Hydrodynamic interactions in particle suspensions: Learning Stokesian dynamics

Stokesian dynamics (SD) is a numerical framework used for simulating hydrodynamic interactions in particle suspensions at low Reynolds number. It combines far-field approximations with near-field lubrication corrections, offering a balance between accuracy and efficiency. This work introduces SD and provides a perspective on future directions for this approach. We outline the mathematical foundations, the method’s strengths and weaknesses, and the computational challenges that need to be overcome to work with SD effectively. We also discuss recent advancements that improve the algorithm’s efficiency, including the use of iterative solvers and matrix-free approaches. In addition, we highlight the limitations of making stronger, albeit more cost-effective approximations to studying hydrodynamic interactions in dense suspensions than made in SD, such as the two-body Rotne–Prager–Yamakawa (RPY) approximation. To overcome these issues, we propose a hybrid framework that replaces SD’s full many-body computations with a neural network trained on SD data. That is, we correct the RPY approximation, while avoiding costly matrix inversions. We demonstrate the potential of this method on a simple system, where we find a close match to SD data while algorithmically outperforming RPY. Our work provides an outlook on the way in which large-scale simulations of particle suspensions can be performed in the foreseeable future.