Simulated Annealing and Tabu Search in the Long Run: a Comparison on Qap Tasks

Simulated Annealing (SA) and Tabu Search (TS) are compared on the Quadratic Assignment Problem. A recent work on the same benchmark suite argued that SA could achieve a reasonable solution quality with fewer function evaluations than TS. The discussion is extended by showing that the conclusions must be changed if the task is hard or a very good approximation of the optimal solution is desired, or if CPU time is the relevant parameter. In addition, a recently proposed version of TS (the Reactive Tabu Search) solves the problem of nding the proper list size with an automatic memory-based reaction mechanism.