{"id":"02b12df9-4b91-4439-8d1d-db00dbd0d32a","arxiv_id":"2507.01248","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"Particle-size segregation velocity scales linearly with local rheology only for 0.01 < I < 0.1, and the standard continuum model underpredicts segregation outside this window.","lead":"A DEM study of shear-driven granular flows finds that the standard linear segregation-velocity scaling works only for a narrow range of inertial numbers (about 0.01 to 0.1), failing in quasi-static and collisional regimes. This regime limitation explains why continuum models underestimate segregation in bidisperse mixtures at extreme compositions.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The regime-cutoff claim is supported only if the ensemble-averaged background fields faithfully represent the local environment of the intruder; the paper's own mixture evidence does not fully close the gap.","rationale":"The reader's weakest_assumption matches the main vulnerability I see: the ensemble-averaged background field mapping. The manuscript in several passages acknowledges the same concern, explicitly attributing deviations to local packing and diffusion (Section IV.B and Section V), and the mixture test (Fig. 8) is the only out-of-sample validation, providing real but partial independent support. The proposed conditional verdict is therefore appropriate: the headline claim is plausible and internally consistent, but its strongest form requires an explicit local-mapping check. I am not proposing REJECT because the paper includes a benchmark against Ferdowsi et al., a coherent DEM setup, and a falsifiable mixture prediction that behaves as claimed. The single most load-bearing concern remains the averaged-field mapping, and the concrete test is a targeted conditioning of the inertial number statistics on the intruder's local neighborhood.","tokens_in":19187,"tokens_out":1386,"duration_ms":15324,"concrete_test":"Isolate the low-I checkpoints (<I> < 0.01) and recompute the local inertial number conditioned on the intruder's instantaneous neighborhood, e.g., using a coarse-graining window of radius ~2d_s around the intruder position and time-averaging only over intervals when the intruder is in a given depth bin. If the conditioned I-bin statistics recover a linear w_seg proportional to I trend that the ensemble-averaged <I> obscures, then the claimed breakdown below I = 0.01 is an artifact of averaging rather than a genuine regime dependence.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that the linear segregation scaling f_sl = F(R)(gamma_dot/p) rho g dbar^2 holds only for 0.01 < I < 0.1, with breakdown in quasi-static and collisional regimes. The load-bearing condition is that the local flow state experienced by the traced intruder is well described by <gamma_dot> and <p> evaluated at the intruder's height from the ensemble-averaged background field (Fig. 1, Appendix B). The paper's own text weakens this premise: in Section IV.A the authors state that small intruders move 'more erratically' and 'more scattered,' and in Section V they attribute deviations to 'local packing configurations' and 'diffusion.' These statements concede that the averaged fields may not capture the actual control parameter in the low-I regime, where the data's nonlinear behavior could reflect an intruder-specific local environment rather than a genuinely regime-dependent scaling law. The claim 'the scaling breaks down' is therefore underdetermined: the reported nonlinearity could be an artifact of plotting against averaged conditions. The one direct, independent check is the mixture comparison in Figs. 6-8; its stated errors (epsilon = 0.215 and 0.300 for the extreme mixtures) support the main claim qualitatively, which is genuine supporting evidence, but this comparison uses the same fitted F(R) and the same averaged profiles, so it does not independently validate the local-environment mapping.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses DEM simulations of a sheared granular bed with a moving, pressure-controlled top boundary to track single intruder particles of various size ratios, and measures their segregation velocities as functions of the ensemble-averaged local shear rate, pressure, and inertial number. The authors report that the linear scaling f_sl = F(R)(γ̇/p)ρg d̄² holds only in a moderate inertial-number window, approximately 0.01 < I < 0.1, and breaks down in both quasi-static and collisional regimes. They then feed the fitted F(R) into a continuum segregation model, compare its predictions with DEM simulations of bidisperse mixtures at three concentrations, and find that the model underestimates segregation, especially for extreme compositions. The paper concludes that existing linear segregation scalings need generalization beyond the intermediate regime.","tokens_in":19469,"tokens_out":6976,"duration_ms":85926,"significance":"If the central claim is correct, this paper would usefully delimit the range of validity of a widely used segregation scaling law and of continuum models based on it, which is an important step for geophysical and industrial applications. The DEM setup is benchmarked against published experiments and prior DEM work, the data are deposited on Zenodo, and the mixture simulations provide a partially independent test of the intruder-fitted scaling. The regime-window claim is, however, not yet supported with the rigor it needs: the scaling parameters and the 0.01 < I < 0.1 boundaries are obtained from the same intruder data that are used to display the collapse, and the load-bearing assumption that ensemble-averaged background fields represent the intruder's local environment is not independently verified, despite the paper itself noting the importance of local packing and fluctuations in the regimes where the scaling fails.","major_comments":[{"comment":"The central quantitative claim, 'the scaling holds only within 0.01 < I < 0.1 and breaks down outside', is not established as an out-of-sample result. F(R) and G(R) are fitted to the entire intruder dataset, and the boundaries I_low and I_high are then chosen post hoc from the visual onset of deviations in the same data. Consequently, the mid-range linear collapse is partly a consequence of the fitting procedure, not an independent validation. I ask for an objective breakpoint analysis (e.g., piecewise regression with an information criterion or a cross-validated threshold search) and for prediction errors reported separately for I < 0.01, 0.01 < I < 0.1, and I > 0.1, with parameters fitted only on the mid-range data. This is necessary to support 'holds only within' rather than 'is fitted within'.","section":"§IV B, Fig. 5"},{"comment":"The load-bearing premise that the intruder experiences the ensemble-averaged background state is not verified in the regimes where the scaling is said to break down. Each measured w_seg is paired with ⟨γ̇⟩ and ⟨p⟩ evaluated at the intruder's height from profiles computed without accounting for the intruder's presence. The paper itself states in §IV A that small intruders move 'more erratically' and 'more scattered', and §V attributes deviations to 'local packing configurations', 'diffusion', and 'collisional dissipation'. In quasi-static and collisional conditions, local force chains, voids, and velocity fluctuations can make the ensemble-mean I a poor proxy for the true local control parameter; the nonlinear behavior outside the window could then be an artifact of plotting against averaged fields. I request a direct test: either intruder-centered measurements of local coordination, void fraction, and local stresses, or uniform-shear DEM simulations at prescribed I spanning the same range, to confirm that the linear scaling genuinely fails in a locally uniform environment rather than merely failing to correlate with the ensemble-mean state.","section":"§IV A, §IV B, Appendix B"},{"comment":"The mixture comparison is the most independent evidence in the paper, but it does not fully close the gap identified above. It uses the same fitted F(R), the same ensemble-averaged shear-rate and pressure profiles, and the same assumed linear scaling, so the reported errors ε = 0.081, 0.215, and 0.300 could also arise from the diffusivity model, from the background-profile assumption, or from the choice of fitted parameters, rather than specifically from failure of the scaling outside 0.01 < I < 0.1. I suggest three quantitative checks: (i) report the fraction of time each segregation interface spends inside versus outside the claimed window; (ii) run the continuum model with a piecewise or saturated f_sl for I outside the window and show whether ε decreases; (iii) report sensitivity of the results to the diffusivity coefficient A. Without such checks, the attribution of the mixture misprediction to the regime dependence of the scaling is underdetermined.","section":"§IV C, Figs. 6–8"},{"comment":"The notation for u0 is ambiguous and, as written, dimensionally inconsistent. In Table II, u0 = 2.5 m/s is the top boundary velocity, but in the fitted shear-rate model ⟨γ̇⟩ = (u0/λ)exp(z/λ) the fitted value is reported as u0 = (9.49 ± 5.00)×10⁻⁶ m/s. These cannot be the same quantity; the fitted prefactor is apparently a velocity scale at z=0 in an exponential velocity profile, not the driving velocity. Since the continuum-model predictions in Figs. 6–8 depend on this fitted profile, the ambiguity must be resolved by renaming the fitted amplitude and stating the velocity profile explicitly with units.","section":"§IV C, Eq. (8) and following text"}],"minor_comments":[{"comment":"The text in §IV B refers to 'red circles' for binned means, while the Fig. 5 caption describes them as 'orange circles'; please unify the color description.","section":"Fig. 5 caption"},{"comment":"The caption states that deeper colors correspond to higher size ratios R, but the two color families (green for large intruders, brown for small intruders) are not mapped monotonically in an obvious way; please specify the colormap and R-value correspondence.","section":"§IV B, Fig. 4 caption"},{"comment":"The benchmark comparison with Ferdowsi et al. is described only qualitatively as showing 'similar temporal trends'; adding a quantitative error measure, such as the normalized root-mean-square difference of armor thickness, would strengthen the validation claim.","section":"Appendix A, Fig. A1"},{"comment":"The gray band is described as 'the standard deviation from the ensemble-averaging procedure', but it is not stated whether this is a pointwise standard deviation of the field or of the binning process; please clarify.","section":"§III, Fig. 1 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper is a serious candidate for publication if the authors can provide the requested out-of-sample validation of the regime window and some direct evidence on the local-state mapping. I do not see a fatal flaw, but the central claim is currently supported largely by in-sample fits and post hoc boundaries, which the mixture comparison alleviates only partially. The notation issue with u0 in §IV C should be fixed regardless; it is currently confusing enough to affect reproducibility."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid, honest DEM paper with a genuinely new regime claim. The authors show the Trewhela/Jing linear scaling for intruder segregation velocity holds only for 0.01 < I < 0.1, and they demonstrate the consequence in bidisperse mixtures. That is worth taking seriously. The mixture test is the real payload: F(R) fitted to intruder data is applied to a continuum model for mixtures and compared to DEM, and the misfit grows exactly when the interface leaves the moderate-I band. That is independent support, not just in-sample collapse.\n\nWhat it does well: benchmarked against two independent experiments (Ferdowsi et al. armor thickness and standard DEM segregation), data on Zenodo, both pressure-based and stress-based scalings tested, and the asymmetry between small and large intruders is handled without overclaiming. The paper is also unusually explicit about where it is weak—Section IV.A admits the small intruder moves erratically, and Section V names local packing and diffusion as confounds.\n\nWhere it is soft: the central breakdown claim is conditioned on the ensemble-averaged background fields being the right local control parameter for a single intruder. The stress-test note has this right. For a small intruder percolating through gaps, the local environment can be very different from <gamma_dot> and <p> at that height, especially at low I where packing configurations dominate. The paper's own language concedes this. So the apparent nonlinearity below I=0.01 could be partly an artifact of plotting against averages. That said, the mixture comparison does not disappear: it uses the same averaged profiles, but the qualitative pattern—underestimation in low-I regions—is what the intruder data predict, and the errors (epsilon=0.08–0.30) are consistent. The regime boundaries 0.01 and 0.1 look post hoc; no statistical test separates regimes from gradual curvature. The F(R) fit is in-sample, though the functional form is minimal.\n\nCitation pattern: fine. It uses prior Trewhela/Jing work appropriately, cites recent mixture modeling, and shows no red flags.\n\nWho this is for: anyone modeling segregation in geophysical or industrial flows, and people building continuum segregation models. It deserves peer review; a good referee will push for out-of-sample validation and perhaps a local-measurement check (e.g., conditioning on the intruder's actual local coordination). I'd send it to JFM or Phys. Rev. Fluids with a request for those additions. My own verdict is conditional, but the paper is a genuine step toward a regime-aware scaling law.","headline":"A careful DEM study with a genuinely new regime claim for segregation scaling, but the local-environment mapping needs out-of-sample support before I'd bet on the 0.01–0.1 window.","tokens_in":20031,"tokens_out":2222,"would_cite":true,"duration_ms":25309,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The standard scaling for granular size segregation works only in a mid-range shear window.","keywords":["granular segregation","inertial number","discrete element method","intruder segregation velocity","continuum segregation model","sheared granular flow","quasi-static regime","collisional regime"],"falsifier":"Run a shear cell at fixed, spatially uniform inertial numbers below $0.01$ (e.g., $I=0.005$) with a single intruder, and measure segregation velocity as $\\dot{\\gamma}/p$ is varied; if $w_{\\mathrm{seg}}$ remains linear in $\\dot{\\gamma}/p$ in this homogeneous quasi-static setting, then the claimed breakdown is not a property of quasi-static granular segregation itself but of the depth-varying, averaged setup used here.","tokens_in":68,"feed_emoji":"","tokens_out":7590,"duration_ms":209645,"temperature":0.7,"pith_summary":"This paper uses discrete-element simulations of a sheared granular column to test whether the widely used scaling for particle-size segregation velocity—linear in the local shear rate over pressure, or in the inertial number—holds when flow conditions vary widely. It finds that the scaling is reliable only for inertial numbers between about 0.01 and 0.1, and breaks down in quasi-static creep and in collision-dominated collisional flow. Because continuum segregation models rely on this linear law, they mispredict how fast bidisperse mixtures separate, underestimating segregation in extreme compositions. A sympathetic reader would take the paper as establishing that segregation laws need regime-dependent corrections to be useful across natural and industrial granular flows.","feed_headline":"Granular segregation law holds only for inertial number 0.01–0.1","feed_subtitle":"DEM simulations show the standard linear scaling fails in quasi-static and collisional flows, skewing mixture predictions.","key_machinery":"The central object is the inertial number $I = \\dot{\\gamma} d / \\sqrt{p/\\rho}$, the dimensionless ratio of inertial to confining forces that classifies granular flow regimes. The argument compares intruder segregation velocities measured from checkpoint-crossing times against two linear scaling laws in terms of $\\dot{\\gamma}/p$ and $I$, and then uses the fitted size-ratio function $F(R)$ inside a continuum segregation equation (advection, segregation flux, and diffusion) to predict mixture evolution. The regime boundaries $0.01$ and $0.1$ are identified from where the DEM data depart from the linear collapse.","core_discovery":"In a horizontally sheared granular medium where depth-dependent pressure and shear rate produce inertial numbers from roughly $4\\times10^{-4}$ near the bottom to $0.25$ near the top, the authors track single intruder particles and measure their vertical segregation velocity. They show that the data collapse onto the proposed linear scalings, $|w_{\\mathrm{seg}}| = F(R)(\\dot{\\gamma}/p)\\,\\rho g \\bar{d}^2$ and $|w_{\\mathrm{seg}}|/\\sqrt{g\\bar{d}} = G(R)\\,I$, only when the local inertial number lies between $0.01$ and $0.1$. At lower $I$, segregation proceeds faster than the linear law predicts, especially for small intruders percolating through large-particle gaps; at higher $I$, segregation is slower than predicted, consistent with collisional dissipation and enhanced diffusion. Feeding the same scaling into a continuum advection-diffusion-segregation model for bidisperse mixtures captures the qualitative direction of segregation but underestimates its rate, with the largest errors (relative center-of-mass error up to about $0.30$) when the segregation interface spends time outside the valid inertial band.","pith_inferences":["I read the paper as implying that single-intruder calibrations performed under moderate shear may not transfer directly to geophysical flows where much of the bed moves by creep or collisions; this is an extension beyond the paper's explicit statements.","A testable next step would be to compute segregation velocity against the intruder's local instantaneous inertial number from its immediate neighborhood rather than the height-averaged background field; if linearity is restored, part of the reported breakdown would be an averaging artifact.","The same scaling framework might be extended to density segregation, where the analogous buoyancy-drag balance could show regime boundaries at different inertial numbers; the paper does not make this claim."],"forward_implications":["Continuum segregation models should only be trusted when the local inertial number of the flow stays within 0.01–0.1; applying them broadly will underestimate segregation in slow, creeping regions and overestimate it in highly agitated regions.","The fitted size-ratio function $F(R)$, a power law in $(R-1)$, gives a concrete, testable correction for how segregation velocity depends on particle size ratio in the moderate regime.","Extreme mixture compositions (e.g., 25% or 75% small particles) produce the largest model errors because their segregation interface spends significant time outside the valid band, so composition matters as much as shear conditions for prediction quality.","Future segregation laws will need to include regime-specific mechanisms—creep and local packing in quasi-static flow, collisional dissipation and diffusion in collisional flow—rather than a single linear rheological response.","Local inertial number, not just shear rate or pressure, is the practical diagnostic that determines whether existing scaling laws apply."],"supporting_citations":[{"why":"Supplies the experimental scaling f_sl = F(R)(\\dot gamma/p) rho g \\bar d^2 that the paper tests and generalizes.","marker":"[1]"},{"why":"Provides the alternative segregation-velocity scaling based on shear stress \\tau that the paper evaluates in Appendix D and finds behaves similarly.","marker":"[27]"},{"why":"Gives the non-dimensionalization and inertial-number-based scaling framework used for bedload transport comparisons.","marker":"[17]"},{"why":"Establishes the DEM setup and segregation-force scaling laws that inform the simulation parameters and the intruder-force picture.","marker":"[24]"},{"why":"Documents the asymmetric segregation behavior and the transition length scale that motivates the checkpoint spacing d_cp.","marker":"[8]"},{"why":"Provides pressure-dependent segregation scaling in shear flows that supports the inertial-number formulation in Eq. (7).","marker":"[31]"},{"why":"Presents drag, diffusion, and segregation scaling in inertial granular flows used to justify the linear-in-I form.","marker":"[41]"},{"why":"Laboratory armoring experiment used to benchmark the DEM framework's ability to reproduce segregation dynamics.","marker":"[44]"},{"why":"Recent continuum-vs-DEM discrepancy study that the paper contrasts with, arguing for alternative mechanisms beyond the reported dimensional-instability explanation.","marker":"[60]"}],"fun_headline_variants":["Segregation scaling valid only for inertial number 0.01–0.1","Granular segregation law fails outside moderate shear rates","DEM: size segregation scaling limited to mid-inertial flows","Sheared granular flows: segregation law works only in narrow band","Particle segregation scaling breaks down at low and high shear"],"cache_read_input_tokens":22144,"weakest_assumption_plain":"The load-bearing premise is that the local flow state felt by the intruder is accurately captured by the ensemble-averaged background shear rate, pressure, and inertial number at the intruder's height—if strong local fluctuations, force chains, or packing voids make those averages unrepresentative, the apparent breakdown of the scaling outside 0.01–0.1 could be partly an artifact of the averaging.","fun_headline_variants_meta":{"raw":{"variants":["Segregation scaling valid only for inertial number 0.01–0.1","Granular segregation law fails outside moderate shear rates","DEM: size segregation scaling limited to mid-inertial flows","Sheared granular flows: segregation law works only in narrow band","Particle segregation scaling breaks down at low and high shear"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1431,"prompt_tokens":928,"completion_tokens":503,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":544,"completion_tokens_details":{"reasoning_tokens":417}},"tokens_in":544,"tokens_out":503,"duration_ms":6252,"temperature":1.0,"reasoning_tokens":417,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T20:56:38.064484+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a shear cell at fixed, spatially uniform inertial numbers below $0.01$ (e.g., $I=0.005$) with a single intruder, and measure segregation velocity as $\\dot{\\gamma}/p$ is varied; if $w_{\\mathrm{seg}}$ remains linear in $\\dot{\\gamma}/p$ in this homogeneous quasi-static setting, then the claimed breakdown is not a property of quasi-static granular segregation itself but of the depth-varying, averaged setup used here.","supporting_citations":[{"cited_title":"Trewhela , author C","cited_arxiv_id":null,"evidence_quote":"Supplies the experimental scaling f_sl = F(R)(\\dot gamma/p) rho g \\bar d^2 that the paper tests and generalizes."},{"cited_title":"Jing , author J","cited_arxiv_id":null,"evidence_quote":"Provides the alternative segregation-velocity scaling based on shear stress \\tau that the paper evaluates in Appendix D and finds behaves similarly."},{"cited_title":"Chassagne , author R","cited_arxiv_id":null,"evidence_quote":"Gives the non-dimensionalization and inertial-number-based scaling framework used for bedload transport comparisons."},{"cited_title":"Guillard , author Y","cited_arxiv_id":null,"evidence_quote":"Establishes the DEM setup and segregation-force scaling laws that inform the simulation parameters and the intruder-force picture."},{"cited_title":"van der Vaart , author P","cited_arxiv_id":null,"evidence_quote":"Documents the asymmetric segregation behavior and the transition length scale that motivates the checkpoint spacing d_cp."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides pressure-dependent segregation scaling in shear flows that supports the inertial-number formulation in Eq. (7)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Presents drag, diffusion, and segregation scaling in inertial granular flows used to justify the linear-in-I form."},{"cited_title":"Ferdowsi , author C","cited_arxiv_id":null,"evidence_quote":"Laboratory armoring experiment used to benchmark the DEM framework's ability to reproduce segregation dynamics."},{"cited_title":"Kumawat , author V","cited_arxiv_id":null,"evidence_quote":"Recent continuum-vs-DEM discrepancy study that the paper contrasts with, arguing for alternative mechanisms beyond the reported dimensional-instability explanation."}],"review_version":1}