Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
CPP: Fachverband Chemische Physik und Polymerphysik
CPP 46: Poster II
CPP 46.54: Poster
Thursday, March 12, 2026, 09:30–11:30, P5
Fourth-Generation High-Dimensional Neural Network Potentials for Molecular Chemistry in Solution — •Djamil A. A. Maouene1,2, Moritz R. 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 potentials have become essential tools in chemistry and materials science, offering accurate, efficient representations of high-dimensional potential energy surfaces for atomistic simulations. Here we compare two generations of high-dimensional neural network potentials (HDNNPs) 2G-HDNNPs and 4G-HDNNPs in their ability to model organic molecules in aqueous solution. 2G-HDNNPs perform well for systems dominated by local interactions because they rely on descriptors of the immediate atomic environment. However, in cases of long-range charge transfer 4G-HDNNPs provide a more reliable description by explicitly accounting for charge redistribution in the system as a function of its global structure. We illustrate these differences for organic molecules in water.
Keywords: Potential Energy Surface; Atomistic Simulations; Machine Learning Potentials; High-Dimensional Neural Network Potentials; 4G-HDNNPs