Source
Pecora, Sorrentino, Hagerstrom, Murphy and Roy (2014), Nature Communications 5:4079, LIT-960, read in full in NOTE-775 (the published article and the arXiv preprint; the Supplementary Information was not obtained).
What was actually shown
- The class. ẋᵢ = F(xᵢ) + σ Σⱼ Aᵢⱼ H(xⱼ), with identical F and H, and A a symmetric 0/1 adjacency matrix (eq. 1). The form is needed only near the synchronization manifold.
- Clusters (a theorem). Every symmetry R_g of the graph commutes with A, so the equations are unchanged when a symmetry permutes the nodes. The orbits of the symmetry group are therefore sets of nodes that can move identically. Each orbit’s synchronous motion is a solution.
- Stability (a theorem plus computation).
- The IRR transformation T makes TAT⁻¹ block-diagonal. One M × M block lies in the synchronization manifold, and the transverse blocks are one per IRR.
- A cluster is stable where the maximum Lyapunov exponent of its transverse block is negative.
- Clusters moved by one non-decomposable factor of the group share a block. They are “intertwined” and “will always either synchronize together or not at all”.
- Isolated desynchronization (a derivation, eq. 6).
- Suppose the group is a direct product of commuting factors. A node outside cluster C_m is then coupled to every node of C_m with the same weight.
- So it receives the same summed input from C_m whether C_m is synchronized or not.
- The other clusters’ synchronous states stay invariant when C_m loses synchrony, and if they are also stable, that is isolated desynchronization.
- What could have failed.
- In an 11-node electro-optic network with 32 symmetries, the clusters were computed before the experiment, and the bifurcations were predicted from the simulated transverse MLEs.
- The experiment, with real heterogeneity and noise, showed the four nontrivial clusters synchronize at β = 0.72π. At β = 1.4π the four-node cluster split into two pairs while the others held. The two intertwined clusters desynchronized together.
- The experiment could have shown clusters other than the orbits, or desynchronizations the blocks did not predict. It did not.
- How common.
- In 10,000 random, Barabási–Albert and bipartite 100-node graphs of each kind, almost every graph had several nontrivial clusters and several factors.
- Two power grids also had many symmetries.
- This is a statement about topology. The authors say stability “will depend on the dynamical systems of the network nodes”.
What this does not say
- It does not say that clusters require symmetry. The published text notes that clusters also arise “where input and output couplings match in a cluster, but there is no symmetry”. So a synchrony pattern that is not a symmetry orbit is not evidence against coupling.
- It does not cover nonidentical units, weighted or directed coupling. The authors say these “can all be handled in much the same way” but that “it is not clear that those more general cases will have as many symmetries”. None is worked through.
- It does not say which clusters will be seen. The graph fixes what can synchronize. The node dynamics and parameters fix what does.
- It says nothing about external drive. The record registered LIT-960
on QUESTION-063 as a coupling signature that a shared drive does not
predict.
- That is an inference from this theorem, not a claim of the paper.
- Within the class it is sound. Identical uncoupled units under one drive have no graph to give their synchrony a pattern of orbits. So synchrony that follows the orbits of a known coupling graph, and above all an isolated desynchronization of one orbit, is a mark of coupling.
- It needs the graph known in advance and the units near-identical.
- It is not the single-member intervention QUESTION-063 asks for.
- It does not say that a synchronized cluster is a unit of any kind. Members of an orbit need not be linked to each other (“remote synchronization”). They are in step because they receive equal input, not because they entrain each other.
Connections
No relations are declared.
- QUESTION-063. It bears on the question as above. It is not an answer, because it neither distinguishes drive from coupling by measurement nor supplies an intervention design.
- CLAIM-172. A mutually coupled network yields several synchronized sub-collectives, and members of one need not be coupled to each other. So synchrony does not mark the boundary of the coupled system, nor entrainment among the members in step.
- CLAIM-157. A cluster that stays synchronized through another’s desynchronization is invariant because its input is equal, not because the coupling across its edge is weak. The near-decomposability test and the symmetry test pick out different objects.
- QUESTION-147. Whether the stability predictions hold in a second system of near-identical units: the condition this account’s promote_when carried until 2026-10-19.
Note of 2026-10-19: the invariance, derived directly
Let π be a graph automorphism, so A_{π(i)π(j)} = Aᵢⱼ, and let P permute node states, (Px)ᵢ = x_{π(i)}. Write f for the right-hand side of eq. (1).
- Equivariance. f(Px)ᵢ = F(x_{π(i)}) + σ Σⱼ Aᵢⱼ H(x_{π(j)}) = F(x_{π(i)}) + σ Σⱼ A_{π(i)π(j)} H(x_{π(j)}) = (Pf(x))ᵢ.
- Fixed-point subspaces are invariant. For any subgroup Σ of the automorphism group, let Fix(Σ) be the states with Px = x for every π ∈ Σ. If x ∈ Fix(Σ), then f(x) = f(Px) = Pf(x), so the vector field is tangent to Fix(Σ), and by uniqueness a trajectory that starts there stays there.
- Clusters. Fix(G), for the whole group G, is the set of states equal on every orbit. So orbit-wise synchrony is invariant.
- Isolated desynchronization. Let G = H_m × K, where H_m moves only the nodes of C_m and K, the product of the other factors, fixes C_m pointwise. For a node i outside C_m, H_m fixes i, so G·i = K·i. The K-orbits are the G-orbits outside C_m and singletons inside it. Fix(K) is then the set of states synchronized on every other cluster with C_m free, and it is invariant. If H_m also moved another cluster, the two would be intertwined, and they lose synchrony together, as the paper says.
Whether these states are stable is the separate question the transverse IRR blocks answer. The title claims only possibility, and the 11-node experiment exhibits the stable case.
Note of 2026-10-27 (evening): the readings of the U15–U23 works
Stewart, Golubitsky and Pivato (NOTE-1217; THEORY-1065) generalize this reading’s result. The orbit partitions of a symmetry group are balanced, so orbit clusters are a special case of their Thm 6.5. That theorem covers non-additive coupling, several cell and arrow types, and clusters with “input and output couplings match … but there is no symmetry” (Figures 3 and 19). Necessity holds only over the whole class of admissible dynamics, and stability, the other half of this reading, is not treated. Torres-Hugas et al. (NOTE-1218) give a further instance. In a seven-node motif with two commuting reflections, the pentagonal block loses transverse stability first while the triangular mode stays stable.