Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 30: Poster Session II
MM 30.39: Poster
Dienstag, 17. März 2020, 18:15–20:00, P4
Symmetry-adapted Hamiltonian representations for machine-learning-based tight-binding parametrization — •Michael Luya1 and Reinhard Maurer2 — 1Department of Mathematics, University of Warwick, Coventry, UK — 2Department of Chemistry, University of Warwick, Coventry, UK
To tackle modern materials challenges, high efficient and accurate electronic structure methods need to be available that reliably predict the atomic structure, electronic and spectroscopic properties of materials at ever larger scales. Machine-learning methods have recently revolutionised the construction of interatomic potentials in computational materials science and are increasingly considered for the efficient construction of effective electronic structure methods such as tight-binding. Here we explore two-centre, three-centre and crystal field parametrisations of Hamiltonians in local basis representations that conserve important symmetries including rotational equivariance properties and permutational invariance. We show that this representation can accurately map on-site and off-site Hamiltonian contributions extracted from Density Functional Theory. We also investigate the physical significance of these parameters and the prospect of integration into deep-learning based parametrisation schemes.