Atomistic Simulation of Plasmonic Hot Carrier Dynamics Using Machine Learning

Understanding the dynamics of plasmonic hot carriers in metal nanostructures provide pathways for designing efficient energy harvesting devices. However, studies of the dynamics at atomistic level using full quantum-mechanical simulation tools such as nonadiabatic molecular dynamics or real-time density functional theory (rt-TDDFT) have been computationally expensive. For example, rt-TDDFT simulation of just photon absorption, plasmon formation and its subsequent decay to generate hot carriers takes several CPU hours in a nanocluster with less than 100 atoms. In this talk, I present our studies of plasmonic hot carrier dynamics using machine learning models in various metal nanoclusters of hundreds of atoms as prototypical examples. We demonstrate that atomistic neural network (NN) architectures that enable cost-effective molecular dynamics with machine-learned (ML) interatomic potentials can be adapted to model electron dynamics. We will show that atomistic NNs can estimate a time dependent electron density capable of capturing the relevant properties in the evolution of dipole moment at fractions of the quantum-mechanical simulation time and with minimal quantum-mechanical input data. Our goal is to explore the transferability of our workflow in pursuit of a scheme for affordable modeling of hot carrier dynamics in systems with thousands of atoms.