{"id":"0e911cf7-dea6-468e-a0ca-78c02c33ecd8","arxiv_id":"2509.01239","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Using polynomial chaos surrogates of DNS data, the authors report that spanwise spacing is 6 to 14 times more influential than streamwise spacing on drag reduction across Re_tau = 180 to 590.","lead":"This paper builds a unified geometric parameterization of superhydrophobic surface patterns and applies uncertainty quantification to direct numerical simulations, finding that spanwise post spacing dominates drag reduction. The result could guide which manufacturing tolerances matter most for drag-reducing surfaces.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"d2-dominance ranking rests on an unvalidated Gaussian input distribution; Sobol indices depend on this assumed measure.","rationale":"The reader's weakest_assumption identifies exactly the most load-bearing concern: the input distribution for d1 and d2. The Sobol indices, which constitute the paper's strongest quantitative claim, are only meaningful relative to the specified input law. The chosen Gaussian with σ=0.33μ is arbitrary and unsupported by empirical data, and the samples do not cover the full unified geometric range the paper advertises. This concern directly threatens the central claim, whereas the other issues (drag-reduction definition, Re_tau computation with asymmetric walls, code availability) affect interpretation and reproducibility but are less likely to change the d2-dominance ranking. The proposed concrete test leverages the already-built PCE surrogates to assess distributional robustness cheaply; if the ranking persists under uniform and narrower distributions, the conditional verdict can be upgraded. Since the paper is already judged CONDITIONAL, no verdict adjustment is needed.","tokens_in":12619,"tokens_out":5190,"duration_ms":64596,"concrete_test":"Using the existing PCE surrogates (UQLab models) without new DNS, recompute Sobol indices under alternative input distributions: (i) uniform d1,d2 over [0,2w], and (ii) independent truncated Gaussians with σ=0.15w. Implement by drawing 10^5 samples from each candidate distribution, evaluating the PCE surrogate, and computing variance-based Sobol indices via Monte Carlo. If S_d2 remains ≥5×S_d1 at all three Reynolds numbers, the d2-dominance claim is robust to the assumed distribution; if the ratio falls below 2 or the ranking reverses, the central claim is an artifact of the Gaussian σ=0.33μ assumption.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—that spanwise spacing d2 is 6–14 times more influential than streamwise spacing d1—is a Sobol-index ranking, and Sobol indices are functionals of the input probability distribution. The paper assumes d1 and d2 are independent Gaussians with mean equal to the post width w=0.1875h and standard deviation σ=0.33μ, justified only by the desire to keep samples non-negative (Section II.C.3, Table III). No manufacturing tolerance data, correlation measurements, or physical argument are given. With this σ, the LHS samples cluster near d1=d2=w, so the UQ explores only a local neighborhood around the post width, not the full 'unified' range 0≤di≤2w claimed in the Introduction. If the true geometric variability is smaller, larger, uniform, correlated, or non-Gaussian, the variance decomposition changes and the S_d2/S_d1 ratio—and possibly the ranking—could change. Thus the headline dominance of d2 is conditional on an arbitrary and unverified input law.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a unified geometric parameterization of superhydrophobic surface (SHS) patterns in terms of streamwise and spanwise post spacings (d1, d2), with post width fixed. It performs direct numerical simulations (DNS) of turbulent channel flow at Re_tau = 180, 395, and 590 using NekRS, validates against Martell et al. for representative ridge, transverse-ridge, and post patterns, and builds polynomial chaos expansion (PCE) surrogates from Latin hypercube samples of d1 and d2. The central quantitative claim is that the spanwise spacing d2 dominates the uncertainty in drag reduction, with Sobol index ratios S_d2/S_d1 of roughly 6–14 across Reynolds numbers. The paper also presents uncertainty propagation for mean velocity, Reynolds stresses, and the barycentric anisotropy invariant map, and recommends controlling d2 for robust SHS design.","tokens_in":82,"tokens_out":3198,"duration_ms":79533,"significance":"If the central claim holds, the paper provides a useful unified framework for comparing SHS geometries and identifies a practically important sensitivity ranking for manufacturing tolerance. The DNS methodology is standard and the validation against established literature is a clear strength. The PCE/LHS workflow is conventional and the use of LOO errors to select polynomial order is reasonable. However, the headline result is a Sobol-index ranking, which is a functional of the assumed input probability distribution. That distribution is chosen for convenience rather than based on physical or manufacturing data, and the quantity of interest (drag reduction) is never explicitly defined. These gaps make the quantitative ranking conditional on untested assumptions. The uncertainty propagation results and the qualitative insight that spanwise spacing matters more than streamwise spacing are plausible, but the paper needs additional robustness checks before the specific 6–14x claim can be accepted.","major_comments":[{"comment":"The Sobol indices S_d1 and S_d2, and hence the headline claim that d2 is 6–14 times more influential than d1, are functionals of the joint input distribution. The paper assumes independent Gaussian distributions for d1 and d2 with mean equal to the post width and sigma = 0.33mu, chosen only to avoid negative samples. No manufacturing tolerance data, physical correlation argument, or sensitivity analysis with respect to the distribution family or sigma is provided. If the true geometric variability is uniform, correlated, or of different magnitude, the variance decomposition and even the ranking could change. Please add a robustness study (e.g., uniform or truncated distributions, different sigma values, and a correlated case) or provide empirical justification for the Gaussian assumption. Without this, the quantitative dominance claim is conditional on an arbitrary input law.","section":"Section II.C.3, Table III; Section IV.A"},{"comment":"The quantity of interest, 'drag reduction,' is never explicitly defined in the manuscript. The reader cannot tell whether it is computed from the mean wall shear stress, the pressure-gradient imbalance, or a formula such as DR = (tau_0 - tau_SHS)/tau_0, nor whether it uses the top or bottom wall. Since Table IV and all Sobol indices are based on this QoI, the missing definition is load-bearing. Please state the exact formula and the averaging procedure used.","section":"Section IV.A"},{"comment":"The claimed 'unified' parameterization encompasses the full range 0 <= d_i <= 2w and includes ridge-type (d1=0) and transverse-ridge-type (d2=0) geometries. However, the UQ input distribution is centered at w = 0.1875h and the LHS samples are clustered around that mean; no samples at the boundaries d1=0 or d2=0 are included in the PCE training set. The response surface in Fig. 6 is extracted from the PCE surrogate, not from boundary DNS runs. Thus the evidence that the surrogate is accurate over the entire unified parameter space, and particularly at the ridge/transverse-ridge limits, is missing. Please either include boundary cases in the LHS design or explicitly assess extrapolation error at d=0.","section":"Section I and IV.A"}],"minor_comments":[{"comment":"The second row of the matrix appears to have a typo: it lists Ψ0(ξ0) instead of Ψ0(ξ1). Please correct.","section":"Eq. (7)"},{"comment":"The notation '±0.33μ' for standard deviation is misleading; standard deviation is positive. Use σ = 0.33μ.","section":"Table III"},{"comment":"The row label 'The number of cells, EN' is unclear; it seems to list total grid points. Please clarify notation and units.","section":"Table I"},{"comment":"The response surface is described as a third-order polynomial fit to 1000 surrogate samples, but this is separate from the PCE surrogate. Clarify the relationship between this visualization and the PCE model, and report the residual error of the response surface.","section":"Section IV.A / Fig. 6"},{"comment":"The phrase 'unified' is used broadly, but the study only varies d1 and d2 at a fixed post width w. Please state this limitation explicitly in the conclusion.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for physics of fluids / CFD and the DNS workload is substantial. The main risk is not the DNS itself but the lack of robustness of the Sobol ranking to the assumed input distribution. I would encourage the editor to request the authors to add a distribution sensitivity analysis and an explicit definition of drag reduction. If those are provided convincingly, the paper could become acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe paper is worth a look for anyone doing UQ for textured surfaces: it runs a clean DNS campaign, validates against Martell, and uses a standard PCE/LHS pipeline to get Sobol indices for streamwise vs spanwise post spacing. The main finding—spanwise spacing d2 drives drag reduction sensitivity, sometimes tenfold over d1—is plausible and actionable for fabrication tolerances, and the uncertainty bands for velocity, Reynolds stresses, and anisotropy maps are genuinely new for this problem.\n\nThat said, the central quantitative claim is softer than the abstract suggests. The Sobol indices are functionals of the input distribution, and the authors simply assume independent Gaussians for d1 and d2 with mean equal to the post width and sigma=0.33mu to keep samples non-negative. No manufacturing data, no correlation, no alternative distributions tested. With that sigma, the LHS samples cluster tightly around the post width, so the 'unified' range 0 <= d_i <= 2w advertised in the introduction is not actually explored in the UQ. If the true manufacturing variability is different, the S_d2/S_d1 ratio could easily shift. The paper does state the assumption in Section II.C.3, so it's not hidden, but the conclusion does not carry the needed caveat.\n\nOther soft spots: drag reduction is never explicitly defined (percent change in friction coefficient? based on which reference?), and the computation of Re_tau with an asymmetric SHS/no-slip channel is not explained. No code or sample data is provided, which limits reproducibility. The 'unified parameterization' is a nice conceptual frame, but the paper only varies post spacing; it does not actually simulate ridge/transverse-ridge limits, so calling it unified is a bit strong.\n\nI disagree with the stress-test note that the entire result is invalid—the DNS and surrogate are sound, and the d2 dominance likely holds for a reasonable local perturbation. But the paper would be stronger with a sensitivity check on the input distribution (e.g., uniform or beta with same mean/sigma) and an explicit acknowledgment that the Sobol ranking is conditional on the assumed law.\n\nFor a reader: this is a solid applied CFD/UQ study, not a breakthrough. It deserves peer review, because the methodology is standard and the result is useful for SHS design. With revisions, the claim about d2 dominance should be positioned as conditional on the assumed distribution.","headline":"Solid DNS-based UQ study showing spanwise spacing dominates drag reduction on textured posts, but the Sobol ranking rests on an arbitrary Gaussian input assumption and the 'unified' framing oversells the coverage.","tokens_in":13334,"tokens_out":2364,"would_cite":false,"duration_ms":27464,"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":"Spanwise spacing, not streamwise spacing, governs how much drag reduction varies across superhydrophobic surfaces.","keywords":["superhydrophobic surfaces","drag reduction","uncertainty quantification","polynomial chaos expansion","Sobol sensitivity analysis","turbulent channel flow","microstructure spacing","direct numerical simulation"],"falsifier":"Run two pairs of DNS cases at Re_tau=590: first hold d2 at its mean and vary d1 by ±0.33 times the mean value, then hold d1 at its mean and vary d2 by ±0.33 times the mean value. If the drag-reduction spread from varying d2 is not roughly fourteen times the spread from varying d1, the claimed dominance of spanwise spacing is refuted at that Reynolds number.","tokens_in":12520,"feed_emoji":"🌊","tokens_out":12210,"duration_ms":131364,"temperature":0.7,"pith_summary":"This paper aims to show that a single geometric axis—the spanwise gap between microscopic posts—controls how much a superhydrophobic surface's drag reduction is affected by manufacturing variability. The authors unify post-type, ridge-type, and transverse-ridge-type surfaces into one two-parameter family (d1, d2), run direct numerical simulations of turbulent channel flow at three Reynolds numbers, and fit a polynomial-chaos surrogate to the results. Their sensitivity analysis finds that the spanwise spacing d2 is the dominant input at every Reynolds number studied, with its sensitivity index roughly six to fourteen times larger than that of the streamwise spacing d1. If this holds, designers should treat the sideways spacing as the primary tolerance to control, and the same conclusion applies across the whole family of textured surfaces, not just one pattern. The paper also shows the resulting uncertainty in velocity and Reynolds stresses concentrates near the wall and shrinks as Reynolds number increases.","feed_headline":"Spanwise spacing rules drag-reduction uncertainty on textured surfaces","feed_subtitle":"Sideways post spacing, not streamwise, controls drag-reduction scatter by up to 14 to 1.","key_machinery":"The central device is the unified two-parameter post-spacing family (d1: streamwise gap, d2: spanwise gap), which turns the three standard superhydrophobic pattern types—posts, longitudinal ridges, transverse ridges—into continuous limiting cases of one geometry. The load-bearing machinery is a polynomial-chaos expansion surrogate trained on Latin-hypercube-selected DNS samples: it maps (d1,d2) to drag reduction, and its coefficients directly yield Sobol' sensitivity indices that decompose the variance into d1, d2, and interaction contributions. This lets a small number of expensive DNS runs (12–30 per Reynolds number) generate continuous response surfaces and uncertainty statements across t","core_discovery":"The paper's central claim is a sensitivity ranking: for a post-type superhydrophobic surface with uncertain streamwise spacing d1 and spanwise spacing d2, the drag-reduction response is overwhelmingly more sensitive to d2. At Re_tau=180 the Sobol index (variance-based sensitivity measure) of d2 is about seven times that of d1; at Re_tau=590, about fourteen times. The interaction term grows with Reynolds number and exceeds the single d1 contribution at Re_tau=590. The paper links this to near-wall turbulence: uncertainty is largest in the viscous sublayer, the shear-stress component R12 is suppressed near the wall, and turbulence-anisotropy trajectories show weakened coherent structures that","pith_inferences":["Because the assumed input distribution is unmeasured, repeating the analysis with correlated distributions estimated from profilometry could change the Sobol ranking; the 14-to-1 ratio is a property of that assumption as much as of the physics.","Since d2 is the dominant axis and ridge-type surfaces hold d1=0, ridge-type textures should exhibit less drag-reduction scatter under the same manufacturing noise than transverse-ridge textures, which vary d2 by construction.","A practical shortcut suggested by this result: for design exploration, treat d2 as the random variable and hold d1 at its nominal value; the surrogate would lose little accuracy while requiring fewer DNS runs."],"forward_implications":["Manufacturing control should target spanwise spacing first; tightening d2 will reduce scatter in drag-reduction performance more than tightening d1.","The unified parameterization means design guidance transfers across post, ridge, and transverse-ridge surfaces rather than being limited to one pattern.","At higher Reynolds numbers, the same geometric uncertainty produces less variation in near-wall turbulence anisotropy, so superhydrophobic-surface performance becomes more predictable in high-speed regimes.","The growing interaction between d1 and d2 at high Reynolds numbers means the two spacings cannot be treated as independent design levers when Re_tau is large."],"supporting_citations":[{"why":"Supplies the reference superhydrophobic channel-flow DNS configuration and the velocity/Reynolds-stress validation data.","marker":"[15,24]"},{"why":"Supplies the GPU-accelerated spectral element solver used to generate all DNS flow fields.","marker":"[22]"},{"why":"Supplies the polynomial chaos expansion formulation that the surrogate models are built on.","marker":"[19,20]"},{"why":"Introduces the point-collocation non-intrusive polynomial chaos method and the oversampling choice that sets the number of DNS samples.","marker":"[29]"},{"why":"Supplies the Latin hypercube sampling strategy used to select the training configurations.","marker":"[30]"},{"why":"Supplies the software implementation used to carry out the polynomial chaos and Sobol-index calculations.","marker":"[32]"}],"fun_headline_variants":["Spanwise spacing dominates drag-reduction uncertainty on superhydrophobic surfaces","Sideways spacing drives drag-reduction scatter 14x more than streamwise","Uncertain drag reduction? It's all about spanwise post spacing","For superhydrophobic drag, spanwise spacing is 14x more crucial","Sensitivity analysis: spanwise spacing rules drag-reduction variance"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The ranking of spanwise over streamwise spacing rests on the unmeasured assumption that real manufacturing deviations in d1 and d2 are independent, bell-shaped, and fixed in size relative to the post width; if actual scatter is correlated or differently sized, the ranking could shift.","fun_headline_variants_meta":{"raw":{"variants":["Spanwise spacing dominates drag-reduction uncertainty on superhydrophobic surfaces","Sideways spacing drives drag-reduction scatter 14x more than streamwise","Uncertain drag reduction? It's all about spanwise post spacing","For superhydrophobic drag, spanwise spacing is 14x more crucial","Sensitivity analysis: spanwise spacing rules drag-reduction variance"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000233,"raw_usage":{"total_tokens":1339,"prompt_tokens":761,"completion_tokens":578,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":505,"completion_tokens_details":{"reasoning_tokens":482}},"tokens_in":505,"tokens_out":578,"duration_ms":5576,"temperature":1.0,"reasoning_tokens":482,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T12:43:37.078596+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run two pairs of DNS cases at Re_tau=590: first hold d2 at its mean and vary d1 by ±0.33 times the mean value, then hold d1 at its mean and vary d2 by ±0.33 times the mean value. If the drag-reduction spread from varying d2 is not roughly fourteen times the spread from varying d1, the claimed dominance of spanwise spacing is refuted at that Reynolds number.","supporting_citations":[{"cited_title":"Xiu ,\\ @noop journal journal Communications in computational physics \\ volume 2 ,\\ pages 293 ( year 2007 ) NoStop","cited_arxiv_id":null,"evidence_quote":"Supplies the GPU-accelerated spectral element solver used to generate all DNS flow fields."},{"cited_title":"Euler-Lagrange study of Microbubble-Laden Turbulent Flow over Superhydrophobic surfaces","cited_arxiv_id":"2504.07377","evidence_quote":"Introduces the point-collocation non-intrusive polynomial chaos method and the oversampling choice that sets the number of DNS samples."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Latin hypercube sampling strategy used to select the training configurations."},{"cited_title":"Hosder , author R","cited_arxiv_id":null,"evidence_quote":"Supplies the software implementation used to carry out the polynomial chaos and Sobol-index calculations."}],"review_version":1}