Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 42: Poster Session III: Poster to Mini-Symposium: Machine learning applications in surface science I
O 42.5: Poster
Dienstag, 2. März 2021, 10:30–12:30, P
Symmetry-Equivariant Representations of Ab Initio Hamiltonians for Machine Learning Purposes — •Michael Luya1 and Reinhard Maurer2 — 1Department of Mathematics, University of Warwick — 2Department of Chemistry, University of Warwick
High-dimensional machine learning is consistently improving the quality of condensed matter simulations. These simulations provide us with the Hamiltonian matrix, useful for calculating a wide variety of material properties. Predicting these matrices requires a deep understanding of covariance properties and directional coordinate dependence.
In an attempt to capture rotation invariance we introduce methods involving crystal field theory and bijective minimal basis transformations, in order to extract symmetry invariant parameters that can describe high-order contributions to the Hamiltonian, with application towards modelling a variety of molecules, metallic clusters, and eventually metal-organic interfaces. We also investigate the physical significance of these parameters and present a scheme as to how they should be best extracted from data, for the purposes of providing significant training data for neural network models.