Faster MIP Solutions via New Node

Rules Daniel T. Wojtaszek John W. Chinneck dtwojtas@sce.carleton.ca chinneck@sce.carleton.ca Systems and Computer Engineering Carleton University Ottawa, Ontario K1S 5B6 Canada November 12, 2009 Abstract When a branch and bound method is used to solve a linear mixed integer program (MIP), the order in which the nodes of the branch and bound tree are explored signi cantly a ects how quickly the MIP is solved. In this paper, new methods are presented that exploit correlation and distribution characteristics of branch and bound trees to trigger backtracking and to choose the next node to solve when backtracking. A new method is also presented that determines when the cost of using a node selection method outweighs its bene t, in which case it is abandoned in favor of a simpler method. Empirical experiments show that these proposed methods outperform the current state of the art.