A GENERALIZED CONVERGENCE RESULT FOR THE GRAPH-BASED ANT SYSTEM METAHEURISTIC

  • Authors:
  • Walter J. Gutjahr

  • Affiliations:
  • Department of Statistics and Decision Support Systems, University of Vienna, Vienna, Austria, E-mail: walter.gutjahr@univie.ac.at

  • Venue:
  • Probability in the Engineering and Informational Sciences
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

It is shown that on fairly weak conditions, the current solutions of a metaheuristic following the ant colony optimization paradigm, the graph-based ant system, converge with a probability that can be made arbitrarily close to unity to one element of the set of optimal solutions. The result generalizes a previous result by removing the very restrictive condition that both the optimal solution and its encoding are unique (this generalization makes the proof distinctly more difficult) and by allowing a wide class of implementation variants in the first phase of the algorithm. In this way, the range of application of the convergence result is considerably extended.