Linearity embedded in nonconvex programs

Nonconvex programs involving bilinear terms and linear equality constraints often appear more nonlinear than they really are. By using an automatic symbolic reformulation we can substitute some of the bilinear terms with linear constraints. This has a dramatically improving effect on the tightness of any convex relaxation of the problem, which makes deterministic global optimization algorithms like spatial Branch-and-Bound much more efficient when applied to the problem.