It is known a class of methods that do not explicitly use
derivatives - direct search methods for unconstrained
optimization. Recent researches have shown the global
convergence of pattern search algorithms (a class of
direct search methods) for the case when function f is
continuously differentiable (Aude & Dennis, 2003,
Torczon, 1997). The advantage of the multidimensional
bisection method presenting in this paper is that it
convergence does not require an assumption about
differentiability of function f and method allows to find
the minimizer of nonsmooth functions.
In point 2 we describe the multidimensional bisection
method (MBM) for minimizing function over simplex. In
point 3, the details of the extension of MBM for solving
the unconstrained minimization problem are presented. In
point 4 some numerical results illustrate the robustness of
the method.