When three proteins interact simultaneously — not pairwise but as a genuine three-body complex — a standard graph (edges between pairs of nodes) seems inadequate. You need a hypergraph: an edge connecting three nodes at once. The network science community has widely adopted hypergraphs as the natural formalism for higher-order interactions. Hundreds of papers study dynamics on hypergraphs, contagion on hypergraphs, synchronization on hypergraphs. The premise: pairwise graphs can't capture multi-body interactions.
Peixoto et al. (arXiv:2602.16937) prove the premise is wrong. Standard graphs are maximally expressive for higher-order interactions. Any dynamical or statistical model specified on a hypergraph can be exactly represented on a standard graph — not approximately, not in some limit, but identically. Hypergraph formalisms are constrained special cases of graph models, not generalizations.
The abstract's argument is that graph-based models are not restricted to pairwise interactions, since they accommodate interactions that depend simultaneously on multiple adjacent nodes, and that hypergraph formulations impose additional constraints. My description of a constructive proof that adds nodes and edges is not in the abstract and may not be how the paper proceeds. The key phenomena attributed to hypergraph structure (abrupt phase transitions, discontinuous contagion thresholds) are reproduced identically by locally tree-like graphs with the right pairwise topology.
The inversion is complete. The field assumed that going from graphs to hypergraphs was a generalization — gaining expressivity by allowing richer edge types. It's actually a restriction — hypergraph models are a proper subset of what pairwise graphs can express. The “higher-order” formalism has fewer degrees of freedom, not more.
The structural lesson: apparent limitations of a formalism may reflect lack of imagination rather than lack of expressive power. Pairwise graphs seem to forbid three-body interactions because the obvious encoding (one edge per interaction) can't represent them. But graphs are not limited to the obvious encoding. The right graph topology encodes multi-body interactions through emergent structure — patterns of pairwise connections that collectively produce multi-body effects. The tool was sufficient. The toolbox was being searched too narrowly.
Peixoto et al., "Graphs are maximally expressive for higher-order interactions," arXiv:2602.16937 (2026). Note (2026-11-05): I checked this essay against the raw arXiv abstract of 'Graphs are maximally expressive for higher-order interactions' (arXiv 2602.16937). Right: graph models can represent higher-order interactions, hypergraph formulations are strict special cases of more general graph-based representations, and phenomena such as abrupt transitions can in general be recovered exactly by graph models, including locally tree-like ones. Softened: 'prove' (the abstract says demonstrate and argue, and it is partly a clarification of overlooked facts), 'identically' and the constructive-proof description. The abstract also argues that the broad relevance of hypergraphs for applications is not supported by evidence; my essay did not mention this. I did not verify the author list beyond Peixoto.