GPU accelerated vortex-induced vibration simulation using JAX: Efficiency and accuracy strategies

Abstract

This study presents an immersed boundary-lattice Boltzmann method (IB-LBM) simulation framework for vortex-induced vibration (VIV), implemented using the JAX framework to exploit GPU acceleration. The code is specifically structured to meet JAX’s functional and static requirements, incorporating an efficient multi-block grid refinement scheme and a novel dynamic region approach for immersed boundary calculations. Through systematic benchmarking and convergence studies, we revealed that the present dynamic region approach achieves improved efficiency by reducing computational workload while avoiding excessive dynamic array operations. We demonstrated that the grid refinement setting should be adjusted according to the Reynolds number to maintain accuracy. The results also showed that while grid refinement saves total time-to-solution, parallel efficiency is reduced due to the stalls caused by frequent inter-block communications. Furthermore, we compared fluid-structure coupling strategies, finding that while weak coupling is adequate for amplitude prediction, strong coupling with at least two iterations is necessary to eliminate spurious frequency artifacts in the response. These findings offer practical guidelines for achieving efficient and accurate VIV simulations with IB-LBM on modern GPU platforms.