Machine Learning Coarse Grained Bond Order Models for Characterizing Mesoscale Aggregation in Liquid Mixtures

All-atom molecular dynamics simulations provide crucial insights into a variety of physical phenomena. Despite its numerous successes, all-atom models can be extremely computationally intensive for mesoscale systems. Coarse grained models have been able to overcome the computation bottleneck, but the description of interaction between organic molecules, aqueous media, and inorganic ions is often non-trivial. In this work, we present a coarse grained model for accurately capturing the phase behavior of multicomponent liquid-liquid mixtures. We use TrappE-UA model as a stencil and develop a bond-order based description of interactions between the components. Using the developed model we explore phase behavior and critical phenomena in mixtures consisting of organic solvents, surfactants, and water. Furthermore, aided by the reduced computational cost of our CG model we investigate mesoscale structuring in our systems, for e.g., formation of reverse micellar structures. Insights obtained from our simulations are utilized to interpret experiments performed on such systems.