Concepts in Optimizing Simulated Annealing Schedules: An Adaptive Approach for Parallel and Vector Machines

Abstract

Simulated Annealing (Čemy 1983, Kirkpatrick et al. 1983) is a technique which allows to find optimal or near optimal solutions to difficult optimization problems. It has been especially successful in applications to NP-complete or NP-hard problems, which occur in a variety of fields (Garey and Johnson 1979). These include mathematics with many graph problems (e.g. Brelaz 1979, Bonomi and Lutton 1987, Andresen et al. 1988), condensed matter physics, e.g. with the problem of finding the ground state of spin glasses (Ettelaie and Moore 1985), with the problem of solving the Ginzberg-Landau equations (Doria et al. 1989), engineering problems with the design of integrated circuits including the partitioning as well as the wiring problem (Vecchi and Kirkpatrick 1983, Sechen and Sangiovanni-Vincentelli 1985, Siarry et al. 1987), the design of binary sequences with low autocorrelation (Beenker et al. 1985, Bernasconi 1987, 1988), image processing (Carnevali et al. 1985), design of X-ray mirrors (Würtz and Schneider 1989), statistics with the application as a learning paradigm in neural network theory (Bernasconi 1990) and economics for instance with the travelling salesman problem (e.g. Bonomi and Lutton 1984, Kirkpatrick and Toulouse 1985, Hanf et al. 1990). Naturally, these are only some selected examples, since it is not possible here to give reference to the few hundred simulated annealing papers which appeared during the last years.