{"id":"05be7497-68c9-4b95-9d59-78b34fe72915","arxiv_id":"2509.04567","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Green Pea galaxies with high O32 ratios, an indicator of Lyman continuum leakage, have significantly lower atomic hydrogen masses, gas fractions, and depletion timescales than those with low O32 ratios.","lead":"Astronomers used the Green Bank Telescope to measure cold hydrogen gas in 30 Green Pea galaxies and combined those data with earlier surveys to study 60 nearby dwarf starbursts. They find that galaxies with high O32 ratios, a marker of ionizing photon escape, contain far less atomic hydrogen than galaxies with low O32 ratios.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The HI-paucity explanation for LyC leakage is undercut by the paper's own admission that single-dish detections can be companion emission; the low-O32 vs high-O32 dichotomy may instead trace merger stage.","rationale":"The reader identified two fragile premises: the O32-to-LyC calibration and the association of single-dish HI with the GP. I agree with both, but I elevate the second to primary because it threatens the empirical HI measurement itself, not merely the LyC proxy. The paper is internally careful with survival analysis and Monte Carlo error propagation, and the raw detection-rate contrast (53% vs 7%) is striking. However, the strongest claim converts a correlation between O32 and single-dish HI detection into a causal statement about HI consumption enabling LyC escape. That conversion breaks if the low-O32 detections are companion HI, a possibility the authors explicitly acknowledge and for which they cite two resolved examples. Removing or reclassifying those two systems is a cheap first check; a positive result would strengthen the paper, while a null result would require moving the inference to 'unverified pending resolved HI mapping.' The O32 threshold concern is real but secondary here: even if O32 perfectly traced LyC leakage, the HI content comparison would still be contaminated by the beam-association ambiguity. I therefore keep the reader's CONDITIONAL verdict rather than moving to reject or accept: the data are valuable and the analysis is careful, but the central causal interpretation is not yet secure.","tokens_in":14114,"tokens_out":6415,"duration_ms":63741,"concrete_test":"Recompute the Peto-Prentice survival analyses for M_HI, f_HI, and tau_dep after removing J0213+0056 and J1148+2546, whose single-dish signals were resolved into companions, and after replacing their fluxes with only the GP-associated HI component if such a component exists. If the significances drop below ~2 sigma or the Kaplan-Meier means shift by more than a factor of 2, the headline anti-correlation is not robust for GP-hosted HI. A decisive follow-up would then be VLA/GMRT mapping of ~10 low-O32 single-dish detections to determine what fraction of the 21 cm flux is associated with the GP rather than an interacting companion.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires the single-dish 21 cm detections to be intrinsic to the GP. In the final paragraph of Section 3, the authors state that the GBT, Arecibo, and FAST beams are 180-550 kpc at z~0.05 and that 'some (or even all) of the observed H I 21 cm emission... could arise from a companion galaxy.' For J0213+0056 and J1148+2546, Purkayastha et al. (2022, 2024) resolved the single-dish signals and found most HI comes from merging companions rather than the GP. These systems are part of the literature sample feeding the 19 detections, yet the survival analysis does not exclude them. If low-O32 detections are preferentially HI-rich companions in early-stage mergers, while high-O32 non-detections are late-stage mergers, the reported 3-4 sigma differences in M_HI, f_HI, and tau_dep would not measure the atomic gas content of the GP itself. That alternative is not ad hoc; it is the explanation the authors themselves offer in the same paragraph. Because both scenarios fit the published data, the 'immediate explanation' for LyC leakage via HI exhaustion is not yet established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a new 193-hour Green Bank Telescope HI 21 cm emission survey of 30 Green Pea galaxies (GPs) at z≈0.012–0.045, yielding 7 detections and 17 upper limits. Combining these with literature measurements gives a sample of 60 GPs at z<0.05 with 19 detections and 41 non-detections. Using the extinction-corrected line ratio O32 ≡ [OIII]λ5007+λ4959/[OII]λ3727,3729 with a threshold of 10 as an indicator of Lyman continuum (LyC) leakage, the authors compare the HI mass, HI-to-stellar mass ratio, and HI depletion timescale between the O32<10 and O32>10 subsamples. Survival analysis with Peto-Prentice tests and Monte Carlo error propagation finds that the two subsamples differ at about 3.3–3.7σ significance, with high-O32 GPs having lower HI content, lower HI-to-stellar mass ratios, and shorter HI depletion timescales. The paper interprets this as evidence that HI paucity causes LyC leakage in GPs and predicts a spatial anti-correlation between HI 21 cm and Lyα emission at z≳6.","tokens_in":14437,"tokens_out":5321,"duration_ms":51442,"significance":"If the central claim holds, this is a valuable observational step connecting the neutral gas content of low-redshift LyC-leaking analogs to the escape of ionizing photons. The paper's strengths include a substantial new GBT data set, an appropriately chosen survival-analysis framework for censored HI measurements, Monte Carlo treatment of measurement errors, and an unusually explicit discussion of caveats, especially the single-dish beam-confusion problem. The reported differences are significant in a statistical sense, and the comparison of stellar mass and SFR distributions between the two O32 subsamples is a useful control. However, the physical conclusion that HI paucity is the cause of LyC leakage rests on two load-bearing assumptions that are not established within the paper: (1) that O32>10 reliably identifies LyC leakers in this population, and (2) that the measured single-dish HI emission is associated with the GP itself rather than with merging companions.","major_comments":[{"comment":"The significance of the survival analysis depends on the assumption that the censoring mechanism (upper limits) is comparable between the two O32 subsamples and across the three single-dish telescopes. The paper asserts that restricting to z<0.05 ensures 'similar sensitivity,' but the upper limits in Table 1 span a wide range (e.g., M_HI upper limits from <0.5×10^8 to <5.5×10^8 M⊙), and the telescopes have different beam sizes, system temperatures, and RFI environments. If the high-O32 subsample is preferentially observed with less sensitive data or at larger distances, the detection-rate difference could be partly a sensitivity artifact. In addition, the assumed Gaussian FWHM of 50 km/s for non-detections is a free parameter; the detected HI lines in Table 1 have W50 values of 40–140 km/s, so a different assumed FWHM could change the upper limits and hence the Kaplan-Meier means and test significances. Please provide a quantitative comparison of the upper-limit distributions, distances, and RMS sensitivities of the two subsamples, and test the sensitivity of the Peto-Prentice results to the assumed FWHM (e.g., 50, 100, and 150 km/s).","section":"Section 3, first paragraph; Section 4"}],"minor_comments":[{"comment":"The column list in the caption contains a numbering error: column (9) is listed twice (absolute B-band magnitude and O32), and the FWHM column for HI detections is not assigned a number. Please fix the column numbering.","section":"Table 1 caption"},{"comment":"The phrase 'Earlier studies have shown that galaxies with O32>10 tend to show significant LyC leakage: our results indicate that this is due to the lack of HI' should be phrased conditionally, because the paper does not measure LyC escape in this sample; recommend 'if O32 traces LyC leakage, our results suggest...'.","section":"Abstract"},{"comment":"The velocity axis range in the third panel of Figure 1 (±1000 km/s relative to redshift) is inconsistent with the other panels, which use ±400 km/s. Please unify the velocity ranges or explain the difference.","section":"Section 2, Figure 1"},{"comment":"For J0840+5333, the HI emission is offset by ≈−100 km/s from the optical redshift, and the text notes it may arise from a companion. This object is counted as a detection; please flag it clearly in Table 1 or Figure 1 so that readers can identify potentially contaminated detections.","section":"Section 2, J0840+5333"},{"comment":"There is a typo in the Summary: 'GPs wih O32 > 10' should be 'GPs with O32 > 10'.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest and the statistical analysis is competently executed, but the final-paragraph caveat about companion contamination is serious enough that the central claim as currently worded is not supported. The proposed reanalysis excluding the resolved companion-dominated detections is straightforward and should be feasible within a revision. If the significance survives that test, the paper would be a solid contribution; if not, the interpretation needs to be substantially reframed. The O32-proxy issue is a correctness risk that can be addressed by softening the causal language and adding tests with other LyC indicators."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the new GBT survey and the 60-galaxy HI-O32 comparison are worth a look; the detection-rate difference (53% for O32<10 vs 7% for O32>10) is striking, and the survival analysis is careful. But the causal headline - low HI causes LyC leakage - is not supported, and the paper's own last paragraph says why.\n\nWhat's new: 7 new GBT detections, and a first combined analysis linking HI content to O32 across GPs. The Peto-Prentice tests with upper limits, Monte Carlo error propagation, and the check that stellar mass and SFR distributions match across the two subsamples are all done properly. The Kaplan-Meier means differ by factors of roughly 2.5 to 6 in the expected direction. That is a genuine empirical result.\n\nSoft spots: two. First, O32 is not a direct LyC measurement; the threshold O32>10 comes from LzLCS. That is a reasonable proxy but it is a proxy, and the paper says so. Second, and more seriously, the single-dish beams are 180-550 kpc. The authors concede that some or all of the HI could be from companions, and for J0213+0056 and J1148+2546 the resolved maps show most of the HI is in merging companions. Those two are in the literature sample. If low-O32 detections are preferentially HI-rich early-stage mergers and high-O32 non-detections late-stage mergers, the same data would look exactly like what they see. That alternative is not ad hoc - it is the explanation the authors themselves float. So the abstract's claim that the results indicate the lack of HI is the cause overreaches; the paper establishes a correlation between O32 and single-dish HI flux, not a causal link via HI exhaustion.\n\nMinor: sensitivity equivalence across GBT/Arecibo/FAST is asserted rather than quantified, and I would have liked a machine-readable table of the full 60-galaxy sample, not just the new 24.\n\nBottom line: for a referee, yes. The empirical pattern is new and likely to be reproduced; the interpretation needs softening, and the companion question needs to be addressed with resolved HI maps or explicit exclusion. The paper is honest about its own weakness, which counts for something.","headline":"A careful new HI survey shows a sharp O32 dichotomy in Green Pea HI content, but the paper's own beam-contamination caveat means the LyC-leakage explanation is not established.","tokens_in":14938,"tokens_out":2077,"would_cite":true,"duration_ms":19174,"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":"Green Pea galaxies with high O32 ratios have largely consumed their neutral hydrogen, and this gas loss is what lets their ionizing light escape.","keywords":["Green Pea galaxies","neutral hydrogen","H I 21 cm emission","Lyman continuum leakage","O32 ratio","starburst dwarf galaxies","cosmic reionization","H I depletion timescale"],"falsifier":"Resolved 21 cm mapping of the high-$O_{32}$ Green Peas would settle the gas-paucity claim: if interferometric observations reveal substantial neutral hydrogen in the galaxies or in companions, the low single-dish detection rate would reflect beam dilution or merger stage rather than exhausted gas. Alternatively, direct Lyman continuum measurements of Green Peas at $z\\approx0.3$, where LyC can be observed, could test whether the $O_{32}>10$ subsample really has the high escape fractions assumed by the threshold.","tokens_in":13913,"feed_emoji":"🌌","tokens_out":12986,"duration_ms":101432,"temperature":0.7,"pith_summary":"Using a 193-hour radio search for neutral hydrogen in 30 Green Pea galaxies and combining it with earlier surveys, this paper assembles a sample of 60 such galaxies at $z<0.05$. It finds that galaxies with $O_{32}>10$, a line ratio previously calibrated to mark Lyman continuum leakage, are detected in 21 cm emission only about 7% of the time, versus about 53% for galaxies with $O_{32}<10$, and have lower H I masses, lower H I-to-stellar mass ratios, and shorter H I depletion timescales. The stellar masses and star formation rates of the two groups are statistically indistinguishable, isolating the gas as the variable. The paper concludes that the high-$O_{32}$ galaxies leak ionizing radiation because their atomic gas has been largely consumed by the starburst, and predicts that 21 cm and Ly$\\alpha$ emission should be spatially anti-correlated in the galaxies that reionized the universe.","feed_headline":"Hydrogen-poor Green Peas are the galaxies leaking ionizing light","feed_subtitle":"Sixty-galaxy survey: high O32 means spent atomic gas, the likely gate for ionizing escape.","key_machinery":"The load-bearing object is the extinction-corrected line ratio $O_{32} = [\\mathrm{O\\,III}]\\lambda5007+\\lambda4959 / [\\mathrm{O\\,II}]\\lambda3727,3729$, used with a threshold of 10 as a proxy for Lyman continuum escape; the threshold comes from external calibration showing that more than half of galaxies with $O_{32}\\gtrsim10$ leak LyC, while fewer than about 20% below it do. The measurement side is single-dish H I 21 cm emission, which gives H I masses via the standard luminosity-distance relation, with survival analysis (Kaplan-Meier means and Peto-Prentice two-sample tests, with Monte Carlo error propagation) handling the many upper limits. The physical mechanism invoked is the density-bounded H II region: high $O_{32}$ means the ionized region ends where the gas runs out rather than where the photons run out, so ionizing photons escape and little neutral gas remains.","core_discovery":"On the paper's own terms, the discovery is a statistical dichotomy: Green Pea galaxies above $O_{32}=10$ are almost never seen in H I 21 cm (2 of 28 systems, one tentative), while those below the threshold are frequently detected (17 of 32), with the Kaplan-Meier mean H I mass a factor of about six higher in the low-$O_{32}$ group. Because the two subsamples have matching stellar masses and star formation rates, the difference is attributed to the gas: high-$O_{32}$ galaxies have consumed most of their neutral hydrogen, shortening their depletion timescales by a factor of about four. The paper takes this as the direct explanation for Lyman continuum leakage in these galaxies, and as evidence that the reionization-era analogs of Green Peas would be gas-poor, producing an anti-correlation between H I 21 cm and Ly$\\alpha$ emission at $z\\gtrsim6$.","pith_inferences":["A direct test of the merger interpretation the paper leaves open: compare optical morphologies of high- and low-$O_{32}$ Green Peas; if the high-$O_{32}$ objects show late-stage merger features, the gas difference may track merger phase rather than internal exhaustion.","The predicted 21 cm / Ly$\\alpha$ anti-correlation at $z>6$ could be checked with current-generation interferometers by stacking 21 cm spectra toward faint Ly$\\alpha$ emitters, where a deficit relative to field averages would support gas-poor leakers.","If the $O_{32}$-H I relation extends to lower stellar masses, the $O_{32}=10$ threshold may mark a universal gas-exhaustion state for dwarf starbursts, offering a redshift-independent way to identify galaxies near the end of their reionization-contributing phase.","Because the paper's sample is limited to $z<0.05$, extending H I stacking to slightly higher redshift Green Peas would test whether the detection-rate gap persists at larger distances, separating a real gas dichotomy from a sensitivity effect."],"forward_implications":["High-$O_{32}$ Green Peas are detected in H I 21 cm in only about 7% of cases, against about 53% for low-$O_{32}$ galaxies, so gas paucity is a statistical property of likely LyC leakers.","The H I mass, H I-to-stellar mass ratio, and H I depletion timescale are all lower in the $O_{32}>10$ group at roughly 3.2-3.7 sigma, with matched stellar properties, so the effect is specific to the atomic gas.","Because density-bounded H II regions let ionizing photons escape once the neutral gas is consumed, low H I content is presented as the immediate cause of Lyman continuum leakage in these galaxies.","If the same applies at $z\\gtrsim6$, searches for 21 cm emission from the epoch of reionization should find a spatial anti-correlation between H I 21 cm and Ly$\\alpha$ emission.","Some of the low-$O_{32}$ detections may be companions rather than the Green Pea itself, so the true H I content of the galaxies could be even lower than the single-dish numbers suggest."],"supporting_citations":[{"why":"Supplies the external calibration that O32 greater than 10 corresponds to significant Lyman continuum leakage, the threshold the paper adopts.","marker":"S. R. Flury et al. 2022"},{"why":"Prior 21 cm survey of 40 Green Peas contributing 19 detections and 21 upper limits to the combined sample.","marker":"N. Kanekar et al. 2021"},{"why":"Prior 21 cm survey of 28 Green Peas contributing 2 detections and 26 upper limits.","marker":"Y. Chandola et al. 2024"},{"why":"Tentative 21 cm detection in one Green Pea that is included among the high-O32 detections.","marker":"S. Dutta et al. 2024"},{"why":"Source of the stellar masses, star formation rates, and O32 values used to construct and match the two subsamples.","marker":"T. Jiang et al. 2019"},{"why":"21 cm mapping showing that single-dish emission from one Green Pea arises from a merging companion, motivating the beam caveat.","marker":"S. Purkayastha et al. 2022"},{"why":"Maps a second Green Pea and likewise finds companion emission, reinforcing the caveat.","marker":"S. Purkayastha et al. 2024"},{"why":"Proposes the density-bounded H II region interpretation that connects high O32 to neutral-gas paucity and LyC escape.","marker":"K. Nakajima & M. Ouchi 2014"},{"why":"Documents Lyman continuum leakage with escape fractions of 2.5-72% in Green Peas at z~0.3, establishing them as local LyC leakers.","marker":"Y. I. Izotov et al. 2018"},{"why":"Provides the survival-analysis approach used to compare H I distributions in the presence of upper limits.","marker":"E. D. Feigelson & P. I. Nelson 1985"}],"fun_headline_variants":["Spent hydrogen lets Green Peas leak ionizing light","Hydrogen-starved Green Peas are the LyC leakers","Green Peas with low H I leak more ionizing light","Spent gas opens ionizing leaks in Green Peas"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument stands or falls on whether $O_{32}>10$ really marks Lyman continuum escape in Green Pea galaxies, since no direct LyC measurement is possible at these redshifts; it also assumes that the single-dish 21 cm detections belong to the target galaxies rather than to merging companions in beams spanning 180 to 550 kpc.","fun_headline_variants_meta":{"raw":{"variants":["Spent hydrogen lets Green Peas leak ionizing light","Hydrogen-starved Green Peas are the LyC leakers","Green Peas with low H I leak more ionizing light","Spent gas opens ionizing leaks in Green Peas"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000931,"raw_usage":{"total_tokens":4122,"prompt_tokens":1220,"completion_tokens":2902,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":836,"completion_tokens_details":{"reasoning_tokens":2832}},"tokens_in":836,"tokens_out":2902,"duration_ms":18104,"temperature":1.0,"reasoning_tokens":2832,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:28:56.818018+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Resolved 21 cm mapping of the high-$O_{32}$ Green Peas would settle the gas-paucity claim: if interferometric observations reveal substantial neutral hydrogen in the galaxies or in companions, the low single-dish detection rate would reflect beam dilution or merger stage rather than exhausted gas. Alternatively, direct Lyman continuum measurements of Green Peas at $z\\approx0.3$, where LyC can be observed, could test whether the $O_{32}>10$ subsample really has the high escape fractions assumed by the threshold.","supporting_citations":[{"cited_title":"J., et al","cited_arxiv_id":null,"evidence_quote":"Prior 21 cm survey of 28 Green Peas contributing 2 detections and 26 upper limits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the stellar masses, star formation rates, and O32 values used to construct and match the two subsamples."},{"cited_title":"2024, ApJ, 977, 68, doi: 10.3847/1538-4357/ad8dd2","cited_arxiv_id":null,"evidence_quote":"Maps a second Green Pea and likewise finds companion emission, reinforcing the caveat."}],"review_version":1}