Recent Developments in Algebraic Multigrid Methods

John Ruge

1005 Gillaspie Dr., Boulder, CO 80303


Abstract

There has been a resurgence of interest in Algebraic Multigrid Methods (AMG) over the last several years as computers become more powerful and the problems to be solved become larger and more complex, making most conventional solvers inadequate. AMG is a "black-box" linear solver based on multigrid principles, in which the coarser levels and necessary operators are constructed automatically based on the matrix itself. In this talk, we present the main idea of AMG, present some results on recent applications such as 3D elasticity, and discuss some aspects of parallel implementation of AMG.