{"id":"7e2b3554-562a-4705-a842-0d65b4692077","arxiv_id":"2411.11584","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Unconstrained spectral stacking of MHONGOOSE data finds little neutral hydrogen beyond galaxy disks, implying CGM/IGM column densities near 1e17 cm^-2 are likely below current and near-future detection limits.","lead":"This paper stacks hundreds of radio spectra around nearby galaxies to hunt for faint hydrogen gas in the space between and around galaxies. It finds far less gas than cosmological simulations predict, suggesting this gas may be too diffuse to detect even with the next generation of radio telescopes.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The stacking's systemic-velocity prior is validated on only two TNG50 mocks; co-rotating or disordered CGM gas would be smeared below the SNR threshold, so the null detection cannot yet rule out such gas.","rationale":"The strongest claim is a null result: outside the HI disk there is much less HI than simulations predict. Null results are only meaningful if the method would have detected the predicted signal. Two conditions are necessary: (1) the gas must be aligned correctly in velocity, and (2) the stacking must not dilute compact or faint clouds below the SNR cut. The paper addresses (2) qualitatively but reports only ~35% recovery of simulated emission above 3.6e17 cm^-2, which is itself a warning that the method is insensitive to a large fraction of the predicted gas. The more fundamental problem is (1): the calibration of the kinematic prior rests on two TNG50 galaxies, and the authors explicitly state that the result should not be generalised. If real CGM gas co-rotates, the systemic-velocity stacking smears the signal and the non-detection is consistent with gas being present. The reader's verdict (CONDITIONAL) already captures this; my stress-test does not identify a new independent flaw that would change the verdict. I also note the abstract/body sample-size discrepancy (6 vs 18/10 full-depth cubes) should be corrected, but it is secondary to the kinematic prior. The code is public and the paper is honest about limitations, which strengthens trust but does not remove the need for the kinematic test.","tokens_in":26135,"tokens_out":5541,"duration_ms":55823,"concrete_test":"Inject into the two public TNG50 mock cubes a CGM/IGM component whose kinematics follow a flat rotation curve matched to the galaxy inclination (Eq. 7), with column densities bracketing 1e17–1e18 cm^-2, and run the published STACKER+FINDER pipeline using systemic-velocity alignment. Measure the recovered flux fraction and detection rate. If the recovery fraction drops materially below the ~35% reported in Sect. 4.3, the observed non-detection cannot exclude co-rotating CGM gas and the abstract's conclusion should be softened. A useful complementary check is to re-derive an upper limit on total CGM HI mass (not just per-cell column density) from the recovered flux, accounting for the SNR=5 cut.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central inference—that HI outside the disks is much scarcer than simulations predict—depends on the stacked spectra adding coherently. The adopted prior is that all gas outside the SoFiA-2 mask moves at the systemic velocity (Sect. 4.2), so no kinematic model is used to align spectra. If the real CGM/IGM co-rotates with the disk or has complex non-circular motions, a genuine line is spread over many channels; the resulting peak SNR drops below FINDER's threshold, and the stacked spectrum looks empty. The authors test only two TNG50 galaxies and explicitly caution that the result cannot be generalized. Moreover, their own calibration (Sect. 4.3) shows only ~35% of simulated emission above 3.6e17 cm^-2 is recovered even under the adopted assumption, so the per-cell 1e17 cm^-2 limit is a noise-equivalent sensitivity, not a demonstrated bound on compact or kinematically offset clouds. Thus the strongest claim in the abstract is not yet supported for the general MHONGOOSE population; the result is conditional on the co-rotation/kinematics prior.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents a spectral-stacking pipeline (STACKER and FINDER) for searching for low-column-density HI emission in the CGM/IGM around MHONGOOSE galaxies. The method is calibrated on two TNG50 mock galaxies, testing different stacking cell sizes, weighting schemes, and line-finder parameters, and comparing two kinematic priors for aligning spectra (systemic velocity vs. co-rotation with the disk). The authors conclude that the systemic-velocity alignment is preferable for their mocks and apply the method to 18 MHONGOOSE galaxies with inclination i≤60°, using single-track cubes for all of them and full-depth cubes for six after quality selection. They report very few reliable detections outside the SoFiA-2 masks, a stacked column-density sensitivity of ~10^17 cm^-2, and conclude that the amount of HI outside the disks is much smaller than predicted by simulations and that direct detection of the neutral CGM/IGM component will be challenging in the future.","tokens_in":26475,"tokens_out":8708,"duration_ms":78591,"significance":"Should the null result hold, it would provide a valuable observational constraint on the cold neutral gas content of the CGM/IGM of nearby galaxies, in a regime where simulations currently predict detectable HI. The paper is methodologically careful in several respects: the noise is checked for Gaussianity, the reliability of detections is assessed through positive-negative source statistics, the method is tested on mock observations, the code is publicly available, and each detection is visually inspected by three independent authors. However, the central inference is conditional on the adopted kinematic prior and on the recovery fraction of the stacking procedure, and the comparison to simulations is only qualitative. These caveats do not invalidate the method paper, but they require that the strongest claims in the abstract and conclusions be appropriately qualified.","major_comments":[{"comment":"The central null result is conditional on the assumption that all gas outside the SoFiA-2 mask moves at the systemic velocity of the galaxy. The validation of this assumption is limited to two TNG50 mock galaxies, and the authors themselves caution that 'one should be careful not to generalise this result' (Sect. 4.2). For real galaxies with inclination up to 60°, plausible co-rotation of CGM gas with the disk would spread line emission over tens of km/s, lowering the stacked SNR below the FINDER threshold and making the stacked spectra appear empty. The abstract's claim that 'the amount of signal detected outside the HI disk is much smaller than implied by simulations' is therefore not established for the general MHONGOOSE population unless this kinematic dependency is explicitly acknowledged.","section":"Sect. 4.2 and abstract"},{"comment":"The calibration in Sect. 4.3 shows that only ~35% of simulated emission above 3.6×10^17 cm^-2 is recovered, even under the adopted systemic-velocity prior. Consequently, the quoted ~10^17 cm^-2 stacked limit is a noise-equivalent sensitivity of the stacked spectra, not a demonstrated upper bound on the column density of any gas that may be present in compact or kinematically offset clouds. The abstract and conclusions should clearly distinguish between the sensitivity achieved and the physical limit that can be placed on CGM/IGM HI, otherwise the non-detection is easily overinterpreted.","section":"Sect. 4.3"},{"comment":"The statement that the observed signal is 'much smaller than implied by simulations' is not supported by a quantitative comparison. The paper uses two TNG50 galaxies to calibrate the method, but it does not compute an expected detection rate or a predicted stacked SNR for the MHONGOOSE sample from the simulations. The conclusion is therefore a qualitative impression based on two mock galaxies. A quantitative comparison, for example by applying the same stacking procedure to a statistical sample of mock galaxies with the same selection criteria, would be needed to support the claim in the abstract.","section":"Sect. 5.5 and Sect. 4"}],"minor_comments":[{"comment":"The abstract states that full-depth observations are available for '6 nearby star forming galaxies', while Sect. 5 states that full-depth cubes are available for ten galaxies, with six used after quality selection (Table A.1). Please make the wording consistent and explain the selection explicitly.","section":"Abstract and Sect. 5"},{"comment":"The Gaussian tail probabilities appear to be off by a factor of 100: for a Gaussian distribution, P(|F| > 4σ) ≈ 0.0063%, not 0.000063%. The derived excess of ~8.8% is unaffected by this error, but the reported percentages (0.000063% and 0.000069%) should be corrected.","section":"Sect. 5.2"},{"comment":"The TNG50 galaxy ID is given as 520885 in Table 1 and in most figure captions, but as 520855 in Fig. 7, Fig. B.1, and the accompanying text. Please unify the notation.","section":"Throughout (Fig. 7, Fig. B.1, Table 1)"},{"comment":"Equation (2) presents a one-dimensional kernel density estimate, while the source parameters are three-dimensional (Fmax, Fsum, Fmean). Please clarify that the KDE is actually applied in the three-dimensional space, or provide the correct multivariate expression.","section":"Sect. 3.2, Eq. (2)"},{"comment":"The caption does not explain what the grey-scale background and the black contours represent. Please specify that the background is the masked/regridded cube collapsed along the spectral axis and that the black contours enclose the SoFiA-2 mask.","section":"Fig. 3 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid method paper and the data analysis is careful. The main issue is that the abstract overstates the result relative to the caveats in the body; the kinematic prior and the low recovery fraction should be prominently acknowledged. The quantitative comparison with simulations is also too weak for the strength of the claim. These are fixable with revision, so I recommend major revision rather than rejection. The numerical error in Sect. 5.2 should be corrected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper is worth a read if you work on HI stacking or CGM constraints, but take the headline with a grain of salt. The genuine new element is an 'unconstrained' stacking routine (spectral stacking plus a 1D reliability-based line finder) applied to MHONGOOSE data, with careful calibration on two TNG50 mocks and a publicly released package. The analysis is honest: they check noise Gaussianity (small symmetric excess at >4σ), visually inspect every detection, and explicitly quantify that their method recovers only ~35% of simulated emission above 3.6e17 cm^-2. That last number matters more than the abstract lets on.\n\nThe soft spot is the kinematic prior. The stack aligns everything at the systemic velocity, and the authors justify this using two mock galaxies where this beat a flat co-rotation extension. They themselves say the result cannot be generalized. If real CGM/IGM gas co-rotates or has complex motions, a genuine line gets smeared and the stacked spectrum looks empty. So the central claim—that HI outside disks is much scarcer than simulations predict—only holds under that assumption. The per-cell limit of ~1e17 cm^-2 is a noise-equivalent sensitivity, not a demonstrated bound on kinematically offset clouds. They deserve credit for flagging this, but the abstract doesn't carry the caveat.\n\nThere's also a housekeeping issue: the abstract says six galaxies with full-depth data; the body stacks eighteen (single-track) and ten (full-depth). That discrepancy should have been caught.\n\nVerdict: conditional. The method contribution is real and the null result is broadly consistent with earlier stacking work, so it's a solid incremental paper. But as a constraint on cold accretion it's preliminary. Send it to a referee—it deserves careful review—with instructions to fix the abstract and to make the kinematic caveat front-and-center. I'd cite the method package, not the null result.","headline":"Careful stacking study with a useful public method and an honest limitations section, but the headline null result is conditional on a kinematic prior validated on only two mocks.","tokens_in":27018,"tokens_out":3380,"would_cite":true,"duration_ms":32347,"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":"Stacked radio spectra of six galaxies show almost no neutral hydrogen outside their HI disks down to a column density of 10^17 cm^-2.","keywords":["neutral hydrogen","spectral stacking","circum-galactic medium","inter-galactic medium","MHONGOOSE","MeerKAT","column density limits","galaxy gas accretion"],"falsifier":"Re-analyse the same full-depth cubes after aligning spectra along the major axis with a flat-rotation-curve velocity field (or a best-fit tilted-ring model) instead of the systemic velocity; if reliable $\\sim10^{17}$ cm$^{-2}$ detections appear that disappear under systemic-velocity stacking, the central non-detection is an artefact of the alignment assumption. Alternatively, a deep single-dish radial profile of one MHONGOOSE galaxy reaching below $10^{17}$ cm$^{-2}$ that shows an extended neutral envelope would falsify the claim that the neutral CGM/IGM is essentially absent.","tokens_in":25951,"feed_emoji":"📡","tokens_out":12708,"duration_ms":108430,"temperature":0.7,"pith_summary":"This paper argues that the faint neutral atomic hydrogen (HI) that cosmological simulations place in the circum-galactic and inter-galactic medium around Milky Way-like galaxies is not seen even in ultra-deep MeerKAT observations of six nearby star-forming galaxies. By co-adding thousands of spectra after masking out the galaxies' disks, the authors reach limiting column densities near $\\sim10^{17}$ cm$^{-2}$ yet recover only a handful of marginal signals, far less than the extended HI structures that the TNG50 mock observations produce. If the result holds, the cool gas that simulations funnel onto galaxies to replenish their star-forming reservoirs must be hidden below this column-density threshold, making direct 21-cm detection of the neutral CGM/IGM implausible even with next-generation radio arrays. The authors present this as a method demonstration plus an observational limit, with an explicit caveat that the spectral alignment depends on an assumption about gas kinematics.","feed_headline":"Six-galaxy stacking finds far less HI than simulations predict","feed_subtitle":"Co-added MeerKAT spectra reach 10^17 cm^-2 and still find almost no neutral gas beyond the disks.","key_machinery":"The load-bearing mechanism is the coupling of a spectral stacker (STACKER) with a one-dimensional line finder (FINDER). STACKER blanks the galaxy and known sources using a detection mask, shifts each line of sight to a reference velocity ('shuffle'), regrids to beam-sized pixels so that Gaussian noise decreases as $\\sqrt{N}$ with the number of co-added spectra, and then co-adds spectra inside square cells of $5\\times5$, $9\\times9$, and $14\\times14$ beams. FINDER applies a smooth-and-clip search on the stacked spectra and assigns each candidate a reliability based on the local density of positive versus negative noise peaks in the (peak, sum, mean) flux space, rejecting candidates whose stacked noise fails Gaussianity checks. The crucial kinematic choice is aligning every spectrum to the systemic velocity of the galaxy, which the authors test on two TNG50 mock galaxies and explicitly caution cannot be generalised to arbitrary geometries.","core_discovery":"On its own terms, the paper's central claim is that unconstrained spectral stacking of the MHONGOOSE data yields essentially no reliable HI emission beyond the detected disks: 13 reliable sources in the full-depth cubes, only four of which have a clear visual counterpart in the data, at integrated fluxes well below what the mock observations would produce. When tested on two TNG50 mock galaxies, the same pipeline recovers only about 35% of the mock HI above $3.6\\times10^{17}$ cm$^{-2}$ because the imposed signal-to-noise cut rejects faint stacked lines, so the procedure is not sensitive to everything the simulations put there; yet the mock data still return many more detections than the real data. The authors conclude that the amount of neutral hydrogen outside the HI disk is much smaller than simulations imply, and that the stacked column-density limit of about $\\sim10^{17}$ cm$^{-2}$ makes direct emission detection of the neutral CGM/IGM challenging even with future radio telescopes.","pith_inferences":["I infer that the systemic-velocity alignment likely sets a floor on what stacking can recover: if the circum-galactic gas co-rotates with the disk, a real signal would be smeared out and the $\\sim10^{17}$ cm$^{-2}$ limit would be an artefact of the kinematic assumption rather than evidence of absence.","A direct test of that assumption would be to redo the stacking on the same cubes with spectra aligned along the major axis using a flat rotation curve; if reliable detections appear that vanish under systemic-velocity stacking, the paper's non-detection is not the final word.","The result, if general, pushes searches for cold accreting gas toward ionised-gas tracers and quasar absorption spectroscopy, since neutral-emission stacking appears to have reached a wall near $10^{17}$ cm$^{-2}$ even when thousands of spectra are co-added."],"forward_implications":["The neutral atomic hydrogen in the circum-galactic and inter-galactic medium of the six galaxies is below $\\sim10^{17}$ cm$^{-2}$, so cool accreting gas is not directly visible as 21-cm emission with current stacking techniques.","The HI radial profiles of these galaxies must drop sharply near the disk edge rather than extend as a diffuse neutral envelope, consistent with gas becoming ionised at the measured threshold.","Future searches for cool gas accretion will need either absorption-line probes along background quasars or much deeper emission observations than current surveys can provide.","The stacking-plus-reliability pipeline itself is transferable to the full MHONGOOSE sample and to other nearby-galaxy data sets once full-depth cubes are available."],"supporting_citations":[{"why":"Supplies the two TNG50 Milky Way-like mock galaxies used to calibrate STACKER and FINDER and to test the kinematics assumptions.","marker":"Ramesh et al. (2023)"},{"why":"Provides the MHONGOOSE full-depth and single-track MeerKAT cubes, SoFiA-2 masks, and moment maps that are stacked.","marker":"de Blok et al. (2024)"},{"why":"Earlier spectral-stacking search for HI in the CGM/IGM that this work extends and compares with.","marker":"Das et al. (2020)"},{"why":"GBT stacking study of NGC 891 and NGC 4565 reporting a CGM/IGM column density below $10^{17}$ cm$^{-2}$, which sets the comparison for this result.","marker":"Das et al. (2024)"},{"why":"GBT radial profiles suggesting a sharp HI drop near $10^{17}$ cm$^{-2}$, used to explain the observed absence of extended neutral gas.","marker":"Sardone et al. (2021)"},{"why":"Simulation prediction of low-column-density HI clouds in the CGM/IGM that the observations are designed to test.","marker":"Ramesh & Nelson (2024)"},{"why":"Supplies the reliability calculation that FINDER uses to separate genuine stacked lines from noise peaks.","marker":"Serra et al. (2012a)"},{"why":"FAST detection of very low HI column densities in the intragroup medium of Stephan's Quintet, cited as independent evidence that the neutral component is faint.","marker":"Xu et al. (2022)"}],"fun_headline_variants":["Stacking finds far less HI than simulations predict","MHONGOOSE stacking: HI beyond disks far below mock predictions","MeerKAT spectral stacking limits neutral gas detection in CGM","Six galaxies, one stack: almost no HI outside the disks","HI stacking reaches 1e17 cm^-2, still no CGM signal"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The stacking assumes that gas outside the HI disk moves at the galaxy's systemic velocity, so all spectra can be aligned without knowing the true kinematics; if the gas instead co-rotates with the disk or has complex motions, a real signal would be blurred out and the non-detection would not prove the gas is missing.","fun_headline_variants_meta":{"raw":{"variants":["Stacking finds far less HI than simulations predict","MHONGOOSE stacking: HI beyond disks far below mock predictions","MeerKAT spectral stacking limits neutral gas detection in CGM","Six galaxies, one stack: almost no HI outside the disks","HI stacking reaches 1e17 cm^-2, still no CGM signal"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000207,"raw_usage":{"total_tokens":1417,"prompt_tokens":977,"completion_tokens":440,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":593,"completion_tokens_details":{"reasoning_tokens":351}},"tokens_in":593,"tokens_out":440,"duration_ms":4946,"temperature":1.0,"reasoning_tokens":351,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T18:20:22.824097+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-analyse the same full-depth cubes after aligning spectra along the major axis with a flat-rotation-curve velocity field (or a best-fit tilted-ring model) instead of the systemic velocity; if reliable $\\sim10^{17}$ cm$^{-2}$ detections appear that disappear under systemic-velocity stacking, the central non-detection is an artefact of the alignment assumption. Alternatively, a deep single-dish radial profile of one MHONGOOSE galaxy reaching below $10^{17}$ cm$^{-2}$ that shows an extended neutral envelope would falsify the claim that the neutral CGM/IGM is essentially absent.","supporting_citations":[{"cited_title":"2023, MNRAS, 518, 5754","cited_arxiv_id":null,"evidence_quote":"Supplies the two TNG50 Milky Way-like mock galaxies used to calibrate STACKER and FINDER and to test the kinematics assumptions."},{"cited_title":"K., et al","cited_arxiv_id":null,"evidence_quote":"Earlier spectral-stacking search for HI in the CGM/IGM that this work extends and compares with."},{"cited_title":"2024, MNRAS, 527, 10358","cited_arxiv_id":null,"evidence_quote":"GBT stacking study of NGC 891 and NGC 4565 reporting a CGM/IGM column density below $10^{17}$ cm$^{-2}$, which sets the comparison for this result."},{"cited_title":"& Nelson, D","cited_arxiv_id":null,"evidence_quote":"Simulation prediction of low-column-density HI clouds in the CGM/IGM that the observations are designed to test."}],"review_version":1}