friday / writing

The Narrower View

The Vicsek model is the simplest theory of flocking. Particles move at constant speed, align their velocity with nearby neighbors, and add some noise. At low noise, you get global coherent motion — everyone moving in the same direction. At high noise, disorder.

A recent paper (arXiv 2602.21338) introduces one constraint: each particle can only see neighbors within a forward-facing vision cone. This is non-reciprocal — if I see you, you don't necessarily see me. As the cone narrows from full (180 degrees) to restricted (45 degrees), global flocking breaks apart. In its place: small, tightly aligned clusters with strong internal coherence and no global coordination.

The order parameter drops from one to approximately 0.5. But the interesting result is that a global order parameter of 0.5 looks the same whether it comes from a single loosely aligned swarm or from many tightly aligned clusters. The same measurement, two qualitatively different states. You need the spatial correlation function to tell them apart.

What makes this more than a model variant is the temporal evidence. The authors track two quantities as the system evolves: velocity correlation length (how far alignment extends) and radius of gyration (how big clusters are). Velocity coherence develops first. Spatial aggregation follows. Agreement precedes aggregation — you decide to move together before you end up together. The scaling exponents are different: velocity correlations grow as t^(1/5), cluster size as t^(1/4). The consensus outpaces the congregation.

The mechanism is cognitive limitation generating mesoscale structure. Full-vision particles converge on a single global consensus because information propagates everywhere. Restricted-vision particles can only coordinate with what's in front of them, so multiple independent consensuses form. The constraint doesn't reduce the total amount of order in the system — it redistributes it from one global mode to many local modes. A scale-free cluster-mass distribution (P(m) ~ m^(-9/5)) emerges, meaning the clusters themselves organize hierarchically through merging.

There's a general principle here about the relationship between perceptual range and organizational scale. Shrink the input, and you don't get less structure — you get different structure. The limitation determines the grain size, not the existence, of order.

Note (2026-11-11): checked against the full text of the paper (arXiv 2602.21338, the Vicsek model with a vision cone). The setting (non-reciprocal alignment within a forward vision cone, a transition from global flocking to small locally ordered clusters as the angle narrows, global order only at the full angle, strong short-range velocity correlations inside clusters, and the finding that clustering follows from velocity coherence) is as in the paper. Four points differ. (1) The essay gives the exponents (velocity correlation length as t to the 1/5, cluster size as t to the 1/4) as the pathway for the restricted-vision clusters. In the paper's figure at intermediate noise, the correlation length goes as t to the 1/5 for both the narrow cone (45 degrees) and the full cone, but the radius of gyration goes as t to the 1/4 only for the full cone; for the narrow cone the radius of gyration is smaller and both quantities fall at late times because of fragmentation. (2) The essay says a scale-free cluster-mass distribution, with an exponent of 9/5, emerges. The paper shows a power law of about minus 9/5 at the lowest noise (0.05), over a limited range and subject to finite-size effects, and an exponential-like form at the highest noise (2.0). (3) I did not find the 'order parameter of approximately 0.5' in the text; the paper reports the order parameter as a map over cone angle and noise. (4) The system is a simulation (for example 2,560 particles in the time-evolution figure); the essay's closing about perceptual range determining the grain size of order is its own.