[Paper Review] New Canonical Decomposition in Matching Theory

Nanao Kita|arXiv (Cornell University)|Aug 3, 2017

Limits and Structures in Graph Theory3 citations

TL;DR

This paper introduces the basilica decomposition, a new canonical decomposition in matching theory that applies to all graphs and provides a finer structural analysis than existing methods. By defining a canonical partial order over factor-components and generalizing the Kotzig-Lovász decomposition, it unifies the organization of graphs and their internal matching structure, offering a self-contained, O(nm)-time algorithm for computation and advancing foundational understanding in combinatorics.

ABSTRACT

In matching theory, one of the most fundamental and classical branches of combinatorics, {\em canonical decompositions} of graphs are powerful and versatile tools that form the basis of this theory. However, the abilities of the known canonical decompositions, that is, the {\em Dulmage-Mendelsohn}, {\em Kotzig-Lovász}, and {\em Gallai-Edmonds} decompositions, are limited because they are only applicable to particular classes of graphs, such as bipartite graphs, or they are too sparse to provide sufficient information. To overcome these limitations, we introduce a new canonical decomposition that is applicable to all graphs and provides much finer information. We focus on the notion of {\em factor-components} as the fundamental building blocks of a graph; through the factor-components, our new canonical decomposition states how a graph is organized and how it contains all the maximum matchings. The main results that constitute our new theory are the following: (i) a canonical partial order over the set of factor-components, which describes how a graph is constructed from its factor-components; (ii) a generalization of the Kotzig-Lovász decomposition, which shows the inner structure of each factor-component in the context of the entire graph; and (iii) a canonically described interrelationship between (i) and (ii), which integrates these two results into a unified theory of a canonical decomposition. These results are obtained in a self-contained way, and our proof of the generalized Kotzig-Lovász decomposition contains a shortened and self-contained proof of the classical counterpart.

Motivation & Objective

• To overcome limitations of existing canonical decompositions—such as restricted applicability to bipartite or factor-connected graphs and insufficient structural detail.
• To develop a unified, canonical decomposition applicable to all graphs that captures both the hierarchical organization of factor-components and their internal matching structure.
• To provide a self-contained theory that generalizes classical results like the Kotzig-Lovász decomposition and integrates them into a coherent framework.
• To enable deeper structural insights into maximum matchings, barriers, and saturated graphs through a new decomposition framework.

Proposed method

• Introduces factor-components as fundamental building blocks for analyzing matchings in any graph.
• Defines a canonical partial order (the basilica order) over factor-components using a binary relation that reflects structural dependencies.
• Generalizes the Kotzig-Lovász decomposition to arbitrary graphs, revealing the internal structure of each factor-component within the global graph context.
• Establishes a canonical relationship between the basilica order and the generalized Kotzig-Lovász decomposition, unifying both into a single coherent decomposition.
• Develops an O(nm)-time algorithm to compute the basilica decomposition using maximum matching computation and component analysis.
• Proves all results independently, including a shortened and self-contained proof of the classical Kotzig-Lovász decomposition.

Experimental results

Research questions

• RQ1 How can a canonical decomposition be constructed that applies universally to all graphs, rather than being limited to specific classes like bipartite or factor-connected graphs?
• RQ2 What is the canonical hierarchical structure that organizes factor-components in a way that reflects the global matching properties of a graph?
• RQ3 How can the internal structure of each factor-component be described in a way that is consistent with the global graph context, generalizing the Kotzig-Lovász decomposition?
• RQ4 What is the canonical relationship between the hierarchical organization of factor-components and their internal structural decomposition?
• RQ5 Can the new decomposition be computed efficiently, and what is the computational complexity of such an algorithm?

Key findings

• The basilica decomposition is a canonical partial order over the set of factor-components, providing a hierarchical view of how a graph is constructed from its components.
• The generalized Kotzig-Lovász decomposition extends the classical result to all graphs, describing the internal structure of each factor-component in the context of the entire graph.
• A canonical relationship unifies the basilica order and the generalized Kotzig-Lovász decomposition into a single, coherent canonical decomposition.
• The entire decomposition can be computed in O(nm) time, making it practically efficient for algorithmic applications.
• The proof of the generalized Kotzig-Lovász decomposition is self-contained and significantly shorter than previous proofs.
• The framework enables new results, including a purely graph-theoretic proof of the tight cut lemma and a new characterization of saturated graphs.