A Probabilistic Analysis of Local Search

We present a theoretical average-case analysis of a 2-opt algorithm for the traveling salesman problem. First, we give a model which allows us to compute the required number of steps and the distribution of final solutions found by a best improvement algorithm. This model is empirically validated for a restricted version of the 2-opt neighborhood. Secondly, we present a semi-empirical analysis of the average-case performance of an iterated 2-opt and Lin-Kernighan algorithm based on empirically obtained parameters.