{"id":"164dae20-d4c4-47c9-84b8-8367498b68b2","arxiv_id":"2411.12801","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Adding HII, H2, and clustering-based HI decomposition improves gas-based dust templates only slightly, with residuals unexplained by any single physical effect, and yields a new upper limit of sigma_T < 1.28 K on dust temperature variation.","lead":"This paper builds new Galactic dust extinction maps by combining neutral, ionized, and molecular hydrogen gas data, and tests whether adding these tracers and automatic cloud-finding reduces the scatter between gas and dust. It finds only modest improvement, and derives an upper limit of 1.28 K on temperature variations of high-latitude dust.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"In-sample model selection and mask pre-cleaning can bias the sigma_T < 1.28 K upper limit downward, undermining its status as a robust bound.","rationale":"The reader's weakest_assumption is the constant-emissivity-per-cloud assumption, and their rationale separately mentions in-sample optimization and mask pre-cleaning as reasons for CONDITIONAL. I agree that the constant-emissivity assumption is important and is explicitly acknowledged by the authors as a limitation, but it does not threaten the sigma_T limit in the direction of invalidity; if emissivity varies within clouds, residuals are inflated, making the upper limit more conservative. The in-sample model selection and residual-based masking are more load-bearing because they bias sigma_T downward, so the quoted upper limit may not be a true bound. The central negative result—that no single physical mechanism explains the bulk of residuals—is robust to this concern because it is a qualitative statement supported by multiple independent cross-checks (stellar extinction, polarization, CIB lensing). The new extinction map and template improvements are valuable and likely correct. Thus the CONDITIONAL verdict remains appropriate, with the conditionality now specifically tied to the robustness of the sigma_T limit. I recommend no change to the reader's verdict.","tokens_in":42041,"tokens_out":6528,"duration_ms":68419,"concrete_test":"Perform a holdout cross-validation. Split each of the NGC and SGC into two disjoint sky halves (e.g., by Galactic longitude). Optimize k and the attribute weights on half A, fit the linear emissivities on half A, then compute the residual map and sigma_T on half B; swap roles and repeat. If the out-of-sample sigma_T exceeds 1.28 K, the quoted upper limit is biased downward by in-sample selection. Additionally, rerun the analysis using only the NHI threshold and Galactic plane mask, omitting the Lenz et al. (2019) residual-based mask criteria, to quantify the pre-cleaning effect on sigma_T.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline upper limit on dust temperature variation, sigma_T < 1.28 K (Section 7.4), is computed from the residual map of the fiducial model via Equation (14). However, that residual map is in-sample in two compounding ways. First, the model complexity (number of clusters k and attribute weights wv, wrho) is selected by minimizing chi2 against the very same Planck 857 GHz data that later defines the residuals (Section 5.2, Figure 4). This is a direct overfitting channel: the optimizer will exploit any noise or systematics to shrink the residual variance, and hence shrink the inferred sigma_T. Second, the analysis mask (Section 2.5) inherits from Lenz et al. (2019) a criterion that removes pixels with residual emission from earlier HI fits, pre-cleaning the largest outliers before the model is even fit. Both effects can only reduce the in-sample residual scatter. The paper presents sigma_T as an upper limit, implying conservatism, but these choices push the estimate in the opposite direction: they make the limit artificially low. The constant-emissivity assumption (Section 3.1), which the reader emphasizes, would, if violated, tend to add variance and thus make the limit more conservative; the in-sample selection and mask pre-cleaning are the threats that can make the quoted limit non-conservative. Therefore, the specific claim that sigma_T < 1.28 K, lower than previous studies, is not secured by the presented analysis.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper constructs high-latitude Galactic dust emission and reddening templates from HI4PI velocity-resolved HI, an all-sky dispersion-measure-based HII map, and a CO-based H2 map. The HI is decomposed with a k-means algorithm in position-position-velocity space, and the resulting gas templates are linearly fitted to Planck 353/545/857 GHz data and to the SFD reddening map. The authors find that adding HII improves the fit only modestly, that the H2 template has negligible impact, that the clustering-based HI decomposition yields only modest improvements over simple velocity cuts, and that large-scale residuals at the tens-of-degrees scale remain at roughly the 20% level. They investigate dust-to-gas ratio, temperature, opacity, magnetic-field orientation, and CIB contamination as explanations, concluding that no single mechanism dominates. They also derive a bound on dust temperature variation of sigma_T < 1.28 K under the assumption of a constant line-of-sight temperature, and release the resulting reddening map and templates.","tokens_in":42400,"tokens_out":4695,"duration_ms":53825,"significance":"If the conclusions hold, the paper is a useful contribution: it provides a public gas-based reddening map, systematically quantifies the limited gains from multi-phase tracers and data-driven HI decomposition, and documents a series of honest negative results that are relevant for the many cosmological analyses that rely on Galactic extinction corrections. The external checks against DESI stellar reddening and CMB lensing cross-correlations are valuable and partially mitigate concerns about in-sample fitting. However, the headline upper limit on dust temperature variation is not secured by the analysis as presented, because the clustering hyperparameters are selected on the same 857 GHz residuals that define sigma_T and because the adopted mask removes pixels with large residuals before the fit.","major_comments":[{"comment":"The sigma_T < 1.28 K limit is computed from the residuals of a model whose hyperparameters were selected in-sample on the same Planck 857 GHz map. In Section 5.2 and Figure 4, the number of clusters k and the attribute weights (w_v, w_rho) are chosen by minimizing the chi^2 against Planck 857 GHz, and in Section 7.4 these same 857 GHz residuals are converted into a temperature map via Equation (14). Any overfitting or noise exploitation in the hyperparameter search will shrink the residual variance and hence bias sigma_T downward, which is the opposite of what an 'upper limit' should do. In addition, the mask described in Section 2.5 inherits from Lenz et al. (2019) a criterion that removes pixels with residual emission found in earlier HI fits, pre-cleaning the largest outliers before the model is even built. The manuscript acknowledges both choices, but neither is revisited when the temperature limit is presented in the abstract. To support the headline claim, the authors should repeat the analysis with cross-validation or an independent calibration region, and quantify how sigma_T changes when the residual-masking criterion is relaxed.","section":"Sections 5.2, 7.4, and 2.5"},{"comment":"Equation (14) is presented as an upper limit on dust temperature variation, but it is only a rescaling of the model residual map under a strong null hypothesis that all residuals are caused by temperature fluctuations with no variations in dust-to-gas ratio, opacity, unaccounted gas, CIB, or magnetic-field orientation. The paper itself finds evidence for several of these other contributions (Sections 7.2, 7.3, 7.6, 7.7), so the derived sigma_T is not an upper bound on the physical temperature dispersion of the interstellar dust; it is an upper bound on the temperature variation that would be inferred if the model were exact in every other respect. The abstract's statement that this limit is 'lower than the temperature variation reported in previous studies' therefore overstates the robustness of the result. The text should either explicitly frame sigma_T as a residual-based statistic under a stated model, or propagate the contributions of the other mechanisms to obtain a genuinely conservative bound.","section":"Section 7.4, Equations (13)-(14)"},{"comment":"The line-of-sight constant-temperature assumption is stated, but its consequences for the claimed upper limit deserve more emphasis. If a sight line contains multiple dust components at different temperatures, the same observed intensity can be produced with a larger temperature dispersion than Equation (14) would infer. The paper notes this in passing, but the abstract and conclusions present sigma_T < 1.28 K without this caveat. At minimum, the authors should state explicitly that the limit applies only to single-temperature lines of sight and does not constrain multi-component temperature structure.","section":"Section 7.4"}],"minor_comments":[{"comment":"The units of X_CO are written as 'cm−2 K−1 km−1 s', which appears to be missing a division by (km/s); the standard unit is cm^-2 (K km s^-1)^-1.","section":"Section 2.3"},{"comment":"The SGC HII emissivity at 545 GHz is listed as 0.0137 MJy sr^-1 (10^20 cm^-2)^-1, which is inconsistent with the 353 and 857 GHz values in the same row (0.040 and 0.374); this is likely a typographical error and should be checked against the actual fit output.","section":"Table 1"},{"comment":"The coefficients reported for the SGC are labeled with 'IVC−' and 'LVC', but the fiducial SGC templates defined in Sections 5.3 and 6 are 'LVC−' and 'LVC+'; the labels in the release description should be corrected.","section":"Section 8.2"},{"comment":"There is a duplicated word in the sentence 'it is difficult to explain the entirety of our fitting residuals residuals by dust-to-gas ratio variations'; the repetition should be removed.","section":"Section 7.2"},{"comment":"The top panel is labeled 'T Model [K]', but the map is not a measured temperature map; it is a temperature inferred under the assumption that all residuals are thermal. The caption should make this construction clear in the figure label as well as in the text.","section":"Figure 12"}],"recommendation":"major_revision","confidential_remarks":"The central negative results of the paper appear sound and are reported honestly, and the released data products are a useful community resource. The main concern is the sigma_T upper limit: as written, it is an in-sample residual statistic with a pre-cleaned mask, and the abstract presents it as a robust physical constraint. I believe this is fixable with cross-checks and reframing, but it is load-bearing enough that the paper should not be accepted in its current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a careful, honest paper with a mostly useful negative result: adding HII and H2 phases and clustering the HI spectral cube into discrete components improves HI-based dust emission fits only modestly, and no single physical mechanism explains the large-scale residuals that remain. The work is thorough on the data side, the released reddening map is a real public product, and the external checks (DESI reddening, CMB-lensing cross-correlation, frequency-dependent residual tests) are the right way to gain confidence in the qualitative conclusions. The claim that dust-to-gas ratio, temperature, opacity, magnetic field orientation, and CIB each contribute but none dominates is plausible and will be a useful reference for cosmological extinction corrections.\n\nThe soft spot is the headline sigma_T < 1.28 K upper limit. It is derived from the residuals of the fiducial model using Equation (14), but those residuals are in-sample in two compounding ways: the clustering hyperparameters (number of clusters, attribute weights) are chosen to minimize chi-squared against the very same 857 GHz Planck map that later defines the residual scatter, and the adopted mask inherits criteria from Lenz et al. (2019) that remove pixels with high residual emission from earlier HI fits. Both effects push the residual variance down, so sigma_T is not a conservative upper bound; if anything, it is biased low. The paper acknowledges the constant-temperature-along-the-line-of-sight and constant-emissivity-per-cloud assumptions, which would tend to make the limit conservative, but it does not acknowledge the in-sample selection or the mask pre-cleaning. I would ask the authors to reframe the limit as a model-dependent estimate and to validate it with a held-out validation split or an out-of-sample residual map. The rest of the analysis does not hinge on this specific number, so the main negative result survives.\n\nMinor notes: the HII template is degenerate with H2 and with the total dust column, as the authors themselves say, and the inferred HII emissivity is twice as large in the SGC as in the NGC, which they attribute to an excess in one region. That is honest but it does limit how much weight one can put on the multiphase interpretation. The clustering algorithm is a reasonable new application, and the paper does not oversell its benefit.\n\nBottom line: this paper deserves a serious referee. It is a solid, useful contribution to the galactic-dust-template literature, and the released map will be cited. I would recommend acceptance after a revision that addresses the in-sample nature of the temperature limit and perhaps adds cross-validation on the clustering step.","headline":"Honest negative-result study of gas-based dust templates; the new temperature-variation limit is intriguing but is the weakest link because it comes from in-sample residuals and a mask that pre-removes high-residual regions.","tokens_in":42922,"tokens_out":2304,"would_cite":true,"duration_ms":27130,"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":"Additional gas phases leave HI-based dust maps with 20 percent residuals and no single dominant cause.","keywords":["Galactic dust","dust extinction","neutral hydrogen","ionized gas","molecular gas","HI clustering","dust temperature","far-infrared emission"],"falsifier":"Fit the same multi-phase templates to the far-infrared maps while including the full pixel covariance matrix, including clustering of the cosmic infrared background, instead of diagonal noise; if the large-scale residuals collapse to the noise level, the inferred emissivity variations are an artefact of ignored correlations. Separately, measure dust temperatures along high-latitude sight lines with a method independent of single-temperature modified-blackbody fitting, such as resolved multi-band photometry of background stars, and check whether the spread exceeds 1.28 K; a larger spread would falsify the paper's upper limit under its stated assumptions.","tokens_in":1950,"feed_emoji":"🌌","tokens_out":4676,"duration_ms":100508,"temperature":0.7,"pith_summary":"Neutral hydrogen emission is a standard proxy for Galactic dust, but fits of HI templates to far-infrared dust maps leave large-scale residuals of roughly 20 percent over tens of degrees. This paper tests whether the remaining mismatch can be absorbed by adding ionized and molecular gas templates and by splitting HI into data-driven clouds. It finds only modest improvement, and it shows that no single physical mechanism—variations in dust-to-gas ratio, dust temperature, opacity law, magnetic-field orientation, or extragalactic background—accounts for most of the residual pattern. Under the assumption that dust temperature is constant along each line of sight, the residuals imply an upper limit of $\\sigma_T < 1.28\\,\\mathrm{K}$ on high-latitude temperature variation, which is tighter than earlier estimates. The result matters because gas-based extinction maps are used to correct cosmological survey data, and the paper's released reddening map offers an independent alternative to existing dust maps.","feed_headline":"HI dust maps keep 20 percent mismatch with extra gas phases","feed_subtitle":"Ionized and molecular gas plus cloud clustering shrink residuals only slightly; dust temperature variation is capped at 1.28 K.","key_machinery":"The load-bearing object is the linear template equation $I_\\nu = \\sum_i \\epsilon_{\\nu,i} N_{\\mathrm{HI},i} + \\epsilon_{\\nu,\\mathrm{HII}} N_{\\mathrm{HII}} + 2\\epsilon_{\\nu,\\mathrm{H}_2} N_{\\mathrm{H}_2} + b_\\nu$, fitted to the far-infrared maps by weighted least squares. Each HI cloud is assigned a constant emissivity per H atom, while the HII and H2 phases each receive a single emissivity. k-means clustering on normalized angular coordinates, velocity, and HI channel density supplies the HI templates, with attribute weights and cluster number chosen to minimize chi-squared at 857 GHz. The machinery isolates what the gas templates can and cannot explain, turning leftover structure into a map of emissivity variation that the paper then compares against external tracers of dust-to-gas ratio, temperature, polarization, and extragalactic background.","core_discovery":"The paper's central claim is that a linear template model in which each gas phase (HI, HII, H2) has a constant dust emissivity per hydrogen atom, with HI decomposed into a handful of clouds by k-means clustering in position-position-velocity space, still leaves residuals of less than 20 percent of the observed intensity in the 353, 545, and 857 GHz bands. Adding the HII template reduces outlier sight lines but only modestly; the H2 template built from CO is noise-dominated at high latitude and does not help. The clustering finds that a simple velocity cut already captures most of the signal, since three clusters in the north and five in the south mostly share similar emissivities, so the improvement from clustering is also modest. The residuals are coherent across frequencies and align only partially with stellar reddening residuals, dust-to-gas ratio maps, polarization-fraction fluctuations, and CMB lensing. The paper concludes that spatially varying dust emissivity not captured by discrete clouds is real, and that under a constant line-of-sight temperature assumption the implied temperature scatter is at most $\\sigma_T < 1.28\\,\\mathrm{K}$.","pith_inferences":["If the constant-emissivity-per-cloud assumption is the real bottleneck, then three-dimensional extinction maps that place dust along the line of sight could break the degeneracy and produce the larger improvement that additional gas phases did not.","The tight 1.28 K bound should be read as much as a statement about model misspecification as about temperature; an independent temperature tracer, such as near-infrared colors of many background stars, could settle which interpretation is correct.","The pattern of residuals points to dust-to-gas ratio variations that are real but spatially patchy, especially toward known low-metallicity gas like the Magellanic Stream, so targeted studies of those regions could quantify the effect.","A natural next test is to fit the same template set to future high-latitude CO surveys or to CIB-subtracted versions of the SFD map; if residuals shrink, the missing piece is gas phase rather than dust physics."],"forward_implications":["Gas-based extinction and emission templates can be pushed only modestly by adding HII and clustering; users should expect roughly 10 to 20 percent large-scale residuals from any such template in the diffuse high-latitude regime.","Current CO-based H2 templates are not sensitive enough to improve high-latitude dust fits; higher-sensitivity CO mapping is needed before molecular gas can be incorporated into gas-based extinction maps.","If the constant-line-of-sight-temperature assumption holds, high-latitude dust temperature variations are below 1.28 K, implying that earlier temperature maps with roughly 1.5 to 2.5 K variation are dominated by systematics or parameter degeneracy.","No single mechanism explains the residual pattern, so future improvements require joint multi-tracer analyses of gas, stellar extinction, polarization, and three-dimensional distance information rather than another single tracer.","The released reddening map, built from multi-phase templates fitted to the SFD map, provides an independent extinction template with residual scatter of order 3 to 3.5 millimag against stellar reddening."],"supporting_citations":[{"why":"Establishes the close coupling between HI column density and dust emission that motivates gas-based dust templates.","marker":"Boulanger et al. (1996)"},{"why":"Demonstrates the strong linear HI-dust correlation at low column densities and sets the NHI threshold used for the high-latitude mask.","marker":"Planck Collaboration XI (2014)"},{"why":"Provides the earlier single-HI-template dust model whose residuals serve as the baseline for this paper's comparison.","marker":"Lenz et al. (2017)"},{"why":"Supplies the all-sky HI spectral data cube from which the clustering templates are built.","marker":"HI4PI Collaboration et al. (2016)"},{"why":"Provides the 41-channel velocity-binned HI maps and the optically thin conversion factor used in the fits.","marker":"Clark & Hensley (2019)"},{"why":"Supplies the all-sky dispersion-measure map that serves as the HII column density template.","marker":"Hutschenreuter et al. (2024)"},{"why":"Supplies the reprocessed full-sky CO map used to construct the H2 template.","marker":"Ghosh et al. (2024)"},{"why":"Provides the Planck PR3 maps whose 353, 545, and 857 GHz bands are fitted.","marker":"Planck Collaboration III (2020)"},{"why":"Supplies the CMB-subtracted Planck maps and the CIB map used for residual cross-checks.","marker":"Lenz et al. (2019)"},{"why":"Provides the DESI stellar reddening map used to test whether dust-to-gas ratio variations explain the residuals.","marker":"Zhou et al. (2024)"}],"fun_headline_variants":["Even with HII and H2, HI dust maps keep 20% residual","Adding gas phases barely shrinks HI dust map residuals","Clustering gas clouds won't fix HI dust map residuals","Dust temperature scatter capped at 1.28 K in HI maps","More gas tracers, same HI dust map 20% gap"],"cache_read_input_tokens":44928,"weakest_assumption_plain":"The model assumes that each HI cloud has one constant dust emissivity per hydrogen atom across its entire angular extent.","fun_headline_variants_meta":{"raw":{"variants":["Even with HII and H2, HI dust maps keep 20% residual","Adding gas phases barely shrinks HI dust map residuals","Clustering gas clouds won't fix HI dust map residuals","Dust temperature scatter capped at 1.28 K in HI maps","More gas tracers, same HI dust map 20% gap"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000359,"raw_usage":{"total_tokens":2005,"prompt_tokens":1069,"completion_tokens":936,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":685,"completion_tokens_details":{"reasoning_tokens":846}},"tokens_in":685,"tokens_out":936,"duration_ms":9724,"temperature":1.0,"reasoning_tokens":846,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T17:11:07.904498+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fit the same multi-phase templates to the far-infrared maps while including the full pixel covariance matrix, including clustering of the cosmic infrared background, instead of diagonal noise; if the large-scale residuals collapse to the noise level, the inferred emissivity variations are an artefact of ignored correlations. Separately, measure dust temperatures along high-latitude sight lines with a method independent of single-temperature modified-blackbody fitting, such as resolved multi-band photometry of background stars, and check whether the spread exceeds 1.28 K; a larger spread would falsify the paper's upper limit under its stated assumptions.","supporting_citations":[],"review_version":1}