Parallel linear multigrid by agglomeration for the acceleration of 3D compressible flow calculations on unstructured meshes (Q1587039)

The authors describe the development and evaluation of parallel linear multigrid algorithms on unstructured tetrahedral meshes. The multigrid strategy relies on a volume agglomeration principle for construction of coarse grids starting from a fine discretization of computational domain. A part of the methodology considered here has been previously designed by the first author [Comput. Fluids 26, No. 3, 299-320 (1997; Zbl 0884.76049)] for two-dimensional flows. Here the authors demonstrate a successful extension and application of the multigrid volume agglomeration principle to acceleration of complex three-dimensional compressible flow through its adaptation to parallel architectures. An important aspect of the present work is that the underlying software developments take place in the framework of an existing industrial flow solver. The proposed multigrid algorithms are assessed by applying them to two representative turbulent and steady flow calculations within complex geometries. The flow solver is based on an averaged form of multi-component Navier-Stokes equations coupled to \(k-\varepsilon\) turbulence model. The spatial discretization combines finite element and finite volume concepts, and is designed on unstructured tetrahedral meshes. Steady-state solutions of the resulting semi-discrete equations are obtained using an Euler implicit time advancing strategy which is based on the following features: linearization (approximate linearization of convective fluxes and exact differentiation of viscous terms), local time stopping, and CFL law (a local time step is computed on each control volume). Then, each pseudo-time step can require the solution of as many as three linear sparse systems for mean flow variables, turbulent variables and chemical species.