Stabilization of algebraic multilevel iteration methods; additive methods (Q1817774)

This paper presents an excellent survey on the algebraic multilevel (AML) iteration methods developed mainly by the author. Basically there are two versions, namely the additive and multiplicative form of the AML preconditioner. The author focuses on the additive version because it has many advandages from the parallelization point of view. Although the AML iteration is applicable to a broader class of problems, the author considers only symmetric and positive definite systems arising from the finite element discretization of selfadjoint, second-order elliptic boundary value problems. For this class of problems, a rigorous analysis is possible. The two-level as well as the multilevel version mainly depends on one parameter, namely the angle between the coarse grid space and the space spanned only by the fine grid functions. The cosine of this angle is else than the constant in the strengthened Cauchy-Bunjakowski-Schwarz inequality. In the multilevel version, one needs some stabilization in order to obtain an efficient method. The stabilized AML iteration can also be made robust against coefficient jumps and anisotropies.