Regensburg 2025 – scientific programme

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

HL: Fachverband Halbleiterphysik

HL 29: Poster II

HL 29.84: Poster

Tuesday, March 18, 2025, 18:00–20:00, P1

Accelerated Electrical Transport Predictions of the Non-Perturbative ab initio Kubo-Greenwood Method via a Deep-learned Hamiltonian technique: Strongly Anharmonic Material Cases — •Juan Zhang, Jingkai Quan, Matthias Scheffler, and Kisung Kang — The NOMAD Laboratory at the FHI of the Max Planck Society, Berlin, Germany

High-performance thermoelectric materials are characterized by low thermal and high electrical conductivities. Thermal insulators with ultra-low thermal conductivity feature strong anharmonicity, also significantly affecting electronic transport [1]. Strong anharmonicity impedes the application of perturbative methods due to the breakdown of the quasi-particle picture. This challenge can be overcome by the non- perturbative ab initio Kubo-Greenwood approach (aiKG)[1]. However, the aiKG method requires substantial computational cost due to its extensive supercell electronic structure evaluations at each step during ab initio Molecular Dynamics (aiMD). A recent deep neural network technique to train and predict the Kohn-Sham Hamiltonian, implemented by DeepH[2], can provide an efficient bypass to extract the electronic structure of each MD step with nearing ab initio accuracy. This work introduces an AI-assisted aiKG formalism with accelerated carrier mobility evaluations, exemplified by its application to strongly anharmonic materials. We thoroughly examine DeepH's capability for electronic property predictions with its spatial scalability.

[1] J. Quan et al., Phys. Rev. B, accepted (arXiv:2408.12908).

[2] X. Gong, et al., Nat Commun. 14, 2848 (2023).

Keywords: Electrical Transport; Kubo-Greenwood method; Deep-learning Hamiltonian