1: \begin{abstract}

2: In this paper, we consider a framework of projected gradient iterations for linear programming (LP) and an augmented lagrangian two-stage algorithm for strongly convex quadratic programming (SCQP). Based on the framework of projected gradient, LP problem is transformed to a finite number of SCQP problems. Furthermore, we give an estimate of the number of the SCQP problems. We use augmented lagrangian method (ALM) to solve SCQP and each augmented lagrangian subproblem is solved by a two-stage algorithm exactly, which ensures the superlinear convergence of ALM for SCQP. The two-stage algorithm consists of the accelerated proximal gradient algorithm as the first stage algorithm which provides an approximate solution, and a simplified parametric active-set method as the second stage algorithm which gives an exact solution. Moreover, we improve the parametric active-set method by introducing a sorting technique to update the cholesky factorization. Finally, the numerical experiments on randomly generated and real-world test problems indicate that our algorithm is effective, especially for random problems.

3: \keywords{linear programming\and strongly convex quadratic programming\and augmented lagrangian method\and proximal gradient method \and parametric active-set method}

4: % \PACS{PACS code1 \and PACS code2 \and more}

5: \subclass{90C05 \and 90C20 }

6: \end{abstract}