Ultrafast Nonadiabatic Dynamics in 2D Perovskites assisted by Machine Learning

An exploration of the on-the-fly non-adiabatic couplings (NAC) for nonradiative relaxation and recombination of excited states in 2D Dion Jackobson Lead-halide perovskites is accelerated by a machime learning approach to ab initio molecular dynamics. Molecular dynamics of nanostructures composed of heavy elements is performed with use of machine learned force-fields (MLFF), as implemented in Vienna Ab initio Simulation Package (VASP). The force field parameterization is establised using on-the-fly learning, which continuously builds a force field using <i>ab initio</i> MD data. At each step of MD it is determined whether to perform an <i>ab initio</i> calculation or to use the force field and skip learning for that step using a Bayesian-learning algorithm. The total energy and forces are predicted based on the machine-learned force field at each time step of the MD simulation and if the Bayesian error estimate exceeds a threshold an <i>ab initio</i> calculation is performed. Model training and evaluation were performed for a range of for a 2D Dion-Jaconson lead halide perovskite models of different thickness and composition. The MLFF-MD trajectories were evaluated against AIMD trajectories to asses level of discrepancy and error accumulation. To examine the practical effectiveness of this approach we have used the MLFF-based MD trajectories to compute NAC and excited-state dynamics. At each stage, results based on machine learning are comapred to traditional ab initio-based electronic dissipative dynamics. We find that MLFF-MD provides comparable results to <i>ab initio</i> MDs when the MLFF is trained in a NpT ensemble.