Michal Outrata

Jose Pablo Lucero Lorca, Conor McCoid, Michal Outrata

We consider an algebraic multigrid (AMG) scheme for the direct solution of complex- valued square linear systems based on a recursive 2 x 2 block partitioning of the coefficient matrix and study the optimal choices of its components. In particular, we complement existing results that characterize the optimal choices for a nonsymmetric cycle method by analyzing the spectral behavior of its symmetric cycle variant. We analyze the error propagation operator of a specific two-level symmetric cycle method by calculating its invariant subspaces and its nonzero eigenvalues that influence the behavior of the error after a single cycle. We show that the error propagation operator can be studied separately for pairs of modes, working as bases of the invariant subspaces. The main result is an explicit choice of smoothing parameters that makes all the pairs of modes respond identically, forcing the nontrivial eigenvalues of the error propagation operator to collapse to a single, a priori known value. As a consequence, we give a new closed form formula for the inverse of a general square matrix. Additionally, the framework provides a clear, self-contained description of an ideal AMG K-cycle, offering a concrete target for the design of related schemes. We illustrate the theory with direct applications to general matrices and with analyses of representative matrices arising in numerical methods.

Lenka Ptackova, Michal Outrata

We examine the use of domain decomposition for potentially more efficient mean curvature flow of surface meshes, whose faces are arbitrary simple polygons. We first test traditional domain decomposition methods with and without overlap of deconstructed domains. And we present adapted Robin transmission conditions of optimized Schwarz method. We then analyze the resulting smoothing from the point of view of shape quality and texture deformation. By decomposing the initial mesh into two sub-meshes, we solve two smaller boundary value problems instead of one big problem, and we can process these two tasks almost entirely in parallel.