1: \begin{abstract}

2: We address the minimization of a smooth objective function under an $\ell_0$-constraint and simple convex constraints.

3: When the problem has no constraints except %for 

4: the $\ell_0$-constraint,

5: %there are 

6: some 

7: efficient algorithms 

8: are available; for example, Proximal DC (Difference of Convex functions) Algorithm (PDCA) %are proposed

9: repeatedly %computes 

10: evaluates closed-form solutions of convex subproblems, 

11: %which leads 

12: leading to a stationary point of the $\ell_0$-constrained problem. 

13: However, when the problem has additional convex constraints, they become inefficient

14: because it is difficult to obtain closed-form solutions of the %resulting 

15: associated subproblems. 

16: In this paper, we reformulate the problem %using 

17: by employing 

18: a new DC representation of the $\ell_0$-constraint, 

19: %by the difference of convex (DC) functions, which 

20: so that %the 

21: PDCA can %retains 

22: retain the efficiency %of PDCA 

23: by reducing its subproblems to the projection operation onto a convex set. 

24: Moreover, inspired by the Nesterov's acceleration technique %in the context of 

25: for proximal methods, we propose the Accelerated PDCA (APDCA), which %has 

26: attains the optimal convergence rate if applied to convex programs, and performs well in numerical

27: experiments. %terms.

28: \end{abstract}

29: