Fast Flexible LSQR with a Hybrid Variant for Efficient Large-Scale Regularization

TL;DR

This paper introduces FaFLSQR, a novel, efficient hybrid regularization algorithm for large-scale ill-posed problems, combining a new bidiagonalization method with proven mathematical equivalence and superior computational performance.

Contribution

The paper presents a new Fast Flexible Golub-Kahan bidiagonalization method and the FaFLSQR algorithm, enhancing efficiency and flexibility in large-scale regularization tasks.

Findings

FaFLSQR matches FCGLS in computational cost

FaFLSQR outperforms FCGLS and FLSQR in efficiency

Mathematically equivalent to FCGLS

Abstract

A wide range of applications necessitates solving large-scale ill-posed problems contaminated by noise. Krylov subspace regularization methods are particularly advantageous in this context, as they rely solely on matrix-vector multiplication. Among the most widely used techniques are LSQR and CGLS, both of which can be extended with flexible preconditioning to enforce solution properties such as nonnegativity or sparsity. Flexible LSQR (FLSQR) can also be further combined with direct methods to create efficient hybrid approaches. The Flexible Golub-Kahan bidiagonalization underlying FLSQR requires two long-term recurrences. In this paper, we introduce a novel Fast Flexible Golub-Kahan bidiagonalization method that employs one long-term and one short-term recurrence. Using this, we develop the Fast Flexible LSQR (FaFLSQR) algorithm, which offers comparable computational cost to FCGLS while also supporting hybrid regularization like FLSQR. We analyze the properties of FaFLSQR and prove its mathematical equivalence to FCGLS. Numerical experiments demonstrate that in floating point arithmetic FaFLSQR outperforms both FCGLS and FLSQR in terms of computational efficiency.