Parts | Days | Selection | Search | Updates | Downloads | Help

O: Fachverband Oberflächenphysik

O 58: Poster Wednesday: New Methods and Developments, Frontiers of Electronic Structure Theory

O 58.16: Poster

Wednesday, September 7, 2022, 18:00–20:00, P4

Improving the transferability of high-dimensional neural network potentials by low-order terms — •Alea Miako Tokita and Jörg Behler — Georg-August-Universität Göttingen, Institut für Physikalische Chemie, Theoretische Chemie, Tammannstraße 6, 37077 Göttingen, Germany

High-dimensional neural network potentials (HDNNPs) are able to provide accurate potential energy surfaces suitable for atomistic simulations of large systems. The key to this accuracy is the high flexibility of the atomic neural networks allowing to reproduce energies and forces from reference electronic structure calculations with very small errors. At the same time, this flexibility is limiting the transferability of HDNNPs to atomic configurations that are very different from the reference geometries. Here, we investigate possible improvements in transferability of HDNNPs by the explicit inclusion of low-order terms in the functional form of the potential. The performance is demonstrated for a series of molecular model systems.