Title: Coarse-graining and designing liquids with the Ornstein-Zernike equation and machine learnt closures

Abstract: A key challenge for soft materials design and coarse-graining simulations is determining interaction potentials between components that would create the desired condensed phase structure. In theory, the Ornstein-Zernike framework provides an elegant route for solving this inverse problem. The missing link is the so-called closure relation. Pioneering work in liquid state theory derived analytical closures, but they are approximations valid only for specific classes of interaction potentials. In this paper, we combine the physics of liquid state theory with deep learning. We use machine learning to infer a closure directly from simulation data. Our resulting closure is broadly accurate across the space of interaction potentials. We show that our approach solves the inverse problem in simple liquids with an accuracy higher than commonly used closures, as well as tackles prototypical problems in fitting coarse-grained simulations.