An inexact Newton method combined with Hestenes multipliers' scheme for the solution of Karush-Kuhn-Tucker systems (Q2572712)

This work is concerned with the numerical solution of large scale nonlinear programming (NLP) problems arising, for instance, from the discretization of optimal control problems with partial differential equations and control and state constraints. The numerical solution of this large scale NLP problem can be determined by solving the constrained system of nonlinear equations obtained by the Karush-Kuhn-Tucker optimality conditions of the problem. The nonlinear system is solved by an inexact Newton scheme, that includes the strategy of the interior-point method for the treatment of the constrained variables. For solving the linear system arising at each step of the iterative scheme, the authors used an inner iterative method devising an adaptive stopping rule that allows to avoid unnecessary inner iterations for the initial outer iterations and, at the same time, assures global convergence and local superlinear convergence of the whole method. Under suitable conditions that assure the nonsingularity of the inner linear system and the boundness of its inverse, the system can be viewed as the Lagrange necessary conditions for the minimum point of a convex quadratic programming problem and it can be efficiently solved by Hestenes multipliers' method. A deep analysis on the choices of the parameters of the method (perturbation and damping parameters) is done.