Finite master programs in regularized stochastic decomposition (Q1341566)

In a series of papers the authors have suggested a stochastic decomposition method for the approximate solution of stochastic linear programs with recourse. Since the approximate master program is a piecewise linear function, the algorithm leads to the growth of the master program's dimension without bounds. In this paper the authors suggest to add to the master program a quadratic regularizing term. Then the master program can be limited to a finite number of cuts (at most to \(n_1+ 3\) cuts, where \(n_1\) is the number of first stage decision variables), while ensuring accumulation at an optimal solution.