs, investigate the impact of equivariance, present a systematic account of their convergence behavior and performance, and compare them to a simpler baseline model.
[1]: C. Carbogno, R. Ramprasad, and M. Scheffler, Phys. Rev. Lett. 118 175901 (2017) [2]: P. Korotaev et al., Phys. Rev. B 100 144308 (2019); C. Mangold et al., J. Appl. Phys. 127, 244901 (2020); C. Verdi et al., NPJ Computer. Mat. 7 156 (2021)