Pattern search methods for finite minimax problems (Q2380874)

Many problems of interest and applications in the real world can be modeled as finite minimax problems. This class of problems arises for instance in the solution of approximation problems, systems of nonlinear equations, nonlinear programming problems, and multiobjective problems. Many algorithms have been developed for the solution of finite minimax problems, which require the knowledge of first and second-order derivatives of the functions involved in the definition of the problem. Unfortunately, in some engineering applications, the first order derivatives cannot be explicitly calculated or approximated. Finite minimax problems have the valuable feature that they can be approximated by smooth problems. This smooth approximation to the minimax problem can be achieved by using different techniques. In particular, the authors consider an approximation approach based on a so-called smoothing function, the exponential penalty function of \textit{B. W. Kort} and \textit{D. P. Bertsekas} [SIAM J. Control Optimization 14, 268--294 (1976; Zbl 0332.90035)], which depends on a precision parameter. The proposed pattern search methods for finite minimax problems are based on a sampling of the smooth function along a set of suitable search directions and on an updating rule for the step-control parameter. The global convergence results are obtained, too.