{"id":"a3e21a60-3ae3-4977-8e8e-cebdc43e6448","arxiv_id":"2507.16102","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The milling-to-schooling transition in the three-A fish model is claimed to be a noisy transcritical bifurcation, with collective memory arising from noise-induced delay rather than structural bistability.","lead":"This paper studies a standard computer model of schooling fish and argues that the switch from milling to schooling is a noisy transcritical bifurcation: the two patterns trade stability, and randomness decides which one wins. The result challenges the usual view that collective memory requires two stable states, and it may help interpret hysteresis in other animal groups.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The transcritical mechanism is not yet established: Eq. 3 is posited rather than derived, and the Kramers fit in Eq. 4 uses a well formula on a negative barrier, leaving a narrow subcritical pitchfork as a viable alternative.","rationale":"The reader's weakest assumption correctly identifies the load-bearing issue: the normal form in Eq. 3 is assumed, not derived, and the central conclusion about bifurcation type and hysteresis follows from that assumption rather than from a direct analysis of the agent-based model. My stress-test confirms this and adds a concrete technical flaw in Eq. 4: for r > 0, the purported Kramers escape rate is computed across a negative barrier, so the fit in Fig. 4F does not provide independent support for the transcritical normal form. The paper is honest about not deriving Eq. 3, and the qualitative match with the fish simulations is genuine, but the evidence presented cannot distinguish a transcritical bifurcation from a subcritical pitchfork with a narrow bistable interval. Since the reader's verdict is already CONDITIONAL, no verdict adjustment is needed; the proposed drift-reconstruction test is the natural next step to settle the mechanism.","tokens_in":13562,"tokens_out":4263,"duration_ms":48499,"concrete_test":"Measure the effective drift from agent-based simulations at fixed zl: run many short trajectories, bin P, compute b(P) = lim_{dt->0} <Delta P | P>/dt, and integrate b to obtain U(P), following the coarse-graining approaches cited in Refs. 24/34. Do this for zl = 2.8, 3.1, and 3.4 at sigma = 0.01, and repeat at sigma = 0.005 as a control. Transcritical predicts exactly one minimum of U on each side of zl* (near 0 for zl < z*, near 1 for zl > z*), whereas a subcritical pitchfork predicts two local minima on the milling side. If the reconstructed potential is single-welled for zl = 2.8, the paper's mechanism survives; if double-welled, the central conclusion must be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that the milling-to-schooling transition is a noisy transcritical bifurcation, not structural bistability. The load-bearing evidence is the phenomenological SDE dP = rP(1-P)dt + sigma dW (Eq. 3) and the associated Kramers fit (Eq. 4). However, Eq. 3 is posited, not derived from the agent rule (Eq. 1), and no test is performed that would falsify the competing subcritical-pitchfork picture. The bimodality in Fig. 4C and the 0/50/100% schooling fractions in Fig. 4B are also what a subcritical pitchfork with a narrow bistable interval would produce in finite simulation time, especially with noise sigma = 0.01. The Kramers argument does not remove this ambiguity: for r > 0 the potential U = -r(P^2/2 - P^3/3) has its minimum at P = 1 and a maximum at P = 0 (negative barrier), so Eq. 4 is not the standard escape rate from a metastable well; it is a rate of leaving an unstable state, and its fit in Fig. 4F is therefore not a valid discriminator. The additive noise in Eq. 3 also has no reflecting boundaries, so P can leave [0,1], reinforcing that Eq. 3 is a cartoon rather than a validated reduction. Because the mechanism claim rests on this unvalidated reduction, the distinction between 'noisy transcritical bifurcation' and 'narrow structural bistability' remains open.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper revisits the three-A model of fish schooling by Couzin et al. and examines the transition from milling to schooling as the alignment zone z_l is varied. Through Monte Carlo simulations, the authors find an abrupt transition near z_l* ≈ 3.1, with apparent bistability in a narrow window, and noise-dependent hysteresis. They propose that these features are signatures of a noisy transcritical bifurcation rather than structural bistability, and introduce a phenomenological stochastic differential equation for the polarization P, dP = rP(1-P)dt + σdW, with r = z_l - z_l*. They fit a Kramers escape-rate formula to the residence-time data and argue that the hysteresis behavior of the fish model matches that of the SDE.","tokens_in":13911,"tokens_out":4949,"duration_ms":46418,"significance":"If the mechanism were established, the paper would resolve a long-standing ambiguity about the origin of hysteresis in a widely used model of collective motion. The manuscript is also valuable for its systematic phase diagram and for characterizing how hysteresis depends on noise level. However, the central identification is not currently established: the SDE in Eq. (3) is posited rather than derived, the Kramers comparison in Eq. (4) has a sign-inconsistent barrier, and the presented observables do not discriminate between a transcritical bifurcation and a narrow subcritical pitchfork. The manuscript therefore requires additional analysis before its central claim is supportable.","major_comments":[{"comment":"The central mechanism is not derived from the agent-based update rule in Eq. (1). The text explicitly states that a formal proof would require deriving the equations for P and M and that, instead, the authors 'sought a phenomenological model' and 'postulate an effective potential.' Because Eq. (3) is assumed, the agreement of its predictions with fish simulations (bimodal histograms, noise-dependent hysteresis) is not a validation of the mechanism; the escape-rate comparison is a two-parameter fit (Eq. (4) and Fig. 4F), so it is not an independent prediction. Moreover, the bimodal histograms in Fig. 4C and the 0/50/100% schooling fractions in Fig. 4B are also what a subcritical pitchfork with a narrow bistable interval would produce in finite simulation time, especially with σ = 0.01. To establish the transcritical mechanism, the paper needs a test that discriminates between the two normal forms, such as measuring the deterministic drift from simulation data or demonstrating that the apparent bistability vanishes in the infinite-time limit (or persists as genuine two-well behavior).","section":"Section 'Signature of Noisy Transcritical Bifurcation', Eq. (3)"},{"comment":"The Kramers formula is applied with an incorrect potential barrier. For U = -r(P^2/2 - P^3/3), one has U(1) - U(0) = -r/6, not +r/6 as stated in the text. For r > 0, P = 1 is the global minimum and P = 0 is a local maximum; there is no potential barrier to escape from P = 0 to P = 1. The expression κ = A r e^{-r/(3σ^2)} is therefore not the standard Kramers escape rate from a metastable well, and the fit in Fig. 4F does not provide evidence supporting the transcritical normal form. This is a load-bearing error because the escape-rate comparison is one of the two quantitative links between the fish simulations and the SDE in Eq. (3).","section":"Section 'Residence Time and Escape Rate', Eq. (4)"},{"comment":"The additive noise in Eq. (3) has no reflecting or absorbing boundaries, so P can leave the interval [0,1] where the polarization order parameter defined in Eq. (2) is confined. The Kramers escape-rate calculation in Eq. (4) likewise presumes well-defined local minima of a confining potential; with unbounded diffusion, P = 0 is a singular point rather than a metastable state. This reinforces that Eq. (3) is a coarse phenomenological 'cartoon' (the authors' own characterization is 'phenomenological model') and the paper should specify how the SDE is meant to be interpreted, e.g., with boundaries or as an approximation valid only in the interior of [0,1]. Without such specification, the quantitative agreement in Fig. 4F and the hysteresis loops in Fig. 5D-E are not well-defined predictions.","section":"Eq. (3) and interpretation of P"}],"minor_comments":[{"comment":"The section heading 'T ransition from Milling to Schooling' contains a stray space before 'r'.","section":"Section heading, p. 7"},{"comment":"Reference 51 (Kramers, Nature 1924) is the wrong citation for the escape-rate formula used in Eq. (4); the standard reference is Kramers, Physica 7, 284-304 (1940).","section":"Reference 51"},{"comment":"The caption calls panel D a 'Transcritical bifurcation plot,' but the panel shows Monte Carlo trajectories of the SDE in Eq. (3) in finite time, not a deterministic bifurcation diagram; this labeling could mislead readers into thinking the panel is derived from the fish simulations.","section":"Fig. 4D caption"},{"comment":"The caption states 'In all simulations, noise intensity σ = 0.01,' but the paper later reports results at σ = 0.05 and 0.10; the caption should say 'unless otherwise stated' or specify the default value.","section":"Fig. 2 caption"},{"comment":"Reference 12 has a truncated DOI: 'https://doi.org/10.1073/pnas.0711' is incomplete.","section":"Reference 12"}],"recommendation":"major_revision","confidential_remarks":"The manuscript would benefit from a data-driven derivation or reconstruction of the effective SDE (e.g., via Kramers-Moyal expansion or a projection of the agent-based dynamics) or a direct quantitative comparison with the subcritical-pitchfork alternative. Without such work, the central claim is not supported beyond qualitative agreement. The authors' own admission that the normal form is 'postulated' should be taken seriously in the revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"What you should know: this paper argues that the well-known hysteresis in the three-A model is not structural bistability but a noise-induced effect near a transcritical bifurcation. That is a worthwhile and genuinely different way of reading the old Couzin et al. results. But the supporting evidence is not yet good enough to close the question.\n\nThe paper does several things well. The sweep over noise levels (Fig. 5) is clearly presented, and the observation that hysteresis is strongest at intermediate noise and disappears at high noise is a real, nontrivial result that deserves attention. The authors are also explicit that Eq. 3 is postulated, not derived, and they frame it as a phenomenological model. The conceptual point that hysteresis alone does not imply structural bistability is correct and worth making in this literature.\n\nThe soft spots are concentrated in the identification of the bifurcation type. First, the Kramers calculation in Eq. 4 has a sign problem: for r>0 the effective potential U=-r(P^2/2-P^3/3) has U(1)-U(0) = -r/6, not +r/6. So the 'barrier' they put into the escape rate is negative, meaning there is no activated escape from P=0 to P=1; the flow is already down hill. Fitting A*r*exp(-r/3σ^2) to the measured escape rate is therefore not a standard Kramers fit and cannot be taken as evidence of noise-activated crossing. Second, the normal form in Eq. 3 is assumed, and the comparison in Fig. 4D is a two-parameter fit of this same SDE to the fish data, so it cannot independently validate the mechanism. Third, the data in Fig. 4C show a narrow parameter window around z_l* where both states are visited; that is exactly what a subcritical pitchfork with a small bistable region would produce over finite simulation times. To distinguish the two, one would need very long runs to see whether the system eventually converges to a single globally stable state (transcritical) or keeps switching between two true wells (subcritical pitchfork). The paper does not provide that test.\n\nSo the central claim remains open. But it is an important question, and the paper is honest enough to warrant engagement. I would send it to peer review with a request for the authors to fix the sign error and add a discriminating simulation or a proper reduction from the microscopic rules. As it stands, the mechanistic attribution outruns the evidence.","headline":"Claims noisy transcritical bifurcation for hysteresis in three-A fish model, but the Kramers fit has a sign error and the normal form is assumed, leaving a narrow subcritical pitchfork still viable.","tokens_in":14425,"tokens_out":4089,"would_cite":false,"duration_ms":41986,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37N25","60H10","92D50"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper argues that collective memory in the three-A fish-school model comes from noise near a transcritical bifurcation, not from two coexisting stable states.","keywords":["collective motion","fish schooling","milling","hysteresis","collective memory","transcritical bifurcation","stochastic differential equations","three-A model"],"falsifier":"Measure the residence time in the milling state at alignment-zone sizes just above the transition, for several noise levels, and fit the escape rate to the Kramers form $\\kappa = A\\,r\\,e^{-r/(3\\sigma^2)}$. The paper's mechanism predicts the effective barrier $r/6$ vanishes as $z_l$ approaches $z_l^*$ from above, while structural bistability would keep a finite barrier at the transition; a nonzero extrapolated barrier would falsify the noisy-transcritical claim.","tokens_in":13320,"feed_emoji":"🐟","tokens_out":15454,"duration_ms":155219,"temperature":0.7,"pith_summary":"Schooling fish in a popular agent-based model, the three-A model, appear to remember their past: slowly changing individual behavior from milling to schooling and back gives different patterns in each direction. The paper argues this hysteresis is not evidence that milling and schooling are two equally stable states. Instead, the two states exchange stability at a critical alignment-zone size, and finite noise lets the group linger in the state that has just lost stability, creating memory-like path dependence. The claim is supported by Monte Carlo simulations, a one-variable stochastic model for group polarization, and a Kramers escape-rate fit. If correct, it resolves a long-standing ambiguity about the origin of collective memory in a widely used model and cautions that observed bistability-like behavior can arise without structural bistability.","feed_headline":"Fish school memory is a noise effect, not true bistability","feed_subtitle":"A noisy transcritical bifurcation explains milling-to-schooling hysteresis in the three-A fish model.","key_machinery":"The carrying object is the one-dimensional stochastic differential equation $dP = rP(1-P)\\,dt + \\sigma\\,dW$ (Eq. 3), with effective potential $U = -r(P^2/2 - P^3/3)$: the normal form of a transcritical bifurcation perturbed by white noise. Here $P$ is group polarization, $r = z_l - z_l^*$ is the distance from the milling-to-schooling transition, and $\\sigma$ is collective noise. The equation is posited as a phenomenological reduction, not derived from the agent-based update rule; at $r=0$ its drift vanishes, the effective potential is flat, and the dynamics are purely diffusive. It carries the argument by generating the observed finite-time bistability, the Kramers escape-rate scaling $\\kappa = A\\,r\\,e^{-r/(3\\sigma^2)}$, and the noise-dependent hysteresis, which lets the paper attribute the fish-model behavior to the bifurcation type of this SDE.","core_discovery":"Rotationally milling and polarized schooling, measured by polarization $P$ and rotation $M$, appear to coexist in a narrow window of alignment-zone sizes around $z_l^* \\approx 3.1$ (with attraction zone $z_a=10$ and noise $\\sigma=0.01$): some simulations remain in milling while others switch to schooling. The paper's central claim is that this window is not structural bistability but the signature of a noisy transcritical bifurcation. It posits the phenomenological stochastic differential equation $dP = rP(1-P)\\,dt + \\sigma\\,dW$, with $r = z_l - z_l^*$ and effective potential $U = -r(P^2/2 - P^3/3)$. In the deterministic limit, $P=0$ (milling) and $P=1$ (schooling) persist for all $r$ and exchange stability at $r=0$; with noise, the group can escape from the newly unstable milling state, with an escape rate matching Kramers' formula $\\kappa = A\\,r\\,e^{-r/(3\\sigma^2)}$. This reproduces the key observations: transient milling beyond the transition, a noise-dependent hysteresis loop that is strongest at intermediate noise and disappears at high noise, and collective memory without a double-well potential.","pith_inferences":["A natural next step is to construct the group-level effective potential directly from the agent-based trajectories by coarse-graining; the transcritical claim predicts a single-well potential whose minimum flattens and shifts near $z_l^*$, whereas structural bistability would show a double well. The paper gestures at this comparison but does not carry it out.","This suggests an experimental handle: if confined fish schools show intermittent milling-schooling transitions, adding or reducing sensory noise should sharpen or broaden the transition in a specific way under the transcritical mechanism, separating it from the double-well mechanism in real animals.","The same normal-form reasoning could be applied to self-propelled particle models with rotational inertia or boundary confinement; their hysteresis should scale differently with noise if their transitions are truly pitchfork-like, making noise dependence a fingerprint of the bifurcation type.","If Eq. (3) is taken literally, it predicts a specific residence-time distribution for the milling state past the transition; fitting that full distribution, rather than only the average escape rate, would test whether the assumed one-variable additive-noise reduction is the right one."],"forward_implications":["The milling-to-schooling transition in the three-A model is controlled by the alignment-zone radius $z_l$, with the attraction-zone radius having little influence once cohesion is maintained; the abrupt switch occurs near $z_l^* \\approx 3.1$ at $z_a=10$.","Hysteresis and collective memory in this model are noise-dependent: weak noise gives memory mainly on the backward sweep, intermediate noise gives a strong hysteresis loop, and strong noise destroys milling and the memory effect.","Observed bistability near the transition is a finite-time, noise-driven phenomenon; in infinite time the system would follow the newly stable schooling state for $r>0$, so bistability is degenerate and localized at the bifurcation point.","The escape rate from milling to schooling obeys Kramers' formula $\\kappa = A\\,r\\,e^{-r/(3\\sigma^2)}$, with a barrier that shrinks to zero at the bifurcation, quantitatively supporting the transcritical normal form.","Collective path dependence alone does not diagnose structural bistability: hysteresis can reflect noise near a stability exchange, so observed bistability-like behavior in animal groups must be interpreted through the underlying mechanism."],"supporting_citations":[{"why":"Introduces the three-A model and the collective-memory/hysteresis behavior that the paper sets out to explain.","marker":"[1]"},{"why":"Draws the analogy between collective bistability and neural multistability that the paper revisits, and reports hysteresis at comparable noise.","marker":"[4]"},{"why":"Provides the contrast case of structural bistability in confined fish schools and the effective-potential methods the paper distinguishes from its single-well picture.","marker":"[24]"},{"why":"Supplies the empirical observations of collective states and transitional behavior in schooling fish that motivate the biological question.","marker":"[31]"},{"why":"Provides the coarse-grained stochastic-switching and effective-potential approach the paper cites as the alternative route to deriving group-level dynamics.","marker":"[34]"},{"why":"Supplies the random-dynamical-systems background for interpreting stability exchange under noise.","marker":"[44]"},{"why":"Supplies the standard normal form and exchange-of-stability description of the transcritical bifurcation.","marker":"[45]"},{"why":"Supplies Kramers' formula used to fit the escape rate from milling to schooling.","marker":"[46]"}],"fun_headline_variants":["Fish memory is a noisy bifurcation, not bistability","Noise, not bistability, drives fish collective memory","Fish school memory arises from noise, not structure","Noisy transcritical bifurcation explains fish memory"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes, rather than derives, that group polarization obeys the simple stochastic equation whose deterministic part is the transcritical normal form; if the many-fish dynamics cannot be reduced to that single variable, the claim that the transition is a noisy transcritical bifurcation loses its foundation.","fun_headline_variants_meta":{"raw":{"variants":["Fish memory is a noisy bifurcation, not bistability","Noise, not bistability, drives fish collective memory","Fish school memory arises from noise, not structure","Noisy transcritical bifurcation explains fish memory"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000475,"raw_usage":{"total_tokens":2384,"prompt_tokens":1000,"completion_tokens":1384,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":616,"completion_tokens_details":{"reasoning_tokens":1321}},"tokens_in":616,"tokens_out":1384,"duration_ms":14047,"temperature":1.0,"reasoning_tokens":1321,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:18:27.235833+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the residence time in the milling state at alignment-zone sizes just above the transition, for several noise levels, and fit the escape rate to the Kramers form $\\kappa = A\\,r\\,e^{-r/(3\\sigma^2)}$. The paper's mechanism predicts the effective barrier $r/6$ vanishes as $z_l$ approaches $z_l^*$ from above, while structural bistability would keep a finite barrier at the transition; a nonzero extrapolated barrier would falsify the noisy-transcritical claim.","supporting_citations":[{"cited_title":"D., Krause, J., James, R., Ruxton, G","cited_arxiv_id":null,"evidence_quote":"Introduces the three-A model and the collective-memory/hysteresis behavior that the paper sets out to explain."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Draws the analogy between collective bistability and neural multistability that the paper revisits, and reports hysteresis at comparable noise."},{"cited_title":"& Kanso, E","cited_arxiv_id":null,"evidence_quote":"Provides the contrast case of structural bistability in confined fish schools and the effective-potential methods the paper distinguishes from its single-well picture."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the empirical observations of collective states and transitional behavior in schooling fish that motivate the biological question."},{"cited_title":"& Kevrekidis, I","cited_arxiv_id":null,"evidence_quote":"Provides the coarse-grained stochastic-switching and effective-potential approach the paper cites as the alternative route to deriving group-level dynamics."},{"cited_title":"Random Dynamical Systems (Springer-Verlag, Berlin Heidelberg, 1998)","cited_arxiv_id":null,"evidence_quote":"Supplies the random-dynamical-systems background for interpreting stability exchange under noise."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the standard normal form and exchange-of-stability description of the transcritical bifurcation."},{"cited_title":"Thinking Probabilistically: Stochastic Processes, Disord ered Systems, and Their Applications isbn: 9781108479523","cited_arxiv_id":null,"evidence_quote":"Supplies Kramers' formula used to fit the escape rate from milling to schooling."}],"review_version":1}