Stable Graphs of Twin-width 2
- The paper establishes that every Hₜ-semi-free graph with twin-width 2 has rank-width at most 22t+170, resolving a central conjecture.
- It employs a detailed contraction sequence analysis and innovative lemmas to link red 4-paths with the forbidden half-graph structure.
- These insights yield practical algorithmic benefits, particularly for graph isomorphism testing via fixed-dimension Weisfeiler–Leman procedures.
Stable graphs of twin-width 2 are a significant subclass in the intersection of structural graph theory and descriptive complexity, characterized by powerful structural restrictions and algorithmic consequences. Their study resolves a central conjecture relating model-theoretic stability, the combinatorial parameter of twin-width, and the graph invariant rank-width, with immediate implications for logical definability and graph isomorphism testing.
1. Fundamental Definitions
A trigraph is a combinatorial structure in which every pair of vertices is either connected by a black edge, a red edge, or is non-adjacent. The red-degree of a vertex in a trigraph is the number of red edges incident to . A contraction step merges two parts of the partitioned vertex set; all non-homogeneous adjacency pairs become red, homogeneous blacks remain black, and non-adjacent pairs remain non-adjacent. The full contraction sequence for a graph is a sequence of trigraphs indexed by coarsenings of the vertex partition, each obtained by a contraction, starting from the discrete partition.
The width of a contraction sequence is defined as the maximal red-degree occurring in any intermediate trigraph of the sequence. The twin-width is the minimal such that admits a contraction sequence of width at most .
Stability is defined through the exclusion of half-graphs. For , the half-graph consists of a bipartition with edges if and only if . A graph is -semi-free if it has no semi-induced copy of , that is, for any bijection of into , the adjacency between the two bipartition classes is not precisely that of . A class is stable (monadically stable) if is contained in the set of -semi-free graphs for some .
Rank-width of a (simple) graph is defined via subcubic rank-decompositions of : trees whose leaves are bijectively labeled by , with each edge of the tree corresponding to some bipartition of . The width is the largest rank of the binary adjacency matrix (over ) across all . The rank-connectivity is . The well-linked-set theorem of Oum–Seymour states that if then contains a -well-linked set of size (Heinrich et al., 9 Jan 2026).
2. Statement of the Main Theorem
Let . For every graph with twin-width that is -semi-free, the rank-width satisfies
Consequently, every monadically stable class of twin-width 2 yields a uniformly bounded rank-width (Heinrich et al., 9 Jan 2026). This directly resolves the conjecture proposed by Bergougnoux, Gajarský, Guspiel, Hlinený, Pokrývka, and Sokolowski. The explicit linear bound in is central for combinatorial and algorithmic applications.
3. Proof Architecture
The proof proceeds by contradiction, assuming a graph with twin-width , -semi-free, but . Using the Oum–Seymour theorem, there exists a -well-linked set of size $11k-5$ for , so $11k-5 = 22t+89$.
Given any width-2 contraction sequence for , two pivotal structural lemmas are invoked:
- Existence of a highly connected red 4-path (Lemma 6.4): At some intermediate quotient trigraph, four parts appear forming a red path –––––– with no red edge , and
- Red 4-path implies a half-graph (Lemma 6.5): The presence of such a connected red 4-path with high rank-connectivity ensures contains a semi-induced .
The core of the argument requires careful partition sequence analysis: if rank-width is large but no exists, the properties of the contraction sequence and the well-linked set force the existence of a forbidden half-graph, yielding a contradiction. The explicit steps involve tracking how red-edges evolve during contractions and leveraging subadditivity of rank.
4. Quantitative Bound on Rank-width
For each , the proven structural result is
with the parameter of excluded half-graphs. This linear bound in terms of is a key quantitative advance for the classification of stable graph classes of twin-width 2 and permits parameterized algorithmic conclusions (Heinrich et al., 9 Jan 2026).
5. Algorithmic Implications and Weisfeiler–Leman Procedures
Bounded rank-width plays a direct role in the complexity of the graph isomorphism problem. It is known that for every , the -dimensional Weisfeiler–Leman (WL) algorithm correctly decides isomorphism for all graphs of rank-width at most (Lemma 2.5 in (Heinrich et al., 9 Jan 2026)). Thus, for -semi-free graphs with twin-width 2, having , the isomorphism problem is solvable algorithmically using the -WL procedure. The implication is that the isomorphism problem on stable twin-width 2 graphs admits a fixed-dimension WL algorithm, and in fact, FP+C-definable canonization.
6. Broader Structural and Logical Significance
The interplay between model-theoretic stability, twin-width, and rank-width highlights the deep connections between logical tameness and combinatorial parameters. Stable twin-width 2 graphs unify notions from first-order model checking, coloring, and isomorphism testing. The boundedness of rank-width enables the import of algorithmic results from the theory of graph decompositions, logical definability (FP+C), and combinatorial optimization into this class. A plausible implication is that further properties of stable twin-width 2 classes—beyond isomorphism and canonization—should be accessible via meta-theorems of bounded rank-width and logical interpretability.
7. Connections and Consequences for Ongoing Research
The result provides a sharp structural foundation for further investigations into the relationship of twin-width with other fundamental invariants, potentially generalizing to higher twin-width under stringent stability or exclusion conditions. The resolution of the Bergougnoux et al. conjecture establishes a key benchmark for classes where low twin-width and logical stability coincide, and it reinforces the program of classifying graph isomorphism complexity in terms of width parameters (Heinrich et al., 9 Jan 2026).