1: \begin{abstract}

2: This paper improves the algorithms based on supporting halfspaces

3: and quadratic programming for convex set intersection problems in

4: our earlier paper in several directions. First, we give conditions

5: so that much smaller quadratic programs (QPs) and approximate projections

6: arising from partially solving the QPs are sufficient for multiple-term

7: superlinear convergence for nonsmooth problems. Second, we identify

8: additional regularity, which we call the second order supporting hyperplane

9: property (SOSH), that gives multiple-term quadratic convergence. Third,

10: we show that these fast convergence results carry over for the convex

11: inequality problem. Fourth, we show that infeasibility can be detected

12: in finitely many operations. Lastly, we explain how we can use the

13: dual active set QP algorithm of Goldfarb and Idnani to get useful

14: iterates by solving the QPs partially, overcoming the problem of solving

15: large QPs in our algorithms.

16: \end{abstract}

17: