Algebraic Multigrid (AMG) methods are a powerful class of accelerators for iterative linear solvers such as Jacobi, Gauss--Seidel, and Successive Over-Relaxation (SOR). This dissertation presents several classical AMG methods as well as introducing two new methods: Smoothed Aggregation and AMGe. This dissertation also introduces AMGLab. AMGLab is a software package written in MATLAB intended to simplify the process of choosing the best AMG method for any particular application. A practical approach is used with a focus on implementation rather than on theorems and proofs. |