Global Optimization for Generalized Geometric Programs with Mixed Free-Sign Variables

Many optimization problems are formulated as generalized geometric programming (GGP) containing signomial terms f(X)·g(Y), where X and Y are continuous and discrete free-sign vectors, respectively. By effectively convexifying f(X) and linearizing g(Y), this study globally solves a GGP with a lower number of binary variables than are used in current GGP methods. Numerical experiments demonstrate the computational efficiency of the proposed method.