Core Concepts
We present a new algorithm that maintains a (1-ε)-approximate maximum matching in a fully dynamic graph, where the update-time depends on the density of a certain class of graphs called Ordered Ruzsa-Szemerédi (ORS) graphs. We also provide improved upper bounds on the density of both ORS and Ruzsa-Szemerédi (RS) graphs with linear size matchings.
Abstract
The paper studies the fully dynamic maximum matching problem, where the goal is to efficiently maintain an approximate maximum matching of a graph that is subject to edge insertions and deletions. The focus is on algorithms that maintain the edges of a (1-ε)-approximate maximum matching for a small constant ε > 0.
The key contributions are:
Dynamic Matching Algorithm:
We present a new algorithm that maintains a (1-ε)-approximate maximum matching, where the update-time is parametrized by the density of Ordered Ruzsa-Szemerédi (ORS) graphs.
ORS graphs are a generalization of the well-known Ruzsa-Szemerédi (RS) graphs, where the edges are decomposed into an ordered list of edge-disjoint matchings such that each matching is induced with respect to the previous matchings.
If the existing constructions of ORS graphs are optimal, our algorithm runs in n^(1/2+O(ε)) time per update, which would be significantly faster than the previous near-linear in n time algorithms.
We show that proving conditional lower bounds for the dynamic (1-ε)-approximate maximum matching problem either requires a strong lower bound on the density of ORS graphs, or requires a conjecture that implies such a lower bound.
Improved Upper Bounds for ORS and RS Graphs:
We provide a better upper bound on the density of both ORS and RS graphs with linear size matchings.
The previous best upper bound was due to a proof of the triangle-removal lemma from more than a decade ago.
Our new upper bound shows that for any constant c > 0, the density of ORS graphs with linear size matchings is O(n/log^(ℓ) n) for some ℓ = poly(1/c), improving the previous bound.
This also immediately implies an improved upper bound for RS graphs.
The paper presents the technical details of these contributions, including the algorithms, analysis, and connections to the density of ORS and RS graphs.
Stats
The maximum matching size in the graph is O(n').
The update-time of the algorithm is O(√n^(1+ε) log n · ORSn(εn')).
Quotes
"Let ε > 0 be fixed. There is a fully dynamic algorithm that maintains the edges of a (1-ε)-approximate maximum matching in O(√n^(1+ε) · ORSn(Θ(ε^2n)) poly(log n)) amortized update-time."
"For any c > 0, it holds that ORSn(cn) = O(n/ log^(ℓ) n) for some ℓ = poly(1/c), where log^(x) is the iterated log function."