Obtaining pseudoinverse solutions with MINRES (Q6936077)

This interesting paper studies ways of obtaining pseudoinverses solutions with the minimum residual method (MINRES) which was first proposed by Paige and Saunders to study Hermitian and complex-symmetric systems \(Ax = b\). Unless the system is consistent, MINRES is not guaranteed to obtain a pseudoinverse solution.\N\NThe basic refinement problem the authors consider is to consider the following linear least-squares problem: \(\min_{x\in \mathbb C^{d}}||b-Ax||^2\) where \(A\in \mathbb C^{d\times d}\), \(b\in \mathbb C^d\) is a vector and \(b\notin\mathrm{Range}A\). The matrix \(A\) is either Hermitian or complex symmetric. The authors focus on cases where the system is inconsistent, that is the case when \(b\notin \mathrm{Range}A\). This problem appears in many applications, for example optimization of nonconvex problems, numerical solution of partial differential equations, low-rank matrix computations and many others.\N\NThe authors prove several interesting results for their refinement problem and provide numerical experiments to support their theoretical results.\N\NThe paper is well written with a very good set of references.