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

O: Fachverband Oberflächenphysik

O 31: Clean surfaces II

O 31.6: Vortrag

Dienstag, 27. März 2012, 11:45–12:00, A 060

Construction of High-Dimensional Neural Network Potentials for Surfaces: Applications to Copper and Zinc Oxide — •Nongnuch Artrith, Björn Hiller, and Jörg Behler — Lehrstuhl für Theoretische Chemie, Ruhr-Universität Bochum, D-44780 Bochum, Germany

Molecular dynamics simulations of large systems critically depend on the accurate description of the underlying potential energy surface (PES). First-principles methods like DFT can provide very accurate energies and forces, but at high computational costs. Therefore, the development of more efficient potentials is a very active field of research. High-dimensional Neural Networks (NN) trained to first-principles data have been shown to provide accurate PESs for systems containing a single atomic species, while they are several orders of magnitude faster than electronic structure calculations. Now, we have generalized this method to multicomponent systems with arbitrary chemical composition. This is achieved by introducing physically motivated terms to deal with long-range interactions and charge transfer. Here we report the capabilities of the NN method for systems containing a single atomic species as well as for multicomponent systems. We present structural energy differences, vacancy formation energies, and surface energies for different copper and zinc oxide surfaces. The predicted geometries, energies, forces, and atomic charges are in excellent agreement with reference DFT calculations.