Classes of Methods

Historically the first global optimization method used is probably multistart, i.e., local minimization is started from several points and the best local minimum found is taken as an estimate of the global minimum.

A lot of global optimization methods have been suggested. The ideas behind these algorithms are fewer than the methods so we may describe a number of classes covering most of the methods. Such a classification is of course not unique, several classification schemes could be used depending on which features that are used in the classification. For the largest class "Search Methods" we base our classification on the strategies local-global-adaptive in choosing points. The classes in bold are examples only.

A crude classification is

    Methods with Guarantees
                Bisection Methods
                Interval Methods

    Search Methods
       Global
                Random Search
                Mode-Seeking
       Adaptive
           Single Working Point Methods
                Simulated Annealing
           Converging Set Methods
                Controlled Random Search
                Genetic Algorithms
       Global-Local
                Multistart
                Clustering Methods

    Bayesian Methods
                P-Algorithms