{"id":"369040b8-5c90-4181-9c40-f21db6abb3aa","arxiv_id":"1909.02604","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A Levenberg-Marquardt based global inversion method that applies per-dataset linear degradation operators to jointly invert solar spectra at different spatial resolutions.","lead":"This paper introduces a way to combine solar observations from instruments with very different spatial sharpness into a single atmospheric model. A generalist would care because multi-telescope data can now be used without forcing everything to the lowest common resolution.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The recovery claim is contingent on exactly known degradation operators, and no test quantifies the effect of PSF misestimation, the main real-data obstacle.","rationale":"The reader's weakest assumption—that all instrumental differences are known, space-invariant linear operators—is indeed the most load-bearing point for the central claim. The paper's strongest evidence is the near-unity continuum contrast recovery in Experiment 3, but that result is achieved under idealized conditions where the degradation operator used to generate the data is exactly the one assumed by the inversion. The real-data motivation (CHROMIS/IRIS) involves uncertain, time-varying PSFs and temporal mismatches, both flagged in Section 5 but not tested. Without a sensitivity analysis to operator errors, the demonstrated recovery does not establish robustness for the intended applications. The reader's CONDITIONAL verdict is appropriate: the method is mathematically sound and the proof-of-concept is convincing, but the practical claim is conditional on accurate operator knowledge and co-temporality. No new concern was identified beyond the reader's; therefore the verdict remains unchanged.","tokens_in":17620,"tokens_out":14882,"duration_ms":164997,"concrete_test":"Re-run Experiment 3 but with the inversion assuming a PSF whose FWHM is 10% larger than the PSF used to generate the data (other operators unchanged). Measure the recovered continuum RMS contrast and the slope of the B_parallel scatter in the upper 2/3 of the FOV. If the contrast deviates by more than about one percentage point from 14.9% or the scatter slope deviates by more than 10% from unity, the method is sensitive to PSF misestimation, and the central claim is not robust to the primary uncertainty in real multi-resolution datasets. Optionally, repeat with a 5% FWHM error to map the sensitivity growth.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that multi-resolution inversion recovers the original continuum contrast almost entirely and all derived quantities follow the one-to-one line—is established in Experiment 3 using synthetic data degraded by an operator that is exactly known to the inversion (the same PSF and rebinning kernels, Sections 3.2–3.3). This proof-of-concept cannot distinguish the method's intrinsic performance from its reliance on exact operator knowledge. In real applications, such as ground-based CHROMIS with residual seeing after MOMFBD combined with IRIS, the effective PSF is not known to this precision and varies in time, as Section 5 acknowledges. If the assumed degradation operator is mismatched, the inverse problem is misspecified: the inversion will attempt to absorb the discrepancy into the retrieved atmospheric parameters, biasing the recovered contrasts and fields. No sensitivity analysis quantifies this effect. The demonstrated 14.8% versus 14.9% contrast recovery is thus contingent on the idealized condition of a perfectly known, space-invariant linear degradation chain, which is not available for the motivating datasets. The co-temporal assumption is likewise unrelaxed, though explicitly acknowledged.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a generalization of van Noort's spatially-coupled Levenberg-Marquardt inversion to datasets acquired at different spatial resolutions and on different spatial grids. The key idea is to represent all instrumental degradation steps—telescope PSF, rebinning/resampling, interpolation, shifts, and rotation—as a chain of linear operators applied to the synthetic spectra before comparison with the observations, with the merit function written as a sum over windows that share a common model grid. Spatial Tikhonov regularization is added, and the resulting sparse Hessian system is solved with BiCGStab. The method is tested in three numerical experiments using ME inversions of STiC LTE synthetic spectra computed from a Rempel (2012) rMHD snapshot: a 1D inversion of only the high-resolution Stokes I dataset, a 1D inversion of the high-resolution I plus a low-resolution full-Stokes dataset naively resampled to the high-resolution grid, and the proposed multi-resolution global inversion. The global inversion recovers a continuum intensity RMS contrast of 14.8% against a 14.9% reference and shows one-to-one scatter for the magnetic field, velocity, and intensity, while the 1D benchmarks show degraded contrast and amplitude. Limitations regarding co-temporality, time-varying residual seeing, and the assumption of known degradation operators are acknowledged in Section 5.","tokens_in":17841,"tokens_out":6300,"duration_ms":73303,"significance":"If the method is robust beyond the idealized tests, it addresses a real and timely need: combining high-resolution ground-based data (CHROMIS, CRISP, and future DKIST/EST) with lower-resolution space-borne data (IRIS, Hinode) in a single self-consistent inversion. The linear-operator formalism is clean and plausibly generalizes the van Noort framework, and the test design is non-trivial because the forward model (STiC, LTE) differs from the inversion model (Milne-Eddington), so the good recovery is not a trivial self-inversion. The explicit quantitative contrast recovery (14.8% vs 14.9%) is a strong point, as is the paper's candid listing of limitations. However, the central demonstration depends on the exact knowledge of the degradation operators and on a hand-tuned regularization weight; without sensitivity analysis to PSF misestimation and an objective regularization-selection method, the practical claims are not yet fully supported. The paper is a solid proof-of-concept that, with the requested additions, would be a valuable contribution to the multi-resolution inversion literature.","major_comments":[{"comment":"The successful recovery in Experiment 3 presumes that the degradation operator used in the inversion is identical to the one that generated the synthetic data. Section 5 acknowledges that ground-based data are affected by residual, time-varying seeing after MOMFBD, which is not known to the precision assumed here. This is load-bearing because the method's advantage over simple resampling is precisely that it corrects a known linear operator; an incorrectly assumed PSF (or a misregistered shift/rotation) will be absorbed into the retrieved atmospheric parameters and bias the recovered contrasts and fields. Please add a numerical experiment that perturbs the assumed operator (e.g., a PSF with a different FWHM, a small unknown image shift, or a residual seeing term), and quantify the resulting bias in the recovered continuum intensity contrast and magnetic-field parameters.","section":"Sections 3.2–3.3 and 5"},{"comment":"The two-stage regularization schedule uses a 'largely overestimated (x50)' weight in the first cycle and then 'a final value' chosen because it 'allowed to converge the problem without showing artifacts from the deconvolution.' This is a subjective, hand-tuned free parameter. Since the central quantitative result (14.8% vs 14.9% contrast) and the one-to-one scatter plots may depend on this choice, the paper should either provide an objective recipe for selecting the regularization weight (e.g., L-curve or cross-validation) or demonstrate explicitly that the recovery is stable over a reasonable range of weights.","section":"Section 3.3 and Section 4.1"},{"comment":"The claim that 'all derived quantities follow the one-to-one line' is supported only by qualitative density scatter plots and a single scalar RMS contrast value. No slopes, intercepts, correlation coefficients, or RMS differences are reported for B_parallel, |B_perp|, the azimuth, or v_los. Because the stated conclusion is that the amplitudes and contrast of all quantities are correct, please provide quantitative agreement metrics for each observable, and include uncertainties on the retrieved parameters (at least for the main experiment) so the reader can evaluate the precision of the recovery.","section":"Section 4, Figures 11 and 12"}],"minor_comments":[{"comment":"The abstract contains a doubled article: 'for the the effects' should read 'for the effects.' In Section 2.3.1, 'the subindexes ji spam from' should read 'span,' and the heading of Section 2.3.2 spells 'Rebining' instead of 'Rebinning.'","section":"Abstract and Section 2.3.1"},{"comment":"The text and the figure caption use 'Azimutal' instead of 'Azimuthal' in the description of the power-spectrum integration.","section":"Section 4.2 and Figure 14"},{"comment":"The caption contains the typo 'mutli-resolution' instead of 'multi-resolution.'","section":"Figure 13 caption"},{"comment":"The normalization factor Nn in Eq. (7) is ambiguous: it appears both inside and outside the window sums, and it is not stated whether Nn is the total number of data points or the number of windows. Please define it explicitly.","section":"Equation (7)"},{"comment":"The description of the two-stage regularization would benefit from giving the actual numerical values of the regularization weight used in the first and second cycles, rather than describing them only as 'largely overestimated' and 'correct value,' to aid reproducibility.","section":"Section 3.3 and Table 1"}],"recommendation":"major_revision","confidential_remarks":"This is a useful methods paper with a strong proof-of-concept, but the missing sensitivity to PSF misestimation is the main obstacle to accepting it as a general method for real multi-instrument data. The hand-tuned regularization is a secondary concern. I do not see circularity or a fundamental flaw in the linear-algebra framework, so revision rather than rejection is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on the de la Cruz Rodríguez paper. The key thing to know: this is a genuine, useful extension of van Noort's global inversion, and the proof-of-concept is solid. The main caveat, which the stress-test note correctly identifies, is that the experiments assume the degradation operators are known exactly. That's a real limitation for ground-based data, but it's not a reason to dismiss the method.\n\nWhat's new: the paper generalizes van Noort (2012) from a single PSF on a common grid to arbitrary chains of linear operators—PSF convolution, rebinning, interpolation, rotation—and decouples the model grid from the observation grids. That's a real step forward. Adding spatial Tikhonov regularization is useful, and the two-cycle approach (heavy regularization first, then light) is a practical way to avoid local minima. The numerical experiments are well designed: the forward model is STiC with LTE, the inversion is Milne-Eddington, so the test is not circular. The comparison with two 1D experiments is illuminating: naive resampling of the low-res dataset into a common grid degrades the continuum contrast to 9.2%, while the multi-resolution method recovers 14.8% against the reference 14.9%. The power-spectrum analysis in §4.2 is a nice addition, showing that the high-res Stokes I dataset constrains small-scale magnetic fields when Zeeman splitting is detectable.\n\nThe soft spots are in proportion. The stress-test concern about known degradation operators is the main one. The paper does not test what happens when the PSF is misestimated, which is the typical situation for ground-based CHROMIS data after MOMFBD. The author acknowledges this in §5 but doesn't quantify it. Also, the regularization weight is hand-tuned in two stages; no automated selection is given. The proof-of-concept code is not released, which limits reproducibility. No error bars on retrieved parameters are reported. These are all minor-to-moderate for a method paper; they don't undermine the central claim, but a careful referee should ask for a sensitivity test on PSF mismatch and ideally a code release.\n\nWho would benefit: anyone doing multi-instrument inversions, especially chromospheric studies combining IRIS with ground-based telescopes. I'd cite it if I worked in that area. The writing is clear, the math is straightforward, and the figures support the conclusions.\n\nMy recommendation: definitely send it to peer review. The method deserves serious referee attention, with the expectation of revisions to address the PSF-mismatch question and reproducibility. It's a solid contribution, not a finished product.","headline":"Genuinely useful extension of van Noort's global inversion to multi-resolution data with a clean proof-of-concept; the main caveat is that real-data PSF misestimation is not tested.","tokens_in":18337,"tokens_out":2785,"would_cite":true,"duration_ms":28472,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that spectropolarimetric data at different spatial resolutions can be inverted jointly by applying dataset-specific linear degradation operators to synthetic spectra, and shows numerically that the recovered continuum…","keywords":["high angular resolution","radiative transfer","polarization","Sun: magnetic fields","Sun: chromosphere","multi-resolution inversion","point-spread function","Levenberg-Marquardt"],"falsifier":"Run the paper's Experiment 3 with a PSF width deliberately wrong by 10%, or with a small artificial time offset between the two spectral lines, and check whether the retrieved B||, v_los, and continuum-contrast scatter plots still follow the one-to-one line; any systematic bias away from that line would show that the assumed known-operator premise is essential.","tokens_in":17442,"feed_emoji":"🔭","tokens_out":7987,"duration_ms":79894,"temperature":0.7,"pith_summary":"The paper is trying to establish that observations taken at different spatial resolutions do not have to be resampled to a common grid before inversion; instead, each dataset can constrain the spatial scales it actually contains. It generalizes the Levenberg-Marquardt inversion algorithm by inserting chains of linear operators—telescope PSF convolution, rebinning, shifts, rotations—that degrade the synthetic spectra of a finely gridded model atmosphere to match each observation. In numerical experiments with two simulated Fe I lines observed through different apertures and sampled on different grids, the joint inversion recovers the original granulation contrast almost exactly (14.8% RMS versus 14.9% in the reference) and keeps all derived quantities on the one-to-one line. If correct, this makes it feasible to combine ground-based and space-borne chromospheric diagnostics without sacrificing the resolution of the best dataset.","feed_headline":"One joint inversion recovers full solar contrast from mixed-resolution data","feed_subtitle":"High- and low-resolution spectra inverted together recover the true contrast, 14.8% versus 14.9%.","key_machinery":"The central object is a chain of linear degradation operators in sparse matrix form, D = D1D2...Dn, applied to the synthetic spectra of a model computed on a fine grid before comparison with each observational dataset. The merit function becomes a sum of per-dataset terms, the Hessian becomes a sum of sparse spatially coupled Hessians weighted by the PSF autocorrelation, and the correction step is solved with an iterative sparse linear solver; spatial Tikhonov regularization adds narrow bands to the Hessian and is used in a two-cycle scheme to stabilize convergence.","core_discovery":"The central claim is that a single model atmosphere can be fitted simultaneously to multiple observations with different spatial resolutions, provided each forward prediction is passed through a known linear degradation operator specific to that dataset. The paper demonstrates this with a Milne-Eddington-based proof of concept: a high-resolution Stokes I dataset in Fe I 6301 Å is combined with a lower-resolution full-Stokes dataset in Fe I 6302 Å, each degraded by a different telescope PSF and resampling, and the joint inversion recovers the continuum intensity contrast at 14.8% versus 14.9% in the reference model. Scatter plots of the longitudinal and transverse magnetic field, azimuth, line-of-sight velocity, and continuum intensity all follow the one-to-one line, indicating that amplitudes and contrasts are preserved rather than blurred away. Each dataset constrains only the spatial scales present in it, so the magnetic field is limited by the lower-resolution polarimetric data while the intensity and velocity maps retain the higher resolution.","pith_inferences":["If the method is applied to real ground-based data, residual seeing not captured by a static PSF will be absorbed into the retrieved model; a testable prediction is that retrieved Doppler widths and field strengths will be biased in exactly the spatial regions where the seeing PSF is worst.","The same operator-chain formalism could absorb temporal misalignment by inserting time-interpolation operators between datasets, extending the method beyond the strictly co-temporal assumption the paper states as a limitation.","The power-spectrum comparison suggests that weak, small-scale magnetic fields remain the hardest part to recover; adding a third full-Stokes dataset at the highest available resolution should push the recovered frequency band higher, an experiment the paper's setup could run directly."],"forward_implications":["A joint multi-resolution inversion keeps each dataset's native resolution: the high-resolution intensity dataset constrains velocity and temperature at fine scales while the low-resolution polarimetric dataset constrains the magnetic field at its own scales.","Because the model grid is decoupled from the observation grids, the model can be sampled denser than the data, reducing pixel discretization and high-frequency damping near the Nyquist cutoff.","The two-cycle regularization strategy, starting with an overestimated smoothness weight and then reducing it, lets the global inversion converge in fewer than about ten iterations per cycle.","Any degradation that can be written as a linear operator—PSF convolution, rebinning, shifts, rotations—can be chained and handled in a single inversion, so the method applies broadly to multi-facility data.","The method opens a practical path for combining ground-based and space-borne chromospheric diagnostics, which is increasingly relevant as the resolution gap grows with future large-aperture telescopes."],"supporting_citations":[{"why":"Introduces the spatially coupled PSF-based inversion that this method generalizes to multiple spatial resolutions.","marker":"van Noort (2012)"},{"why":"Provides the Levenberg-Marquardt with l2 regularization derivation and the notation used throughout the paper.","marker":"de la Cruz Rodríguez et al. (2019)"},{"why":"Supplies the 3D rMHD active-region simulation snapshot used as the ground-truth atmospheric model.","marker":"Rempel (2012)"},{"why":"Provides the MURAM code that produced the simulation snapshot used in the numerical experiments.","marker":"Vögler et al. (2005)"},{"why":"Supplies the circular-aperture telescope PSF model, with central obscuration, used to degrade the synthetic data.","marker":"Danilovic et al. (2008)"},{"why":"Defines the Milne-Eddington model atmosphere on which the proof-of-concept inversion code is based.","marker":"Auer et al. (1977)"},{"why":"Gives the analytical derivatives of intensity with respect to Milne-Eddington parameters, used to build the Jacobian.","marker":"Orozco Suárez & Del Toro Iniesta (2007)"},{"why":"Origin of the spatial smoothness regularization that the paper adds to stabilise the global inversion.","marker":"Tikhonov & Arsenin (1977)"},{"why":"Provides the sparse iterative linear solver used to compute the Levenberg-Marquardt correction step.","marker":"Guennebaud et al. (2010)"}],"fun_headline_variants":["Joint inversion recovers solar contrast across resolutions","Multi-res solar data fused in one coherent inversion","One inversion, many resolutions: solar spectra unified","Mixed-resolution solar data inverted as a single model","Solar data from different telescopes, one joint inversion"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that every difference between the datasets can be represented by a known, space-invariant linear operator applied to the synthetic spectra, and that the datasets are strictly co-temporal; if the real degradation, including time-varying seeing residuals, or the alignment differs from that operator, the inversion will fold the mismatch into the retrieved atmospheric quantities.","fun_headline_variants_meta":{"raw":{"variants":["Joint inversion recovers solar contrast across resolutions","Multi-res solar data fused in one coherent inversion","One inversion, many resolutions: solar spectra unified","Mixed-resolution solar data inverted as a single model","Solar data from different telescopes, one joint inversion"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001338,"raw_usage":{"total_tokens":5459,"prompt_tokens":981,"completion_tokens":4478,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":4408}},"tokens_in":597,"tokens_out":4478,"duration_ms":31078,"temperature":1.0,"reasoning_tokens":4408,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:45:18.319048+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the paper's Experiment 3 with a PSF width deliberately wrong by 10%, or with a small artificial time offset between the two spectral lines, and check whether the retrieved B||, v_los, and continuum-contrast scatter plots still follow the one-to-one line; any systematic bias away from that line would show that the assumed known-operator premise is essential.","supporting_citations":[{"cited_title":"2012, ApJ, 750, 62 Riethmüller, T","cited_arxiv_id":null,"evidence_quote":"Supplies the 3D rMHD active-region simulation snapshot used as the ground-truth atmospheric model."},{"cited_title":"2008, A&A, 484, L17","cited_arxiv_id":null,"evidence_quote":"Supplies the circular-aperture telescope PSF model, with central obscuration, used to degrade the synthetic data."},{"cited_title":"H., Heasley, J","cited_arxiv_id":null,"evidence_quote":"Defines the Milne-Eddington model atmosphere on which the proof-of-concept inversion code is based."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Origin of the spatial smoothness regularization that the paper adds to stabilise the global inversion."}],"review_version":1}