Parallel Domain-Decomposition Preconditioning for Computational Fluid Dynamics

Algebraic preconditioning algorithms suitable for computational fluid dynamics (CFD) based on overlapping and non-overlapping domain decomposition (DD) are considered. Specific distinction is given to techniques well-suited for time-dependent and steady-state computations of fluid flow. For time-dependent flow calculations, the overlapping Schwarz algorithm suggested by Wu et al. [28] together with stabilised (upwind) spatial discretisation shows acceptable scalability and parallel performance without requiring a coarse space correction. For steady-state flow computations, a family of non-overlapping Schur complement DD techniques are developed. In the Schur complement DD technique, the triangulation is first partitioned into a number of non-overlapping subdomains and interfaces. The permutation of the mesh vertices based on subdomains and interfaces induces a natural 2×2 block partitioning of the discretisation matrix. Exact LU factorisation of this block system introduces a Schur complement matrix which couples subdomains and the interface together. A family of simplifying techniques for constructing the Schur complement and applying the 2×2 block system as a DD preconditioner are developed. Sample fluid flow calculations are presented to demonstrate performance characteristics of the simplified preconditioners.