A universal deep learning-based framework towards fully ab initio simulation of atmospheric aerosol nucleation

Atmospheric aerosol nucleation contributes to around half of cloud condensation nuclei globally. Despite the importance for climate, detailed nucleation mechanisms are still poorly understood. Understanding aerosol nucleation dynamics is hindered by non-reactivity of force fields and high computational costs due to rare event nature of aerosol nucleation. Developing reactive force fields for nucleation systems are even more challenging than covalently bonded materials because of wide size range and high dimensional characteristics of non-covalent hydrogen bonding bridging clusters. Here we proposes a system transferable framework to train an accurate reactive force field (FF) based on deep neural network (DNN) and further bridges the DNN-FF based molecular dynamics (MD) with cluster kinetics model based on Poisson distributions of reactive events to overcome high computational costs from direct MD. We found that previously reported acid-base formation rates tend to be underestimated several times, emphasizing acid-base nucleation observed in multiple environments should be revisited.