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

O: Fachverband Oberflächenphysik

O 46: Nanostructures and Clusters

O 46.2: Vortrag

Mittwoch, 13. März 2013, 10:45–11:00, H45

A full-dimensional and reactive neural network potential for water clusters based on first-principles — •Tobias Morawietz and Jörg Behler — Lehrstuhl für Theoretische Chemie, Ruhr-Universität Bochum, D-44780 Bochum, Germany

Artificial neural networks (NNs) provide an unbiased way to construct accurate interatomic potentials for a wide range of systems [1]. For this purpose a set of energies and forces obtained from electronic structure calculations is interpolated. The obtained NN potential can be evaluated several orders of magnitude faster than the underlying electronic structure data allowing to perform accurate simulations on extended length and time scales. Here, we report a full-dimensional NN potential for water clusters containing up to ten water molecules trained to density-functional theory (DFT) data [2,3]. Unlike other potentials for water based on first-principles, our potential is not constructed employing a truncated many-body expansion and is thus able to describe reactions involving proton transfer. Binding energies for global and local minima obtained with the NN potential typically deviate by less than 1 % from the DFT values.

[1] J. Behler, PCCP 13, 17930 (2011).

[2] T. Morawietz, V. Sharma, and J. Behler, JCP 136, 064103 (2012).

[3] T. Morawietz and J. Behler, submitted (2012).