Statistical Physics of the Travelling Salesman Problem

Abstract

If one places N cities on a continuum in an unit area, extensive numerical results and their analysis (scaling, etc.) suggest that the best normalized optimal travel distance becomes 0.72 for the Euclidean metric and 0.92 for the Manhattan metric. The analytic bounds, we discuss here, give 0.5 and 0.92 as the lower and upper bounds for the Euclidean metric, and 0.64 and 1.17 for the Manhattan metric. When the cities are randomly placed on a lattice with concentration p, we find that the normalized optimal travel distance vary monotonically with p. For p=1, the values in both Euclidean and Manhattan metric are 1, and as p tends to zero, the values are 0.72 and 0.92 in the Euclidean and Manhattan metrics respectively.The problem is trivial for p=1 but it reduces to the continuum TSP as p tends to zero. We do not get any irregular behaviour at any intermediate point, e.g., the percolation point. The crossover from the triviality to the NP- hard problem seems to occur at p<1.