A structure theorem for rooted connectivity in bidirected graphs

Recently, bidirected graphs have received increasing attention from the graph theory community with both structural and algorithmic results. Bidirected graphs are a generalization of directed graphs, consisting of an undirected graph together with a map assigning each endpoint of every edge either sign $+$ or $-$. The connectivity properties of bidirected graphs are more complex than those of directed graphs and not yet well understood. In this paper, we show a structure theorem about rooted connectivity in bidirected graphs in terms of directed graphs. As applications, we prove Lovász’ flame theorem, Pym’s theorem and a strong variant of Menger’s theorem for a class of bidirected graphs and provide counterexamples in the general case.

💡 Research Summary

This paper presents a foundational structure theorem that demystifies the complex connectivity properties of bidirected graphs, a generalization of directed graphs where each endpoint of an undirected edge is assigned a ‘+’ or ‘-’ sign. The core challenge addressed is that walks (and thus connectivity) in bidirected graphs are governed by a sign-alternation rule at vertices, making their behavior more intricate and less understood than their directed counterparts.

The authors’ key strategy is to work within a focused setting: they consider rooted bidirected graphs (with a designated root vertex r) that are clean, meaning they exclude certain problematic types of walks. Within this framework, the main result is a decomposition theorem that models any given rooted clean bidirected graph B on a directed graph D. The construction hinges on a crucial classification of edges in B with respect to r:

1. Trail-directable edges: Edges that can be traversed in only one specific orientation on any trail starting from r. This unique orientation is called the natural orientation.
2. Trail-undirectable edges: Edges for which trails from r exist using both orientations.

The set of trail-undirectable edges forms connected components C_i. The structure theorem (Lemma 3.1) reveals that each such component C_i not containing r is connected to the rest of the graph by a unique trail-directable edge f_i, whose natural orientation points into C_i. Furthermore, inside each C_i, all edges are bidirectionally traversable from this entry point, and one can reach any vertex within C_i in both positive and negative sign states.

This insight allows for the construction of the model digraph D. The vertex set of D consists of the original vertices of B plus a new vertex representing each trail-undirectable component C_i. The edge set of D comprises all naturally oriented trail-directable edges of B, plus an edge representing the unique connector f_i for each component C_i. The power of this transformation is that r-based connectivity in the original bidirected graph B—such as the existence of edge-disjoint or vertex-disjoint trails/paths—is faithfully mirrored by r-based connectivity in the directed graph D. Effectively, it reduces problems about bidirected graphs to problems about directed graphs.

The paper provides two versions of this structure theorem: one tailored for analyzing edge-connectivity (the Edge-Decomposition) and another for vertex-connectivity (the Vertex-Decomposition).

As major applications, the authors leverage this decomposition to prove several classic graph theory theorems for the class of clean bidirected graphs:

• A strong variant of Menger’s Theorem for edge-disjoint and vertex-disjoint paths.
• Lovász’s Flame Theorem, which concerns the existence of certain connectivity-preserving substructures.
• Pym’s Theorem, related to linking structures between sets.

For each, they demonstrate that the theorem holds for clean bidirected graphs via their structure theorem and the corresponding known result for directed graphs. Crucially, they also provide counterexamples showing that these theorems can fail for general (non-clean) bidirected graphs, precisely delineating the scope of their results.

Finally, the paper notes the algorithmic implications. Given that checking if an edge is trail-directable can be done in polynomial time (using ideas from Edmonds’ Blossom Algorithm), the entire decomposition process and the algorithms derived from the applications (e.g., finding maximum numbers of disjoint paths) are also polynomial-time computable. Thus, the work establishes not only a deep structural understanding but also a pathway for efficient computation in the realm of bidirected graph connectivity.