Data-assisted, physics-informed propagators for recurrent flows

Computational fluid dynamics simulations of dynamic flows usually entail large numerical costs. Over the last few years, several data-driven methods, partly of significant complexity, have been devised to mitigate CPU times while still capturing the relevant physics. In this work, a very simple approach with a clear physical interpretation is presented that splits the problem into a linear and a nonlinear part. Based on a database of precomputed reference states, predictions are made using the method of analogues (nonlinear dynamics) together with physics-informed propagators that capture and correct for any deviation from the nearest reference state in a linear fashion.