ABSTRACT This study implements the immersed boundary method (IBM) coupled with the moment representation lattice Boltzmann method (MR‐LBM) for simulating fluid–structure interactions. Due to the nature of the MR‐LBM framework, which only stores moments, it is possible to eliminate some steps of the coupling with IBM compared to traditional LBM methods that store populations. As a result, the performance improves, and memory bandwidth requirements are reduced, a crucial factor in GPU implementations. We present results demonstrating the impact of parameters such as node spacing, kernel size, and node distribution on the computational speed of the method. Analyzing these metrics enables the assessment of which parts of the method impact performance and, therefore, should be targeted for future development.
Ferrari et al. (Thu,) studied this question.