Influence of rounding errors on the quality of heuristic optimization algorithms

Search space smoothing and related heuristic optimization algorithms provide an alternative approach to simulated annealing and its variants: while simulated annealing traverses barriers in the energy landscape at finite temperatures, search space smoothing intends to remove these barriers, so that a greedy algorithm is sufficient to find the global minimum. Several formulas for smoothing the energy landscape have already been applied, one of them making use of the finite numerical precision on a computer. In this paper, we thoroughly investigate the effect of finite numerical accuracy on the quality of results achieved with heuristic optimization algorithms. We present computational results for the traveling salesman problem.