Generalized saddle version of the level method. (Q1395228)

The author describes an iterative algorithm for finding a saddle point of a convex-concave function whose effective set \(G^1\) is contained in the Cartesian product \(G\) of polyhedra. When \(G^1=G\) the algorithm reduces to the saddle version of the level method. The convergence rate of the algorithm is estimated. A problem that can be solved by this algorithm arises in implementations of the primal-dual decomposition approach in linear or convex programming.