A Quadratic Approximation Method Based on Augmented Lagrangian Functions for Nonconvex Nonlinear Programming Problems

857 22 871 3 1727 document 871 WP-78-061.pdf application/pdf 0a6347eafc47912ce25453efa3232f2d MD5 1796034 2016-01-15 01:44:50 https://pure.iiasa.ac.at/id/eprint/857/1/WP-78-061.pdf 857 1 1 text en public

WP-78-061.pdf

22 4bfef553842965f71420456aaf6706b3652cd18b 27861 2 39593 document 27861 lightbox.jpg image/png 4f07f45e7ac2828253db40cfbc0e3cff MD5 12757 2016-01-15 21:31:07 https://pure.iiasa.ac.at/id/eprint/857/2/lightbox.jpg 857 2 2 image/png other Thumbnails conversion from text to thumbnail_lightbox en public

lightbox.jpg

preview.jpg

medium.jpg

small.jpg

indexcodes.txt

WP-78-061.pdf

http://eprints.org/relation/isVersionOf https://pure.iiasa.ac.at/id/document/871 http://eprints.org/relation/isVolatileVersionOf https://pure.iiasa.ac.at/id/document/871 http://eprints.org/relation/isCoversheetVersionOf https://pure.iiasa.ac.at/id/document/871 archive 351 disk0/00/00/08/57 2016-01-15 01:44:49 2021-08-27 17:08:51 2016-01-15 01:44:49 monograph show 2 Wierzbicki A.P. 7380 A Quadratic Approximation Method Based on Augmented Lagrangian Functions for Nonconvex Nonlinear Programming Problems pub iis_das iis_met prog_sds The paper describes an algorithm for solving nonlinear programming problems of fairly general type, in particular nonconvex problems that can be convexified via augmented Lagrarlgian functions. The algorithm consists of two phases. In the first phase, a point in a rather large neighborhood of an optimal solution is crudely but effectively estimated by a shifted-increased penalty function algorithm. In the second phase, a saddle point of an augmented Lagrangian function and thus an optimal solution together with corresponding Lagrange multipliers are found by a rapidly convergent method of successive quadratic approximations. The switch between these two phases is adaptive. According to the existing experience in nonlinear programming algorithms, the proposed algorithm combines the best local effectiveness of a quadratic approximation method with the global robustness of a penalty method; due to the convexifying properties of augmented Lagrangian functions, the algorithm can solve nonconvex problems which satisfy second-order sufficient conditions of optimality at the solution. For the sake of a clear presentation, a short review of some basic facts in the theory of Lagrangian functions, quadratic approximations, penalty functions and augmented Lagrangian functions is given in the first part of the paper. Then the quadratic approximations for augmented Lagrangeans are discussed in detail, in particular, in the case when these functions are not twice differentiable which corresponds to the lack of strict complementarity at the optimal solution. The double-phase algorithm is presented and commented upon. The proofs of convergence of the algorithm are given. 1978 published WP-78-061 WP-78-061 10 1663 public working_paper IIASA, Laxenburg, Austria 52 FALSE info:eu-repo/semantics/book