Source
Federico Battiston, Giulia Cencetti, Iacopo Iacopini, Vito Latora, Maxime Lucas, Alice Patania, Jean-Gabriel Young and Giovanni Petri (2020), “Networks beyond pairwise interactions: Structure and dynamics”, Physics Reports 874, 1–92, LIT-091, read in full in NOTE-057.
What was actually shown
Linear group terms reduce.
- §V (p. 37): “traditional diffusion is a linear process and consequently the simplest generalization to higher-order structures can always be reduced to equations involving only pairwise couplings”. Three-body linear diffusion becomes a weighted Laplacian L^△ with ℓ^△_ij = δ_ij Σ_k a^△_ijk − Σ_k a^△_ijk (Eq. 35).
- §V B (p. 41): random walks on hypergraphs with edge-independent vertex weights are equivalent to walks on a weighted projected graph (Chitra and Raphael, reported).
- Near synchrony, identical oscillators with higher-order coupling are governed by a weighted pairwise multi-order Laplacian (Lucas et al., Eqs. 49–51).
Nonlinear group terms do not. The reported effects all come from coupling functions in which the group enters nonlinearly:
- triadic Kuramoto, sin(θ_j + θ_k − 2θ_i) (Eq. 43): abrupt desynchronisation and an incoherent state stable for all coupling;
- simplicial SIS, where a 2-simplex transmits only when both other members are infected (Eq. 58): a discontinuous, bistable transition. NOTE-057’s mean-field check finds an endemic branch below λ = 1 exactly when λ_Δ > 1;
- three-body consensus with an influence function s(|x_j − x_k|) (Eq. 61): the mean is conserved only when s is constant;
- replicator dynamics with b_ijk x_j x_k terms, and Grilli et al.’s three-seedling competition, where the coexistence point that pairwise competition leaves neutral becomes globally stable (Eqs. 67–70).
The representation matters. A simplicial complex assumes every sub-group of an interacting group also interacts; a hypergraph assumes nothing. The graph projection loses group identity and the clique complex invents groups (Fig. 1). Simplicial contagion (Eq. 58) and contagion on d-uniform hypergraphs (Eq. 59) behave differently.
What could have come out otherwise. Linear diffusion with group terms could have had dynamics no weighted graph reproduces. It does not: the reduction is direct algebra.
What this does not say
- That linearity is the only escape. Walks with edge-dependent vertex weights are not time-reversible and cannot be reduced to weighted-graph walks, though the walk is still linear.
- That higher-order structure as such changes dynamics. That is the introduction’s and §X’s framing (C10, C11). §X’s list of “profoundly changed” critical behaviour cites an edge-independent hypergraph walk and linear simplex diffusion, which §V itself reduces.
- That the nonlinear effects are general. Most are mean-field, all-to-all or for identical oscillators, and the review says the effect of symmetric against asymmetric coupling functions has “not been investigated systematically” (p. 47).
- That group interactions can be detected in data. The review says no inference scheme yet distinguishes a genuine k-body coupling from pairwise coupling plus correlations at scale (§X).
- That the brain is shown to be higher-order. The review reports the neuroscience evidence as mixed.
- That a pairwise description is more fundamental. Phase reduction of pairwise-coupled nonlinear oscillators produces three- and four-body phase terms at second order (C9), so the arrow can run the other way.
Connections
- Forman curvature (THEORY-502, from LIT-020). Forman’s curvature lives on cell complexes and decomposes the Hodge Laplacian this review defines; the network version keeps only the 1-skeleton.
- Spectral and Fiedler dimensions (THEORY-576, from LIT-028). The review’s spectral dimension is the density-of-states one, which that account separates from the gap exponent.
- Information decomposition (LIT-025, LIT-042). The review’s O-information pointers belong to the same programme of measuring high-order dependence, used here only in passing.