Regensburg 2022 – wissenschaftliches Programm

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

O: Fachverband Oberflächenphysik

O 32: Poster Tuesday: Adsorption and Catalysis 1

O 32.4: Poster

Dienstag, 6. September 2022, 11:00–13:00, P3

Adaptive training of a machine-learned model for nonadiabatic hydrogen chemistry on multiple facets of Copper. — •Wojciech G. Stark, Julia Westermayr, Oscar A. Douglas-Gallardo, James Gardner, and Reinhard J. Maurer — University of Warwick, Coventry, United Kingdom

Traditionally, molecular dynamics methods utilise the Born-Oppenheimer approximation and dynamics are governed by a single potential energy surface. However, on metallic surfaces often the energy exchange between adsorbate and electronic excitations in the metal is significant and causes the breakdown of the Born-Oppenheimer approximation. There are multiple methods to include such nonadiabatic effects, with one of the most efficient being molecular dynamics with electronic friction (MDEF). MDEF introduces nonadiabatic effects via additional electronic friction forces, which can be calculated with time-dependent perturbation theory based on Density Functional Theory. However, a meaningful comparison between computational simulations and experiments demands the capability to run tens of thousands of MDEF trajectories. We present high-dimensional machine-learning based interatomic potential and electronic friction models that enable the simulation of nonadiabatic molecular dynamics of hydrogen scattering and associative desorption at different copper surfaces. We construct deep neural network representations via iterative adaptive sampling based on the target dynamical observables, namely the scattering and reaction probabilities.