Global optimization of nonconvex factorable programming problems (Q5930738)

In this paper is presented a global optimization approach for solving a class of nonconvex factorable programming problems, that arise in a variety of engineering process control and design problems. McCormick introduced the nonconvex factorable programming problem in 1976 in a different, but equivalent, form. The proposed procedure is a branch and bound algorithm that employs two levels of approximation. First the original nonconvex factorable functions are approximated using appropriately constructed lower/upper bounding polynomial functions. Then an LP relaxation is generated for resulting nonconvex polynomial programming problem via the application of a reformulation-linearization technique. A suitable partitioning process is proposed that induce convergence to a global optimum. The algorithm is illustrated by a numerical example. Also the algorithm was implemented in C++ and tested on a set of fifteen test problems.