The uptake of atmospheric pollutants such as N2O5 and HOCl to aerosol particles impacts air quality and climate, yet key aspects of the relevant physical chemistry remain unresolved. [1, 2] For instance, the experimentally measured reactive uptake of N2O5shows a large variation that cannot be explained with the currently widely employed resistor model.  To this end we develop highly accurate many-body potentials and perform molecular dynamics simulations of N2O5 and HOCl on water clusters and at the bulk air/water interface to resolve atomistic details of the mechanism of solvation. In particular, we obtain the potential of mean force for the solvation of N2O5 from which we determine its solubility, a quantity that is currently not accessible experimentally but a key ingredient in atmospheric models of reactivity including the resistor model.<br/><br/>Our potentials are built on the highly successful MB-pol water model, [3, 4] which is based on the many-body expansion of the total energy, combining a physically motivated representation of long-range interactions with machine learning techniques that accurately represent the quantum mechanical short-range interaction energy terms. We employ an active learning procedure via query-by-committee using ANI-like neural networks  to generate training sets for one-body, two-body, and three-body potentials, which are then used to train efficient representations in terms of permutationally invariant polynomials based on coupled cluster reference data. By construction, the new potentials accurately represent both short- and long-range interactions, which is important for a proper description across gas and liquid phases and interfaces. We demonstrate the numerical accuracy of the many-body potentials with respect to test sets from cluster and bulk simulations and as part of this work also assess the accuracy of a range of widely used density functional approximations.<br/><br/>
No datasets are available for this submission.