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

MM: Fachverband Metall- und Materialphysik

MM 10: Poster Session 1

MM 10.1: Poster

Montag, 5. September 2022, 18:00–20:00, P2

Training Gaussian Approximation Potentials for Aqueous Systems — •Nikhil Bapat, Martin Vondrák, Johannes T. Margraf, Hendrik H. Heenen, and Karsten Reuter — Fritz-Haber-Institut der MPG, Berlin, Germany

An accurate and efficient description of aqueous systems via atomistic computer simulations is of high relevance for many applications. Machine learning potentials (MLPs) trained on first principles data have demonstrated promising accuracy and computational efficiency for the length and time scales critical to the description of water. But even with the compelling advancements in MLPs, building a successful data-efficient atomistic model for complex aqueous systems remains a challenging task. The training of such MLPs can be notoriously difficult and so far required either negligence of chemical reactivity in the MLP or excessive amounts of training data.

In this work we propose an efficient training procedure specifically designed for aqueous systems. To that end, we employ the widely applicable Gaussian approximation potential MLP and leverage it with a workflow for generating training data which ensures systematic inclusion of the bulk water configuration space. We calculate and compare ensemble properties of bulk water like its equilibrium density and diffusion coefficient to validate the MLP. The resulting model, when coupled with an added stimuli from a solid surface, can provide insights into many technologically important solid-liquid systems which are difficult to simulate otherwise.