The k-OPT class of moves takes the tour and removes k edges from it, then reglues the pieces in some other way.
Picture...
Theoretically, even in metric TSPs, ratio between and
in the worst case.
Furthermore, it can take O(2n/2) moves to get there!
Good performance in practice, though.