1: \begin{abstract}

2: We propose a new method for preconditioning Kaczmarz method by sketching.

3: Kaczmarz method is a stochastic method for solving overdetermined linear systems based on a sampling of matrix rows. 

4: The standard approach to speed up convergence of iterative methods is using preconditioner.

5: As we show the best possible preconditioner for this method can be constructed from QR decomposition of the system matrix, but the complexity of this procedure is too high.

6: Therefore, to reduce this complexity, we use random sketching and compare it with the Kaczmarz method without preconditioning.

7: The developed method is applicable for different modifications of classical Kaczmarz method that were proposed recently.

8: We provide numerical experiments to show the performance of the developed methods on solving both random and real overdetermined linear systems.  

9: % This paper studies different approaches to speed up convergence of Kaczmarz method using preconditioners.

10: % Preconditioning is well-known approach based on reduction condition number of the matrix in linear system. 

11: % We employ this approach to speed up Kaczmarz method since existing convergence results claim dependence of convergence speed on condition number of the matrix.  

12: % Kaczmarz method is the method to solve overdetermined linear systems.

13: % These systems also can be solved in the least-squares manner, therefore we compare Kaczmarz method with other stochastic methods that solve least squares problem associated with considered linear system.

14: \end{abstract}

15: