Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 60: Phase Transitions II

MM 60.6: Vortrag

Donnerstag, 25. März 2010, 16:45–17:00, H5

Neural Network Potential-Energy Surfaces for Materials Simulations — Nongnuch Artrith, Tobias Morawietz, and •Jörg Behler — Lehrstuhl für Theoretische Chemie, Ruhr-Universität Bochum, D-44780 Bochum, Germany

Artificial neural networks represent a very flexible class of mathematical functions, which is well suited for the construction of potential-energy surfaces by interpolating a set of reference energies obtained from accurate electronic structure calculations. Recently, the applicability of neural network potentials has been extended to high-dimensional energy surfaces of condensed systems. By incorporating long-range electrostatic interactions also multicomponent systems can be addressed. Using semiconductors, oxides and metals as benchmark systems we show that neural networks provide reliable potentials for a wide range of materials. Since analytic gradients are readily available to calculate the forces, neural network potentials can be used to carry out efficient molecular dynamics simulations.