Long-Lived Hot Electron in a Metallic Particle for Plasmonics and Catalysis: Ab Initio Nonadiabatic Molecular Dynamics with Machine Learning.

Multiple experiments provide evidence for photovoltaic, catalytic, optoelectronic, and plasmonic processes involving hot, i.e., high energy, electrons in nanoscale materials. However, the mechanisms of such processes remain elusive, because electrons rapidly lose energy by relaxation through dense manifolds of states. We demonstrate a long-lived hot electron state in a Pt nanocluster adsorbed on the MoS2 substrate. For this purpose, we develop a simulation technique, combining classical molecular dynamics based on machine learning potentials with ab initio nonadiabatic molecular dynamics and real-time time-dependent density functional theory. Choosing Pt20/MoS2 as a prototypical system, we find frequent shifting of a top atom in the Pt particle occurring on a 50 ps time scale. The distortion breaks particle symmetry and creates unsaturated chemical bonds. The lifetime of the localized state associated with the broken bonds is enhanced by a factor of 3. Hot electrons aggregate near the shifted atom and form a catalytic reaction center. Our findings prove that distortion of even a single atom can have important implications for nanoscale catalysis and plasmonics and provide insights for utilizing machine learning potentials to accelerate ab initio investigations of excited state dynamics in condensed matter systems.