Simulated annealing for the QAP-Optimal tradeoff between simulation time and solution quality

This paper reports on the optimization of solution quality for a Simulated Annealing algorithm applied to the Quadratic Assignment Problem. The optimization is performed by varying two of the parameters in the algorithm: the size of the ensemble (i.e. the number of simulations), and the length of each simulation. The variation is performed in such a way that the algorithm always uses the same amount of time for a given problem size. It is found that the simulation length is indeed optimizable, but also that a rather broad spectrum of values produce a near-optimal solution quality. Generally, it seems that the optimal simulation length is rather short compared to results found for other optimization problems.