Greedy descent outperforms other metaheuristics in routing problems

[Petri Määttä]

Hi,

In the routing library, is it possible in general for the greedy descent heuristic to outperform another metaheuristic (e.g. Guided local search) in the same exact problem instance with the same initial solution? I have found this to be possible for some instances when the solver time limit is low enough, but is this intended behavior or a bug on my end?

I am asking since in the OR-tools user manual it says that "Our meta-heuristics only kick in when we already have reached a local optimum with the Local Search."  Since greedy descent stops at the first local optimum and other metaheuristics kick in after reaching a local optimum, it would seem that the metaheuristics would always be at least as good as greedy descent. Or is the local search different in greedy descent compared to the other metaheuristics, which could give this result?

Regards,

Petri Määttä