Towards fully ab initio simulation of atmospheric aerosol nucleation
Atmospheric aerosol nucleation contributes to more than half of cloud condensation nuclei globally. The emissions, properties and concentrations of atmospheric aerosols or aerosol precursors could respond significantly to climate change. Despite the importance for climate, the detailed nucleation mechanisms are still poorly understood. The ultimate goal of theoretical understanding aerosol nucleation is to simulate nucleation in ambient condition, hindered by lack of accurate reactive force field. Here we propose the reactive force field for nucleation systems with good size scalability based on deep neural network. The huge computational costs from direct molecular dynamics in ambient conditions are surmounted by bridging the simulation in the limited box with cluster kinetics, facilitating the aerosol nucleation simulation to be fully ab initio. We found that the acid-base formation rates previously based on hard sphere collision rate constants tend to be underestimated up to several times. These findings show that the widely recognized acid-base nucleation observed in the CLOUD (Cosmics Leaving OUtdoor Droplets) chamber experiments, pristine and polluted environments should be revisited to considering the contribution of collision enhancement. Besides, the framework here is transferable to other nucleation systems, potentially boosting the nucleation parameterizations accuracy generally to effectively advance the climate model predictions reliability.