{"id":"930f6030-d96a-4a98-9dea-ca1054152366","arxiv_id":"2501.11910","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Stacking velocity-misaligned HI spectra makes secondary components broader and shallower, but spin temperatures remain unbiased; stacked GASKAP data reveal a 1320 K unstable neutral medium component and spatially stable cold gas temperatures over about 100 pc.","lead":"This paper stacks 462 Milky Way hydrogen absorption and emission spectra and 2240 non-detection spectra from the GASKAP survey, then uses toy models to show how velocity misalignment biases measurements made on stacked spectra. It reports a warm unstable-medium component at 1320 K in absorption and shows that cold gas optical depth declines with distance from known cold-gas locations while spin temperature stays stable.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The load-bearing step is the claim that the emission peak traces the dominant cold absorber in all 2240 non-detection sightlines; this is validated only on the 462 detections, and if a warm peak misaligns the primary component the unbiased spin-temperature result and the 100-pc stability…","rationale":"The paper's methodological conclusion about secondary components (lower-limit optical depths and well-correlated spin temperatures) is well supported: the lower-limit statement follows from averaging non-negative shifted profiles, and the toy models in Appendix A explicitly build in equal emission-absorption offsets. The weakest point is not this internal argument but its application to the non-detection sample, where the alignment is an untested prior. The reader's conditional verdict already captures this. A failed concrete test would degrade the headline observational claims, but would not overturn the paper's central stacking considerations, and the paper openly notes some limitations (e.g., bin-1 non-detection, single emission-only component bias). Therefore I do not move the verdict; UNCHANGED is appropriate.","tokens_in":24004,"tokens_out":11173,"duration_ms":135523,"concrete_test":"Perform an emission-only multi-Gaussian decomposition of the 2240 non-detection spectra (e.g., with ROHSA or a fixed three-component fit) and measure, for each sightline, the velocity offset between the global TB peak and the centroid of the narrowest (coldest) component. If more than ~10% of sightlines have offsets >2 km/s, re-run the total and spatially binned non-detection stacks after shifting by the narrow-component centroid instead of the global peak. Compare the resulting primary component's peak optical depth and Ts with Table 2 and Fig. 14. A significant increase in tau_peak or a change in Ts beyond the quoted 1-sigma would show that the emission-peak alignment assumption is not valid for the non-detection sample.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.1 justifies emission-peak shifting by stating that 'the peak in emission typically traces cold gas' and by citing 1-2 km/s offsets seen in GASKAP detections (Nguyen et al. 2024). That validation uses spectra with detected absorption, where the cold component can be identified. The 2240 non-detection spectra have no such confirmation, yet all of Section 5 (total stack, spatial bins, and the ~100 pc temperature-stability claim) assumes the same alignment. If a non-detection sightline's brightness-temperature peak is produced by warm or unstable gas, or by a blend of components, the 'primary' component is not centred at zero; the cold absorber is partially assigned to the broad secondary component. The argument that velocity offsets cancel in Ts assumes the offset is identical in emission and absorption, which is true for a single isothermal component but not when the selected emission peak is a different physical phase than the dominant absorbing gas. Section 3.4's comparison with individual sightline properties covers only the 462 detections and therefore cannot certify the non-detection sample. The central observational result, Ts=98±12 K stable across bins, would then be an artifact of the alignment rule rather than a property of the CNM.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies biases in stacking HI absorption-emission spectra, using toy models and 2702 GASKAP Phase II Pilot sightlines toward the Magellanic system. The authors argue that shifting spectra by the emission brightness-temperature peak correctly aligns the dominant 'primary' cold component, while offset 'secondary' components combine into a broader, shallower Gaussian whose peak optical depth is a lower limit and whose FWHM is an upper limit, but whose spin temperature remains a good estimator of the average because the offset affects emission and absorption equally. They apply this to stacking 462 detected absorption sightlines after subtracting Nguyen et al. (2024) Gaussian models, detecting a broad residual component with Ts = 1320 +/- 263 K, and to stacking 2240 non-detection sightlines, finding primary and secondary components with Ts = 98 +/- 12 K and 255 +/- 106 K. Spatial binning of the non-detections shows an increase in primary optical depth toward regions of high detection density and a primary spin temperature consistent with no spatial variation over roughly 100 pc.","tokens_in":24245,"tokens_out":9007,"duration_ms":93310,"significance":"The paper's main methodological point is useful and largely correct: flux conservation makes the stacked peak optical depth of velocity-offset secondary components a lower limit, while the Ts estimate is more robust because the velocity offset enters emission and absorption together. The toy model in Appendix A, the comparison with individual sightline properties in Section 3.4, and the bootstrap and half-inverted controls in Sections 4.2 and 5.2 are carefully done and go beyond what is common in stacking papers. If the observational results stand, the detection of an average UNM component at 1320 K in residual spectra and the apparent stability of CNM Ts across roughly 100 pc are interesting additions to the GASKAP pilot analysis. However, the strongest observational claims rest on assumptions about the emission-peak alignment of non-detection sightlines and on Gaussian decomposition systematics, both of which need stronger support.","major_comments":[{"comment":"The alignment rule, shifting by the emission peak so that the 'primary' component sits at 0 km/s, is established on the detection sample, where the cold absorber can be identified in absorption, but it is applied to the 2240 non-detection sightlines without independent verification. If in even a fraction of those sightlines the emission peak is produced by warm or blended gas rather than by the dominant cold absorber, the primary component is not centred at zero and the cold gas is partly redistributed into the broad secondary component, so the quoted Ts values in Section 5.2 and Table 2 and the ~100 pc stability claim in Section 6.2 would be affected. I request a validation on the non-detection spectra themselves, for example by generating synthetic absorption spectra from the non-detection emission profiles with assumed spin temperatures and checking whether the emission-peak-aligned stack recovers the injected primary Ts and tau.","section":"Section 3.1 / Section 5.1"},{"comment":"The spin temperatures are derived from a decomposition in which a single broad component is added to the emission to complete the fit, and the manuscript explicitly notes that this choice can under-represent the spin temperature of the absorption feature. The 1320 +/- 263 K UNM result in Section 4.2 and the 98/255 K CNM values in Section 5.2 therefore carry a systematic error from the number and shape of the emission-only components that is not included in the bootstrap uncertainties, which only resample sightlines. Please quantify this systematic by varying the decomposition configuration, for example the number of emission-only Gaussians and the allowed parameter ranges, and report the resulting range of Ts for each component.","section":"Section 3.3 / Section 4.2"},{"comment":"The residual stack is formed by subtracting Nguyen et al. (2024) Gaussian fits from individual detection spectra. Errors in the fitted velocities, widths, or amplitudes of those cold components enter the residual spectra coherently, so the half-inverted bootstrap in Section 4.2, which tests only sign-symmetric random noise, cannot exclude a spurious broad absorption feature produced by systematic model-subtraction residuals. Please propagate the published fit uncertainties through the subtraction and re-stacking procedure, or otherwise demonstrate that the 1320 K component is not produced by residual cold-gas signal.","section":"Section 4.1 / Section 4.2"},{"comment":"The claim that the spin temperature 'remains stable in both components' over ~100 pc is not supported by Table 2 and the caption of Fig. 12: a secondary component is detected in only two of the six spatial bins (bins 2 and 4), and bin 4 has Ts = 135 +/- 215 K. The spatial stability claim is therefore established only for the primary component; the abstract and summary should be revised to state this limitation, unless additional bins with detected secondary components can be provided.","section":"Abstract / Section 5.3.1 / Table 2"}],"minor_comments":[{"comment":"The statement that 'the peak in emission typically traces cold gas' is an assertion that should be supported by a quantitative check within the detection sample, for example the fraction of sightlines where the nearest absorption component is within half a FWHM of the emission peak.","section":"Section 3.1"},{"comment":"The weighting is described as 'weights tau_res / sigma_tau^2' and then said to simplify to 1/sigma_tau; please write the actual weight used for the stacks explicitly and define tau_res in that context, since the same symbol is used later for model-subtracted residuals.","section":"Section 3.2"},{"comment":"The phrase 'allowing for a +/-1 km/s deviation in their central velocities and 10 per cent deviation in their widths' should clarify whether this allowance is a fitting constraint or an acceptance criterion for the Gaussian decomposition.","section":"Section 3.3"},{"comment":"The notation '>20 +/- 10' for the secondary optical depth lower limits is confusing as printed; a footnote explaining that the quoted values are the observed stacked values and the inequality indicates the relation to the true average would help.","section":"Table 2"},{"comment":"The conversion n_H = C x A'_ZGR23 with C = 1653 cm^-3 needs the units of A'_ZGR23 stated and a reference to O'Neill et al. (2024) for the conversion constant; as written the units on the two sides do not visibly match.","section":"Section 6.2 / Eq. (1)"},{"comment":"The vertical dotted lines representing stacked values are not labelled in the legend; please add labels or a caption note identifying which colour and dash style corresponds to the primary and secondary stack values.","section":"Fig. 5"}],"recommendation":"major_revision","confidential_remarks":"This is a solid and readable paper with a sound core toy-model result, but the observational claims in the abstract outrun the evidence in three places: alignment validation for non-detection sightlines, decomposition systematics in the spin-temperature estimates, and the claimed spatial stability of the secondary component. These concerns are fixable within the manuscript's scope by adding validation experiments and revising the claims, so I recommend major revision rather than rejection. No concerns about novelty or author conduct."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core methodological result here is real and useful: when you stack HI absorption-emission pairs by shifting to the emission peak, the peak optical depth and brightness temperature of secondary components are biased low, but the spin temperature stays well correlated with the true average because the velocity offset hits emission and absorption equally. The toy model in Appendix A is simple and convincing, and the comparison of the detection stack against individual sightline properties in Section 3.4 is a solid validation. The bootstrapping and half-inverted controls are careful, and the paper is honest about what is a lower limit and what is not. The UNM detection at 1320 K in the residual stack is plausible and well supported by the controls.\n\nThe main soft spot is the alignment assumption for the non-detection sample. The paper justifies shifting by the emission peak by saying it typically traces cold gas, citing 1-2 km/s offsets seen in the detection sample. But the 2240 non-detections have no absorption detections to confirm that the emission peak really is the dominant cold absorber. If, on some sightlines, the peak is actually a warm component or a blend, then the primary component in the stack is not centred correctly, and the cold gas leaks into the broad secondary component. In that case the headline result—Ts = 98 ± 12 K stable over ~100 pc—could be an artifact of the alignment rule rather than a real property of the CNM. I think the paper would be strengthened by an explicit test of this, e.g. using the detection-sample emission profiles to estimate how often the peak actually corresponds to the absorption component, and to quantify the effect when it doesn't. That said, the dust density trends and the fact that the primary optical depth increases toward known cold gas sightlines give some indirect support, so I would call this a caveat rather than a fatal flaw.\n\nMinor issues: the model subtraction relies on Nguyen et al. (2024) Gaussian fits, which could propagate fitting errors into the residual stack, and no code is shipped, with data only on request. Neither is disqualifying.\n\nThis paper is for anyone who stacks HI spectra, Galactic or extragalactic; it gives a clear warning about interpreting secondary components. It deserves a serious referee. I would recommend peer review, with a request for the non-detection alignment test and a somewhat deeper discussion of the assumptions.","headline":"A careful stacking analysis that convincingly shows spin temperatures survive velocity-offset biases, with an untested but plausible alignment assumption for non-detection sightlines.","tokens_in":24877,"tokens_out":2410,"would_cite":true,"duration_ms":27635,"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":"Stacking HI spectral pairs keeps spin temperatures true while making off-centre components' peak optical depths strict lower limits — a bias used to detect 1320 K unstable gas and stable cold gas over about 100 pc.","keywords":["ISM: structure","solar neighbourhood","galaxies: ISM","radio lines: ISM","spin temperature","spectral stacking","cold neutral medium","GASKAP survey"],"falsifier":"On the 462 detection sightlines the true average is already known component by component, so the claim is directly testable: stack the secondary components after emission-peak shifting exactly as the paper does, then bin the sightlines by the measured size of the emission–absorption velocity offset and compare the recovered $T_s$ against the noise-weighted mean of the individually measured components. If the recovered spin temperature drifts with the offset or deviates beyond the bootstrap uncertainties, the claimed cancellation fails; a toy-model version with offsets deliberately correlated with component temperature rather than drawn at random would show whether the unbiasedness is a general property or only holds for the assumed symmetry.","tokens_in":23823,"feed_emoji":"🌌","tokens_out":19537,"duration_ms":173132,"temperature":0.7,"pith_summary":"This paper establishes what spectral stacking does to measurements when thousands of Milky Way HI absorption–emission pairs are averaged together to reveal gas too faint for any single sightline. Shifting each spectrum by its emission peak centres the dominant cold component correctly, but every additional component keeps its own velocity offset, and averaging those misaligned features produces a broad, shallow 'secondary' component whose peak optical depth and brightness temperature are mathematically guaranteed lower limits of the true average, because total flux is conserved and the offsets broaden the stack. The spin temperature escapes this bias: the velocity offset moves the emission and absorption spectra equally, so it cancels in the ratio $T_s = T_{\\rm B,peak}/(1-e^{-\\tau_{\\rm peak}})$ that defines the temperature, a property the paper verifies with toy models and then exploits on real data. Stacking 462 detection sightlines after subtracting the known cold components yields a $1320 \\pm 263$ K component, gas of the unstable neutral medium; stacking 2240 non-detections uncovers cold neutral medium gas at $98 \\pm 12$ K (narrow primary component) and $255 \\pm 106$ K (broad secondary component), whose optical depth falls with distance from known cold-gas regions while its temperature stays constant over about 100 pc.","feed_headline":"Stacking HI spectra dims optical depth, keeps spin temperature","feed_subtitle":"Off-centre components broaden the stack, making peak depths lower limits but leaving spin temperatures unbiased.","key_machinery":"The load-bearing mechanism is the ratio symmetry of the spin-temperature formula, $T_s = T_{\\rm B,peak}/(1 - e^{-\\tau_{\\rm peak}})$, combined with flux conservation in the average. When a Gaussian component sits at a different velocity on each sightline, the stack broadens by an amount set by the spread of central velocities; with the integrated flux fixed, a wider component must have a lower peak, which makes the stacked $\\tau_{\\rm peak}$ and $T_{\\rm B,peak}$ strict lower limits of the true noise-weighted averages and the FWHM a strict upper limit. Because the velocity offset of a given component is the same in emission and in absorption, the same broadening factor enters the numerator and the denominator of the $T_s$ ratio and cancels, leaving a spin temperature linearly correlated with the true average — a property the paper verifies for both primary and secondary components. The alignment choice that makes this split clean is shifting each spectrum by the peak of the emission profile rather than by the first velocity moment: the primary component, tied to the emission peak, is centred correctly, while all remaining components are relegated to the secondary component of the stack.","core_discovery":"The paper's central claim is a pair of statements about stacking HI emission–absorption spectral pairs after aligning each spectrum by the peak of its emission. First, the component that follows the emission peak — the 'primary' component — is correctly centred, and its stacked peak optical depth, width, and spin temperature reproduce the noise-weighted average of the individual sightlines. Second, every additional component on a sightline — the 'secondary' component — keeps a non-zero, sightline-dependent velocity offset after this shifting; when stacked, those offsets broaden the feature, and because total flux is conserved, a broader Gaussian must have a lower peak, so the measured peak optical depth and brightness temperature of the secondary component are mathematically guaranteed lower limits of their true averages, while its FWHM is an upper limit. The spin temperature remains well correlated with the truth because the velocity offset affects emission and absorption equally and therefore cancels in the ratio that sets $T_s$; the paper demonstrates this cancellation both by comparing the stacked values with the individually measured components on the same 462 sightlines and with toy models of 2000-component ensembles. It then exploits the sensitivity gain: stacking the 462 detection spectra after subtracting the modelled cold gas reveals a broad component with $T_s = 1320 \\pm 263$ K, attributed to the unstable neutral medium; stacking 2240 non-detections reveals narrow and broad cold neutral medium components at $98 \\pm 12$ K and $255 \\pm 106$ K; and spatially binning the non-detections shows the primary component's optical depth decreasing by almost an order of magnitude with distance from known cold gas while both components' spin temperatures stay constant over roughly 100 pc of sky. The stacked spectra reach optical-depth noise near $10^{-3}$ yet do not detect the warm neutral medium, setting an upper limit $\\tau_{\\rm peak,WNM} \\lesssim 3.3 \\times 10^{-3}$ for this survey field.","pith_inferences":["The lower-limit theorem is generic, not HI-specific: any stacking or averaging procedure that aligns spectra by a proxy feature will under-estimate peaks and over-estimate widths of components not tied to that proxy, while any ratio diagnostic whose two quantities share the same misalignment stays protected; the same caveat should attach to molecular-line stacks, background-absorption surveys, and","Because the observed secondary optical depth is a lower limit that scales linearly with the true average, the plateau of the secondary component across the spatial bins is evidence that the warm-cold gas column responsible for it is roughly uniform over the ~100 pc field, even though the dense cold gas traced by the primary component is not — a distinction the current data cannot fully resolve.","The constancy of CNM spin temperature alongside a changing optical depth points to local pressure balance, not proximity to the large filaments, as the controller of cold-gas temperature; comparing the measured $T_s$ with pressure-regulated two-phase model predictions along the same sightlines would test this directly.","The claimed cancellation assumes emission–absorption velocity offsets are random; if deeper surveys find offsets correlated with optical depth or gas temperature — for example in denser, more turbulent regions — the stacked spin temperature would carry a residual bias, so the current result is best read as holding for gas with genuinely random offsets."],"forward_implications":["Peak optical depths and brightness temperatures of secondary components in any emission-peak-aligned stack are rigorous lower limits of the true averages, and their FWHMs are upper limits; quantities derived from those peaks, such as turbulent Mach numbers, column densities, and mass fractions, become unreliable and should not be reported from stacked spectra alone.","Stacked spin temperatures remain trustworthy: for the 462 detection sightlines the stacked values ($68 \\pm 7$ K primary, $159 \\pm 58$ K secondary) agree with the noise-weighted means of the same components measured individually, and the toy models show the cancellation holds for any spread of central velocities.","Gas of the thermally unstable neutral medium is present in this cirrus region at an average $T_s = 1320 \\pm 263$ K, detected in absorption only after subtracting the cold components and stacking; this is the highest-temperature component found so far with GASKAP in this region.","Cold neutral medium exists even along the 2240 sightlines where no single absorption detection was possible, at $98 \\pm 12$ K (narrow) and $255 \\pm 106$ K (broad) — the broad component sitting at the CNM/UNM boundary and likely mixing warmer CNM with cooler unstable gas.","Over roughly 100 pc of sky the amount of cold gas varies strongly — the primary component's peak optical depth falls from about $187 \\times 10^{-3}$ to $46 \\times 10^{-3}$ with distance from known cold-gas regions — while the spin temperature of cold gas stays constant within about $1\\sigma$ across all bins."],"supporting_citations":[{"why":"Supplies the stacking methodology the paper follows: noise weighting of spectra, equivalent-width definition, and bootstrap Monte Carlo validation of stacked signals.","marker":"Murray et al. (2014)"},{"why":"The completed 21-SPONGE stacking analysis whose CNM spin temperature (~104 K) and WNM detection are the baselines the new values are compared against.","marker":"Murray et al. (2018)"},{"why":"Provides the Gaussian-decomposition and radiative-transfer recipe, including foreground/background warm-gas permutations, used to fit spin temperatures of stacked components.","marker":"Heiles & Troland (2003)"},{"why":"Source of the 462 detection sightlines, their individual Gaussian fits that are subtracted before residual stacking, and the individual component properties used to validate the stacks.","marker":"Nguyen et al. (2024)"},{"why":"Describes the GASKAP absorption pipeline that produced the optical-depth spectra and defined the detection criteria for the two catalogues.","marker":"Dempsey et al. (2022)"},{"why":"Origin of the inverse-variance weighting scheme ($\\tau_{\\rm res}/\\sigma_\\tau^2$) applied to every stacked emission and absorption spectrum.","marker":"Treister et al. (2011)"},{"why":"Theoretical two-phase-medium results that fix the CNM/UNM temperature boundary (~250 K) used to classify the stacked components.","marker":"Wolfire et al. (2003)"},{"why":"Defines the GASKAP survey, its fields, and its observational parameters.","marker":"Dickey et al. (2013)"}],"fun_headline_variants":["HI stacking: secondary peak depths are lower limits, Ts hold","Stacking HI spectra keeps spin temps, dims optical depths","HI stacks: depths become floors, but spin temps stay true","Unstable neutral medium seen via HI stacking at 1320 K","HI stacking reveals unstable gas with unbiased spin temps"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole method assumes the brightest feature of each emission spectrum is the cold gas that produces the absorption, with only a small random offset between the two; if the emission peak instead comes from warm gas, or if the emission and absorption probe different gas because they are taken on slightly different sightlines, even the main component is misaligned and the spin temperatures the stacking recovers would no longer be trustworthy (Section 3.1).","fun_headline_variants_meta":{"raw":{"variants":["HI stacking: secondary peak depths are lower limits, Ts hold","Stacking HI spectra keeps spin temps, dims optical depths","HI stacks: depths become floors, but spin temps stay true","Unstable neutral medium seen via HI stacking at 1320 K","HI stacking reveals unstable gas with unbiased spin temps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000961,"raw_usage":{"total_tokens":4243,"prompt_tokens":1242,"completion_tokens":3001,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":858,"completion_tokens_details":{"reasoning_tokens":2917}},"tokens_in":858,"tokens_out":3001,"duration_ms":21990,"temperature":1.0,"reasoning_tokens":2917,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T17:42:17.168541+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On the 462 detection sightlines the true average is already known component by component, so the claim is directly testable: stack the secondary components after emission-peak shifting exactly as the paper does, then bin the sightlines by the measured size of the emission–absorption velocity offset and compare the recovered $T_s$ against the noise-weighted mean of the individually measured components. If the recovered spin temperature drifts with the offset or deviates beyond the bootstrap uncertainties, the claimed cancellation fails; a toy-model version with offsets deliberately correlated with component temperature rather than drawn at random would show whether the unbiasedness is a general property or only holds for the assumed symmetry.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Origin of the inverse-variance weighting scheme ($\\tau_{\\rm res}/\\sigma_\\tau^2$) applied to every stacked emission and absorption spectrum."}],"review_version":1}