Sciweavers

COR
2011

Faster integer-feasibility in mixed-integer linear programs by branching to force change

12 years 11 months ago
Faster integer-feasibility in mixed-integer linear programs by branching to force change
Branching in mixed-integer (or integer) linear programming requires choosing both the branching variable and the branching direction. This paper develops a number of new methods for making those two decisions either independently or together with the goal of reaching the first integer-feasible solution quickly. These new methods are based on estimating the probability of satisfying a constraint at the child node given a variable/direction pair. The surprising result is that the first integer-feasible solution is usually found much more quickly when the variable/direction pair with the smallest probability of satisfying the constraint is chosen. This is because this selection forces change in many candidate variables simultaneously, leading to an integer-feasible solution sooner. Extensive empirical results are given.
Jennifer Pryor, John W. Chinneck
Added 13 May 2011
Updated 13 May 2011
Type Journal
Year 2011
Where COR
Authors Jennifer Pryor, John W. Chinneck
Comments (0)