Source
Stewart, Golubitsky and Pivato (2003), LIT-1674, §§1–6 (pp. 609–632 of SIAM J. Appl. Dyn. Syst. 2(4), read from the accepted version); read in NOTE-1217.
What was actually shown
The setting. A coupled cell network is a finite directed graph with typed cells and typed arrows (Definition 2.1, p. 616). An admissible vector field gives each cell an equation depending only on its input set, the same function for cells whose input sets are isomorphic by a type-preserving map, symmetric under permutations of equivalent inputs (Definition 4.1, p. 622). Coupling need not be additive: a cell’s evolution “can in principle be any (smooth) function” of its inputs (p. 610).
The theorem. For an equivalence relation ⋈ on the cells, the polydiagonal {x : x_c = x_d whenever c ⋈ d} is invariant under every admissible vector field “if and only if ⋈ is balanced” (Theorem 6.5, p. 630). Balanced: equivalent cells c, d have an input isomorphism γ with i ⋈ γ(i) for every input i (Definition 6.4, p. 629). The proof’s counting condition, |U ∩ K_s| = |U ∩ L_s| for every class U and arrow type s (eq. 6.2), is the “same number of inputs from each class” form. Lemma 6.3 (p. 629): only cells with isomorphic input sets can be robustly synchronous at all.
What could have come out otherwise. Necessity is not obvious, and the paper proves it by constructing, for any unbalanced relation, an admissible field that leaves the polydiagonal (eq. 6.4, pp. 631–632). Its worked case: moving one cell to another class in Example 5.1 changes “multiplicity”, three inputs of one colour against two, and the subspace is no longer invariant (p. 629).
Without symmetry. In the three-cell network of Figure 3 (eq. 1.5, p. 612), “This network has no symmetry, but the network structure forces the “polydiagonal” subspace Y = {x : x₁ = x₂} to be flow-invariant.” Example 4.4 (p. 623), the five-cell network of Example 7.1 (p. 633) and the three-cell master–slave network of Figure 19 (p. 635) are others. A network symmetry, restricted to an input set, is an input isomorphism that keeps each cell in its orbit, so orbit partitions are balanced and the symmetry-group case (THEORY-276) is contained in this one; the paper builds on the fixed-point-subspace argument (p. 611) but does not state this inclusion as a theorem.
Without coupling among the synchronized cells. In Example 5.1 cells 4, 5, 7 and 9 have no arrows between them; each receives one input of the round type and one from the class {2, 3, 6, 8}, and they are robustly synchronous (Figures 11–12, p. 626). In Figure 19, cell 3 is only driven by cell 1 while cell 2 is driven by and drives cell 1; 2 and 3 are robustly synchronous.
What this does not say
- That a balanced pattern is stable or will be seen. “in this paper we are studying existence, not stability, and we do not wish to rule out unstable synchrony because it still corresponds to an invariant subspace for the dynamics” (p. 626). The authors’ example, identical cells with no inputs, is invariant and, they say, “dynamically unstable” (p. 626); that holds for some internal dynamics, not all. Which patterns attract is a model-dependent question the paper leaves alone.
- What is invariant for one model. “Robust” quantifies over the whole admissible class, and the necessity proof uses linear admissible fields (eq. 6.4). A narrower family, such as identical diffusive coupling with one fixed coupling function, can have invariant subspaces that are not balanced in this sense.
- That synchronized cells interact. Balance is equal input; cells can be robustly synchronous with no arrow between them, so a synchronized class is not shown to be coupled, entrained or a unit of any kind.
- That a synchrony pattern distinguishes coupling from common drive. A drive is another cell, and balance counts inputs regardless of source (Figure 19). The paper does not discuss inference from observed synchrony; QUESTION-063’s case is outside it.
- Anything about weighted or multiple arrows. The 2003 definition takes arrows as a subset of C × C (Definition 2.1), one per ordered pair.
- Anything about nonidentical cells within a class. ⋈ refines cell type (p. 628); synchrony of cells of different types is not defined here.