{"id":"46a2c3f4-1951-415e-9b24-f62f1529e639","arxiv_id":"2504.16243","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A nonlocal advection-diffusion PDE for vision-controlled Janus particles reproduces agent-based clustering and reveals an annular density structure inside the swarm.","lead":"This paper builds a partial differential equation model for swarms of Janus particles that move randomly, sense neighbors in a vision cone, and activate forward motion when enough neighbors are seen. The model reproduces the clustering seen in earlier simulations and reveals a ring-shaped density pattern inside the cluster that prior work did not report.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The annular peak sits exactly at the region where the model's admitted particle overlap is most severe; without a short-range repulsion test, the paper's novel central structure may be a no-excluded-volume artifact.","rationale":"The reader's weakest assumption, missing excluded volume, is the same one I judge load-bearing, and the paper itself flags it. My stress test adds detail: the singular 1/d perception kernel makes the overlap assumption interact directly with the activation mechanism that produces the ring, so a qualitative 'we do not anticipate' statement is not enough. The positive side is that the PDE is a natural Fokker-Planck form of the agent-based rules, the code is public, mass conservation is checked, and the agent-based versus PDE histograms agree at increasing n; these are real independent supports. Neither a formal derivation of the n-to-infinity limit nor direct quantitative comparison to experimental data from [31] is present, but those are secondary to the novel annular claim. Because the weakness is addressable by a well-specified numerical experiment and the rest of the model behavior is plausible, the appropriate stance remains conditional: accept only after the repulsion test is run. Consequently I do not change the reader's verdict.","tokens_in":10254,"tokens_out":7919,"duration_ms":86147,"concrete_test":"Run the agent-based simulator with a hard-disk or Weeks-Chandler-Andersen excluded-volume interaction of particle diameter 4.28 um for the nominal case (alpha = pi/2, P* = Pc; n = 200, 1000, 5000; 250 um domain), using the same time-averaged radial histogram protocol as Fig. 6(A). Compute the radial density profile and compare the peripheral peak height and position with the no-repulsion run; if the peak drops below the interior density by more than the histogram noise, or shifts by more than one particle diameter, the annulus is not robust to excluded volume.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central novelty is the annular density profile, and its own Discussion concedes that the model 'ignore[s] this size and therefore allow[s] for (unrealistic) particles to overlap.' This assumption is not cosmetic because the perception kernel in Eqs. (1), (5)-(6) is singular, with w proportional to 1/d. In the agent-based model, overlapping particles at unphysically small separations contribute very large weights to the perception sum, which can flip the activation threshold S from 0 to 1 in Eq. (7). In the PDE mean-field equation (9), the same 1/d convolution feeds the threshold, and the advection term then drives these particles outward; this is precisely a mechanism that could pile density into the reported peripheral ring. The ring is the high-density region where overlap is most probable, so the missing excluded volume is not a small quantitative correction: it is potentially the origin of the structure being claimed. The paper offers only an unsupported 'we do not anticipate' statement, not a test. Since the claim that the PDE highlights a real annulus depends on the ring surviving a physically realistic hard-core or short-range repulsive interaction, this is the load-bearing weak point.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a nonlocal advection-diffusion PDE, Eq. (9), for the density ρ(θ, x, t) of orientable Janus particles whose motility is controlled by vision-cone perception and a threshold activation rule, and it compares this PDE with an agent-based simulation of the same rules. The paper reports that both models produce disk-shaped cohesive groups for moderate parameters and that, in addition, the time-integrated agent histograms and the PDE end-states exhibit a peripheral density peak, giving an annular structure not highlighted in the prior experimental study [31]. A 24×24 parameter sweep over vision angle α and threshold ratio P*/Pc_α yields phase portraits for group formation and for cohesion, and the authors conclude that the PDE recreates the behavior seen in experiments and simulations.","tokens_in":10497,"tokens_out":8021,"duration_ms":88694,"significance":"If the central claims hold, the paper offers a computationally cheap continuum description of an experimental active-particle system whose interactions are known, with all physical parameters inherited from [31] rather than fitted, openly available code, and a falsifiable prediction (the annular density profile) that can be checked with time-integrated imaging or larger particle numbers. The time-averaged histogram approach for comparing discrete and continuum models is a useful methodological contribution. However, the annular structure is the paper's main novel finding, and its physical robustness is not yet established; in particular, the effect of omitted excluded volume on the peripheral density peak is a load-bearing open question. The significance of the paper therefore depends critically on the response to Major Comment 1.","major_comments":[{"comment":"The model omits excluded volume even though the experimental particles have diameter 4.28 µm, and the perception kernel in Eq. (1) is singular, w ~ 1/d. In the agent-based model, overlapping particles at unphysically small separations contribute very large weights to the perception sum and can flip the activation state S in Eq. (7); in the PDE, the same 1/d convolution feeds the threshold in Eq. (6), and active particles advect outward in Eq. (9), which is exactly a mechanism that can pile density into a peripheral ring. Since the ring is the high-density region where particle overlap would be most common, the omitted excluded-volume interaction is not a cosmetic correction: it may be the origin of the claimed annular structure. The sentence \"we do not anticipate that this would substantially affect the nature of the results reported here\" is not a substitute for a test. The authors should add a short-range repulsion or hard-core-like regularization to both the agent-based model and the PDE and show that the annulus persists, in both height and existence, over the parameter ranges shown in Figs. 2 and 7; alternatively, they should provide a quantitative argument for why overlap is negligible in the relevant density regime.","section":"Discussion (final paragraph); Eqs. (1), (6), (9)"},{"comment":"Eq. (9) is introduced as the PDE \"that approximates the limit n→∞\" and §III B 1 concludes that \"the PDE model does indeed represent a continuum limit of the agent-based model,\" but no mean-field or kinetic derivation of Eq. (9) from Algorithm 1 is given. The evidence in Fig. 6(A) is a single parameter set, compared visually and with time-averaged histograms normalized to the PDE mass. To support the \"continuum limit\" wording, either provide a formal derivation (e.g., from an interacting-particle or kinetic formulation) or reframe Eq. (9) as a phenomenological continuum model whose validity is demonstrated numerically. In either case, the convergence in Fig. 6(A) should be quantified with a metric such as the relative L1 or L2 difference between the PDE cross-section and the agent histograms for n = 500, 1000, 2000, and 5000.","section":"§II B and §III B 1"}],"minor_comments":[{"comment":"The definition K_θ(s) := w(0, -s) is hard to follow because w is originally defined with two position arguments; please write the kernel explicitly in terms of |s|, the half-angle α, and the orientation θ, and clarify the sign convention leading to the convolution in Eq. (6).","section":"§II B, Eq. (5)"},{"comment":"The total mass N used in the PDE and its relationship to the agent number n are not stated explicitly. Since V in Eq. (6) scales with the total density, the choice of N and the precise setting of P* (absolute value versus ratio P*/Pc_α) are needed for reproducibility and for interpreting the normalization used in Fig. 6(A).","section":"§II C and §III B 1"},{"comment":"The phase diagrams in Fig. 7 are produced from single PDE runs at Nx = 64, and the cohesion panel explicitly acknowledges discretization artifacts (square/rectangular collapse aligned with the grid). Please include a reproducibility check for at least the phase boundaries, such as a second run with different initial conditions or a coarser/finer grid, to show that the reported boundaries are not numerical artifacts.","section":"§III C, Fig. 7"},{"comment":"The claim of \"near perfect rotational symmetry\" for Nθ ≥ 30 is based on visual inspection of cross-sections; a quantitative measure, such as the amplitude of low-order angular Fourier modes of ρ̄ around the center of mass, would make this claim precise and verifiable.","section":"§III A 4, Fig. 5"},{"comment":"The phrase \"bifurcation structure is then extensively explored\" overstates what is done in the paper; the results are phase portraits of end-states, not computed bifurcation curves. Please rephrase to match the actual content.","section":"§I and §III C"},{"comment":"The abstract's statement that the PDE \"highlights an annular structure\" should make clear that this structure is apparent in time-integrated agent histograms and continuum densities, not in raw experimental snapshots with n = 75; otherwise readers may expect the annulus to be directly visible in the original experiments of [31].","section":"Abstract and §IV"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a reasonable fit for nlin.PS and the code availability plus parameter-free comparison to [31] are strengths. My main concern is not circularity or novelty disclosure but physical robustness: the central new observation, the annular density peak, may be an artifact of allowing unlimited particle overlap in a model with a 1/d perception kernel. I would be willing to accept after the authors provide the excluded-volume test and a quantitative consistency measure; without those, the headline claim is not yet supported at the level a journal should require."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does something worth having: it writes down a nonlocal advection-diffusion PDE for the vision-cone activation rule from Lavergne et al. and shows that it reproduces the agent-based model, both in aggregate density and in the active/inactive split. The genuinely new item is the annular density peak in the continuum solution, which the original experimental paper did not report. The numerics are careful: time-averaged histograms, cross-section comparisons, and convergence checks in particle number, time, and angular/spatial resolution all support the claim that the PDE is a faithful large-population description of the agent rule. The code is public. That is real work, and the phase diagrams for formation versus cohesion are a useful map for anyone who wants to test this system experimentally.\n\nThe soft spots are real but not fatal. First, the PDE is posed as a mean-field limit rather than derived; the convergence evidence is good, but a rigorous derivation would remove a lingering gap. Second, the comparison to experiments is qualitative only — the abstract says the PDE \"can recreate\" experimental behavior, but there is no quantitative fit to the particle positions in [31], and the experimental paper did not report the annulus, so the ring remains a prediction, not a validated observation. Third, the missing excluded-volume interaction is exactly where the stress lands: the perception kernel is 1/d, so overlapping particles in the agent model contribute huge weights, and the annular peak is a high-density periphery where overlap would be most severe in reality. The authors acknowledge this and say they do not anticipate a big effect, but they do not test it. That is the one load-bearing assumption I would want addressed — a short-range repulsion in both models, or even a heuristic estimate of how much density the ring gains from unphysical overlap.\n\nThat said, the ring is not a numerical artifact: it appears consistently in the low-noise agent histograms and in the PDE across resolutions. The missing piece is whether it survives a physically realistic hard-core interaction, and the paper itself flags that as the key open question. I disagree with any reading that calls the central claim circular — the PDE is not fitted to the annulus; the parameters are inherited from [31], and the ring is an emergent output.\n\nThe citation pattern is fine; the paper builds on standard nonlocal swarm PDEs and credits [31] properly. The writing is clear, the figures convey the evidence, and the limitations section is honest, even if the \"we do not anticipate\" line is weaker than the rest of the paper.\n\nWho should read it: active-matter experimentalists who want a cheap way to explore parameter space before running particles, and PDE modelers working on nonlocal swarming. It deserves a serious referee; the annulus prediction is specific enough to be falsifiable, and the numerical evidence is solid. I would send it out, with a request that the authors either run the excluded-volume test or soften the language about experimental recreation.","headline":"A competent and useful continuum model for a vision-based Janus swarm, with a genuinely new annular-density prediction that deserves a real referee; the main weakness is an untested excluded-volume assumption sitting exactly where the ring lives.","tokens_in":11057,"tokens_out":1553,"would_cite":true,"duration_ms":18359,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q92"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that vision-guided Janus particles, which move forward only when enough neighbours appear in a cone in front of them, form cohesive swarms whose stable shape is an annulus, and that a nonlocal advection-diffusion PDE…","keywords":["Janus particles","active matter","quorum sensing","agent-based model","nonlocal advection-diffusion PDE","swarming","annular density structure","pattern formation"],"falsifier":"Run both models with a short-range repulsion term at the physical particle diameter and compare the radial cross-section of the annulus: if the rim peak flattens, splits, or shifts beyond the discretization error, the annular state is a finite-size artifact rather than the true continuum attractor. A complementary experiment would time-average high-resolution images of the original system and check whether boundary particles are predominantly inactive; if the rim is not predominantly inactive, the mechanism proposed here is wrong.","tokens_in":10040,"feed_emoji":"🌀","tokens_out":10726,"duration_ms":95533,"temperature":0.7,"pith_summary":"This paper argues that a nonlocal advection-diffusion equation for particle density can reproduce the clustering of Janus particles, the half-coated colloids that move forward only when enough neighbours appear inside a vision cone. The reason to care is that the model says cohesion needs no attraction or alignment: Brownian diffusion plus thresholded, perception-triggered forward motion is enough. The new observation is that the cohesive group is not the solid disk reported in earlier work but an annulus, with density elevated at the periphery and active particles concentrated in the interior. The paper verifies this ring appears both in the continuum PDE and in time-averaged agent-based simulations, and maps where in the parameter space it forms and persists.","feed_headline":"Janus-particle swarms form a ring, not a solid disk","feed_subtitle":"A PDE and time-averaged simulations both show inactive particles piling at the rim while active ones cross inside.","key_machinery":"The load-bearing object is the nonlocal advection-diffusion PDE (Eq. 9, with density $\\rho(\\theta,t,\\vec x)$ over orientation, time, and space): $\\rho_t = D_{\\vec x}\\Delta \\rho - \\nabla\\cdot(S v_0 \\hat\\theta \\rho) + D_\\theta \\rho_{\\theta\\theta}$. Perception is a convolution $V = K_\\theta * \\bar\\rho$ between the vision cone $K_\\theta$ (half-angle $\\alpha$, weight $1/d(\\vec x,\\vec y)$) and the orientation-averaged density $\\bar\\rho$; the activation flag $S$ is a step function that is 1 where $V$ exceeds the threshold $P^*$ and 0 elsewhere. The PDE does the argument's work by turning that asymmetric, binary sensing rule into a drift term: active particles advect along their orientation $\\hat\\theta$ while spatial and angular diffusion constantly scramble positions and headings. The ring emerges from the balance of these terms, and the paper uses the PDE's cheap numerical solutions to sweep the two main parameters, vision angle $\\alpha$ and threshold ratio $P^*/P^c_\\alpha$.","core_discovery":"On its own terms, the paper's central discovery is that the attractor of the vision-controlled Janus-particle system is an annulus rather than a uniform disk. Solving the advection-diffusion PDE to a pseudo-steady state yields a nearly rotationally symmetric density profile with a pronounced peak at the rim; the agent-based model, when its histograms are averaged over time, shows the same rim. Breaking the density down by orientation shows each orientation class forms an arc at the periphery where particles face outward, and combining the arcs produces the ring. The active/inactive split is sharp: inactive particles are almost entirely confined to the annulus, while interior particles are active and traverse the group, which explains how the cluster holds together without attractive forces. As the number of agents grows, the agent-based radial cross-sections converge toward the PDE cross-section, supporting the claim that the PDE is the continuum limit of the stochastic process.","pith_inferences":["A natural next test is to add a short-range repulsion of roughly the particle diameter to both models; if the peripheral peak persists with a width comparable to the repulsion length, the annulus is a true continuum prediction, and if it collapses, the rim height is a point-particle artifact.","If the annulus is the attractor, then in larger domains the ring radius should be set by the vision cone and activation threshold rather than by the boundary, which could be checked by repeating the simulations and experiments in fields several times larger than 250 µm.","The same convolution-threshold-advection structure may describe other sensory particles, such as chemotactic bacteria or phototactic colloids, where the perceived signal is a scalar field instead of a neighbour count; the annular profile would then be a generic signature of thresholded directional motility."],"forward_implications":["If the paper is right, the clustering seen in the original physical experiments can be obtained from a continuum model with no attraction or alignment terms; diffusion plus vision-thresholded forward motion is sufficient.","The ring is robust to model choice: it appears in the time-averaged histograms of the stochastic agent-based model and in the deterministic PDE, so it is not an artifact of the continuum approximation.","The active/inactive split is a concrete signature: boundary particles are mostly outward-facing and inactive, interior particles are active, so any experiment that can label activation state should see the rim dominated by inactive particles.","The phase diagrams for formation and cohesion differ: some parameters never form a ring from random initial data but will sustain an already-formed ring, at least on intermediate time scales.","The PDE's low cost makes parameter sweeps and bifurcation estimates practical, giving a route to predict group radius and rim height without running expensive agent-based ensembles."],"supporting_citations":[{"why":"Supplies the experimental Janus-particle system, the vision-cone activation rule, parameter values, and the reference threshold $P^c_\\alpha$ that both models are calibrated against.","marker":"[31]"},{"why":"Provides the nonlocal advection-diffusion equation framework that the paper adapts for vision-based activation.","marker":"[22]"},{"why":"Documents ring patterns and their bifurcations in nonlocal swarm models, giving the annular attractor a known family to join.","marker":"[30]"},{"why":"Shows stability of ring patterns in two-dimensional particle interactions, supporting the interpretation of the annulus as an attractor.","marker":"[23]"}],"fun_headline_variants":["Janus swarm forms a ring, not a disk","PDE shows Janus particles trap inactive at rim","Annular structure emerges in Janus particle swarm","Active Janus particles cross, inactive pile at edge","Simulations and PDE agree: Janus swarm is a ring"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that ignoring the finite size of the particles—treating each as an overlap-allowed point, although the real ones are 4.28 µm across—does not change the qualitative pattern, even though the ring's peak is exactly where crowding and stacking would be most severe in the laboratory.","fun_headline_variants_meta":{"raw":{"variants":["Janus swarm forms a ring, not a disk","PDE shows Janus particles trap inactive at rim","Annular structure emerges in Janus particle swarm","Active Janus particles cross, inactive pile at edge","Simulations and PDE agree: Janus swarm is a ring"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000101,"raw_usage":{"total_tokens":946,"prompt_tokens":796,"completion_tokens":150,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":412,"completion_tokens_details":{"reasoning_tokens":73}},"tokens_in":412,"tokens_out":150,"duration_ms":1957,"temperature":1.0,"reasoning_tokens":73,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T11:09:28.557150+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run both models with a short-range repulsion term at the physical particle diameter and compare the radial cross-section of the annulus: if the rim peak flattens, splits, or shifts beyond the discretization error, the annular state is a finite-size artifact rather than the true continuum attractor. A complementary experiment would time-average high-resolution images of the original system and check whether boundary particles are predominantly inactive; if the rim is not predominantly inactive, the mechanism proposed here is wrong.","supporting_citations":[{"cited_title":"Mogilner and L","cited_arxiv_id":null,"evidence_quote":"Supplies the experimental Janus-particle system, the vision-cone activation rule, parameter values, and the reference threshold $P^c_\\alpha$ that both models are calibrated against."},{"cited_title":"Chen and T","cited_arxiv_id":null,"evidence_quote":"Provides the nonlocal advection-diffusion equation framework that the paper adapts for vision-based activation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents ring patterns and their bifurcations in nonlocal swarm models, giving the annular attractor a known family to join."},{"cited_title":"Helbing, P","cited_arxiv_id":null,"evidence_quote":"Shows stability of ring patterns in two-dimensional particle interactions, supporting the interpretation of the annulus as an attractor."}],"review_version":1}