Table of Contents
Fetching ...

On the Exact Matching Problem in Dense Graphs

Nicolas El Maalouly, Sebastian Haslebacher, Lasse Wulf

TL;DR

This work tackles the Exact Matching problem ($Em$), seeking deterministic polynomial-time algorithms for dense graphs where randomized methods are known to exist. It introduces two complementary frameworks: a simple local-search method that is guaranteed to succeed under the path-shortening property $ extsc{Pshort}(t)$, and a Karzanov-inspired characterization extended to chain graphs, unit interval graphs, and complete $r$-partite graphs via a chord property that reduces Em to Bcpm. The paper proves $Em$ is solvable in deterministic polynomial time for several dense graph classes (complete $r$-partite, unit interval, chain graphs, graphs with bounded neighborhood diversity, and those without a complete bipartite $t$-hole), and achieves quasi-polynomial time for $G(n,p)$ with $p$ constant (notably $p=1/2$). It also identifies important limitations, such as the failure of Local Search on chain graphs and the non-satisfaction of Karzanov's property by interval graphs, and discusses weak variants that may guide future derandomization. Overall, the results advance deterministic approaches for $Em$ in dense regimes and map out clear lines for future extension and open questions, including derandomization for $G(n,1/2)$ and broader classes with the weak Karzanov property.

Abstract

In the Exact Matching problem, we are given a graph whose edges are colored red or blue and the task is to decide for a given integer k, if there is a perfect matching with exactly k red edges. Since 1987 it is known that the Exact Matching Problem can be solved in randomized polynomial time. Despite numerous efforts, it is still not known today whether a deterministic polynomial-time algorithm exists as well. In this paper, we make substantial progress by solving the problem for a multitude of different classes of dense graphs. We solve the Exact Matching problem in deterministic polynomial time for complete r-partite graphs, for unit interval graphs, for bipartite unit interval graphs, for graphs of bounded neighborhood diversity, for chain graphs, and for graphs without a complete bipartite t-hole. We solve the problem in quasi-polynomial time for Erdős-Rényi random graphs G(n, 1/2). We also reprove an earlier result for bounded independence number/bipartite independence number. We use two main tools to obtain these results: A local search algorithm as well as a generalization of an earlier result by Karzanov.

On the Exact Matching Problem in Dense Graphs

TL;DR

This work tackles the Exact Matching problem (), seeking deterministic polynomial-time algorithms for dense graphs where randomized methods are known to exist. It introduces two complementary frameworks: a simple local-search method that is guaranteed to succeed under the path-shortening property , and a Karzanov-inspired characterization extended to chain graphs, unit interval graphs, and complete -partite graphs via a chord property that reduces Em to Bcpm. The paper proves is solvable in deterministic polynomial time for several dense graph classes (complete -partite, unit interval, chain graphs, graphs with bounded neighborhood diversity, and those without a complete bipartite -hole), and achieves quasi-polynomial time for with constant (notably ). It also identifies important limitations, such as the failure of Local Search on chain graphs and the non-satisfaction of Karzanov's property by interval graphs, and discusses weak variants that may guide future derandomization. Overall, the results advance deterministic approaches for in dense regimes and map out clear lines for future extension and open questions, including derandomization for and broader classes with the weak Karzanov property.

Abstract

In the Exact Matching problem, we are given a graph whose edges are colored red or blue and the task is to decide for a given integer k, if there is a perfect matching with exactly k red edges. Since 1987 it is known that the Exact Matching Problem can be solved in randomized polynomial time. Despite numerous efforts, it is still not known today whether a deterministic polynomial-time algorithm exists as well. In this paper, we make substantial progress by solving the problem for a multitude of different classes of dense graphs. We solve the Exact Matching problem in deterministic polynomial time for complete r-partite graphs, for unit interval graphs, for bipartite unit interval graphs, for graphs of bounded neighborhood diversity, for chain graphs, and for graphs without a complete bipartite t-hole. We solve the problem in quasi-polynomial time for Erdős-Rényi random graphs G(n, 1/2). We also reprove an earlier result for bounded independence number/bipartite independence number. We use two main tools to obtain these results: A local search algorithm as well as a generalization of an earlier result by Karzanov.
Paper Structure (24 sections, 35 theorems, 9 equations, 13 figures, 2 algorithms)

This paper contains 24 sections, 35 theorems, 9 equations, 13 figures, 2 algorithms.

Key Result

Lemma 2

Assume we are given a PM $M$ in a colored graph, and an integer $s \geq 0$. There is an algorithm which runs in $O(m^{s+3})$ time and computes the set $\{ k' \in \mathbb{N} \mid \exists M' \in \mathop{\mathrm{\mathcal{N}}}\nolimits_s(M), r(M') = k' \}$ and for each $k'$ in this set outputs at least

Figures (13)

  • Figure 1: An example of the property $\textsc{Pshort}(4)$ on a path of length 11. Matching edges are bold. Both possibilities of path shortening are highlighted.
  • Figure 2: An example of a chain graph $G$ and a perfect matching $M$ in $G$ (left figure), where there exists an $M$-alternating path with $n$ edges from $M$ (right figure). Edges in $M$ are bold. The property $\textsc{Pshort}(n)$ is violated on this path.
  • Figure 3: An example of a $10$-cycle with three chords. The chords $\{x, y\}$ and $\{u, v\}$ are adjacent and they are both odd chords. Conversely, $\{a, b\}$ is an even chord with split $2$. The split of $\{x, y\}$ is $3$ and the split of $\{u, v\}$ is $5$.
  • Figure 4: On the left we have a colored interval graph on eight vertices which does not satisfy Karzanov's property. In particular, there are PMs with $0$, $1$, $3$, and $4$ red edges but there is no PM with exactly $2$ red edges. The interval representation of the graph is given on the right. Interval $I(v)$ corresponds to vertex $v$ from the left. Note that the vertical position of the intervals is irrelevant, only the relative horizontal position of the intervals matters.
  • Figure 5: By deleting the even chords from the graph in Figure \ref{['fig:counterexample_interval']}, we obtain a bipartite interval graph. It admits the same PMs as the interval graph in Figure \ref{['fig:counterexample_interval']}. Hence, this colored graph also violates Karzanov's property.
  • ...and 8 more figures

Theorems & Definitions (51)

  • Definition 1
  • Lemma 2
  • Definition 3
  • Theorem 4
  • Definition 5: Karzanov's Property
  • Definition 7: Odd Chord, Even Chord, Split of a Chord
  • Definition 8: Adjacent Chords
  • Definition 9: Chord Property
  • Lemma 9: Chord Property is Sufficient
  • Lemma 9: on Graphs with the Chord Property
  • ...and 41 more