Core Concepts
Proving optimal complexity bounds for stable set and knapsack problems in MIP formulations.
Abstract
The article discusses lower bounds on the complexity of mixed-integer programs for stable set and knapsack problems. It presents new results improving previous bounds, focusing on the stable set problem over n-node graphs and the knapsack problem. The study shows that standard MIP formulations already use an optimal number of integer variables. The proof involves information-theoretic methods and extends to approximate extended formulations. The analysis highlights the challenges in determining the number of integer variables needed in small-size MIP formulations for combinatorial optimization problems.
Stats
Standard mixed-integer programming formulations for the stable set problem require n integer variables.
A family of n-node graphs requires Ω(n/ log2 n) integer variables for polynomial-size MIP formulation.
Every subexponential-size MIP formulation for matching or traveling salesman problems has Ω(n/log n) integer variables.