Coarse level Newton-Krylov acceleration of sub-iterations in partitioned fluid-structure interaction

Computational fluid-structure interaction is commonly performed using a partitioned approach. For strongly coupled problems sub-iterations are required, increasing computational time as flow and structure have to be resolved multiple times every time step. Reductions in computing times can be achieved by e.g. improving the convergence of the sub-iteration technique and/or performing sub-iterations on a coarse level, but also by improving the iterative solver used in the flow solver. In this paper we investigate the combination of a multilevel acceleration technique for sub-iterations which employs a Newton-Krylov solver on the coarse level to obtain high convergence for the correction term and a multigrid solver which performs only a limited amount of iterations on the fine level to reduce memory and computing requirements. For switching between a coarse grid correction and fine grid solve, an automated coarse grid ACG(r) selection algorithm is proposed. The algorithm is applied to an academic, two dimensional test case with incompressible flow. Compared to sub-iterating with a (memory intensive) JFNK algorithm on the fine mesh, the hybrid algorithm already requires 10% less computing time. When compared to sub-iterating with multigrid, the performance increase for the hybrid scheme is a factor 3