1: \begin{abstract}

2: The problem of finding a vector with the fewest nonzero elements 

3: that satisfies an underdetermined system of linear equations is 

4: an NP-complete problem that is typically solved numerically via 

5: convex heuristics or nicely-behaved nonconvex relaxations.  In   

6: this work we consider elementary methods based on projections 

7: for solving a {\em sparse feasibility} problem  

8: without employing convex heuristics.  In a recent paper Bauschke, 

9: Luke, Phan and Wang (2014) showed that, locally, the fundamental 

10: method of alternating projections {\em must} converge linearly to 

11: a solution to the sparse feasibility problem with an affine constraint. 

12: In this paper we apply different analytical tools that allow us to 

13: show {\em global} linear convergence of  alternating projections  

14: under familiar constraint qualifications.

15: These analytical tools can also be applied to other algorithms.  This is 

16: demonstrated with the prominent Douglas-Rachford algorithm where 

17: we establish local linear convergence of this method applied 

18: to the sparse affine feasibility problem.  

19: % Additionally, we show that stronger 

20: % assumptions than the sufficient conditions for 

21: % linear convergence of alternating projections  are in fact {\em necessary} for mere convergence

22: % of another prominent algorithm, the Douglas-Rachford algorithm.  These 

23: % necessary assumptions are {\em not} satisfied for most 

24: % sparsity problems of interest.  Still, we recover positive results locally for

25: % the Douglas-Rachford algorithm in some instances of the sparse feasibility 

26: % problem with respect to the {\em shadows} of the iterates. 

27: \end{abstract}

28: