Global optimization of signomial programming problems

In this presentation, an overview of a signomial global optimization algorithm is given. As the name indicates, the algorithm can be used to solve mixed integer nonlinear programming problems containing signomial functions to global optimality. The method employs singlevariable power and exponential transformations for convexifying the nonconvex signomial functions termwise. By approximating the transformations using piecewise linear functions, piecewise convex underestimators for the nonconvex signomial functions as well as a relaxed convex problem can be obtained. In the algorithm, the approximations resulting from the piecewise linear functions are subsequentially improved resulting in a set of subproblems whose optimal solution converges to that of the original nonconvex problem. Finally, some recent theoretical results regarding the underestimation properties of the convexified signomial terms obtained using different transformations are also given