A mathematical preprint reports a theorem about the Kuramoto model, in which identical phase oscillators sit at the vertices of a graph and interact along its edges. It says that there is an absolute constant η > 0 such that every finite simple graph on n vertices whose minimum degree is at least (3/4 − η)n has no local energy minimum except a fully synchronized state.
The result concerns the energy landscape of a mathematical oscillator model. In this paper, global synchronization has a precise meaning: the fully synchronized states are the only local minima of the Kuramoto energy.
The gap below the threshold
The important change is the room below three-quarters. The theorem says that some positive absolute gap η can be subtracted from 3/4 and the conclusion still holds. But the paper does not report an explicit numerical value for η, so the statement establishes that a gap exists without saying how large it is.
The authors describe this as the first strict improvement over an earlier 3/4 upper bound and say that it refutes the conjecture cited in the paper.
How the proof closes the escape routes
To test for a hidden exception, the proof treats every local minimum as a second-order critical point. Such a point must have zero gradient and a positive-semidefinite Hessian, the mathematical object that records the local curvature of the energy. The argument therefore combines equilibrium equations with a check that the second-order curvature has no negative direction.
One part of the proof controls two quantities called the first and second order parameters, ρ1 and ρ2. For every ε > 0, it says η can be chosen so that any nontrivial local minimum at minimum degree at least (3/4 − η)n has |ρ1| ≤ ε and |ρ2| ≤ ε. The analysis does not give an explicit relationship between η and ε, but it shows that both quantities can be made as small as the argument requires.
That smallness leaves little freedom. When the clustered conditions are met, a putative nontrivial minimum must organize the vertices into four disjoint clusters, each containing at least (1/4 − ε)n vertices, with each cluster lying within ε of a quarter-turn center. In ordinary language, a candidate exception would have to look like four large groups gathered around four quarter-turn positions.
The proof next attacks the imperfections that could keep such a pattern from being exact. Its complex-valued test function handles an exceptional set of vertices and offsets in the bulk phases; under the near-threshold clustered conditions, the argument forces E to be empty and the bulk offsets to be zero.
Writing the four clusters as B0 through B3, an exact quarter-turn local-minimum configuration must also satisfy min deg(v) ≤ max_{0≤j≤3}(|Bj| − 1). In plain language, the graph’s minimum degree is bounded by the size of its largest cluster minus one in that configuration.
Finally, the authors perturb the exact arrangement by a sufficiently small negative t. The resulting energy variation is negative, contradicting local minimality and forcing m = 0. With no edge joining distinct clusters, the remaining nonsynchronized arrangement is eliminated.
A result about landscapes, not experiments
The theorem’s units are finite simple graphs on n vertices, and it quantifies over every graph satisfying the degree condition. The oscillators appear inside the model, placed at graph vertices and coupled by graph edges; there is no sampled oscillator population behind the result.
That distinction sets the boundary of the claim. In this paper, global synchronization means that fully synchronized states are the only local minima of the energy. It is therefore a statement about the landscape defined by the model, not a measured outcome from a population.
The stated scope is finite simple graphs in the specified minimum-degree regime. The authors also leave an extension to the exactly clustered case for future work.
Most importantly for anyone hoping to use the theorem as a rule, η remains unspecified. The open mathematical task is to derive and report an explicit usable value; until then, the theorem says that a positive gap exists without quantifying it.
Publication and disclosure
The document is identified as arXiv:2608.20010v1 [math.CO], dated 20 Aug 2026, and marked as a preprint.
The acknowledgments say the research was completed at the SPUR program at Massachusetts Institute of Technology. They also disclose that ChatGPT-5.6 Sol was used for editing, proofreading and refining technical proofs; the authors state that the proof strategy and errors are their own.
Paper data and sources
Original title: Graphs with connectivity $3/4 - \varepsilon$ are globally synchronizing
Authors: Saba Lepsveridze, Sam Zhang
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text