Necessary and sufficient conditions for the convergence of \(\text{Orthomin}(k)\) on singular and inconsistent linear systems (Q1592354)

The authors discuss some properties of the \(\text{Orthomin} (k)\) method on singular and inconsistent linear systems and propose key-criteria for avoiding breakdown of convergence. For singular systems obtained from discretization of partial differential equations with Neumann boundary conditions, most of the conjugate gradient type methods usually diverge. This is the motivation of the authors' research for \(\text{Orthomin} (k)\) on singular and inconsistent systems. Necessary and sufficient conditions for convergence of \(\text{Orthomin}(k)\) algorithm, are proved. Two numerical examples are discussed, one symmetric problem (the discretized Neumann problem for Poisson's equation on unit square) and one non-symmetric problem.