Difference of convex solution of quadratically constrained optimization problems

Recently, global optimization algorithms have been proposed for solving nonconvex programs whose objective and constraint functions are expressed as the difference of two convex functions. A surprisingly large class of functions can be decomposed in this way, so these algorithms can be viewed as general procedures for global optimization. Often, a given problem can have more than one difference of convex (d.c.) formulation. This paper investigates the impact of alternative d.c. formulations of a nonconvex program. In particular, we investigate the computational performance of two d.c. programming algorithms when applied to two d.c. formulations of quadratically constrained quadratic optimization models. Computational results suggest that the performance of d.c. techniques can be significantly influenced by the nature of the d.c. decomposition.