Differentiable Local Barrier-penalty Paths Barrier-penalty Paths
نویسنده
چکیده
Perturbations of Karush-Kuhn-Tucker conditions play an important role for primal-dual interior point methods. Beside the usual logarithmic barrier various further techniques of sequential unconstrained minimization are well known. However other than logarithmic embeddings are rarely studied in connection with Newton path-following methods. A key property that allows to extend the class of methods is the existence of a locally Lipschitz continuous path leading to a primaldual solution of the KKT-system. In this paper a rather general class of barrier/penalty functions is studied. In particular, under LICQ regularity and strict complementarity assumptions the differentiability of the path generated by any choice of barrier/penalty functions from this class is shown. This way equality as well as inequality constraints can be treated directly without additional transformations. Further, it will be sketched how local convergence of the related Newton path-following methods can be proved without direct applications of self-concordance properties. keywords: Perturbed KKT-systems, general barrier-penalty embedding, differentiable path, path-following methods, interior point methods 1 Barrier/penalty functions and primal-dual paths Barrier/penalty methods and its path-following variants form an important class of numerical methods for constrained optimization problems via a family of unconstrained ones (cf. [2], [8], [9]). While large classes of classical ∗[email protected] TU Dresden, Institut für Numerische Mathematik, D01062 Dresden, Germany; Paper written with financial support of DFG grant GR 1777/2-2
منابع مشابه
Differentiable Local Barrier-Penalty Paths
Perturbations of Karush-Kuhn-Tuckerconditions play an importantrole forprimaldual interior point methods. Beside the usual logarithmic barrier various further techniques of sequential unconstrained minimization are well known. However other than logarithmic embeddings are rarely studied in connection with Newton path-following methods. A key property that allows to extend the class of methods i...
متن کاملPenalty and Barrier Methods for Convex Semidefinite Programming
In this paper we present penalty and barrier methods for solving general convex semidefinite programming problems. More precisely, the constraint set is described by a convex operator that takes its values in the cone of negative semidefinite symmetric matrices. This class of methods is an extension of penalty and barrier methods for convex optimization to this setting. We provide implementable...
متن کاملPenalty/barrier Path-following in Linearly Constrained Optimization
In the present paper rather general penalty/barrier path-following methods (e.g. with p-th power penalties, logarithmic barriers, SUMT, exponential penalties) applied to linearly constrained convex optimization problems are studied. In particular, unlike in previous studies [1, 11], here simultaneously different types of penalty/barrier embeddings are included. Together with the assumed 2nd ord...
متن کاملAn interior point method with a primal-dual quadratic barrier penalty function for nonlinear semidefinite programming
In this paper, we consider an interior point method for nonlinear semidefinite programming. Yamashita, Yabe and Harada presented a primal-dual interior point method in which a nondifferentiable merit function was used. By using shifted barrier KKT conditions, we propose a differentiable primal-dual merit function within the framework of the line search strategy, and prove the global convergence...
متن کاملMathematical Programming Manuscript No. Smooth Exact Penalty and Barrier Functions for Nonsmooth Optimization
For constrained nonsmooth optimization problems, continuously diierentiable penalty functions and barrier functions are given. They are proved exact in the sense that under some nondegeneracy assumption, local optimizers of a nonlinear program are also optimizers of the associated penalty or barrier function. This is achieved by augmenting the dimension of the program by a variable that control...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2005