Lower Bounds for Approximating the Matching Polytope

TL;DR

This paper establishes tight lower bounds on the size of extended formulations approximating the matching polytope within a factor of (1+ε), revealing fundamental complexity limits for such linear programs.

Contribution

It proves new lower bounds for approximate matching polytope formulations that are tight and connects these bounds to the non-negative rank of a specific matrix.

Findings

Lower bounds are tight and match known upper bounds.

Bound depends exponentially on 1/ε for approximation.

Connects polytope approximation complexity to matrix non-negative rank.

Abstract

We prove that any extended formulation that approximates the matching polytope on $n$-vertex graphs up to a factor of $(1+\varepsilon)$ for any $\frac2n \le \varepsilon \le 1$ must have at least $\binom{n}{{\alpha}/{\varepsilon}}$ defining inequalities where $0<\alpha<1$ is an absolute constant. This is tight as exhibited by the $(1+\varepsilon)$ approximating linear program obtained by dropping the odd set constraints of size larger than $({1+\varepsilon})/{\varepsilon}$ from the description of the matching polytope. Previously, a tight lower bound of $2^{\Omega(n)}$ was only known for $\varepsilon = O\left(\frac{1}{n}\right)$ [Rothvoss, STOC '14; Braun and Pokutta, IEEE Trans. Information Theory '15] whereas for $\frac2n \le \varepsilon \le 1$, the best lower bound was $2^{\Omega\left({1}/{\varepsilon}\right)}$ [Rothvoss, STOC '14]. The key new ingredient in our proof is a close connection to the non-negative rank of a lopsided version of the unique disjointness matrix.