Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 32: Poster: Solid-Liquid Interfaces
O 32.17: Poster
Dienstag, 19. März 2024, 18:00–20:00, Poster C
Fourth-Generation High-Dimensional Neural Network Potentials for Molecular Chemistry in Solution. — •Djamil Abdelkader Adel Maouene1,2, Moritz Richard Schäffer1,2, Moritz Gubler3, Stefan Goedecker3, and Jörg Behler1,2 — 1Theoretische Chemie II, Ruhr-Universität Bochum, Germany — 2Research Center Chemical Sciences and Sustainability, Research Alliance Ruhr, Germany — 3Department Physik, Universität Basel, Switzerland
Machine learning has found many applications in the fields of chemistry and materials science, and high-dimensional neural network potentials (HDNNPs) have become an accurate tool to represent the multi-dimensional potential energy surface in atomistic simulations.
Here, we compare the performance of two types of HDNNPs, i.e., 2G-HDNNPs and 4G-HDNNPs, in the description of organic molecules in aqueous solution. While it has been shown for many systems that 2G-HDNNPs are well suited to represent local bonding as a function of the atomic environments, they are not applicable to systems in which long-range charge transfer is important. Such systems can be addressed by 4G-HDNNPs, in which the atomic charges depend on structural or electronic changes even very far away in the system. Using typical organic molecules, the performance of both approaches is illustrated.
Keywords: Machine learning; Neural network potentials; 4G; potential energy surface; molecular dynamics