Direct Search Methods on Parallel Machines 论文

摘要

Direct search methods are methods designed to solve unconstrained minimization problems of the form: $$\mathop {\min }\limits_{x \in {\mathbb{R}^n}} f(x),$$ where f: ℝn → ℝ. These methods are distinguished by the fact that they neither use nor require explicit derivative information; the search for a local minimizer is driven solely by function information. Popular methods in this class include the factorial design algorithm of Box [1], the pattern search algorithm of Hooke and Jeeves [4], and the simplex method of Neider and Mead [5].