A new greedy Kaczmarz algorithm for the solution of very large linear systems (Q1726657)

The Kaczmarz numerical algorithm is a very popular iterative projection method for solving very large consistent systems of linear equations. However, this algorithm converges very slowly. A number of attempts have been made to speed up this procedure (see, for example [\textit{Z.-Z. Bai} and \textit{W.-T. Wu}, Linear Algebra Appl. 553, 252--269 (2018; Zbl 1391.65063)]). To improve the convergence of the Kaczmarz algorithm, the author examines a new modification with suitable selection of ``control sequence'' determined by a greedy rule and some suitable relaxation parameters. The algorithms: ``the GRK algorithm'' and ``the GK algorithm'' are described in details. Numerical experiments confirm the guaranteed convergence rate.