2-level polytopes naturally appear in several areas of pure and applied mathematics, including combinatorial optimization, polyhedral combinatorics, communication complexity, and statistics. In this paper, we present a study of some 2-level polytopes arising in combinatorial settings. Our first contribution is proving that f0(P )f d−1 (P ) ≤ d2 d+1 for a large collection of families of such polytopes P . Here f0(P ) (resp. f d−1 (P )) is the number of vertices (resp. facets) of P , and d is its dimension. Whether this holds for all 2-level polytopes was asked in [7], and experimental results from [16] showed it true for d ≤ 7. The key to most of our proofs is a deeper understanding of the relations among those polytopes and their underlying combinatorial structures. This leads to a number of results that we believe to be of independent interest: a trade-off formula for the number of cliques and stable sets in a graph; a description of stable matching polytopes as affine projections of certain order polytopes; and a linear-size description of the base polytope of matroids that are 2-level in terms of cuts of an associated tree.
IntroductionLet P ⊆ R d be a polytope. We say that P is 2-level if, for each facet F of P , all the vertices of P that are not vertices of F lie in the same translate of the affine hull of F . Equivalently, P is 2-level if and only if it has theta-rank 1 [20], or all its pulling triangulations are unimodular [50], or it has a slack matrix with entries in {0, 1} [7]. Those last three definitions appeared in papers from the semidefinite programming, statistics, and polyhedral combinatorics communities respectively, showing that 2-level polytopes naturally arise in many areas of mathematics. 2-level polytopes generalize Birkhoff [53], Hanner [28], and Hansen polytopes [29], order polytopes and chain polytopes of posets [49], spanning tree polytopes of series-parallel graphs [24], stable matching polytopes [27], some min up/down polytopes [37], and stable set polytopes of perfect graphs [10].