Machine-Learning Accelerated Description of Polarons in 2D Lead Halide Perovskites

MLFF-based MD trajectories to compute NAC and excited-state dynamics. At each stage, results based on machine learning are compared to traditional ab initio based electronic dissipative dynamics. The limits of this method are explored as we increase the timescale of the calculation beyond those allowed by ab initio molecular dynamics and to determine the effectiveness of MLFF-MD when used for excited-state nonadiabatic dynamics simulations using atomistic models that include charge injection or when an excited-stated potential energy surface is used to account for changes due to an excitation.