Berlin 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 17: Poster Ib

MM 17.20: Poster

Montag, 18. März 2024, 18:30–20:30, Poster F

Machine learning approach to obtaining the scattering self-energy from transmission calculations — •Fabian Engelke, Markus Kremer, Michael Czerner, and Christian Heiliger — Justus-Liebig-Universität, Institute for Theoretical Physics, Gießen, Germany

Aiming to assist the development of nanoscale electronic devices, we contribute to developing ab initio transport calculations. This work is particularly concerned with treating phase-breaking scattering events due to the electron-phonon interaction. In the Keldysh formulation of the non-equilibrium Green*s function formalism, as implemented in a Korringa-Kohn-Rostoker electronic structure code, those scattering events are characterized by an additional self-energy. Even though it is possible to estimate the self-energy, those calculations involve many approximations, such as k-vector averaging and Wannier-function interpolation of band structures.

Introducing a new way to calculate the self-energy, we train a deep neural network based on conventionally calculated self-energies and transmission calculation results from the Keldysh formulation. We then use this neural network to map transmission results based on the molecular dynamics Landauer approach back to the self-energy.

Keywords: Ab initio transport calculations; Machine Learning; Phase-breaking scattering; Neural network