We identify ejection chain methods for the traveling salesman problem
based on a special reference structure for generating constructions re
lated to alternating paths. Computational tests show that the method p
erforms very effectively, obtaining generally better solutions than im
proved versions of the Lin-Kernighan method within the same time frame
. Our approach, which currently has a simple tabu search guidance comp
onent at a local level, also has the potential to be combined in more
advanced ways with metaheuristics such as genetic algorithms, simulate
d annealing and tabu search.