{"id":"f1abe0f8-73f2-40e1-8d8b-0b6fbfe3647b","arxiv_id":"1908.03437","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A stochastic model that adds acceleration to the usual position-velocity description of fluid particles reproduces acceleration anisotropy and intermittency in a turbulent channel flow.","lead":"The paper adds acceleration to the standard random-walk model of turbulent fluid particles and tests it in a channel flow. The model reproduces key acceleration statistics, which could improve near-wall predictions of dispersion and mixing.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Quantitative predictions hinge on the unverified assumption β = 1/τη; the proportionality constant is assumed equal to one and is not tested.","rationale":"The reader's weakest assumption identifies exactly the condition on which the central quantitative claim depends. The model's predictions for acceleration variance, velocity–acceleration correlations, and decorrelation times all follow from the OU form (2.3) together with the consistency relation (2.8). Once β is written as c/τη, every one of those predictions scales with c, and the paper neither derives c from theory, measures it from the data, nor reports sensitivity to it. The qualitative shapes may be robust, but the claimed quantitative reproduction of variance and correlation times is not. I considered whether the more fundamental concern is that the whole-channel PDFs in Fig. 4 could be non-Gaussian only because they average over inhomogeneous variances, but the paper's stated claim is about channel-averaged acceleration PDFs, and the beta issue remains the primary load-bearing assumption. The proposed test is decisive: it directly varies the untested coefficient and compares against the same reference data used for validation, using a quantitative metric instead of visual inspection. The reader's conditional verdict is appropriate; my read does not move it.","tokens_in":9573,"tokens_out":15172,"duration_ms":173646,"concrete_test":"Re-run the channel-flow simulation at Reτ ≈ 1440 with the same RANS mean fields and numerical scheme, setting β = c/τη for c ∈ {0.5, 1, 2}, and compute a quantitative error metric (e.g., relative L2 error) for ⟨a_i²⟩(y+), the correlation functions ρyy(τ) and ρzz(τ), and the PDF tail |a_i|/arms_i > 3 against the experimental and DNS data of Stelzenmuller et al. (2017). If the c = 1 run is not the best fit, or if the error varies by more than the experimental uncertainty across this range, then the model's quantitative predictions are not established without an independently calibrated β. If all metrics stay within error bars for c = 1, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 2, immediately after Eq. (2.7), the inverse correlation time of the acceleration process is set to the local Kolmogorov rate, β = 1/τη, with the sentence 'assuming a constant of proportionality of one'. This is not a prediction of Kolmogorov theory: the scaling argument fixes the order of magnitude of the acceleration timescale, not its numerical coefficient. The coefficient is also not measured or fitted. The choice is load-bearing because β enters every acceleration statistic. From Eq. (2.3), the stationary acceleration variance is B/(2β); with B fixed by Eq. (2.8), B ≈ C0⟨ϵ⟩β^2, so ⟨a²⟩ ≈ C0 β⟨ϵ⟩/2 = C0 c ⟨ϵ⟩/(2τη). Thus a factor c in β changes all variance profiles by c. Likewise, the acceleration autocorrelation is exp(−βτ), so the decorrelation times shown in Fig. 3 scale as 1/c. The velocity–acceleration correlation in Eq. (2.6) also changes through the β-dependent stationary value. No sensitivity analysis is reported for this coefficient; the statement that results are insensitive to C0 ∈ [0.2, 1.5] does not probe c independently, because for the variance C0 and c are partly degenerate, while for the correlation shape and for consistency with the Rotta constant they are not. Since the validation in Figs. 2–4 is visual rather than metric-based, the quantitative agreement claimed for variance, decorrelation times, and PDF tails is conditional on this assumed order-one constant.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a stochastic Lagrangian model that includes fluid particle position, velocity, and acceleration as dynamical variables for inhomogeneous turbulent flows. The acceleration is modeled as an Ornstein-Uhlenbeck process, and the coefficients are fixed by Kolmogorov scaling (β = 1/τη) and by requiring consistency with the standard velocity-only Langevin model in the limit τη → 0, which fixes B via Eq. (2.8) and C0 from the Rotta constant. The closed model is applied to a channel flow at Reτ ≈ 1440 using a hybrid RANS/PDF approach with the Durbin elliptic-relaxation Reynolds-stress model. Predictions for the mean and variance of acceleration, Lagrangian autocorrelations, and acceleration PDFs are compared with experiments and DNS, and the paper claims good agreement including anisotropy and strong intermittency. The model is shown to reduce to Sawford's isotropic model and to the generalized Langevin model in appropriate limits.","tokens_in":9882,"tokens_out":13660,"duration_ms":133930,"significance":"If the results hold, the model provides a relatively simple, theoretically motivated way to include acceleration as a dynamical variable in inhomogeneous wall-bounded flows, with potential applications to dispersion, mixing, and inertial-particle modeling. A clear strength is that the acceleration statistics are not used to calibrate the model: B is fixed by consistency with the velocity-only Langevin model, and C0 is inherited from the Rotta constant, so the acceleration variance, correlations, and PDFs are genuine outputs of the model rather than fits. The numerical scheme in Appendix A is explicitly given, and the comparisons with experiments and DNS are valuable. The main weakness is that the acceleration timescale β−1 is fixed by an unverified proportionality constant, and the validation is largely qualitative, so the quantitative agreement claimed is conditional on that assumption.","major_comments":[{"comment":"The choice β = 1/τη, made 'assuming a constant of proportionality of one', is not derived from Kolmogorov theory, not measured, and not independently tested. This is load-bearing because β enters every acceleration statistic. From Eq. (2.3), the stationary acceleration variance is B/(2β), and with Eq. (2.8) this gives ⟨a_i²⟩ = C0⟨ε⟩/(2τη) per component, so a factor c in β = c/τη multiplies all variance profiles by c (after re-imposing consistency through Eq. 2.8). The acceleration autocorrelation is exp(−βτ), so the decorrelation times shown in Fig. 3 scale as 1/c, and the velocity–acceleration correlation in Eq. (2.6) also changes with β. The statement that results are insensitive to C0 ∈ [0.2, 1.5] does not probe c independently, because for the variance C0 and c are partly degenerate. A sensitivity analysis on c, or an independent determination of c from DNS/experiment, is required before the quantitative agreement in Figs. 2–4 can be asserted.","section":"Section 2, paragraph following the dimensional analysis of the acceleration correlation (β = 1/τη)"},{"comment":"The validation is essentially qualitative. The text explicitly concedes 'qualitative overall agreement' for the acceleration variances and 'satisfactory' for the correlations, and the PDF tails in Fig. 4 are compared down to 10⁻⁵ without any quantification of the sampling error (the paper states that statistical error is significant but gives no estimate). Since the central claim is that the model reproduces the acceleration statistics, quantitative measures are needed: relative errors for the variance peaks, integral timescales extracted from the correlations, and moment-based statistics such as flatness or kurtosis of the PDFs, together with confidence intervals. This is particularly important for the extreme tails, which are used to support the conclusion that inhomogeneity alone generates the observed intermittency.","section":"Section 4, Figs. 2–4"},{"comment":"The paper describes the hybrid RANS/PDF approach as 'self-consistent', but the stochastic model is not shown to satisfy the well-mixed condition with respect to the prescribed RANS Eulerian fields. With G^a_ij = 0, the stationary velocity statistics of the stochastic particles will generally differ from the RANS Reynolds stresses that determine the model coefficients. Consequently, the spatial distribution of particles, and hence the sampled Lagrangian acceleration statistics (especially the correlations and PDFs that depend on particle paths), may be distorted. The authors should either demonstrate that the model's one-point velocity statistics are consistent with the RANS fields used in the coefficients, or discuss the sensitivity of the acceleration results to this mismatch.","section":"Section 3, Eqs. (2.1)–(2.3)"}],"minor_comments":[{"comment":"The typeset expression 'B 2τη(1/τη + 1/TL)⁻¹' is ambiguous; it should read (B τη/2)(1/τη + 1/TL)⁻¹. The subsequent limit and the final expression for B are consistent with this reading, but the notation should be cleaned up to avoid confusion.","section":"Equation (2.8) and surrounding text"},{"comment":"The caption refers to 'stochastic model simulations (squares)', but the figure legend appears to use continuous 'model' lines; please harmonize the caption with the actual symbols.","section":"Figure 2 caption"},{"comment":"The DNS curve is shown for only one wall distance, with no explanation of which y₀⁺ it corresponds to or why only one case is given; please specify or add DNS data for the other positions.","section":"Figure 3"},{"comment":"There is a typo: 'responsable' should be 'responsible'.","section":"Section 2, text near Eq. (2.5)"},{"comment":"The claim that the model 'generalises both the acceleration-based models for homogeneous flows' appears overstated: the homogeneous limit of Eq. (2.3) is a linear Ornstein-Uhlenbeck process, which does not include the conditional cubic-Gaussian model of Lamorgese et al. (2007) cited earlier. Please clarify the intended sense of 'generalise'.","section":"Introduction and Section 5"},{"comment":"The statement that 'results do not change qualitatively in the range C0 ∈ [0.2, 1.5]' is not supported by any figure or table; please provide a sensitivity plot or a reference.","section":"Section 2, after Eq. (2.8)"}],"recommendation":"major_revision","confidential_remarks":"The paper is potentially interesting and the model derivation is elegant, but the main technical concern is the unverified proportionality constant in β = 1/τη, which all quantitative acceleration predictions depend on. I would also expect reviewers to ask about the well-mixed condition, which is standard for inhomogeneous Lagrangian stochastic models and is not addressed. The validation would be much stronger with quantitative error metrics and a sensitivity study on β. The comparison against the authors' own experimental data is a reasonable starting point, but independent DNS or experimental data would increase confidence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nYou asked about arXiv:1908.03437, the Lagrangian stochastic acceleration model for channel flows. The short version: it's a legitimate and reasonably honest extension of the Sawford/Pope program to inhomogeneous, wall-bounded turbulence, with a real test against experiments and DNS. The main thing to know is that the model's quantitative predictions lean on one unexamined order-one assumption: the inverse acceleration correlation time is set exactly to the local Kolmogorov rate, beta = 1/tau_eta. That choice is not derived, not measured, and not stress-tested. If the true constant is c, the variance and correlation time scale accordingly; the paper doesn't probe c at all.\n\nWhat's new: they generalize the acceleration-based stochastic models from homogeneous/isotropic turbulence to inhomogeneous flows, fixing coefficients via Kolmogorov scaling plus consistency with the standard generalized Langevin model. The resulting closed model, coupled to a RANS mean field via a hybrid PDF/RANS approach, reproduces the mean acceleration (including the negative streamwise peak near the wall), the anisotropy, the acceleration autocorrelations, and the far-from-Gaussian PDF tails in a Re_tau ~ 1440 channel. That is a genuine step beyond Sawford (1991) and Pope (2002), and the comparison against both experiments and DNS is the right way to test it.\n\nThe soft spots, in proportion: first, the beta = 1/tau_eta assumption is exactly what makes the model closed, and dimensional analysis only fixes it to order one. They state it plainly, but they don't justify the constant, don't fit it, and don't report a sensitivity scan. For the variance, C0 and the beta-constant are partly degenerate, so the C0 range they quote doesn't cover this. Second, the validation is visual. Correlations and PDFs are compared by eye; there are no quantitative error metrics. For a model that claims to capture intermittency and anisotropy, a few RMS errors or a collapse check would make the claims more convincing. Third, no code or data are shipped, so an independent check of the numerical scheme or the comparisons isn't possible. None of these are fatal; the core derivation is clean and the physics is sensible. The paper is honest about the assumption, and the qualitative agreement is real.\n\nWho this is for: people working in PDF/RANS modeling, Lagrangian stochastic models, and inertial particle closures. A serious referee should engage with it; I'd accept it for review with a strong request for a sensitivity analysis on beta and quantitative validation metrics.\n\nRecommendation: send to review, but tell the authors the beta issue needs a real treatment before I'd trust the numerical values.","headline":"A credible generalization of the Sawford/Pope acceleration model to wall-bounded flows, but its quantitative predictions rest on an unverified order-one coefficient for the acceleration timescale.","tokens_in":10379,"tokens_out":5403,"would_cite":false,"duration_ms":48503,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A three-variable stochastic model reproduces acceleration statistics in wall-bounded turbulence.","keywords":["Lagrangian stochastic model","fluid particle acceleration","turbulent channel flow","wall-bounded turbulence","acceleration intermittency","anisotropy","RANS-PDF coupling","Ornstein-Uhlenbeck process"],"falsifier":"Measure the Lagrangian acceleration autocorrelation function for fluid particles in a DNS of a channel flow at $\\mathrm{Re}_\\tau \\approx 1440$ at several wall-normal positions, and compare the initial decay time of the correlation to $\\tau_\\eta = \\sqrt{\\nu/\\langle\\epsilon\\rangle}$; if the ratio differs from one by more than the statistical uncertainty, the assumed $\\beta = 1/\\tau_\\eta$ fails and the model's acceleration predictions for that wall distance cannot be correct.","tokens_in":9379,"feed_emoji":"🌀","tokens_out":9114,"duration_ms":86313,"temperature":0.7,"pith_summary":"This paper proposes a Lagrangian stochastic model of turbulent fluid motion in which each fluid particle carries position, velocity, and acceleration as dynamical variables, instead of only position and velocity. The model's coefficients are fixed from Kolmogorov scaling and from the requirement that it reduce to the standard generalized Langevin velocity model when the acceleration decorrelates quickly. Coupled to a Reynolds-averaged Navier-Stokes mean field for a channel flow at friction Reynolds number about 1440, the model reproduces the measured mean and fluctuating acceleration, the anisotropy among streamwise, wall-normal, and spanwise components, the acceleration decorrelation times, and the strongly non-Gaussian tails of the acceleration probability distributions. A reader interested in wall-bounded turbulence would care because this provides a relatively simple, self-contained route to near-wall statistics that velocity-only Lagrangian models miss.","feed_headline":"Stochastic model reproduces turbulent acceleration near walls","feed_subtitle":"Adding acceleration as a variable captures anisotropy, intermittency, and near-wall effects in channel flow.","key_machinery":"The mechanism that carries the argument is the stochastic differential system (2.1)-(2.3): position $x_i$ and velocity $U_i$ evolve with a drift that returns the velocity toward its mean with a matrix $D_{ij}$ taken from the generalized Langevin model, while the acceleration $a_i$ follows an Ornstein-Uhlenbeck process $da_i = -\\beta a_i\\,dt + \\sqrt{B}\\,dW_i$. The key identity is the closure $\\beta = 1/\\tau_\\eta$ and $B = (C_0\\langle\\epsilon\\rangle/\\tau_\\eta)(1/\\tau_\\eta + 1/T_L)$, obtained by matching the moment equations for $\\langle u_i a_j\\rangle$ and $\\langle a_i a_j\\rangle$ to the velocity-based Langevin model in the limit $\\tau_\\eta \\to 0$. This colored-noise construction replaces the white noise of velocity-only models, and the finite memory time $\\beta^{-1}$ is what produces the finite acceleration correlation time, the viscous near-wall acceleration signal, and the modified Reynolds-stress balance.","core_discovery":"On the paper's own terms, the discovery is that a diffusion process on position, velocity, and acceleration with the closure $\\beta = 1/\\tau_\\eta$ and $B = (C_0\\langle\\epsilon\\rangle/\\tau_\\eta)(1/\\tau_\\eta + 1/T_L)$ is enough to capture the statistical signature of fluid-particle acceleration in an inhomogeneous wall-bounded flow. The diffusion coefficient $B$ is fixed by requiring the velocity-acceleration covariance equations to reduce to the standard Langevin model in the limit $\\tau_\\eta \\to 0$, while the acceleration memory rate $\\beta$ is set to the local Kolmogorov time scale. Without low-Reynolds-number corrections and without multiplicative noise, the model gives the correct mean streamwise acceleration near the wall, the anisotropy of the acceleration variance, wall-distance-dependent decorrelation, and heavy-tailed probability distributions consistent with experiments and direct numerical simulation. The authors attribute the heavy tails to inhomogeneity: the spatially varying dissipation rate that enters $B$ along particle trajectories creates the extreme events, a mechanism they note is absent in homogeneous-turbulence models.","pith_inferences":["Beyond the paper: because the heavy tails are attributed to inhomogeneity rather than to fluctuating dissipation, a natural test is to run the same closure in homogeneous turbulence; the model would predict much weaker tails, isolating the inhomogeneity mechanism.","Beyond the paper: the proportionality constant in $\\beta = 1/\\tau_\\eta$ is the model's main free ratio; a direct measurement of the Lagrangian acceleration correlation time at several wall distances would either confirm the value one or supply a spatial correction function that could be inserted without changing the model's structure.","Beyond the paper: the same colored-noise construction can be iterated to include higher velocity derivatives, such as jerk, using the Kolmogorov scaling of derivative correlations that the paper derives; the second-order form in Appendix B shows the extension path."],"forward_implications":["The standard velocity-only Langevin description is insufficient near walls: the acceleration variable is what brings in the near-wall mean acceleration, the anisotropy, and the intermittent tails without extra low-Reynolds-number terms.","The same closed model, with the same $\\beta$ and $B$ formulas, can be coupled to any available RANS mean field in a different inhomogeneous flow; the channel test suggests no flow-specific tuning beyond the choice of $C_0$.","The model supplies explicit transport equations for the velocity-acceleration correlations and the acceleration variance, so the near-wall budget of these correlations can be inspected in the same framework.","Heavy-tailed acceleration probability distributions in wall turbulence can be generated by spatial inhomogeneity of the mean dissipation alone, without modeling the fluctuations of $\\epsilon$; the paper argues this is what the channel data show."],"supporting_citations":[{"why":"Supplies the isotropic acceleration model and the second-order process form that the present model generalises to inhomogeneous flows.","marker":"Sawford (1991)"},{"why":"Earlier stochastic acceleration model in isotropic turbulence that the present model reverts to in the isotropic limit.","marker":"Krasnoff & Peskin (1971)"},{"why":"Provides the generalized Langevin velocity model and the consistency requirement that fixes $D_{ij}$, as well as the RANS/PDF hybrid methodology.","marker":"Pope (2000)"},{"why":"Proposes the general diffusion model for velocity and acceleration in homogeneous anisotropic turbulence that this work extends.","marker":"Pope (2002)"},{"why":"Experiments and DNS of Lagrangian acceleration in a channel at $\\mathrm{Re}_\\tau \\approx 1440$ that provide the comparison data.","marker":"Stelzenmuller et al. (2017)"},{"why":"Elliptic-relaxation Reynolds-stress RANS model used to compute the mean fields in the hybrid approach.","marker":"Durbin (1993)"},{"why":"Documents long-time correlations and intermittency in Lagrangian acceleration, used to interpret the heavy-tailed PDFs.","marker":"Mordant et al. (2002)"},{"why":"The velocity-dissipation PDF model that introduces log-normal multiplicative noise, the alternative intermittency route the present model does not need for wall flows.","marker":"Pope & Chen (1990)"}],"fun_headline_variants":["Stochastic model captures near-wall acceleration intermittency","Lagrangian model reproduces turbulent acceleration anisotropy","New model predicts heavy tails in wall-bounded acceleration","Acceleration diffusion model matches wall-flow experiments","Modeling turbulent acceleration: a wall-bounded success"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model assumes the acceleration memory time is exactly the local viscous time scale $\\tau_\\eta$, with no adjustable factor; if the real ratio differs, the predicted acceleration variance and correlation shape shift.","fun_headline_variants_meta":{"raw":{"variants":["Stochastic model captures near-wall acceleration intermittency","Lagrangian model reproduces turbulent acceleration anisotropy","New model predicts heavy tails in wall-bounded acceleration","Acceleration diffusion model matches wall-flow experiments","Modeling turbulent acceleration: a wall-bounded success"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000176,"raw_usage":{"total_tokens":1294,"prompt_tokens":955,"completion_tokens":339,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":571,"completion_tokens_details":{"reasoning_tokens":268}},"tokens_in":571,"tokens_out":339,"duration_ms":4069,"temperature":1.0,"reasoning_tokens":268,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:13:10.212870+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the Lagrangian acceleration autocorrelation function for fluid particles in a DNS of a channel flow at $\\mathrm{Re}_\\tau \\approx 1440$ at several wall-normal positions, and compare the initial decay time of the correlation to $\\tau_\\eta = \\sqrt{\\nu/\\langle\\epsilon\\rangle}$; if the ratio differs from one by more than the statistical uncertainty, the assumed $\\beta = 1/\\tau_\\eta$ fails and the model's acceleration predictions for that wall distance cannot be correct.","supporting_citations":[{"cited_title":"Geophysical & Astrophysical Fluid Dynamics 2 (1), 123--146","cited_arxiv_id":null,"evidence_quote":"Earlier stochastic acceleration model in isotropic turbulence that the present model reverts to in the isotropic limit."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proposes the general diffusion model for velocity and acceleration in homogeneous anisotropic turbulence that this work extends."},{"cited_title":"Physical Review Fluids 2 (5), 054602","cited_arxiv_id":null,"evidence_quote":"Experiments and DNS of Lagrangian acceleration in a channel at $\\mathrm{Re}_\\tau \\approx 1440$ that provide the comparison data."},{"cited_title":"Journal of Fluid Mechanics 249 , 465--498","cited_arxiv_id":null,"evidence_quote":"Elliptic-relaxation Reynolds-stress RANS model used to compute the mean fields in the hybrid approach."},{"cited_title":"Physical review letters 89 (25), 254502","cited_arxiv_id":null,"evidence_quote":"Documents long-time correlations and intermittency in Lagrangian acceleration, used to interpret the heavy-tailed PDFs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The velocity-dissipation PDF model that introduces log-normal multiplicative noise, the alternative intermittency route the present model does not need for wall flows."}],"review_version":1}