# Dresden 2020 – wissenschaftliches Programm

# Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...

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

# O: Fachverband Oberflächenphysik

## O 42: Focus Session: Innovation in Machine learning PRocEsses for Surface Science (IMPRESS)

### O 42.4: Vortrag

### Dienstag, 17. März 2020, 11:45–12:00, TRE Phy

**Symmetry-adapted neural network representations of electronic friction to simulate nonadiabatic dynamics at metal surfaces** — •Reinhard J Maurer^{1}, Yaolong Zhang^{2}, and Bin Jiang^{2} — ^{1}Department of Chemistry, University of Warwick, United Kingdom — ^{2}Hefei National Laboratory for Physical Science at the Microscale & Department of Chemical Physics, USTC, Hefei, China

In catalytic reactions or molecular scattering, molecules impinging on metal surfaces excite electronic excitations. This leads to nonadiabatic energy transfer between the adsorbate and the metal that can measurably affect reaction outcomes. We study these effects with ab-initio molecular dynamics simulations using Density Functional Theory (DFT), where electron-phonon coupling is modelled as system-bath coupling, so-called electronic friction, in a Generalised Langevin equation framework. [1] To enable statistical averaging over many reaction events, we employ neural-network-based representations of the electronic friction tensor (EFT) calculated with time-dependent perturbation theory and DFT. [2] A particular challenge hereby is to capture the symmetry equivariance properties of the electronic friction tensor as a function of all atom positions by constructing a neural network with a tensor output layer. Our approach achieves an efficient and continuous representation of EFT, which we apply to metal surface scattering of diatomic molecules. [3]

[1] PRL 116, 217601 (2016); [2] Chem. Sci. 10, 1089-1097 (2019); [3] arXiv:1910.09774