A splitting technique for discrete search based on convex relaxation

In mixed integer programming branching methods are a powerful and frequently employed tool. This paper presents a branching strategy for the case that the integer constraints are associated with a finite set of points in a possibly multidimensional space. We use the knowledge about this discrete set represented by its minimum spanning tree and find a splitting based on convex relaxation. Typical applications include design optimization problems where design points specifying several discrete choices can be considered as such discrete sets. c ©2009 World Academic Press, UK. All rights reserved.