Why many travelling salesman problem instances are easier than you think

While there are many inexact heuristics for generating high quality solutions to the Travelling Salesman Problem, our understanding of why these methods are effective and efficient is still limited. This paper looks at two population based heuristics: the EAX algorithm and the Mixing GA using partition crossover. We show that the local optima used to construct the initial population are also sampling edges found in the global optimum at an extremely high rate: in the majority of TSP instances, the number of global edges in the initial population is more than 73%. Next, we look at how recombination operators increase the representation of edges from the global optimum in the population, or increase the number of global edges in the best solutions in the population. We also look at TSP instances that are more difficult to solve, and again we find that edge frequency information can help to explain algorithm performance. Finally we use these result to suggest new strategies for generating high quality solutions for Travelling Salesman Problems.

[1]  Thomas Stützle,et al.  On the empirical scaling of running time for finding optimal solutions to the TSP , 2018, J. Heuristics.

[2]  L. Darrell Whitley,et al.  Efficient Recombination in the Lin-Kernighan-Helsgaun Traveling Salesman Heuristic , 2018, PPSN.

[3]  Gregory Gutin,et al.  The traveling salesman problem , 2006, Discret. Optim..

[4]  Thomas Stützle,et al.  On the empirical scaling of run-time for finding optimal solutions to the travelling salesman problem , 2014, Eur. J. Oper. Res..

[5]  G. Reinelt The traveling salesman: computational solutions for TSP applications , 1994 .

[6]  Andrew B. Kahng,et al.  A new adaptive multi-start technique for combinatorial global optimizations , 1994, Oper. Res. Lett..

[7]  Doug Hains,et al.  Revisiting the big valley search space structure in the TSP , 2011, J. Oper. Res. Soc..

[8]  M Dorigo,et al.  Ant colonies for the travelling salesman problem. , 1997, Bio Systems.

[9]  Keld Helsgaun,et al.  An effective implementation of the Lin-Kernighan traveling salesman heuristic , 2000, Eur. J. Oper. Res..

[10]  L. Darrell Whitley,et al.  Building a better heuristic for the traveling salesman problem: combining edge assembly crossover and partition crossover , 2017, GECCO.

[11]  William J. Cook,et al.  The Traveling Salesman Problem: A Computational Study , 2007 .

[12]  J. Lenstra,et al.  In Pursuit of the Traveling Salesman: Mathematics at the Limits of Computation , 2016 .

[13]  Shigenobu Kobayashi,et al.  A Powerful Genetic Algorithm Using Edge Assembly Crossover for the Traveling Salesman Problem , 2013, INFORMS J. Comput..

[14]  Thomas Stützle,et al.  On the empirical time complexity of finding optimal solutions vs proving optimality for Euclidean TSP instances , 2015, Optim. Lett..

[15]  Doug Hains,et al.  A Hybrid Genetic Algorithm for the Traveling Salesman Problem Using Generalized Partition Crossover , 2010, PPSN.

[16]  Shigenobu Kobayashi,et al.  Edge Assembly Crossover: A High-Power Genetic Algorithm for the Travelling Salesman Problem , 1997, ICGA.

[17]  Gabriela Ochoa,et al.  Deconstructing the Big Valley Search Space Hypothesis , 2016, EvoCOP.

[18]  Thomas Stützle,et al.  On the Empirical Scaling Behaviour of State-of-the-art Local Search Algorithms for the Euclidean TSP , 2015, GECCO.

[19]  Keld Helsgaun,et al.  General k-opt submoves for the Lin–Kernighan TSP heuristic , 2009, Math. Program. Comput..

[20]  Thomas Stützle,et al.  Analysing the Run-Time Behaviour of Iterated Local Search for the Travelling Salesman Problem , 2002 .