1: \begin{abstract}

2: We propose first order algorithms for convex optimization problems

3: where the feasible set is described by a large number of convex inequalities

4: that is to be explored by subgradient projections. The first algorithm

5: is an adaptation of a subgradient algorithm, and has convergence rate

6: $1/\sqrt{k}$. The second algorithm has convergence rate $1/k$ when

7: (1) one has linear metric inequality in the feasible set, (2) the

8: objective function is strongly convex, differentiable and has Lipschitz

9: gradient, and (3) it is easy to optimize the objective function on

10: the intersection of two halfspaces. This second algorithm generalizes

11: Haugazeau's algorithm. The third algorithm adapts the second algorithm

12: when condition (3) is dropped. We give examples to show that the second

13: algorithm performs poorly when the objective function is not strongly

14: convex, or when the linear metric inequality is absent.  

15: \end{abstract}

16: