New Preprint: Uncovering Higher-Order Interactions from Memory in Observed Dynamics

Our Ph.D. student Arthur Rondeau has posted a new preprint on arXiv, together with Alain Barrat (Aix Marseille Université, CNRS, Centre de Physique Théorique). When all we observe is a system’s dynamics — for instance a sequence of visited nodes — how can we tell whether group (higher-order) interactions are really at play, or whether pairwise interactions suffice? Group interactions leave a trace as memory in the observed sequence, but a memory-driven pairwise process can mimic that same footprint. The paper introduces an inference pipeline built on a hidden Markov model whose latent states are the hyperedges of an unobserved hypergraph, turning the question into a falsifiable model-selection test against memoryless and memory-based pairwise baselines. Applied to ball-possession trajectories from professional football matches (StatsBomb open data), the recovered hyperedges map directly onto recognisable tactical team formations.

Uncovering Higher-Order Interactions from Memory in Observed Dynamics
Rondeau A, Barrat A & Bouffanais R
arXiv preprint arXiv:2610.20095 [cs.SI], October 2026. [pdf] [arXiv] [doi]