1: \begin{abstract}

2: The Kaczmarz method for solving linear systems of equations is

3: an iterative algorithm that has found many applications ranging from

4: computer tomography to digital signal processing. Despite the popularity

5: of this method, useful theoretical estimates for its rate of convergence

6: are still scarce. We introduce a randomized version of the

7: Kaczmarz method for consistent, overdetermined linear systems and we prove

8: that it converges with expected exponential rate. 

9: Furthermore, this is the first solver whose {\em rate does not depend on 

10: the number of equations} in the system.  

11: The solver does not even need to know the whole system, but only a small 

12: random part of it.

13: It thus outperforms all previously known methods on general 

14: extremely overdetermined systems.

15: Even for moderately overdetermined systems, numerical simulations as

16: well as theoretical analysis reveal that our algorithm can converge faster 

17: than the celebrated conjugate gradient algorithm.  

18: Furthermore, our theory and numerical simulations confirm a prediction

19: of Feichtinger et al.\ in the context of reconstructing bandlimited

20: functions from nonuniform sampling.

21: \end{abstract}

22: