{"id":"443cb120-cd41-4d0c-ac3f-45769f85da43","arxiv_id":"2411.17427","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A subpixel image analysis method estimates the bimodal size distribution of ligament blobs, interpreted as forming drops, in a rocket-engine coaxial jet under combustion, with fitted parameters that evolve with injector distance.","lead":"This paper measures the shapes of liquid ligaments peeling off a rocket-engine fuel jet under combustion, using a multiscale image analysis method. It extracts the sizes of the 'blobs' that bulge on these ligaments, taken as drops in formation, and shows how those sizes and numbers change with distance from the injector.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The two-component 3pGG extraction is non-unique: at z=12.1 mm the same 'region 2' blob population is fitted twice with materially different distributions (Fig. 13b vs 13d), so the reported bimodality and parameter trends are not uniquely determined by the data.","rationale":"The reader's weakest assumption correctly identified the unvalidated conversion from right-tail scale distributions to blob/drop diameters and the risk of fitting artifacts. The present stress-test sharpens that objection with internal evidence: the paper's own fits at z=12.1 mm give two different 'region 2' distributions depending on whether region 2 is paired with region 1 or with region 3. This is a concrete demonstration of non-identifiability rather than a purely hypothetical concern. It does not invalidate the genuine methodological contribution of the subpixel scale-distribution measurement, which is supported by a synthetic-image check in Fig. 5. However, because the central quantitative deliverable is the bimodal diameter distribution and its spatial evolution, and those numbers depend on a non-unique local fitting procedure, the appropriate verdict remains CONDITIONAL. The paper should either provide a global, model-selected fit with uncertainty quantification, or validate the extraction on synthetic images with known blob populations, before the quantitative distributions can be used as predictions.","tokens_in":16549,"tokens_out":7867,"duration_ms":76934,"concrete_test":"Take the z=12.1 mm measured −e2(d),d curve and fit a single global model consisting of a mixture of stretched exponentials over the full resolved range [16,250] µm, with component assignment and number of components determined by maximum-likelihood model selection (e.g., BIC/AIC) rather than by manual 'linear region' inspection; then compare the inferred region-2 parameters and peak diameters with the two piecewise fits in Fig. 13. If the global optimum gives a different number of components or shifts the region-2 peak by more than about 10% of its reported value, the reported bimodality and spatial parameter trends are not uniquely determined by the data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—bimodal blob/drop diameter distributions with regular spatial evolution—rests entirely on the two-component 3pGG decomposition of −e2(d),d presented in Section 4 (Eqs. 15–16). The paper itself demonstrates that this decomposition is not unique. At z=12.1 mm, the 'region 2' blob population is fitted twice: once as the second component of the [52,124] µm window (Figs. 13a–13b) and once as the first component of the [74,180] µm window (Figs. 13c–13d). The two recovered region-2 distributions differ in width, and the first extends to the region-3 peak at 190 µm. The authors resolve the ambiguity by asserting that the analysis of regions 1 and 2 reports the physically relevant distribution, but no selection criterion, likelihood comparison, or uncertainty estimate supports that choice. Because each component is a stretched exponential (Eq. 16), a smooth decreasing curve can be approximated by many two-component sums over finite intervals; the five free parameters per window are unconstrained by independent blob counts or drop-size measurements. The manual identification of 'linear regions' in Fig. 12, the exclusion of Region 0, and the extrapolated [d1,d2] intervals when Region 1 is absent at z=23.5 and 25.6 mm are likewise unquantified. The synthetic validation in Fig. 5 covers the measurement of scale distributions, not the blob-extraction model, so it provides no support for this step. Thus the reported bimodality, peak diameters (70, 100, 190 µm), blob-number fractions (73%, 74%), and their spatial parameter trends may be fitting artifacts of local window selection rather than physical characterizations.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a multiscale image-analysis method for characterizing the textural atomization ligaments of a cryogenic coaxial rocket injector operating under combustion. It improves the measurement of the scale distribution e2(d) by replacing the integer distance map with an exact distance transform and a 4x4 subpixel interpolation, and it validates this improvement on a synthetic sinusoidally perturbed cylinder. The paper then models the right-hand tail of the second derivative -e2(d),d as a sum of two three-parameter Generalized Gamma (3pGG) components and converts each component into a number-based blob/drop diameter distribution using previously derived equivalent-cylinder/sphere relations. Applying the model as a function of injector distance, it reports bimodal blob diameter distributions whose peak diameters, widths, and numbers evolve regularly with z, and it interprets these as drops in formation.","tokens_in":16926,"tokens_out":6710,"duration_ms":73923,"significance":"The improved scale-distribution measurement is a genuine methodological contribution: the synthetic test in Fig. 5 shows that the subpixel and exact-distance corrections remove pixelization oscillations and sharpen the measured peaks, and the availability of converged averages from 100 images is useful for future work in harsh combusting environments. If the blob-extraction step were independently validated, the paper would offer a route to quantitative estimates of ligament swelling scales and potential drop sizes where conventional droplet diagnostics cannot be used. However, the paper's central quantitative claim, namely the bimodal blob/drop diameter distributions and their spatial evolutions, is not yet supported, because the 3pGG decomposition is non-unique and the conversion has no ground-truth validation. The qualitative result that ligament scales increase with distance and then the textural population ceases is plausible, but the specific peak diameters and parameter trends should be treated as conditional.","major_comments":[{"comment":"The two-component 3pGG extraction is non-unique. At z=12.1 mm the paper fits the 'region 2' blob population twice: as the second component of the [52,124] um fit and as the first component of the [74,180] um fit (Figs. 13a-13d). The two recovered region-2 distributions differ in width, and the first extends to the 190 um region-3 peak. No selection criterion, likelihood comparison, or uncertainty estimate is given for choosing the regions-1-and-2 analysis over the regions-2-and-3 analysis. Because each component is a stretched exponential, many two-component sums can approximate the smooth measured curve over finite intervals, so the reported bimodality and the parameter trends in Fig. 15 are not uniquely determined by the data.","section":"§4, Fig. 13, Eqs. (15)-(16)"},{"comment":"The conversion of the scale-distribution tail into blob/drop diameter distributions relies on assumptions that are not tested. Section 4 assumes that the swelling scales of the ligaments correspond to circular blobs, that alpha=1 holds for the equivalent sphere set, and that the linear regions identified in Fig. 12 correspond to distinct physical blob families. The synthetic validation in Fig. 5 covers only the measurement of -e2(d),d; it does not test the blob-extraction model. The paper therefore needs either a synthetic test with known blob populations or independent droplet-size measurements to support the claim that the fitted components represent drops in formation.","section":"§4, Eqs. (6)-(12) and (15)-(16)"},{"comment":"The manual identification of the linear regions and the handling of absent regions are not quantified. Fig. 12 identifies linear regions by inspection; Region 0 is excluded with only a qualitative justification; and at z=23.5 and 25.6 mm, where region 1 is not found, the scale intervals [d1,d2] are extrapolated from linear trends of d1(z) and d2(z). These choices determine which data enter each fit, but no sensitivity analysis is reported. Consequently, the disappearance of region 1 and the switch to a single-component model at the farthest positions are not robustly established.","section":"§4, Fig. 12 and Fig. 14 (z=23.5, 25.6 mm)"},{"comment":"No uncertainty quantification accompanies the parameter evolutions in Fig. 15. The quantities q1, q2, Ds1, Ds2, Ns1, and Ns2 are reported as functions of z without confidence intervals, bootstrap estimates, or multi-start checks. In view of the non-uniqueness documented in Fig. 13, the 'clear spatial evolutions' may be within the ambiguity of the fitting procedure rather than physical trends. The paper should quantify the fit uncertainty and demonstrate that the trends are robust to initial parameter choices and interval definitions.","section":"§4, Fig. 15"}],"minor_comments":[{"comment":"The sentence 'An example is shown in Fig. 10 for the result obtained at z = 12.1 mm' should refer to Fig. 12, not Fig. 10.","section":"§4"},{"comment":"Equation (14) is used twice: once for the distance-bin repartition in Section 3.2 and again for the log-linearity relation in Section 4; the equations should be renumbered to avoid ambiguity.","section":"§3.2 and §4"},{"comment":"The caption for Fig. 5 should make explicit that panel b is the corrected measurement without subpixel interpolation and that panels c and d use 4x4 and 8x8 interpolation, respectively; the current wording is easy to misread.","section":"Fig. 5 caption"},{"comment":"The notation m(r_I) is introduced as the total distance count in bin r_I, but the preceding text describes splitting each pixel's contribution between bins r_I and r_I+1; rewording this passage would improve clarity.","section":"§3.2, Eq. (16)"}],"recommendation":"major_revision","confidential_remarks":"The measurement part of the paper is solid and worth publishing once the blob-extraction step is brought to the same standard. I recommend major revision rather than rejection because the gap can in principle be closed within the manuscript's scope by adding a synthetic-image validation of the two-component 3pGG extraction, uncertainty quantification, and a comparison against any available droplet-size data. The duplicate equation numbering should also be corrected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's genuine contribution is the improved scale-distribution measurement: subpixel interpolation plus exact Euclidean distances, validated on a synthetic sinusoidally perturbed cylinder. That test shows real gains—cleaner peaks, reduced pixelization artifacts, and correct small-scale behavior. Applying the method to a LOX/methane coaxial injector under combustion is also new, since laser diffraction and phase Doppler can't operate there. The authors extract spatial trends in blob size distributions, which is a plausible route to droplet information in exactly the conditions where nothing else works.\n\nThe soft spot is the conversion from scale distribution to blob/drop diameter distributions. The two-component 3pGG fit has five free parameters per window, the scale intervals are chosen manually, and the paper itself demonstrates non-uniqueness: at z = 12.1 mm the region-2 population is fitted twice with materially different widths depending on whether it is paired with region 1 or region 3. The authors resolve this by asserting that the regions 1 and 2 analysis reports the physically relevant distribution, but no model-selection criterion, likelihood comparison, or uncertainty estimate supports that choice. The synthetic validation covers the measurement of scale distributions, not the blob-extraction model, so it provides no independent support for this step. The extrapolated intervals when region 1 is absent at far positions add another unquantified choice.\n\nTo be fair, the authors do acknowledge the region-2 discrepancy and discuss why they prefer one fit. But they stop short of quantifying the uncertainty or comparing against any independent droplet size measurement, even though the blob-to-drop assumption is load-bearing. The spatial trends in Fig. 15 are parameter evolutions of these fits, not direct measurements, and the reported bimodality and number fractions should be read as conditional.\n\nThe paper is written for atomization and spray researchers, especially those working on injectors in harsh environments. It deserves a serious referee: the measurement advance is solid, the application is valuable, and the analysis framework is worth engaging with. A referee should push for uncertainty quantification, a principled way to choose between fits, or a reframing of the quantitative claims as qualitative. Send it to review, but expect heavy revision.","headline":"The measurement improvement is real and the application is new, but the reported blob/drop size distributions rest on a non-unique fit that needs uncertainty quantification before the quantitative claims should be trusted.","tokens_in":17520,"tokens_out":2089,"would_cite":false,"duration_ms":30681,"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":"Improved multiscale image analysis lets a rocket-engine coaxial jet's textural atomization be quantified from ligament shapes, yielding bimodal blob/drop diameter distributions whose sizes and counts evolve regularly with injector distance.","keywords":["textural atomization","multiscale analysis","scale distribution","blob diameter distribution","coaxial assisted atomization","subpixel image analysis","3pGG distribution","rocket engine combustion"],"falsifier":"Measure the actual droplet size distribution just downstream of the ligament zone, in a cold-flow twin of this injector with matched Weber number and momentum-flux ratio, and compare it with the blob diameter distribution predicted by the two-component 3pGG fit; if the predicted bimodality and its spatial evolution do not appear in the measured spray, the blob-equivalence step is wrong. A simpler check is to apply the extraction to synthetic ligament images with known blob sizes and verify that the fitted parameters recover the injected blob diameters.","tokens_in":16291,"feed_emoji":"💧","tokens_out":8636,"duration_ms":75039,"temperature":0.7,"pith_summary":"Textural atomization—the peeling of short-lived ligaments and droplets from a liquid surface—is usually visible in images but hard to quantify. This paper presents an improved multiscale image-analysis method that measures the scale distribution of those ligaments with subpixel accuracy, then fits the measured distribution with a two-component three-parameter generalized gamma model. The fit returns the sizes and numbers of the 'blobs' (swelling sections of the ligaments) and, treating blobs as drops in formation, turns them into droplet diameter distributions. Applied to a methane–oxygen coaxial jet burning at 7 bar in the fiber-type breakup regime, where the gas flow draws the liquid into long threads, the method reports a bimodal blob diameter distribution whose mean diameters grow linearly with distance from the injector while the blob count first rises and then falls. The authors conclude that the analysis locates the zone of most intense textural atomization (roughly 14–21 mm from the injector) and the point where this process stops, and that the regular parameter trends make mathematical models of textural atomization feasible.","feed_headline":"Image analysis maps where rocket-jet spray forms and stops","feed_subtitle":"Ligament shapes become droplet-size distributions that show where atomization peaks and stops.","key_machinery":"The central object is the scale distribution $e_2(d)$ of the liquid–gas interface, obtained by eroding the segmented liquid system with disks of diameter $d$ and recording the surface area lost at each scale. Its derivative $-e_2(d)_{,d}$ is proportional to the diameter distribution of the equivalent set of cylinders that has the same scale distribution as the real, arbitrarily deformed ligament population. The paper represents that cylinder distribution by a three-parameter generalized gamma (3pGG) function, and in the fully atomized limit ($\\alpha=1$) the blob diameter distribution inherits the same 3pGG form, $f_{0s}(D)\\propto D^{q-1}e^{-(D/D_s)^q}$. A two-component version of this model is fitted to the measured derivative, with each component corresponding to one blob family; the fitted parameters directly give the mean diameter, width, and relative number of each family through $D_s = q^{1/q}D_c$, $N_s \\propto -e_2(0)_{,d}\\,D_s^q/\\Gamma(q)$. On the measurement side, the enabling step is an exact-distance computation with fractional-distance binning combined with $4\\times4$ bilinear subpixel interpolation, which removes the oscillations and pixelization bias that otherwise corrupt the small-scale part of the distribution.","core_discovery":"The paper's central claim is that the multiscale scale distribution of textural ligaments, measured with subpixel fidelity, carries enough information to reconstruct the diameter distribution of the blobs that structure ligament deformation, and that treating these blobs as drops in formation yields quantitative spray information where conventional droplet diagnostics cannot operate. The argument runs through the identity $e_2(d)=L(d)/2$, which relates the scale distribution to the interface length of the eroded liquid system; differentiating once gives a quantity proportional to the diameter distribution of an equivalent set of cylinders, and a second differentiation connects the fully atomized limit to a spherical-drop distribution. Fitting the measured $-e_2(d)_{,d}$ with two 3pGG components separates blob families by scale. In the present reactive coaxial jet the fit identifies two such families, giving a bimodal diameter distribution: small blobs formed by textural deformation of the ligament surface and larger blobs formed by structural deformation of the ligament body. The mean diameters of both families increase linearly with distance from the injector, while the total blob number peaks between roughly 14 mm and 21 mm and the small-blob family disappears at the farthest positions, marking where textural atomization stops.","pith_inferences":["A natural next test would be to run the same analysis on a cold-flow twin with matched Weber number and momentum-flux ratio, where conventional droplet sizing is possible, and compare the predicted blob distributions with measured spray diameters; agreement would validate the blob-as-drop assumption, disagreement would localize the error in the equivalence step.","Because the method needs only images of the liquid interface, it should transfer directly to numerical simulation data: applying the extraction to simulated ligament fields with known blob populations would provide a ground-truth check of the linear-region selection and the two-component fit.","If the linear growth of blob mean diameter with injector distance is confirmed as a general trend, the multiscale parameters could serve as a surrogate for the local turbulent scales that initiate textural ligaments, linking image-derived morphology to turbulence-driven breakup models."],"forward_implications":["Textural atomization can be followed quantitatively along the injector: blob mean diameters grow linearly with distance, and the blob number per unit width rises then falls, locating the intense atomization zone near 14–21 mm.","The bimodal blob diameter distribution points to two coexisting production mechanisms in the textural process, one attached to the ligament surface and one to the ligament body.","The regular spatial evolution of the fitted 3pGG parameters ($q_1,q_2,D_{s1},D_{s2}$) means a closed-form expression for the blob diameter distribution as a function of injector distance is within reach.","In reactive flows where laser-diffraction or phase-Doppler instruments fail, this image-based route provides an alternative estimate of the droplet population being formed and of where its production is strongest."],"supporting_citations":[{"why":"Establishes that any set of deformed ligament structures has an equivalent set of cylinders with the same scale distribution, the foundation for converting scale measurements into diameter distributions.","marker":"Dumouchel et al. 2019"},{"why":"Provides the theoretical links between surface area, interface length, integrated curvature, and scale distribution that justify using e2(d) and its derivatives.","marker":"Dumouchel et al. 2022"},{"why":"Supplies the analytical scale-distribution solution for a sinusoidally perturbed cylinder used to validate the improved measurement procedure.","marker":"Dumouchel et al. 2023"},{"why":"Gives the relation between the second derivative of the scale distribution and the sphere/drop diameter distribution, and the dpc=0 fully atomized limit used in the blob model.","marker":"Thiesset et al. 2019"},{"why":"Introduced the conceptual representation of deformed ligaments as circular blobs of various diameters that the paper turns into a measurable quantity.","marker":"Villermaux et al. 2004"},{"why":"Provides the aerodynamic primary-breakup framework used to interpret the orientation of textural ligaments and to compare the observed growth of textural drop sizes.","marker":"Wu and Faeth 1993"}],"fun_headline_variants":["Subpixel imaging pinpoints where rocket spray breaks into droplets","Bimodal spray analysis reveals atomization peak and cutoff in rocket injectors","Multiscale image analysis tracks ligament-to-droplet transitions in rocket jets","Blob-diameter mapping shows where rocket-jet atomization intensifies and ends","Image-based method locates intense atomization zone in rocket coaxial spray"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire drop-size estimate rests on treating the swelling bumps of a deformed ligament as spherical drops and on reading the selected straight segments of a double-log plot as distinct blob families; the paper does not check that equivalence against independent droplet measurements.","fun_headline_variants_meta":{"raw":{"variants":["Subpixel imaging pinpoints where rocket spray breaks into droplets","Bimodal spray analysis reveals atomization peak and cutoff in rocket injectors","Multiscale image analysis tracks ligament-to-droplet transitions in rocket jets","Blob-diameter mapping shows where rocket-jet atomization intensifies and ends","Image-based method locates intense atomization zone in rocket coaxial spray"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000831,"raw_usage":{"total_tokens":3686,"prompt_tokens":1058,"completion_tokens":2628,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":674,"completion_tokens_details":{"reasoning_tokens":2531}},"tokens_in":674,"tokens_out":2628,"duration_ms":20500,"temperature":1.0,"reasoning_tokens":2531,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T12:07:11.787019+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the actual droplet size distribution just downstream of the ligament zone, in a cold-flow twin of this injector with matched Weber number and momentum-flux ratio, and compare it with the blob diameter distribution predicted by the two-component 3pGG fit; if the predicted bimodality and its spatial evolution do not appear in the measured spray, the blob-equivalence step is wrong. A simpler check is to apply the extraction to synthetic ligament images with known blob sizes and verify that the fitted parameters recover the injected blob diameters.","supporting_citations":[],"review_version":1}