A constraint linearization method for nondifferentiable convex minimization (Q1099787): Difference between revisions

A solution method for nonsmooth convex problems is investigated. In the considered problems the objective function and one constraint are assumed to be nonsmooth, while all further constraints are linear inequalities. The proposed solution method combines principles of bundle methods and of successive quadratic programming approximation methods. The search directions are obtained as solutions of quadratic subproblems, in which the linearizations of the nonsmooth functions are formulated on the basis of the corresponding subgradients. In the line-search phase an exact penalty function is used as merit function. After the description of the method convergence properties are given. For the polyhedral case the author proves that the algorithm will be finite. The paper ends with some numerical test results.