{"id":"50d002b9-5ea0-4d05-8b9a-8ac0d8b3339a","arxiv_id":"1908.08456","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"In synthetic tests, time-distance helioseismic inversions recover the peak depth of non-separable supergranule flow models even when the full velocity profile is not recovered.","lead":"Using simulated seismic waves, the authors test whether helioseismic inversions can recover the depth of supergranule flows that vary with both horizontal position and depth. They find that the full flow profile is hard to recover, but the peak depth of the flow is reliably recovered after a few iterations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Peak-depth recovery is demonstrated only in noise-free inversions with the true surface flow imposed, so the abstract's 'above the noise cutoff' hedge is untested.","rationale":"The paper is a careful synthetic validation, and the reader's CONDITIONAL verdict is appropriate. The central numerical finding, that peak depth is recoverable for the tested non-separable models under this regularization, is supported by the experiments as far as they go. The most load-bearing weakness is that the abstract and conclusion frame the result in terms of measurements 'presumably above the noise cutoff' while no noise is ever injected and the true surface flow is imposed exactly. This is not an internal inconsistency, but it is a correctness risk for the practical claim: if the exact surface boundary condition is doing most of the work, or if realistic noise breaks the fast convergence, the headline result will not transfer to observations. The paper explicitly identifies this limitation in the Conclusion, and I agree with the reader that it warrants a conditional verdict rather than rejection. No additional concern rises to the same level: the synthetic models are plausible, the full-waveform forward modeling is a legitimate validation step, and the use of a smooth basis is a stated regularization choice rather than a hidden assumption. Thus the verdict should remain CONDITIONAL with the recommendation that the authors either add the noise and surface-error experiments or soften the abstract's noise-cutoff language.","tokens_in":13274,"tokens_out":3170,"duration_ms":32311,"concrete_test":"Repeat the SG(d1)-SG(d4) inversions with two modifications: first, contaminate the fixed surface coefficients c_surf_i,ell with Gaussian errors at the level of residual HMI supergranule-averaging uncertainty (roughly 5-10 percent of the surface velocity), and second, add realization noise to the true travel times at the level expected after averaging N supergranule cells, following the noise model of Birch et al. (2006). If the recovered peak depths in Fig. 6 shift by more than about 0.5 Mm, or if the iteration trajectories in Fig. 7 no longer converge within the first two iterations, then the 'above the noise cutoff' claim is unsupported and the result should be reported as a noiseless proof-of-concept rather than as validation for real measurements.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing weakness is the gap between the central claim and the evidence. Section 5 and the abstract assert that peak depth is recovered in a few iterations where the measurements are presumably above the noise cutoff, but every experiment in the paper is noise-free, and the inversion is handed the exact surface stream-function coefficients via Eqs. (6)-(7). The algorithm therefore never solves the coupled problem of inferring the surface flow and the subsurface profile jointly; it only updates below-surface B-spline coefficients from a starting model whose surface already equals the truth. Real HMI-style Doppler measurements carry realization noise, and the ensemble-averaged surface profile of an average supergranule has its own bias and uncertainty, neither of which is modeled here. In addition, the depth metric in Eq. (13) is evaluated from the same stream-function representation used in the inversion, so the good peak-depth agreement may partly reflect the chosen basis and the imposed exact boundary rather than independent information carried by the travel times. The Conclusion itself concedes that validation under noisy measurements and uncertainty estimates are lacking; that concession is precisely what the abstract's 'presumably above the noise cutoff' anticipates, and it belongs in the central claim rather than only in future work.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents synthetic validation tests of time-distance helioseismic inversions for non-separable models of an average supergranule. The flow is written as a sum of Gaussians in depth times Legendre polynomials in the horizontal coordinate, projected onto a B-spline basis. Travel times and sensitivity kernels are computed with the SPARC code, and BFGS and conjugate-gradient optimizers update the subsurface B-spline coefficients to minimize a travel-time misfit. For low-dimensional models (69-161 parameters) the flow profile approaches the truth; for high-dimensional models the exact profile is not recovered, but the peak depth, defined by the maximum of the horizontally averaged squared stream function, is recovered within about two iterations for four models peaking at different depths. All experiments are noise-free and the surface stream-function coefficients are fixed to their true values.","tokens_in":13527,"tokens_out":5161,"duration_ms":55853,"significance":"If the result holds, it is a meaningful step: it extends earlier separable-profile inversions (Bhattacharya et al. 2017) to non-separable models, uses a consistent forward-plus-adjoint framework in the public SPARC code, and cross-validates the BFGS and CG optimizers. The recovery of a depth scale despite failure to recover the full profile is a falsifiable, practically useful outcome. However, the headline claim is currently conditional on two untested idealizations, exact surface flow and noise-free travel times, so the significance is moderate until those gaps are addressed.","major_comments":[{"comment":"The inversion fixes all B-spline coefficients at and above the surface to their true values, so the surface velocity is exactly known. This is a strong assumption that is not available in practice: real HMI-style Doppler measurements carry realization noise, and an ensemble-averaged supergranule surface profile has its own bias and uncertainty. Since the central claim concerns recovering peak depth from measurements 'presumably above the noise cutoff,' the paper needs to test robustness to perturbed or co-inverted surface coefficients and to noisy travel times before that claim can be considered established.","section":"Section 2.2, Eqs. (6)-(7)"},{"comment":"All inversions are noise-free: the misfit in Eq. (9) compares exact synthetic travel times, and no realization noise, data covariance, or measurement uncertainties are introduced. The abstract's hedge 'presumably above the noise cutoff' is therefore not actually tested. The conclusion itself concedes that uncertainties are lacking and that the noisy-measurement case requires further work; that concession should move into the central validation, for example by adding noise at levels typical of HMI ensemble-averaged supergranule measurements and recomputing peak-depth recovery. Without such experiments, the headline result remains conditional.","section":"Section 4 and Section 5, Figs. 3-4"},{"comment":"The peak depth is computed from the horizontally averaged squared stream function of the same B-spline/Legendre representation used in the inversion. Because the surface coefficients are fixed to truth and the B-spline basis is smooth, the depth estimator may inherit information from the prior and the basis rather than from the travel-time data. The paper would be strengthened by a control experiment using a different basis for the inversion or a different, more model-independent definition of depth, such as the depth of maximum vertical velocity, to confirm that the travel-time data alone carry the depth information. This is important because Fig. 2 shows that for models with many parameters the inverted flow profile is not close to the true model, yet the peak depth is recovered.","section":"Section 4.2, Eq. (13)"},{"comment":"For models SG(l3)-SG(l5), the model misfit kappa increases while the travel-time misfit decreases, which the authors interpret as convergence to local minima. If the same local-minimum behavior occurs in Case 2, the recovered peak depth could be a property of the starting model and basis rather than of the data. Please report the travel-time misfit and model-misfit curves for the SG(d1)-SG(d4) inversions and test whether the recovered peak depth depends on the starting model, for example by starting from a nonzero subsurface flow with a different initial depth.","section":"Section 4.1, Figs. 2-3 and Section 4.2, Figs. 5-7"}],"minor_comments":[{"comment":"Equation (8) appears to contain a typo in the advection term: the term 2 rho v . grad partial_t^2 xi should likely be 2 rho v . grad partial_t xi, since the dimensions otherwise do not match the other terms. Please verify and correct the equation or explain the notation.","section":"Section 3, Eq. (8)"},{"comment":"Several figure labels are corrupted or misspelled: 'De th[Mm]', 'In%erted', and 'In)erted' should read 'Depth [Mm]' and 'Inverted'. The figures should be regenerated with correct labels.","section":"Figures 2 and 4"},{"comment":"The axes in Fig. 6 are inverted so that larger depths appear to the left, and the panels lack axis labels and quantitative error bars or scatter estimates; please consider using a standard depth axis with depth increasing downward and adding labels or a caption note explaining the orientation.","section":"Figure 6"},{"comment":"The statement that inversions 'are able to recover their peak depths accurately' should be qualified to indicate that this is demonstrated for noise-free synthetic data with exactly known surface boundary conditions; the current wording overstates what the experiments show.","section":"Section 5, Conclusion"},{"comment":"The phrase 'presumably above the noise cutoff' in the abstract is a hedge that belongs in the results or conclusion rather than in the abstract, since no noise cutoff is computed or applied anywhere in the paper.","section":"Abstract and Introduction"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is an honest validation study, but the abstract overstates the realism of the experiments. The main revision requested, adding noise and relaxing the exact-surface-flow assumption, is feasible within the manuscript's scope and should be required before acceptance. The novelty over the authors' earlier separable-model paper is incremental but real, and the paper is a reasonable fit for a solar-physics journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper for one result: with the B-spline/Legendre regularization carried over from Bhattacharya et al. (2017), travel-time inversions of non-separable supergranule models recover the peak depth of the flow even when they miss the full profile. That is the useful, genuinely new thing here, and it comes out of a cleanly designed synthetic experiment.\n\nWhat is new: previous validation work assumed separable models, and Ferret (2019) showed that this premise is inconsistent with surface observations. This paper drops that assumption, goes to non-separable models with hundreds to thousands of parameters, and asks what can still be reliably inferred. The answer—peak depth, not the exact velocity field—is an important step for the subfield, because discriminating shallow versus deep supergranule models is exactly what people want to do with real data. The comparison between BFGS and conjugate-gradient schemes is a nice touch, and the willingness to report that the optimizer often lands in a local minimum is refreshingly candid.\n\nWhere it is soft: every experiment is noise-free, and the inversion is handed the exact surface stream-function coefficients (Eqs. 6–7). So the abstract's phrase 'measurements are presumably above the noise cutoff' is not actually tested. Real HMI-style Doppler measurements carry realization noise and the average supergranule surface profile has its own bias; without that, we do not know whether peak-depth recovery survives. The lack of any uncertainty estimate compounds this. The depth metric (Eq. 13) is computed from the same stream-function representation used in the inversion, so part of the agreement may reflect the basis rather than independent information. These are real caveats, and the conclusion acknowledges them, but the abstract still leans on a noise-related hedge that the body never confronts.\n\nThat said, the central claim holds for what is actually tested: in an idealized, noise-free, perfectly-surfaced-bounded setting, peak depth is recoverable while the full flow is not. The paper does not overstate beyond that in its conclusion. It is a legitimate, careful validation study, not a breakthrough and not a failure.\n\nWho gets value: local helioseismologists, particularly anyone inverting supergranule flows or planning ensemble-averaged observations. It deserves a serious referee, and I would send it to review. The main revision I would push for is a reframing of the abstract so the noise-free boundary condition is stated up front, or better, a single added experiment with realistic noise. Either way, the paper is worth engaging with for the peak-depth result.","headline":"A careful, honest validation showing that peak depth, not the full flow, is recoverable for non-separable supergranule models; the noise-free setup keeps the claim modest but leaves the abstract's 'above noise cutoff' untested.","tokens_in":14052,"tokens_out":1927,"would_cite":true,"duration_ms":23211,"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":"Time-distance helioseismic inversions recover the peak depth of non-separable supergranule models, even when the full subsurface flow is not retrieved.","keywords":["supergranules","time-distance helioseismology","travel-time inversions","adjoint sensitivity kernels","stream function","B-spline basis","solar convection","non-separable flows"],"falsifier":"Run the same four depth-peaking models with realistic noise added to the surface Doppler velocities, or with the surface coefficients estimated from the data rather than fixed to their true values, and record how often the recovered peak depth stays within the spread of Fig. 6; if modest noise pushes the inferred peak depth away from the true value, the claim that peak depth is reliably recoverable fails.","tokens_in":13039,"feed_emoji":"☀️","tokens_out":6605,"duration_ms":59178,"temperature":0.7,"pith_summary":"Supergranules are 30-Mm-wide divergent flows visible at the solar surface, but their depth is hard to measure, and different techniques disagree. This paper runs a validation test: it builds synthetic, non-separable supergranule flows, simulates seismic wave travel times through them, and asks whether a time-distance helioseismic inversion can find them again. With flows expressed in a smooth basis of horizontal Legendre polynomials and vertical B-splines, the inversion usually cannot reproduce the exact subsurface profile once the model has more than a few hundred parameters. It can, however, recover the peak depth—the depth at which the horizontally averaged squared stream function is largest—in about two iterations, for all tested models. The authors present this as a step toward deciding which supergranule parameters helioseismology can be trusted to infer.","feed_headline":"Supergranule peak depth survives helioseismic inversion test","feed_subtitle":"Synthetic non-separable flow tests show time-distance helioseismology pins down depth in about two iterations.","key_machinery":"The central object is the stream function $\\psi(x,z)$ that generates a mass-conserving flow through $\\mathbf{v} = \\rho^{-1}\\nabla\\times(\\rho\\psi\\,\\hat{\\mathbf{y}})$. It is written as a weighted sum over normalized Legendre polynomials in the horizontal direction and expanded on cubic B-splines in the vertical direction, which encodes an implicit smoothness regularization and shrinks the parameter count from a full gridded flow to hundreds or thousands of coefficients. The inversion iteratively updates only the subsurface B-spline coefficients, while the coefficients at and above the surface are fixed to their true values. The gradient direction comes from adjoint sensitivity kernels, $K_{i\\ell}=\\int dx\\,K_\\psi(x)B_i(z)f_\\ell(x)$, and the optimization is carried out with BFGS or nonlinear conjugate gradients, minimizing the squared travel-time misfit between true and iterated models.","core_discovery":"The paper claims that with the chosen implicit regularization, travel-time inversions of non-separable supergranule models recover their peak depths accurately even though the full velocity field is not recovered. For the low-dimensional case, with fewer than roughly 200 parameters, the iterative inversion approaches the true flow; for larger models the travel-time misfit keeps falling but the solution converges to a local minimum. In all depth-peaking models SG(d1)-SG(d4), the peak depth of the horizontally averaged squared stream function is retrieved within roughly two iterations, using either BFGS or conjugate-gradient optimization. The paper defines this peak depth as a lower-bound estimator of supergranule depth and argues the result is an important step toward identifying what can be reliably inferred from time-distance helioseismology.","pith_inferences":["If the peak-depth recovery is as robust in the presence of noise as it is here, then depth becomes a practical observable for testing supergranule formation theories, such as rotation-suppression or shallow-convection models.","The surface coefficients fixed in Eqs. (6)-(7) are an optimistic premise; a test with noisy or estimated surface velocities is the natural next validation, and the paper itself notes that noise will limit how deeply the kernels can probe.","The local-minimum behaviour suggests that travel-time misfit alone is a weak selector among high-dimensional flow models; adding model-covariance or prior information, as the authors note, could turn the recovered depth into a full profile with uncertainties.","One could test whether the same basis regularization recovers peak depth for non-axisymmetric or time-dependent supergranules, since the present models are azimuthally symmetric and stationary."],"forward_implications":["Peak depth is a robust inferential target: even when the inversion falls into a local minimum, the depth of maximum $\\int\\psi^2\\,dx$ comes out right in about two iterations.","For flows described by fewer than about 200 parameters, the full stream function and velocity components are progressively recovered, suggesting that smaller basis sets are a route to full-profile inversions.","The monotonically decreasing travel-time misfit in high-dimensional runs is not evidence of convergence to the true model, since the model misfit can worsen while travel-time misfit improves.","Ensemble averaging over supergranule cells, which improves signal-to-noise by roughly $\\sqrt{N}$, may allow the depth recovery demonstrated here to be translated into bounds on real supergranule depths.","Depth estimates obtained this way can be compared with mode-coupling and holographic estimates to check consistency among helioseismic techniques."],"supporting_citations":[{"why":"Introduces the Legendre/spline basis decomposition inversion that this paper extends from separable to non-separable flows.","marker":"Bhattacharya et al. 2017"},{"why":"Shows observed surface velocities of an average supergranule are inconsistent with 2D separable models, motivating the non-separable test.","marker":"Ferret 2019"},{"why":"Supplies the adjoint method for computing sensitivity kernels that map stream-function perturbations to travel-time misfit.","marker":"Hanasoge et al. 2011"},{"why":"Establishes the full-waveform inversion framework and documents the non-convergence of high-dimensional unregularized inversions.","marker":"Hanasoge 2014"},{"why":"Provides ensemble-averaged surface Doppler measurements that set the surface velocity values fixed in the inversions.","marker":"Duvall & Birch 2010"},{"why":"Supplies the mass-conserving stream-function form for average supergranules and the shallow-peak depth estimate.","marker":"Duvall & Hanasoge 2012"},{"why":"Quantifies how realization noise limits the depth to which flow inversions are sensitive, framing the noise caveat.","marker":"Dombroski et al. 2013"}],"fun_headline_variants":["Peak depth recovered in helioseismic inversion test","Supergranule depth pinned down despite complex profiles","Helioseismic inversions find supergranule depth fast","Non-separable supergranule profiles: depth still recovered"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The inversion fixes the stream-function coefficients at and above the solar surface to their true values (Eqs. 6-7), so the algorithm is handed the exact surface flow and only needs to find the subsurface part; noisy or biased surface measurements could destroy the peak-depth recovery.","fun_headline_variants_meta":{"raw":{"variants":["Peak depth recovered in helioseismic inversion test","Supergranule depth pinned down despite complex profiles","Helioseismic inversions find supergranule depth fast","Non-separable supergranule profiles: depth still recovered"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000169,"raw_usage":{"total_tokens":1292,"prompt_tokens":999,"completion_tokens":293,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":615,"completion_tokens_details":{"reasoning_tokens":227}},"tokens_in":615,"tokens_out":293,"duration_ms":3364,"temperature":1.0,"reasoning_tokens":227,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:38:47.748672+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same four depth-peaking models with realistic noise added to the surface Doppler velocities, or with the surface coefficients estimated from the data rather than fixed to their true values, and record how often the recovered peak depth stays within the spread of Fig. 6; if modest noise pushes the inferred peak depth away from the true value, the claim that peak depth is reliably recoverable fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows observed surface velocities of an average supergranule are inconsistent with 2D separable models, motivating the non-separable test."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the full-waveform inversion framework and documents the non-convergence of high-dimensional unregularized inversions."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides ensemble-averaged surface Doppler measurements that set the surface velocity values fixed in the inversions."},{"cited_title":"2012, SoPh, 136, doi: 10.1007/s11207-012-0010-0","cited_arxiv_id":null,"evidence_quote":"Supplies the mass-conserving stream-function form for average supergranules and the shallow-peak depth estimate."}],"review_version":1}