Numerical Analysis and Scientific Computing Seminar | Chen Greif, On the iterative solution of block-structured sparse linear systems

For Zoom Link please contact ddelreyfernandez@uwaterloo.ca  

Speaker

Chen Greif, Professor Department of Computer Science at the University of British Columbia

Title

On the iterative solution of block-structured sparse linear systems

Abstract

Saddle-point systems and other types of block-structured linear systems often arise from discretization of partial differential equations with constraints. When the matrices involved are sparse and large, iterative solution methods are often preferred. Their convergence speed primarily depends on the distribution of the eigenvalues and the ability to derive efficient preconditioners that exploit the spectral structure. This is a challenging task, which requires an understanding of the underlying continuous operators and the rank structure of the discrete operators involved. In this talk I will describe some of the numerical challenges that arise in the development of effective preconditioners for this type of problems, and will present some recent work that offers an eigenvalue analysis for discretized multiphysics problems.