{"id":"10b0d4bf-6c54-4277-be65-653497f93c2e","arxiv_id":"2511.00008","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Randomized Kelvin-Helmholtz solutions of the compressible Euler equations can be characterized by Cesàro-averaged numerical solutions, Reynolds-stress and energy defects, PDFs, and POD modes that stabilize as mesh resolution increases.","lead":"This paper tests a statistical way to describe the Kelvin-Helmholtz instability in the compressible Euler equations, where weak solutions are not unique, by averaging many numerical solutions over grids and a random input parameter. The authors find that averaged quantities, fluctuation-driven energy defects, and reduced-order mode counts behave consistently, suggesting a practical route for describing chaotic inviscid flows statistically.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported 'random KH' statistics are conditional on one fixed draw of the perturbation coefficients in Eq. (4.2); 1-D ξ collocation cannot support the general claim.","rationale":"The reader's verdict centers on the unproved Theorem 2.2 and on reproducibility. I agree those are real gaps, but the more load-bearing issue for the paper's headline claim is the sampling of the random input. The manuscript declares a_i^k and b_i^k to be random variables but draws them only once; the stochastic collocation then varies only the scalar ξ inside tanh(ξ). Consequently, all reported statistics are conditional on a single realization of the coefficients. Theorem 2.2, even if fully proved, concerns Monte Carlo/Cesàro convergence for a fixed initial-data law and cannot convert a one-dimensional slice of the input space into an ensemble representative of random KH instabilities. This is not an internal inconsistency: the numerics likely compute the conditional statistics correctly. The risk is external validity—whether the patterns in Figs. 4.4–4.8 and Tables 4.1–4.4 generalize. A multi-draw sensitivity check is inexpensive and decisive. Because the reader already assigned CONDITIONAL, my concern does not change the verdict but sharpens the condition: the authors should demonstrate stability across independent coefficient draws before claiming a general statistical description of random KH instability. I credit the paper for clear definitions and honest convergence diagnostics; the issue is representativeness of the random space, not algorithmic error or scientific misconduct.","tokens_in":14196,"tokens_out":6687,"duration_ms":62401,"concrete_test":"Re-run the full M=5, L=101, τ∈{0.1,0.5,1.1} pipeline for K=10 independent draws of the 20 coefficients (a_i^k,b_i^k) in Eq. (4.2), keeping ξ collocation and all numerical settings identical. For each draw compute: (i) ||E[tr R_5]^{(r)} - E[tr R_5]^{(1)}||_1 and the analogous energy-defect difference; (ii) the window PDFs in Fig. 4.7; (iii) K_0.95 at m=5 for ρ. If the coefficient-induced spread exceeds the mesh-convergence residuals reported in Fig. 4.6 or is comparable to the τ-variation in Tables 4.1–4.4, the reported statistical framework is conditional on one draw and the central claim is not established. If the spread is below those thresholds, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that random KH instabilities can be systematically described by averaging over realizations. The implementation, however, does not average over the realized randomness of the initial perturbation. In Eq. (4.2), the perturbation is Y_i(x,y;ξ)=(1+τ tanh ξ) Σ_{k=1}^{10} a_i^k cos(b_i^k+10kπx), with a_i^k, b_i^k declared uniformly distributed random variables. These 20 coefficients are drawn once ('generated once for repeatability') and are not part of the collocation; the stochastic space is reduced to s=1 by assuming ξ is the only random variable (§3). Hence the mean, variance, PDFs, defects, and POD are all conditional on a single realization of the phase/amplitude vector. A different draw of a_i^k,b_i^k changes the spatial structure of the shear-layer perturbation, not just its global amplitude. There is no argument that one draw is representative or that the quantities reported are insensitive to it. Theorem 2.2, even if proved, concerns convergence of MC/Cesàro averages at fixed initial-data law; it does not justify replacing a 20(+1)-dimensional random input by a one-dimensional slice. Thus the evidence in Figs. 4.4–4.8 and Tables 4.1–4.4 may characterize one deterministic perturbation family, not random KH instability.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a numerical framework for studying random dissipative weak (DW) solutions of the compressible Euler equations, applied to the Kelvin–Helmholtz instability. The method combines A-WENO spatial discretizations on embedded uniform meshes with stochastic collocation in a one-dimensional random variable ξ, forms Cesàro averages over M=5 meshes, and computes diagnostics: Reynolds stress and energy defects, PDFs of averaged quantities, and POD mode counts. The authors claim that these tools reveal turbulence-like statistical structure and that random KH instabilities can be meaningfully characterized by a turbulence-inspired statistical framework. The theoretical justification relies on Theorem 2.2, which is asserted but not proved.","tokens_in":14556,"tokens_out":7724,"duration_ms":69955,"significance":"If the central claim were fully supported, the paper would provide a practical way to extract well-behaved observables from an ill-posed inviscid problem: defect residuals decrease with M (Fig. 4.6), defect ratios remain within the theoretical bounds of (2.2), and POD mode counts appear to saturate around 76–78 for the finest mesh, suggesting a moderate-dimensional reduced basis. The numerical pipeline is clearly described and reproducible in principle, and the diagnostics are internally consistent. However, the stochastic representation used in the experiments is too narrow to support the paper's broad claim about random KH instabilities, and the theoretical basis for interpreting the computed means is not made rigorous.","major_comments":[{"comment":"The stochastic dimension is reduced to s=1 by fixing the random coefficients a_i^k, b_i^k after a single draw ('generated once for repeatability') and treating ξ as the only random variable. Thus the ensemble averages, PDFs, defects, and POD results in Figures 4.4–4.9 and Tables 4.1–4.4 are conditional on one realization of the 20 Fourier phases/amplitudes. A different draw changes the spatial structure of the shear-layer perturbation, not just its amplitude. No sensitivity analysis or argument that the single draw is representative is provided. This directly undermines the §5 conclusion that 'random KH instabilities can be meaningfully characterized using a turbulence-inspired statistical framework.' The authors should either sample a_i^k, b_i^k (e.g., by stochastic collocation or Monte Carlo in the full 21-dimensional space) or explicitly restrict the claim to the one-dimensional famil","section":"§3, Eq. (4.2)"},{"comment":"Theorem 2.2 is asserted without proof; the manuscript says only that its proof can be established analogously to [14, Theorem 5.6] for the barotropic system. This theorem is the theoretical basis for interpreting the computed means of Cesàro averages as approximations of the expected DW solution. In addition, the theorem is formulated for Monte Carlo averages over i.i.d. samples (indexed by ℓ), whereas the numerical method uses deterministic collocation points in ξ with a high-order interpolation/quadrature. No argument is given that the collocation means converge to the same limit as the Monte Carlo means. The authors should provide a proof or a precise citation covering the collocation case, or soften the theoretical framing of §4.","section":"§2.2, Theorem 2.2"},{"comment":"The POD analysis uses L=101 collocation snapshots for each mesh m, so the data matrix has rank at most 101. The claimed 'saturating' mode count of 76–78 at m=5 could be influenced by the fixed finite snapshot count; no study of the dependence on L is reported. A convergence check in L (e.g., L=51, 101, 201) is needed before concluding that the mode count saturates as a property of the flow rather than as an artifact of the snapshot set.","section":"Tables 4.1–4.4"}],"minor_comments":[{"comment":"In the momentum identity, the first term should be m·∂_t φ, not m·φ. As printed, the variational formulation is missing the time derivative.","section":"Appendix A, Eq. (A.2)"},{"comment":"The target space is written as L^q(T^d×(0,T); R^{d+1}); for the state (ρ,m,S) this should be R^{d+2}, matching Theorem 2.1.","section":"§2.2, Theorem 2.2"},{"comment":"The text says the ratio E5/tr(R5) stays within the 'theoretically bounds 0.5 and 1.25, specified in (2.2).' These are actually the inverse bounds 1/d2 and 1/d1 for d=2 and the chosen γ. Please clarify that the plotted ratio is E/trR and relate it explicitly to (2.2).","section":"§4, Figure 4.5"},{"comment":"The definition of R_M and E_M does not make clear whether the expectation over ξ (as in (3.2)) is applied inside the defect formulas or whether the displayed quantities are pointwise in ξ. Please define explicitly, e.g., R_M(x,y) = E[ ... ] or R_M(x,y;ξ).","section":"§3, Eq. (3.5)"},{"comment":"The singular value decay in Figure 4.9 is shown for m=1 and m=5 only; the caption and text would benefit from specifying the mesh sizes N_m used.","section":"§4.1.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is well written and the numerical machinery is credible, but the one-dimensional reduction of the random space is a serious gap in the central claim. If the authors expand the sampling to the full coefficient space or clearly restrict the conclusions, the paper could become publishable. The unproved Theorem 2.2 and the mismatch between the Monte Carlo theorem and the actual collocation algorithm also need to be addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, what you should know: this is an honest numerical study that reports a genuinely new set of observations—Cesàro-averaged dissipative weak solutions of the KH instability computed with stochastic collocation, defect statistics that converge as the number of embedded grids grows, and POD mode counts that saturate around 76–78. The internal evidence is consistent: the defect ratio respects the theoretical bounds, the residuals shrink with M, and the POD saturation is a concrete finding a reduced-order modeler could use.\n\nThe soft spot is in the word 'random.' The stochastic space is reduced to a single scalar ξ, which enters through tanh(ξ) and just inflates a fixed perturbation shape. The amplitudes a_k and phases b_k in Eq. (4.2) are declared random variables, drawn once, and never sampled. So the PDFs, moments, and defects are all conditional on one realization of the perturbation geometry. A different draw would change the spatial structure of the shear layer, not just its amplitude, and the paper gives no reason to expect that the one draw is representative. The claim in §5 that random KH instabilities can be meaningfully characterized outruns the experiments, which really characterize a one-parameter family of perturbations.\n\nWhere the paper earns credit: it is not a curve fit. The Reynolds stress and energy defects are computed directly from definition (3.5), the POD analysis is diagnostic, and the comparison with the bounds (2.2) is the right check. The link between DW solutions and turbulence-style observables is conceptually useful.\n\nThe soft spots beyond the randomness gap: Theorem 2.2 is simply asserted, with a pointer to a proof in a barotropic setting. If the interpretation of the averages depends on it, the proof should be supplied or the result demoted to an assumption. The m0 parameter in the mesh sequence is not specified, which makes the exact grids irreproducible. And the passing remark that unreported window experiments gave 'consistent evidence' is too vague to carry weight. Minor, but worth fixing.\n\nOverall: this paper deserves a serious referee. It is a legitimate numerical study with reproducible-in-principle experiments and a clear diagram of how to compute the observables. But the authors should either sample the coefficient space more honestly or narrow the conclusion to what a single ξ can actually show. Send it to review, with the expectation of a conditional revision.","headline":"Useful numerical observations, but the 'random' in the title is really a one-parameter family; the paper deserves review once the authors close the gap between the sampling and the claim.","tokens_in":15053,"tokens_out":3052,"would_cite":false,"duration_ms":29381,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M70","65M06","76M20","35L65","35R60"],"pacs":[],"model":"deepseek-v4-flash","headline":"Random Kelvin–Helmholtz instabilities in compressible Euler flow have reproducible statistical structure even though individual solutions do not.","keywords":["Kelvin-Helmholtz instability","compressible Euler equations","dissipative weak solutions","stochastic collocation","Cesàro averages","Reynolds stress","proper orthogonal decomposition","turbulence statistics"],"falsifier":"Compute the same statistical setup with far more than 101 collocation points, with multi-dimensional random inputs, or with more than five embedded meshes, and check whether the Reynolds-stress and energy defects continue to decrease. If the defects stop converging or POD mode counts keep growing with resolution, the claimed saturation and stabilization are false. A sharper test: run the same flow with the Navier–Stokes equations at progressively smaller viscosity and compare the averaged defect measures to those predicted here; a mismatch would show that the DW statistical description does no","tokens_in":14043,"feed_emoji":"🌊","tokens_out":4464,"duration_ms":43245,"temperature":0.7,"pith_summary":"Compressible Euler equations admit many weak solutions for the same initial data, so a single deterministic prediction of a Kelvin–Helmholtz shear layer is not well defined. This paper tries to show that the right object of study is not one solution but a statistical ensemble: perturb the shear interface randomly, solve on several nested grids, and average. The claim is that the resulting Cesàro averages approximate a random dissipative weak solution, and that turbulence-style diagnostics—Reynolds stress, energy defect, PDFs, and POD spectra—stabilize under mesh refinement and across perturbation strengths. If true, this gives a practical way to describe chaotic inviscid compressible flows without resolving every small-scale feature.","feed_headline":"Averaging tames non-unique Kelvin-Helmholtz flows","feed_subtitle":"Cesàro averages, stress defects, and POD modes settle into stable patterns across random perturbations and mesh refinements.","key_machinery":"The central objects are random dissipative weak (DW) solutions: defect-modified weak solutions of the Euler equations in which unresolved oscillations are encoded by Reynolds stress and energy defect rather than by a single pointwise solution. The machinery is stochastic collocation in a one-dimensional random space (uniform collocation points, CWENO7 interpolation, exact integration for the uniform measure), embedded uniform meshes whose solutions are combined into Cesàro averages, and defect formulas that turn averaged moments into turbulence diagnostics. Theorems 2.1 and 2.2 assert strong convergence of Cesàro and Monte-Carlo averages to this DW solution, which is what licenses interpreti","core_discovery":"On the paper's own terms, the discovery is that a randomized Kelvin–Helmholtz instability can be meaningfully characterized by averaged quantities. Random interface perturbations (controlled by a scalar ξ and ten random Fourier coefficients per interface) are propagated with a fifth-order A-WENO scheme on five embedded meshes; solutions are interpolated in random space with seventh-order CWENO and averaged to Cesàro means. The Reynolds-stress and energy defects computed from these averages satisfy the theoretical bounds of dissipative weak solutions, converge as the mesh index increases, and the ratio of energy to stress defects approaches the predicted slope. The POD analysis shows that the","pith_inferences":["A natural next test is to replace the single scalar random variable with a genuinely multi-dimensional random input; if the reported statistics change qualitatively, the present results describe only a narrow slice of the randomness.","The saturation of POD mode counts near 80 suggests an intrinsic stochastic attractor dimension for this configuration; if it persists at even finer meshes, it points to a resolution-independent reduced-order model.","One could connect the observed energy and Reynolds defects to the viscous dissipation of an underlying Navier–Stokes flow: at small but finite viscosity, the defect measures should match turbulent dissipation rates, tying the DW framework to physical turbulence modeling.","The unproved Theorem 2.2 is the bridge between the numerical averages and the DW solution concept; until it is supplied, the statistical claims rest on numerical evidence rather than a fully established convergence result."],"forward_implications":["Cesàro-averaged solutions on five embedded meshes approximate a random DW solution, so averaged fields can be used where deterministic weak solutions are non-unique.","Reynolds stress and energy defects, the unresolved-fluctuation signatures, shrink consistently with mesh index and their ratio stays within the theoretical bounds, making them usable as quantitative statistical monitors.","PDFs of averaged density, entropy, and stress in fixed spatial windows have non-Dirac support, indicating genuine stochastic variability rather than convergence to a single realization.","POD mode counts saturate around 76–78 modes for fine meshes across all perturbation strengths, implying a bounded reduced-order complexity for these flows; coarser meshes need only 20–50 modes.","Cesàro averaging reduces the effective dimensionality, with fewer POD modes for averaged density than for raw density, so statistical averaging acts as a practical compression strategy for chaotic flow data."],"fun_headline_variants":["Random KH chaos yields to Cesàro average order","Averaging tames non-unique Kelvin-Helmholtz flows","POD and stress defects pin down random vortex mixing","Statistical mean exposes stable structure in KH turbulence","Random interface bursts collapse into predictable patterns"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The central load-bearing premise is that the Cesàro and Monte-Carlo averages converge to a random dissipative weak solution—a convergence asserted by analogy to a related barotropic result but not proved here—together with the practical assumption that one scalar random variable and five embedded meshes capture the essential randomness of the instability.","fun_headline_variants_meta":{"raw":{"variants":["Random KH chaos yields to Cesàro average order","Averaging tames non-unique Kelvin-Helmholtz flows","POD and stress defects pin down random vortex mixing","Statistical mean exposes stable structure in KH turbulence","Random interface bursts collapse into predictable patterns"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000194,"raw_usage":{"total_tokens":1183,"prompt_tokens":727,"completion_tokens":456,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":471,"completion_tokens_details":{"reasoning_tokens":396}},"tokens_in":471,"tokens_out":456,"duration_ms":5188,"temperature":1.0,"reasoning_tokens":396,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T10:08:25.731015+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the same statistical setup with far more than 101 collocation points, with multi-dimensional random inputs, or with more than five embedded meshes, and check whether the Reynolds-stress and energy defects continue to decrease. If the defects stop converging or POD mode counts keep growing with resolution, the claimed saturation and stabilization are false. A sharper test: run the same flow with the Navier–Stokes equations at progressively smaller viscosity and compare the averaged defect measures to those predicted here; a mismatch would show that the DW statistical description does no","supporting_citations":[],"review_version":1}