The celebrated minimum residual method (MINRES), proposed in the seminal paper of Paige and Saunders, has seen great success and wide-spread use in solving linear least-squared problems involving Hermitian matrices, with further extensions to complex symmetric settings. Unless the system is consistent whereby the right-hand side vector lies in the range of the matrix, MINRES is not guaranteed to obtain the pseudo-inverse solution. Variants of MINRES, such as MINRES-QLP, which can achieve such minimum norm solutions, are known to be both computationally expensive and challenging to implement. We propose a novel and remarkably simple lifting strategy that seamlessly integrates with the final MINRES iteration, enabling us to obtain the minimum norm solution with negligible additional computational costs. We study our lifting strategy in a diverse range of settings encompassing Hermitian and complex symmetric systems as well as those with semi-definite preconditioners. We also provide numerical experiments to support our analysis and showcase the effects of our lifting strategy.
翻译:暂无翻译