Fish schools may be able to stay coordinated without becoming rigid. A review of flocking models, social-network simulations and fish movement data describes a system in which alignment keeps the group together while bursts of collective turning preserve responsiveness. It brings together work on network structure and interaction weights, selective responses to neighbors, temporary leadership and cascade-like turning dynamics. The central idea is interpretive: heterogeneous, changing interactions may allow order and rapid collective adjustment to coexist.
Uneven networks change the threshold
The network models point to a sharp difference in how much noise a group can withstand. In these models, a node represents an individual and its degree is the number of connections it has. For scale-free networks, degree-distribution exponents below 5/2 were associated with critical noise tending to 1. Above 5/2, the models produced a finite critical-noise level in the large-network limit. Critical noise here is the model's threshold for losing ordered motion as random variation increases. This is a theoretical result about the specified networks, rather than a measurement from a single fish school.
Adding weights to social ties changed the picture. In comparisons of empirical network structures, weighted interactions were generally associated with lower critical-noise thresholds than binary interactions, meaning every tie was treated as simply present or absent. The review reports this pattern across 20 empirical weighted social networks, with reductions ranging from about 1% to more than 50%. A finite-size mean-field calculation nevertheless predicted the highest threshold at a weight exponent of one-half. The model's finite-size optimum and the comparisons of observed networks are related findings, but they are not the same test.
Selective attention changes the picture
Selective attention offered an explanation for an apparently contradictory movement pattern. A model that responded to faster neighbors while ignoring slower ones reproduced alignment with the faster fish and apparent anti-alignment with the slower fish. It did so without an explicit repulsive rule. In this account, the effect comes from which neighbors are included in the calculation, not from a separate command to move away. The review presents this as a model-based explanation supported by trajectory analysis.
Speed also emerged as a marker of temporary leadership. Time-delayed velocity correlations showed positive-delay peaks for faster individuals and negative-delay peaks for slower individuals, a pattern consistent with leader and follower roles that changed over time. In that measure, a positive delay means the nearest neighbor's matching orientation appeared later, while a negative delay means it appeared earlier. The result associates relative speed with momentary influence; it does not show that speed alone makes other fish follow.
Turning events arrive in cascades
Turning data supplied a second signal. Turning avalanches, bursts of direction changes that cluster through a school, had power-law tails in both duration and size. The reported exponents were about 2.4 for duration and 2.0 for size, across turning thresholds and school sizes from 8 to 50 fish. These tails describe how event frequency falls as duration or size grows; randomized temporal sequences instead showed exponential decay. The average size-duration relationship produced a scaling exponent of about 1.31 to 1.41. The review reads that contrast as support for critical-like dynamics in the analyzed datasets.
Influence was concentrated rather than evenly shared in the avalanche analysis. Groups of 40 fish were compared with an uncorrelated independent-turn null model, a baseline in which turns are treated as independent. Several fish in most experimental series had leadership probabilities above the null model's 99% confidence interval and were identified as effective leaders. Aftershock-like clustering lasted less than about 250 frames. Within that window, the gap between a parent turn and a later aftershock followed an Omori pattern, in which the probability of another event fell with the time gap. The fitted parameters were c = 4.3 plus or minus 0.4 and p = 2.2 plus or minus 0.1. Since p was greater than 1, the reported aftershock activity decayed rapidly.
A framework still under test
The authors draw these results into a picture of dynamically heterogeneous interaction networks that support stable alignment while critical-like fluctuations preserve responsiveness. The evidence comes from model extensions, numerical simulations, weighted-network analyses and trajectory-based methods. One force-map or averaging approach reconstructs effective interaction forces from trajectories by relating a focal fish's acceleration to selected variables; the paper applies it to Hyphessobrycon herbertaxelrodi. The weighted-network analysis spans species such as bison, dolphins, mice and sea lions. The document is an arXiv version 1 preprint dated 29 September 2026.
Paper data and sources
Original title: Statistical Physics of Fish Collective Motion
Authors: Elena G. de Lamo, M. C. Miguel, R. Pastor-Satorras
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-09-29
DOI: Not available
Original paper · Full text