EVALUATION OF PARALLEL AGGREGATE CREATION ORDERS: SMOOTHED AGGREGATION ALGEBRAIC MULTIGRID METHOD

摘要

The Algebraic MultiGrid method (AMG) has been studied intensively as an ideal solver for large scale Poisson problems. The Smoothed Aggregation Algebraic MultiGrid (SA-AMG) method is one of the most efficient of these methods. The aggregation procedure is the most important part of the method and is the main area of interest of several researchers. Here we investigate aggregate creation orders in the aggregation procedure. Five types of aggregation procedure are tested for isotropic, anisotropic and simple elastic problems. As a result, it is important that aggregates are created around one aggregate in each domain for isotropic problems. For anisotropic problems, aggregates around domain borders influence the convergence much. The best strategy for both anisotropic and isotropic problems in our numerical test is the ag- High Performance Computational Science and Engineering gregate creation method which creates aggregates on borders first then creates aggregates around one aggregate in the internal domain. In our test, the SA-AMG preconditioned Conjugate Gradient (CG) method is compared to the Localized ILU preconditioned CG method. In the experiments, Poisson problems up to 1.6 x 10~7 DOF are solved on 125PEs.