摘要
提出了一种面向对象的代数多重网格(algebraic multi-grid,AMG)算法,以每一层网格作为研究单元.网格粗化过程中,形成各单元,同时记录其前后单元,形成双向链表.粗化过程采用Ruge和Stüben算法,光滑算子用Gauss-Seidel迭代.由于AMG算法与网格信息无关,可以作为"即插即用"型的线性方程组求解器.对CFD计算过程耗时最多的压力修正方程作了研究,分别对二维后台阶流动模型在不同网格划分情况进行了计算,代数多重网格方法与单重网格的不完全分解共轭梯度法对比发现,前者具有明显的优势,随着网格数目增加,优势表现更为明显.最后与AMG1r5相比,开发的程序内存占用较少,最高只有AMG1r5的36%.
A new objected oriented AMG （Algebraic MultiGrid） method is presented, in which every level of grid is treated as operated object. Every level is constructed during coarsening process and records its last and next level. The method of Ruge and Stuhen is used for coarsening and Gauss--Seidel＇s iterative method for smooth operator. AMG is treated as a plug in linear equation solver because it is irrelative with grid information. The result of the back facing step flow with different mesh shows that AMG has a better performance than incomplete LU decompose conjugate gradient method for pressure correction equation. Comparing with AMGlr5, the present method is more efficient on memory usage and is only 36.1% of AMGlr5.
出处
《武汉理工大学学报（交通科学与工程版）》
2009年第1期87-90,共4页
Journal of Wuhan University of Technology（Transportation Science & Engineering）
基金
国防预研基金项目资助(批准号:GF10206030102)
