1: \begin{abstract}

2: In this paper we propose a variant of the random coordinate descent

3: method for solving linearly constrained convex optimization problems

4: with composite objective functions. If the smooth part of the

5: objective function has Lipschitz continuous gradient, then we prove

6: that our method obtains  an $\epsilon$-optimal solution in ${\cal

7: O}(N^2/\epsilon)$ iterations, where $N$ is the number of blocks. For

8: the class of problems with cheap coordinate derivatives we show that

9: the new method is faster than methods based on full-gradient

10: information. Analysis for the rate of convergence in probability is

11: also provided. For strongly convex functions our method converges

12: linearly. Extensive numerical tests confirm that on very large

13: problems, our method is much more numerically efficient than methods

14: based on full gradient information.

15: 

16: \keywords{Coordinate descent \and  composite objective function \and

17: linearly coupled constraints \and randomized algorithms \and

18: convergence rate ${\cal O}(1/\epsilon)$.}

19: \end{abstract}

20: