Parallel Strategies For Algebraic Multi-Grid

The major portion of computing time in a computational fluid dynamic (CFD) flow solver is consumed by the solution of the linear equations, the use of an efficient matrix solver is critical to obtain high computing performance. Currently, Algebraic Multi-Grid (AMG) is recognized in CFD community as one of the best choices to achieve high efficiency for solving linear algebraic equations. As AMG is increasingly used for large problems, its parallelization is highly demanded. In this paper, a parallel algorithm for AMG is proposed with regard to unique characteristics of partial differential equations governing fluid flows. A building strategy was developed to accomplish grid coarsening and build the hierarchy of multi-level grids. Moreover an agglomeration approach was proposed, which comes from the physical and mathematical properties of flow fields. The parallelized AMG has demonstrated excellent performances on many cases. Key-Words: Computational fluid dynamics, Multi-level grid, Parallel algorithms, Numerical algorithms, Parallel computing, Algebraic multi-grid (AMG)