Machine-learning energy-preserving nonlocal closures for turbulent fluid flows and inertial tracers

Abstract

Machine learning turbulence closures for non-homogeneous and non-isotropic flows is a challenging task. The novelty of the presented approach is the adoption of a universal constraint associated with the energy preservation of the nonlinear terms, which is valid for any turbulent system. This constraint is embedded in the training process, and in combination with nonlocal representations in space and time, results in significant improvement for the resulted coarse scale models.