{"id":"17762139-5f7e-4edb-a680-85153767751a","arxiv_id":"2606.24275","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"In 2D Euler flows, the structure-preserving SALT noise concentrates variance at vorticity gradients and scales as |k|² relative to the energy-preserving SFLT noise, which spreads variance diffusely.","lead":"This paper compares two ways of adding randomness to the two-dimensional Euler equations: one that preserves fluid structure (SALT) and one that preserves energy (SFLT). It finds SALT concentrates uncertainty at sharp vorticity gradients and is more sensitive to fine-scale motion, while SFLT spreads uncertainty more smoothly across the domain.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Headline |k|^2 scaling is set by the arbitrary SFLT normalization in Eq. (33), not by the frameworks themselves.","rationale":"The paper is a serious comparison with clear derivations of conservation properties and a plausible mechanism for localization. The single most load-bearing point is the quantitative superiority claim 'scaling by |k|^2'. That scaling is derived in Eqs. (23)–(24) by comparing {F_m,ω_k} with {ψ_k,F_m}, which already contains a factor 4π²||k||². The factor 4π² is removed by the SFLT normalization Eq. (33), but the ||k||² is retained. There is no invariant reason to remove one and not the other; a natural 'same velocity perturbation' normalization would remove both (or replace ||k||² by ||m||²), making the ratio 1 or ||k||²/||m||². Therefore the headline is conditional on the authors' calibration. The numerical results, while visually consistent, cannot settle this because they use the same normalization. This is exactly the reader's weakest assumption; I agree. The appropriate verdict remains CONDITIONAL: the paper should be published with the normalization dependence made explicit and ideally with a robustness check under an alternative normalization. No reason to change the reader's verdict.","tokens_in":21806,"tokens_out":8475,"duration_ms":82440,"concrete_test":"Recompute the spectral ratio and rerun the traveling-dipole and vortex-merger cases with the alternative normalization θ_k = σ4π²||k||²F_k (equivalently θ=-Δζ, so the SFLT forcing corresponds to the same velocity perturbation as the SALT forcing). If the |k|² ratio collapses to 1 and the SALT-vs-SFLT variance localization contrast is eliminated or reversed, then Eq. (33) is the sole source of the headline claim. If the contrast persists quantitatively, the conclusion survives but must be restated as normalization-dependent.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—SALT noise effects scale as |k|^2 relative to SFLT—depends entirely on the calibration θ_k = σ4π²F_k chosen in Eq. (33). In §3.1 the single-mode comparison gives {F_m,ω_k}/{ψ_k,F_m} = 4π²||k||², because ω_k = -4π²||k||²ψ_k. If the SFLT scalar is written θ_k = αF_k, the ratio becomes 4π²||k||²/α. The paper sets α=4π², yielding the headline |k|². But no physical principle fixes α: setting α=4π²||k||² (so θ=-Δζ, i.e., the SFLT noise is the vorticity of the SALT noise velocity) makes the ratio 1; setting α=1 gives 4π²|k|². Thus the claimed scaling is an artifact of normalization, not an intrinsic property of SALT vs SFLT. The numerical experiments inherit this choice, so they cannot independently validate the scaling; they only show the consequences of Eq. (33). The qualitative localization contrast might survive some normalizations, but the quantitative headline is conditional on an arbitrary constant. The paper should state this dependence explicitly or justify α from a physical/statistical principle.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper compares two geometric stochastic frameworks for the 2D incompressible Euler equations: SALT (circulation-preserving, via stochastic advection by Lie transport) and SFLT (energy-preserving, via stochastic forcing by Lie transport), together with their averaged variants LA SALT and EA SFLT. In §3.1 the authors derive a spectral comparison of the two noise terms, obtaining a ratio proportional to ||k||^2, which they interpret as SALT being more sensitive to high-frequency flow components and acting as a localized perturbation, whereas SFLT is a more regularized global forcing. This interpretation is tested numerically in three setups: a traveling dipole, vortex merger, and forced-damped turbulence. The numerical results show that SALT produces variance concentrated near vorticity gradients, while SFLT produces a more diffuse variance field. The paper concludes that the choice of geometric invariant determines the scale sensitivity and spatial distribution of modeled uncertainty.","tokens_in":22184,"tokens_out":10701,"duration_ms":103359,"significance":"If the central claim holds, the paper would provide practically useful guidance for selecting between circulation-preserving and energy-preserving stochastic parameterizations in geophysical fluid dynamics, a question of current interest. The manuscript has clear strengths: the derivations of the bracket identities are explicit and self-contained; no empirical constants are fitted; the numerical implementation uses the open-source Firedrake package and is described in enough detail to be reproducible; and the prediction that SALT localizes uncertainty near vorticity gradients is falsifiable. However, the headline quantitative claim—the ||k||^2 scaling of SALT relative to SFLT—is conditional on an ad-hoc normalization in Eq. (33). The numerical experiments inherit that normalization, so they cannot independently validate the scaling. The significance of the paper is therefore real but lower than the abstract suggests, and the qualitative localization contrast, while plausible, needs to be disentangled from the chosen calibration.","major_comments":[{"comment":"The central claim that 'noise effects scale by |k|^2 relative to SFLT' is not an intrinsic property of the two frameworks; it is determined by the arbitrary normalization θ_k = σ4π²F_k in Eq. (33). The spectral comparison gives a ratio 4π²||k||² between the SALT bracket {F_m, ω_k} and the SFLT bracket {ψ_k, F_m}. If the SFLT scalar is written θ_k = αF_k, the actual SFLT noise amplitude for that mode becomes α{ψ_k, F_m}, and the ratio becomes 4π²||k||²/α. Equation (33) sets α=4π², which converts the ratio to exactly ||k||². No physical or statistical principle is provided for this choice. Setting α=4π²||k||² (so that the SFLT forcing is the vorticity of the SALT noise velocity) would make the ratio 1, eliminating the claimed scale-sensitivity. The numerical experiments use this same normalization and thus cannot independently validate the scaling; they illustrate the consequences of Eq. (","section":"§3.1 and §4.2, Eqs. (23)–(24) and (32)–(33)"},{"comment":"The spectral comparison is local in state space and does not by itself establish the relative magnitude of noise effects in the stochastic dynamics. It compares the magnitudes of the noise vector fields {F_m, ω_k} and {ψ_k, F_m} for a single Fourier mode of the current state, but the actual effect on the ensemble variance is governed by the full nonlinear SPDEs (12) and (15), including the Itô correction and the coupled time evolution of ψ and ω. The claimed 'scaling by |k|^2' is therefore a heuristic indicator, not a proven property of the two frameworks. Because the numerical tests inherit the normalization of Eq. (33), they cannot resolve this gap. I recommend either deriving the variance/covariance dynamics in Fourier space (e.g., from the LA SALT/EA SFLT expectation equations) or softening the wording in the Abstract and §5 so that the scaling is presented as a conditional observati","section":"§3.1, Eqs. (23)–(24)"},{"comment":"The numerical evidence rests on ensembles of only 10 members, with no ensemble-size convergence study and no error bars on the variance maps. The text states that 10 members 'is found to provide a clear insight into the qualitative and quantitative differences', but the palinstrophy and trajectory plots show one-standard-deviation bands whose own sampling error is substantial at this ensemble size. Since the paper makes quantitative statements (e.g., 'the effect of EA SFLT is substantially smaller', §4.3; 'SALT induces variance localized near vorticity gradients'), the absence of sampling-error quantification weakens the empirical support. Please add a convergence check with increasing ensemble size, or at least report bootstrap/standard errors on the key statistics (palinstrophy peak timing, trajectory spread, variance maxima).","section":"§4.1–4.5, Figures 3, 5, 8, 13"}],"minor_comments":[{"comment":"The notation k is used both as a wave vector and as a vector component in Eq. (22) ('k = (k,l)^T'); this is confusing. Use e.g. k = (k_1,k_2).","section":"Throughout"},{"comment":"The author name is misspelled as 'Krajchnan'; it should be 'Kraichnan'.","section":"Reference [37]"},{"comment":"The structure description says 'The numerical tests are described in Section 2'; this should be Section 4.","section":"§1, last paragraph"},{"comment":"Minor wording: 'using the integration by parts identity' should be 'using the integration-by-parts identity'.","section":"Eq. (14)"},{"comment":"The definition of the low-frequency group ('5 ≤ ||k||') is incomplete; it presumably means 5 ≤ ||k|| < 10 or an explicit finite band. Please clarify.","section":"§4.2"},{"comment":"The captions say the common logarithm of the variance is shown, but the colorbar labels (e.g., 13.00, 6.61, 0.22) appear to be variance values, not logarithms. Please make the scale unambiguous.","section":"Figures 3, 5, 8, 13"},{"comment":"There is a typo 'SLFT' in the sentence 'individual SLFT realizations remain more closely aligned'; should be 'SFLT'.","section":"§4.5"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a timely question and the numerical implementation is technically sound, but the headline scaling claim is tied to the arbitrary normalization in Eq. (33). I am not recommending rejection because the qualitative localization contrast may survive a range of normalizations and is interesting in its own right; however, the authors need to either justify the calibration from a physical/statistical principle or explicitly present the results as normalization-dependent. The ensemble-size issue is secondary but should be addressed to make the numerical validation convincing."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Useful paper, and the first systematic numerical side-by-side of SALT and SFLT for 2D Euler vortex flows. The qualitative conclusion—SALT concentrates uncertainty near sharp vorticity gradients, SFLT spreads it more evenly—comes through clearly in the dipole, merger, and forced-turbulence experiments, and the mean-field expectation equations for LA SALT and EA SFLT are handled carefully. The clean single-mode comparison in §3.1 is genuinely instructive, and the numerical setup (mixed CG/DG, Firedrake, SSPRK3) looks sound.\n\nThe soft spot is the quantitative headline. The claim that SALT noise effects scale as |k|^2 relative to SFLT depends on the choice θ_k = σ4π²F_k in Eq. (33). If you write θ_k = α F_k, the single-mode ratio becomes 4π²||k||²/α. Setting α=4π² (the paper's choice) gives |k|²; setting α=4π²||k||² (i.e., SFLT noise is the vorticity of the SALT velocity noise) gives unity. No physical principle fixes α. So the scaling is a property of the chosen calibration, not of the frameworks themselves. The qualitative localization contrast may survive for a range of normalizations, but the paper overstates it as intrinsic. This needs to be flagged and ideally re-run or re-worded.\n\nOther gaps are more minor: ensemble size of 10 with no convergence check, variance maps with no error bars, and no code/data release. The variance maps are convincing as illustrations, but not as quantitative UQ statements. Also, the conclusion that SALT 'localizes' is close to a restatement of the equations: SALT perturbs ∇ω, SFLT perturbs ∇ψ, and ω is the Laplacian of ψ. The experiments confirm it, but they don't independently discover it.\n\nWho this is for: practitioners choosing between SALT and SFLT for subgrid-scale uncertainty in 2D models, and people doing stochastic geometric mechanics. It deserves a serious referee; the normalization issue is addressable, not fatal. I'd send it to review and ask for an explicit statement of the normalization dependence, an ensemble-size sensitivity test, and preferably a code/data pointer.","headline":"Useful first systematic SALT-vs-SFLT comparison for 2D vortex flows, but the headline |k|^2 scaling is an artifact of the noise normalization in Eq. (33), not an intrinsic property.","tokens_in":22605,"tokens_out":5293,"would_cite":true,"duration_ms":46551,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["76B47","60H15","37N10","65M60"],"pacs":["47.10.-g","47.32.C-","47.27.E-"],"model":"deepseek-v4-flash","headline":"Choosing which invariant to preserve — circulation or kinetic energy — determines whether stochastic uncertainty in 2D vortex flows concentrates at sharp gradients or spreads across the domain.","keywords":["stochastic fluid dynamics","SALT","SFLT","2D Euler equations","vortex dynamics","uncertainty quantification","transport noise","Lie–Poisson structure"],"falsifier":"Run the traveling dipole and vortex-merger experiments with the SFLT amplitude θ_k scaled by a wavenumber-independent factor different from 4π². If the ratio of SALT-induced to SFLT-induced variance no longer grows like |k|², the claimed scaling is an artifact of calibration rather than intrinsic. Alternatively, compute the power spectrum of the ensemble variance field: SALT variance should dominate at high wavenumbers and SFLT at low wavenumbers for the same noise basis.","tokens_in":21755,"feed_emoji":"🌀","tokens_out":9131,"duration_ms":79967,"temperature":0.7,"pith_summary":"The paper compares the two canonical geometric ways to add stochastic noise to the two-dimensional Euler equations: SALT (stochastic advection by Lie transport), which preserves circulation and the Lie–Poisson structure, and SFLT (stochastic forcing by Lie transport), which preserves kinetic energy. Its central finding is that these two choices are not equivalent in how they generate uncertainty: a Fourier mode used as SALT noise interacts with the vorticity field and is amplified by the squared wavenumber relative to the same mode used as SFLT noise, which interacts with the smoother stream-function gradient. As a result, SALT produces ensemble variance concentrated where vorticity gradients are sharp, while SFLT produces a broader, more diffuse variance field. Numerical experiments on a traveling dipole, vortex merger, and forced-damped turbulence confirm the analytical scaling and show that the conserved invariant determines the spatial distribution of modeled uncertainty.","feed_headline":"Vortex noise models diverge: SALT is k² stronger at small scales","feed_subtitle":"Preserving circulation instead of energy changes where model uncertainty appears — a choice that shapes forecast ensembles.","key_machinery":"The central object is the comparison of the two stochastic interaction terms in the Lie–Poisson formulation. In spectral space, the SALT bracket {F_m, ω_k} scales as |k|² times the SFLT bracket {ψ_k, F_m}, because vorticity is the Laplacian of the stream function and therefore carries amplified high-wavenumber content. In physical space, the analogous object is the gradient ratio ∂ω/∂r ÷ ∂ψ/∂r = 4c(cr²−2) for a Gaussian vortex, which grows with the sharpness parameter c. This ratio is the mechanism that converts the conservation choice into a scale-sensitivity and localization statement.","core_discovery":"The paper shows that the abstract choice of which geometric invariant to preserve has a concrete, measurable consequence. In spectral space, the SALT term {F_m, ω_k} carries a prefactor 16π⁴|k|² while the SFLT term {ψ_k, F_m} carries 4π², so the same Fourier mode excites flow components a factor |k|² more strongly when added as transport noise than when added as forcing noise. In physical space, an idealized Gaussian vortex gives the same message through the ratio of vorticity to stream-function gradients, 4c(cr²−2), which grows as vortices shrink. The numerical ensembles confirm that SALT localizes uncertainty along active vorticity gradients and SFLT distributes it globally. The authors de","pith_inferences":["A direct test of the mechanism: measuring the wavenumber spectrum of the ensemble variance field in a homogeneous turbulent run should show SALT variance concentrated at higher wavenumbers than SFLT variance at the same noise amplitude.","The claimed |k|² ratio is calibrated by the specific normalization θ_k = σ 4π² F_k; if a different relative amplitude were chosen, the absolute ratio would change, though the qualitative localization contrast may persist.","If the gradient-ratio mechanism is generic, the same localization-versus-spreading contrast should appear in shallow-water and MHD extensions that the paper proposes; these are natural settings to test it.","The paper explicitly defers the turbulent-regime comparison of the averaged frameworks LA SALT and EA SFLT; testing whether their regularization difference survives in forced turbulence is a clear next step."],"forward_implications":["Circulation-preserving SALT places ensemble uncertainty exactly where vorticity gradients are strong; energy-preserving SFLT spreads it broadly, so the two frameworks answer different uncertainty-quantification questions.","Because SALT's noise grows with wavenumber, uncalibrated SALT amplitude can artificially accelerate small-vortex break-up; SFLT is less prone to this when the location of coherent structures is not well known.","The mean-field versions LA SALT and EA SFLT inherit the contrast: LA SALT's diffusive term is driven by expected vorticity (sharp), EA SFLT's by expected stream function (smooth), so their regularization also differs in locality.","Since simultaneously preserving both energy and circulation reduces the noise to a stochastic time reparametrization, modelers must choose one invariant; the paper demonstrates that the choice has observable consequences for forecast spread."],"fun_headline_variants":["SALT amplifies small-scale vortex activity |k|² stronger than SFLT","Choosing circulation over energy concentrates vortex uncertainty","Circulation-preserving noise localizes; energy-preserving spreads","SALT vs SFLT: Small scales decide where vortex uncertainty lives","More noise at small scales if you preserve circulation, not energy"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The quantitative comparison depends on the noise calibration θ_k = σ 4π² F_k chosen to 'ensure a fair comparison' (Eq. 33); with a different relative amplitude between SALT and SFLT forcings, the |k|² ratio and the degree of localization would change.","fun_headline_variants_meta":{"raw":{"variants":["SALT amplifies small-scale vortex activity |k|² stronger than SFLT","Choosing circulation over energy concentrates vortex uncertainty","Circulation-preserving noise localizes; energy-preserving spreads","SALT vs SFLT: Small scales decide where vortex uncertainty lives","More noise at small scales if you preserve circulation, not energy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000715,"raw_usage":{"total_tokens":3046,"prompt_tokens":737,"completion_tokens":2309,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":481,"completion_tokens_details":{"reasoning_tokens":2221}},"tokens_in":481,"tokens_out":2309,"duration_ms":16635,"temperature":1.0,"reasoning_tokens":2221,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T10:19:43.045327+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the traveling dipole and vortex-merger experiments with the SFLT amplitude θ_k scaled by a wavenumber-independent factor different from 4π². If the ratio of SALT-induced to SFLT-induced variance no longer grows like |k|², the claimed scaling is an artifact of calibration rather than intrinsic. Alternatively, compute the power spectrum of the ensemble variance field: SALT variance should dominate at high wavenumbers and SFLT at low wavenumbers for the same noise basis.","supporting_citations":[],"review_version":2}