抄録
By focusing on the two basic search functions, diversification and intensification, the origin of the success of finite-time optimization by simulated annealing (SA) is investigated on the traveling salesman problems. Two and one additional experiments are designed with the help of the methods devised for the studies on liquid and glass. The present experimental analyses show the existence of effective temperature again; in the search process of the Metropolis algorithm running at this temperature, a successive interbasin transition in a downward direction effectively lasts until the end of observation, that is, a good intensification characteristic appears on the observation time scale. In the optimization process of SA, this effective relaxation dynamics and the resulting good performance are not only dependent on but also sensitive to the search around the effective temperature. This influential temperature is determined from the temperature dependence of the Deborah number, which is used to identify glass transition.
