Threshold graphs are globally synchronizing

The Kuramoto model can be formulated as a gradient flow on a nonconvex energy landscape of the form $E(\boldsymbolθ) := \frac{1}{2} \sum_{1\le i,j\le n} A_{ij}\bigl(1-\cos(θ_i-θ_j)\bigr).$ A fundamental question is to identify graph structures for which this landscape is benign, in the sense that every second-order stationary point corresponds to a fully synchronized state. This property guarantees that all trajectories of the Kuramoto model converge to a fully synchronized state except for a measure-zero set of initial conditions, a phenomenon known as global synchronization. Existing guarantees typically require that each node be connected to a sufficiently large fraction of the other nodes, enforcing high graph density. In this work, we show that threshold graphs lie well outside this regime while still exhibiting global synchronization. In particular, threshold graphs realize arbitrary edge densities and have degree sequences that are extremal in the sense of majorization. Our analysis is based on a phasor–geometric characterization of stationary points that exploits the structural and geometric symmetries induced by threshold graphs.

💡 Research Summary

The paper investigates the global synchronization properties of the homogeneous Kuramoto model on a class of graphs known as threshold graphs. The Kuramoto dynamics in the homogeneous case can be written as a gradient flow (\dot{\theta}i = \sum{j} A_{ij}\sin(\theta_j-\theta_i)) with an associated non‑convex energy (E(\boldsymbol\theta)=\frac12\sum_{i,j}A_{ij}(1-\cos(\theta_i-\theta_j))). Global synchronization is equivalent to the statement that the energy landscape contains no spurious second‑order stationary points: the only points where the gradient vanishes and the Hessian is positive semidefinite are the fully synchronized states where all phases are equal (up to a global rotation).

Previous work has largely focused on density‑based conditions, showing that if the minimum degree (\delta(G)) exceeds a constant fraction (approximately 0.75) of the total possible degree, then the graph is globally synchronizing. Such results require the graph to be very dense, essentially close to a complete graph. However, they do not address whether sparse or structurally irregular graphs can enjoy the same guarantee.

Threshold graphs are generated by starting from a single vertex and repeatedly adding either an isolated vertex (connecting to none of the existing vertices) or a dominating vertex (connecting to all existing vertices). This construction is uniquely encoded by a binary sequence of length (n-1). Consequently, threshold graphs can realize any edge density from the star (very sparse) to the complete graph (very dense) and possess degree sequences that are extremal with respect to majorization.

The authors’ main contribution is Theorem 1.2, which states that every connected threshold graph is globally synchronizing. The proof does not rely on the classical “half‑circle lemma” (which asserts that if all phases lie inside an open half‑circle, the dynamics contract toward synchronization). Instead, the authors develop a planar phasor‑geometric framework: each phase (\theta_i) is represented by a unit vector (v_i=(\cos\theta_i,\sin\theta_i)) in (\mathbb{R}^2). The energy becomes (E = \frac12\sum_{i,j}A_{ij}(1-\langle v_i,v_j\rangle)), and the gradient and Hessian can be expressed in terms of vector sums and angles.

Two central notions are introduced: closed twins (pairs of vertices sharing exactly the same neighbor set) and geometric twins (pairs that, at a second‑order stationary point, become collinear vectors even if they are not combinatorially twins). Lemma 5.2 shows that any closed twins must synchronize at any second‑order stationary point. Section 5.3 defines “synchronous pendants,” configurations where a pendant vertex and its neighbor become geometrically equivalent, forcing synchronization despite lacking structural symmetry.

The proof proceeds by establishing a collection of “Local Synchronization Primitives.” These are small substructures (e.g., a star, a complete subgraph) for which the authors can directly verify that any second‑order stationary point forces all involved vertices to share the same phase. Using the hierarchical nature of threshold graphs, they then apply an inductive argument: when a new vertex is added (either isolated or dominating), the previously established primitives guarantee that the new vertex’s phase must align with the existing synchronized core. This inductive step propagates backward through the construction sequence, ultimately showing that every second‑order stationary point of the energy on a threshold graph is a fully synchronized state.

Because the energy landscape is “benign,” gradient descent from almost any initial condition converges to the global minimum, yielding global phase synchronization for all but a measure‑zero set of initial phases. Importantly, this result holds regardless of the graph’s minimum degree or average degree, demonstrating that symmetry—both combinatorial and geometric—can replace density as the driving mechanism for global synchronization.

The paper also discusses the implications of these findings. Threshold graphs, despite having extremal degree sequences, do not require high connectivity; even the sparsest star graph (minimum degree 1) is globally synchronizing under the homogeneous Kuramoto dynamics. This contrasts sharply with prior density‑based thresholds and suggests a new line of inquiry: identifying other graph families where structural symmetries guarantee benign energy landscapes. The authors outline several open problems, including extending the analysis to heterogeneous Kuramoto models (non‑identical natural frequencies), exploring synchronization on graphs with partial symmetry, and applying the geometric twin concept to design robust synchronization protocols in engineered networks such as power grids or neural architectures.

In summary, the paper provides a rigorous, geometry‑driven proof that threshold graphs are globally synchronizing, thereby broadening the understanding of how graph structure influences collective dynamics beyond traditional density considerations.