As the explicit name indicates, the methods of descent, or local search, consist in "sliding" along the objective function until a local optimum is found, that is to say where the descent is no longer possible. There are several kinds of descent method which we will describe here.
The algorithm is as follows:
The algorithm is mainly the same as for the simple descent, only a selection criterion of a neighboring solution is modified:
We therefore choose the neighboring solution offering the best improvement of the current solution.
Multi-start descent performs multiple instances of the single descent or greater descent problem. The algorithm is as follows: