Randomized Kaczmarz methods with beyond-Krylov convergence (Q6881035)

This interesting paper deals with randomized Kaczmarz methods with beyond Krylov convergence. In the study of solving large linear systems of equations, a popular method is the Kaczmarz iterative algorithm which has vast applications for example in irregular sampling, tomography, and many more.\N\NIn this paper the authors, in particular, develop randomized Kaczmarz methods that exploit outlying singular values in the input to achieve fast convergence, going beyond what Krylov subspace methods can achieve for a class of singular value distributions. The author's methods do not require the systems to be very tall or overdetermined to work effectively.\N\NThe paper is well written with a very good set of references.