Optimization by Simulated Annealing: An Experimental Evaluation; Part I, Graph Partitioning

0030-364X/89/3706·0865 $01.25 © 1989 Operations Research Society of America new problems (even in the absence of deep insight into the problems themselves) and, because of its apparent ability to avoid poor local optima, it offers hope of obtaining significantly better results. These observations, together with the intellectual appeal of the underlying physical analogy, have inspired articles in the popular scientific press (Science 82, 1982 and Physics Today 1982) as well as attempts to apply the approach to a variety of problems, in areas as diverse as VLSI design (Jepsen and Gelatt 1983, Kirkpatrick, Gelatt and Vecchi 1983, Vecchi and Kirkpatrick 1983, Rowan and Hennessy 1985), pattern recognition (Geman and Geman 1984, Hinton, Sejnowski and Ackley 1984) and code generation (EI Gamal et al. 1987), often with substantial success. (See van Laarhoven and Aarts 1987 and Collins, Eglese and Golden 1988 for more up-to-date and extensive bibliographies of applications.) Many of