Machine-learning tight-binding model for large-scale electronic-structure calculations

Calculating optoelectronic properties of large-scale systems at different temperatures using typical density functional theory (DFT) can pose a steep computational barrier. By employing machine learning force fields (ML-FFs), trajectories at different temperatures are nowadays readily available. However, calculating the temperature-dependent electronic structure still proves challenging. Following up on our previous work on a dynamic tight-binding (TB) model [1], we present an approach utilizing ML techniques in order to investigate the accuracy and computational cost of the approach. The combination of ML and TB may provide access to material properties at different temperatures at a relatively low computational cost.

[1] M. Schwade, M.J. Schilcher C. Reverón Baecker, M. Grumet, D. A. Egger, arXiv:2308.08897 [cond-mat.mtrl-sci] (2023)