Numerical methods of nonlinear optimization

Numerical mathematical programming methods (65K05) Introductory exposition (textbooks, tutorial papers, etc.) pertaining to operations research and mathematical programming (90-01) Computational methods for problems pertaining to operations research and mathematical programming (90-08) Nonlinear programming (90C30)

Review text

The book is based on a series of lectures on optimization theory as presented by the author during the last 15 years. It is written mainly for students in mathematics and related fields, who want to become familar with modern optimization methods and their numerical features. The book is organized as follows: (1) Introduction: The author introduces the optimization problem and motivates it by a few examples. - (2) Theory: The optimality criteria for the differentiable and convex optimization problems are stated and proved. Moreover a few results on sensitivity analysis are shown. - (3) Methods: The main chapter of the book describes a large variety of different numerical methods for the following problem classes: (a) Unconstrained minimization: The classical unconstrained methods are introduced, e.g. steplength determination, gradient related and conjugate gradient methods, Newton- and quasi-Newton-methods. Moreover methods for solving nonlinear least squares problems are discussed. (b) Quadratic Programming: The main emphasis is set on methods developed in the last ten years, and some new ones, e.g. the dual method of Goldfarb and Idnani, or of the more modern polynomial algorithms. (c) Linear programming: An a few pages the most important numerical algorithms are described, e.g. the simplex method and numerically stable variants. Also new methods of Barnes, Khachiyan, and Karmarkar are treated. (d) Projection and reduction methods: These methods are important for solving problems, where the feasibility of intermediate iterates is required. A couple of approaches are discussed, for example the generalized reduced gradient method or the projection method of Bertsekas. (e) Penalty and multiplier methods: Although of less practical interest, these methods are described for the matter of completeness and because of their importance for understanding the SQP-methods. The algorithms of Hestenes, Rockafellar and some other variants are analyzed from the theoretical and numerical point of view. (f) Sequential quadratic programming methods: The method is probably the most frequently used practical optimization algorithm for solving the general, smooth nonlinear programming problem. The author outlines the method in detail and discussed several theoretical aspects. Most of the theorems, particularly convergence theorems of the algorithms under consideration, are proved in a mathematical rigorous way. Each section of the book contains a couple of exercises and addional remarks on further investigations of other authors or more specialized problems not treated in detail. Since about 50\% of the references were published between 1982 and 1992, the book represents the state-of-the-art in differentiable, small scale nonlinear programming particularly from the theoretical point of view. Thus the book is highly useful also for specialists who are looking for references, hints or alternatives. Whenever possible and reasonable, the theoretical problem is illustrated by a sketch or the actual example by a graphical plot. The performance of the most important numerical algorithms is shown by test runs, where the output listings of the underlying software is reproduced. Some of the codes are taken from software libraries, some others, particularly the SQP-method, were implemented by the author and his students. The book contains numerous remarks concerning the numerical stability and implementation of optimization algorithms.

Recommendations

Cited in

(47)

This page was built for publication: Numerical methods of nonlinear optimization

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5287074)