Berlin 2024 – scientific programme

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

O: Fachverband Oberflächenphysik

O 82: Electronic Structure Theory I

O 82.5: Talk

Thursday, March 21, 2024, 11:30–11:45, MA 043

Prediction of the Single Particle Electronic Hamiltonian for Periodic Systems — •Hanna Türk, Divya Suman, Jigyasa Nigam, and Michele Ceriotti — EPFL, Lausanne, Switzerland

The electronic structure of a material provides essential information on a materials properties. For most materials, it can directly be computed by ab initio calculations. However, for complex systems such as material grain boundaries and interfaces, which often govern relevant chemical processes, the required large simulation cells exceed the feasiblity of such methods.

Here, we take a first step to develop a neural network based model to predict the effective single particle electronic Hamiltonian for periodic systems from given atomic environments. This apporach has proven successful for molecules[1], and we now adapt it to bulk materials. By expanding the framework to learn the realspace Hamiltonian, which includes all relevant periodic translations, sampling of the entire k-phase becomes possible. Our versatile framework yields an accurate description of the relevant electronic structure, which can be used to obtain band structures and electronic conductivities. In perspective, the prediction on local environments allows training of a model on small cells, and the trained model can then be used to obtain electronic structure information on larger, more complex systems.

[1] J. Nigam, M. J. Willatt, M. Ceriotti, J. Chem. Phys. 2022, 156, 014115, DOI 10.1063/5.0072784.

Keywords: Hamiltonian; Electronic Structure; Machine Learning; Solid State