Answers are grounded in this paper and cite their sources.
Let H be a 2-edge-connected cubic graph. Each alternating circuit of H (a component of the symmetric difference of two perfect matchings) is even and so carries its own 2-colouring. We show that these local 2-colourings, restricted to the ends of the ears of each circuit, glue into one global 2-colouring of V(H) if and only if every brick in the tight cut decomposition of H is K4 or the Petersen graph; by a theorem of Joshi, Raghul, Diwan and Kothari, equivalently, iff every brick is θ-free. The question comes from linear algebra: for a graph G let ι0(G) be the minimum number of negative eigenvalues of a real symmetric matrix with zero diagonal and off-diagonal nonzero pattern E(G); the ordered matching number νo(G) is a lower bound. For every connected H with a perfect matching on n vertices, νo(L(H)) = n/2 − 1, with equality ι0 = νo for bipartite H. The gluing condition is equivalent to the consistency of a GF(2) system Σ(H) expressing that all blocks A[V∖W, W] indexed by perfect matchings W can be simultaneously deficient; when it fails, ι0(L(H)) ≥ n/2 > νo(L(H)). This holds for the odd prisms and the Möbius ladders on 4j vertices; consistency does not suffice for equality (an exact counterexample on 14 vertices). Important: this work was produced with substantial assistance from AI systems (Claude, Anthropic): several AI agents worked independently on proofs, prior-art search and verification. Proofs and computations were checked by independent AI reviewing agents (reports included in the archive) and by exact computation, but have NOT been reviewed by a human specialist. The proofs of the main results (Theorem G, Section 5; Theorem F, Section 6) were checked by independent AI reviewers (one definitional defect found and corrected) and by exhaustive computation on all bridgeless cubic graphs up to 18 vertices. A Lean 4 / Mathlib proof of the smallest obstruction (the triangular prism) is included. The prior-art search is necessarily incomplete; its limits are listed in the paper. Code (MIT licence), certificates and reports are included.
No takes yet. Share an insight, caveat, or question.
Antoine Delorme (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: