Source
Torres-Hugas, Duch, Gómez and Arenas (2026), LIT-1668, §II A–B and §II C 3, pp. 2–5 of the arXiv preprint 2604.19682v2; read in NOTE-1218.
What was actually shown
The setting. A connected graph, each edge oriented, with a phase αᵢⱼ on i → j and αⱼᵢ = −αᵢⱼ (eq. 1). The twisted incidence operator is (δα z)ᵢⱼ = zⱼ − e^{iαᵢⱼ}zᵢ (eq. 5), the energy E(z) = ½‖δα z‖² (eq. 6), and Lα = δα*δα (eq. 8). With every αᵢⱼ = 0 these are the ordinary incidence operator and graph Laplacian (p. 2).
The criterion. E(z) = 0 exactly when every edge relation zⱼ = e^{iαᵢⱼ}zᵢ holds (eq. 7), so “ker Lα ≠ {0} if and only if the edge compatibility conditions can be satisfied globally in the relaxed linear setting” (p. 3). The paper then states: “When all cycle holonomies vanish, Eq. (10) admits nontrivial solutions. If some cycle has nonzero holonomy, then ker Lα = {0} and the smallest eigenvalue of Lα is strictly positive” (p. 3), with the holonomy Hol_α(C) = Σ_{(i,j)∈C} αᵢⱼ mod 2π (eq. 2).
The step the paper leaves out. If z meets every edge relation and is not zero, then |zⱼ| = |zᵢ| on every edge, so on a connected graph every entry has the same nonzero modulus. Fix z at one node; the relations then fix z along any spanning tree. Each edge outside the tree closes one fundamental cycle, and its relation holds exactly when that cycle’s holonomy is 0 mod 2π. The fundamental cycles generate all cycles, so the relations have a nonzero solution iff every cycle’s holonomy vanishes mod 2π, and the solution is then unique up to a common factor. On a tree there is nothing to close, so every connection is gauge-trivial, as the paper says (p. 2). This is the standard result for the discrete magnetic Laplacian (Shubin, the paper’s ref. [14]) and the connection Laplacian (refs. [15], [16]).
Residual incompatibility. λ₁(α) = min_{‖z‖=1}⟨z, Lα z⟩ (eq. 26) is zero iff every holonomy is (eqs. 27, 28), and when positive “measures the minimum normalized residual incompatibility after optimization over all node states” (p. 5).
The isolated cycle. For an ℓ-cycle the eigenvalues are 2 − 2cos((2πm + Hol)/ℓ), m = 0, …, ℓ − 1 (eq. 29); λ₁ is zero at Hol ≡ 0 and largest at Hol ≡ π (eqs. 30–32), where “two adjacent winding sectors become degenerate” (p. 5), giving λ₁ = 2 − 2cos(π/ℓ). A numerical check gave λ₁ = 1 for ℓ = 3 and 0.382 for ℓ = 5 at holonomy π, and λ₁ ≈ 10⁻¹⁶ for a 4-cycle with random phases summing to 2π.
What could have failed. Lα could have had a zero mode for some nonzero holonomy, or none for a flat one; the spanning-tree argument rules out both.
What this does not say
- Anything about the unit-modulus phase dynamics directly. The criterion is for complex states zᵢ ∈ ℂ. For the gauge-coupled flow ϕ̇ᵢ = κ Σ sin(ϕⱼ − ϕᵢ − αᵢⱼ) (eq. 12), the paper says only that low eigenmodes “best accommodate the incompatibility” (p. 3) and computes nothing.
- That λ₁ is a stability boundary or a relaxation rate. In the paper’s own motif the stability threshold is set by the Jacobian, not by Lα (THEORY-1087).
- That the obstruction is a failure of the graph or of any edge. Every edge relation is satisfiable on its own; the obstruction is a property of composition around cycles (p. 2).
- Where the connection comes from. The αᵢⱼ are given. Nothing in the result says an interaction produces them (CLAIM-471).
- That this is new mathematics. The paper’s contribution lies in the synchronization reading, not in the criterion.