Constant-Time Algorithms for Sparsity Matroids

TL;DR

This paper presents constant-time approximation algorithms for the rank of sparsity matroids in degree-bounded graphs, enabling efficient property testing for various graph properties such as connectivity and rigidity.

Contribution

It introduces the first constant-time algorithms for approximating the rank of sparsity matroids and testing related properties in bounded-degree graphs.

Findings

Constant-time approximation algorithm for the rank of sparsity matroids.

Constant-time property testers for $(k,\, extell)$-fullness and $(k,\, extell)$-edge-connected-orientability.

Lower bounds showing $ ext{Ω}(n)$ queries are necessary for one-sided error testers.

Abstract

A graph $G=(V,E)$ is called $(k,\ell)$-full if $G$ contains a subgraph $H=(V,F)$ of $k|V|-\ell$ edges such that, for any non-empty $F' \subseteq F$, $|F'| \leq k|V(F')| - \ell$ holds. Here, $V(F')$ denotes the set of vertices incident to $F'$. It is known that the family of edge sets of $(k,\ell)$-full graphs forms a family of matroid, known as the sparsity matroid of $G$. In this paper, we give a constant-time approximation algorithm for the rank of the sparsity matroid of a degree-bounded undirected graph. This leads to a constant-time tester for $(k,\ell)$-fullness in the bounded-degree model, (i.e., we can decide with high probability whether an input graph satisfies a property $P$ or far from $P$). Depending on the values of $k$ and $\ell$, it can test various properties of a graph such as connectivity, rigidity, and how many spanning trees can be packed. Based on this result, we also propose a constant-time tester for $(k,\ell)$-edge-connected-orientability in the bounded-degree model, where an undirected graph $G$ is called $(k,\ell)$-edge-connected-orientable if there exists an orientation $\vec{G}$ of $G$ with a vertex $r \in V$ such that $\vec{G}$ contains $k$ arc-disjoint dipaths from $r$ to each vertex $v \in V$ and $\ell$ arc-disjoint dipaths from each vertex $v \in V$ to $r$. A tester is called a one-sided error tester for $P$ if it always accepts a graph satisfying $P$. We show, for $k \geq 2$ and (proper) $\ell \geq 0$, any one-sided error tester for $(k,\ell)$-fullness and $(k,\ell)$-edge-connected-orientability requires $\Omega(n)$ queries.