Relaxations of Factorable Programs
In Chapter 2, we developed a theory of convex extensions and formalized a technique for building tight relaxations. However, the size of the resulting relaxations can be exponential in the number of variables. In this chapter, we present a slightly modified version of the factorable programming technique due to McCormick (1976) that, when used in conjunction with our relaxation techniques, constructs relaxations that are tight as well as manageable in size and can be generated in an automated fashion. To enable the use of efficient LP software, in Section 4.2 we build linear programming relaxations of the nonlinear convex relaxations using the sandwich algorithm (Rote 1992).
KeywordsConvex Function Supporting Line Symmetric Difference Projective Distance Algorithm Relax
Unable to display preview. Download preview PDF.