Active learning for parameter-free multiscale modeling of boundary lubrication

Abstract: Lubricated friction is a multiscale problem where molecular processes dictate the macroscopic response of the system. Traditional lubrication models rely on semi-empirical constitutive relations, which become unreliable under extreme conditions. Here, we present a simulation framework that seamlessly couples molecular and continuum models for boundary lubrication without fixed-form constitutive laws. We train Gaussian process regression models as surrogates for predicting interfacial shear and normal stress in molecular dynamics simulations. An active learning algorithm ensures that our model adapts in scenarios where common constitutive laws fail, such as near phase transitions. We demonstrate our approach for nanoscale fluid flow over rough and heterogeneous surfaces, paving the way for accurate boundary lubrication simulations at experimental length and time scales.