Global optimization with non-convex constraints. Sequential and parallel algorithms (Q1590262): Difference between revisions

This book gives an account of the approach to global optimization as developed in the group around the first author over the past thirty years. The goal is to minimize a (multi-) objective function of a finite number of variables over a subset defined by inequality constraints. One of the main ideas consists in reducing the dimension to one via space-filling Peano-type curves. The latter technique is improved by simultaneously using several (shifted) Peano scannings. In this way an asynchronous technique for multiprocessor realization becomes available. The book is divided into three parts. The first part is devoted to the development of minimization techniques for the one-dimensional unconstrained case. The objective functions are assumed to be Lipschitzian (or Hölderian) or they are viewed as samples of some known random function. Concerning practical implementations, the Lipschitz constants etc. are adaptively estimated using the data accumulated in the search process. In the second part the techniques from the unconstrained case are extended to the (one-dimensional) constrained case. Here, a specially designed index scheme is introduced which takes separate account of each constraint. The special index scheme is suggested for reducing multiobjective (constrained) problems to unconstrained scalar ones. Specific attention is paid to parallel optimization algorithms and the evaluation of the efficiency of parallelism. The last part of the book deals with dimension reduction (as mentioned in the beginning) together with parallelization techniques.