A local convergence property of primal-dual methods for nonlinear programming (Q948957): Difference between revisions

The authors consider a nonlinear, finite dimensional, smooth, constraint optimization problem, with general equalities and nonnegativity constraints. A standard primal-dual method for solving this problem is introduced. In each iteration the complementary condition for the inequalities is perturbed by a small parameter. During the Newton-type method it gradually is reduced to zero such that in the limit the problem solution fulfills the first order conditions of the optimization problem. Under standard nondegeneracy assumptions, a good first guess, a suitable line-search that guarantees strict feasibility the authors prove that the iterates converge to the optimum with the same rate as the penalty parameter converges to zero. However, to guarantee the iterates to be asymptotically tangent to the central trajectory, the convergence rate of the small parameter must not be faster than superlinear as shown by a simple counter example.