Glassfish school by copying one neighbor at a time, not by averaging the crowd

The distance between the two rules is arithmetic. Give a fish n neighbors and an averaging rule hands each of them a 1/n share of the next heading, so one neighbor's odd turn arrives diluted by the crowd it sits in; the copying rule hands the selected neighbor the whole share and the rest nothing ^[15]. Puri and colleagues report that the pairwise version produces the dynamic behavior they see, including groups forming and breaking apart, and that a classical alignment-to-the-average model predicts something different ^{[4] [8]}.

That is hard to settle from trajectories alone, because in a free-swimming school nobody controls what the neighbors do. Physicists have modelled collectives as active matter for a long time, and the standing complaint has been how weakly those models connect to what animal groups actually do ^[10]. The virtual reality step supplies the missing control: the authors programmed virtual fish to change direction, and the real glassfish responded the way the copying model said they would ^[5].

Timing is the part most likely to bite existing models. Influence here is not a force acting continuously. The end of one fish's turn opens what the authors call a window of opportunity for its neighbors to align with it ^[6], and Lovett-Barron describes each fish as sensing and responding at discrete moments and running a specific computation at those moments ^[12]. Mature glassfish produce these copying turns with fast reaction times ^[7], which is what makes the give-and-take legible turn by turn rather than as a smooth drift.

The neurobiology follows from the timing rather than from the schooling. Aljadeff's stated interest is pairing the identification of precisely timed social interactions with brain-wide imaging to see how processing in one brain supports the behavior of the group ^[11], and Danionella cerebrum is the animal for it because the fish are transparent enough to record across the whole brain ^{[12] [1]}.

The turn-by-turn account spans a range of developmental stages ^[13], and this group had already argued that the copying rule explains how schooling appears as the fish mature ^[9]. What the announcement does not give is a latency in milliseconds or the group sizes examined ^[14], and those are the first two numbers anyone re-fitting an averaging model would ask for.