Multistart Local Search Continuous Global Optimization Method with a Taboo Step and its Condition for Finding the Global Optimum

We introduce a multistart local search-based method with a taboo step for solving continuous global optimization problems with bound constraints. Since this algorithm has a characteristic taboo step[5, 1995] by removing candidate points that converge to the current local optimum in each iteration, the step enables us to avoid repeated convergence to one of an already known optima in a local search. Since a similar step has been proposed by Ursem[13, 1999], known as the hill-valley step, we show the difference between the hill-valley step and the taboo step. Finally, we show that the algorithm stops after a finite number of iterations and finds the global optimum under certain conditions.