Optimal Vertex-Cut Sparsification of Quasi-Bipartite Graphs

Abstract

In vertex-cut sparsification, given a graph $G=(V,E)$ with a terminal set $T\subseteq V$, we wish to construct a graph $G^{\prime }=(V^{\prime },E^{\prime })$ with $T\subseteq V^{\prime }$, such that for every two sets of terminals $A,B\subseteq T$, the size of a minimum $(A,B)$-vertex-cut in $G^{\prime }$ is the same as in $G$. In the most basic setting, $G$ is unweighted and undirected, and we wish to bound the size of $G^{\prime }$ by a function of $k=|T|$. Kratsch and Wahlström [JACM 2020] proved that every graph $G$ (possibly directed), admits a vertex-cut sparsifier $G^{\prime }$ with $O(k^3)$ vertices, which can in fact be constructed in randomized polynomial time. We study (possibly directed) graphs $G$ that are quasi-bipartite, i.e., every edge has at least one endpoint in $T$, and prove that they admit a vertex-cut sparsifier with $O(k^2)$ edges and vertices, which can in fact be constructed in deterministic polynomial time. In fact, this bound naturally extends to all graphs with a small separator into bounded-size sets. Finally, we prove information-theoretically a nearly-matching lower bound, i.e., that $\tilde{\Omega }(k^2)$ edges are required to sparsify quasi-bipartite undirected graphs.

12 pages, 3 figures