Pseudoinverse-free randomized block iterative algorithms for consistent and inconsistent linear systems

TL;DR

This paper introduces two new pseudoinverse-free randomized block iterative algorithms for solving large-scale linear systems, applicable to both consistent and inconsistent cases, with proven linear convergence and demonstrated efficiency.

Contribution

The paper presents novel randomized algorithms that extend existing methods by allowing a broader choice of random matrices, improving flexibility and efficiency in solving linear systems.

Findings

Algorithms achieve linear convergence in mean square sense.

Numerical experiments confirm efficiency on synthetic and real-world data.

Special cases outperform traditional methods in certain scenarios.

Abstract

Randomized iterative algorithms have attracted much attention in recent years because they can approximately solve large-scale linear systems of equations without accessing the entire coefficient matrix. In this paper, we propose two novel pseudoinverse-free randomized block iterative algorithms for solving consistent and inconsistent linear systems. The proposed algorithms require two user-defined random matrices: one for row sampling and the other for column sampling. We can recover the well-known doubly stochastic Gauss--Seidel, randomized Kaczmarz, randomized coordinate descent, and randomized extended Kaczmarz algorithms by choosing appropriate random matrices used in our algorithms. Because our algorithms allow for a much wider selection of these two random matrices, a number of new specific algorithms can be obtained. We prove the linear convergence in the mean square sense of our algorithms. Numerical experiments for linear systems with synthetic and real-world coefficient matrices demonstrate the efficiency of some special cases of our algorithms.