Benchmarking Aggregation Amg for Linear Systems in Cfd Simulations of Compressible Internal Flows

The performance of parallel implementations of three fundamentally different aggregation AMG (algebraic multigrid) solvers, including novel k-cycle methods, for systems of linear equations appearing in industrial CFD simulations are examined. The results show that the k-cycle methods are a good choice for cases with less than 20000 unknowns per process if the cost of the setup tends to become critical; for most other applications, however, established methods proved to be equally efficient or superior.