{"id":"ab2dece3-f650-4207-b230-f61e6c11f468","arxiv_id":"2507.19573","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"In 26 low-mass galaxies, atomic gas turbulence correlates most strongly with star formation 100-500 million years ago, indicating a long-lived coupling between stellar feedback and the interstellar medium.","lead":"Astronomers compared star formation histories of 26 nearby dwarf galaxies with the turbulence of their atomic hydrogen gas. They find that the gas turbulence correlates most strongly with star formation that happened 100 to 500 million years ago, implying stellar feedback energy lingers in the gas for hundreds of millions of years.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 100-500 Myr correlation is shown for HI energy surface density, a quantity proportional to HI column density; the paper does not isolate the velocity-dispersion contribution, so the claim that past star formation drives HI turbulence may rest on a gas-density confound.","rationale":"The reader identified the independence of regions as the weakest assumption, noting inflated P-values and galaxy weighting. That is a real statistical concern, and the paper's Section 5.4 equal-region resampling mitigates the weighting issue but does not remove spatial autocorrelation within galaxies. However, the more fundamental threat to the central claim is that the headline correlation uses HI energy surface density, which is the product of HI column density and velocity dispersion squared. The paper never reports whether the 100-500 Myr correlation survives when controlling for HI column density. Since the velocity-dispersion-only correlations are weak and peak at different timescales, the abstract's generalization to 'atomic hydrogen turbulence measures' may be unsupported. This concern does not change the reader's CONDITIONAL verdict, but it sharpens the required revision: the authors should add a partial-correlation or stratified analysis before the turbulence-timescale interpretation is accepted. I agree with the reader's overall assessment that the qualitative result is plausible but quantitatively overstated; my concern adds a specific confound that should be tested.","tokens_in":53747,"tokens_out":3478,"duration_ms":43994,"concrete_test":"Recompute Spearman partial correlations between SFR(t) and sigma_central (and sigma_m2) controlling for M_HI/A_HI across the full 961-region sample, or bin regions into narrow M_HI/A_HI quartiles and recompute the rho(t) curves within each stratum. If the 100-500 Myr enhancement in rho drops below 0.2 after controlling for HI column density, the claimed turbulence timescale is a gas-surface-density effect; if partial correlations remain rho>=0.3 at 100-500 Myr, the turbulence interpretation survives.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central result (Figures 6 and 8, Section 5.1-5.2) is a Spearman correlation between SFH in the 100-500 Myr bins and HI energy surface density, e.g. Sigma_E,central = (3/2)(M_HI/A_HI)(1-f_wings)(1-f_cold) sigma_central^2. Because Sigma_E is proportional to HI surface density, any regional property that correlates with gas content (radius, past star-forming history, total gas reservoir) will produce a correlation with Sigma_E even if the turbulent velocity dispersion sigma is unrelated to past star formation. The paper's own velocity-dispersion results are much weaker: in the full sample the strongest sigma correlation is rho=0.26 at t>500 Myr, and only after the young-star cut does sigma_central reach rho=0.44 at 200-500 Myr. The abstract's phrase 'atomic hydrogen turbulence measures' conflates energy surface density with turbulence; if the 100-500 Myr signal is driven by M_HI/A_HI rather than sigma^2, the conclusion that stellar feedback sustains HI turbulence on ~100 Myr timescales is not supported. This is load-bearing because it attacks the physical interpretation, not just the significance calculation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper combines CMD-derived star formation histories with VLA HI kinematics and WIYN SparsePak H-alpha kinematics for 26 low-mass galaxies, divided into 400 pc regions, to identify the lookback timescale over which stellar feedback correlates with present-day ISM turbulence. The central empirical result is that ionized gas velocity dispersion shows no correlation with star formation over 5-500 Myr, whereas HI energy surface density correlates most strongly with star formation 100-500 Myr ago (Spearman rho ~0.42-0.52 in the key bins). The authors interpret this as evidence that atomic gas turbulence is coupled to star formation on ~100 Myr timescales, consistent with a long turbulence dissipation timescale, and they contrast this with the shorter global timescale found in earlier work.","tokens_in":53918,"tokens_out":4362,"duration_ms":51740,"significance":"If the interpretation holds, the paper provides a direct observational constraint on the dissipation timescale of HI turbulence in low-mass galaxies and on the efficiency with which stellar feedback can sustain it. The sample is large and homogeneous, the SFH derivation is a well-established method, and the paper makes its regional SFH tables available as machine-readable products. The equal-region re-sampling test in Section 5.4 is a thoughtful robustness check, and the null result for H-alpha is reported honestly. The main risk is that the headline correlation is measured for HI energy surface density, which is proportional to HI column density, so the physical link to turbulence rather than gas content is not yet established.","major_comments":[{"comment":"The central correlation is between past star formation and Sigma_E,central = (3/2)(M_HI/A_HI)(1-f_wings)(1-f_cold) sigma_central^2, which is proportional to HI surface density. A correlation between past SFR and Sigma_E can therefore be produced entirely by spatial variations in gas column density, even if sigma_central is unrelated to star formation. The paper's own velocity-dispersion correlations are much weaker (rho=0.26 in the full sample at t>500 Myr, and rho=0.44 at 200-500 Myr only after the young-star cut), so the abstract's phrase 'atomic hydrogen turbulence measures' conflates energy surface density with turbulence. Please report partial Spearman correlations controlling for M_HI/A_HI, or equivalently show rank correlations separately for sigma and column density; without this, the physical interpretation that stellar feedback sustains HI turbulence on ~100 Myr timescales is not supported.","section":"Section 3.3, Eq. (5); Figures 6 and 8"},{"comment":"The Spearman tests treat the 961 regions as independent measurements, but regions within a galaxy share gas, stars, and large-scale kinematics, and the number of regions per galaxy ranges from 4 to 235. The equal-region re-sampling in Section 5.4 corrects for unequal galaxy weighting but does not remove intra-galaxy spatial autocorrelation, so the reported P-values likely overstate the significance of the correlations. A block bootstrap clustered by galaxy, or a mixed-effects model with galaxy as a random effect, would provide a more defensible significance statement. This is load-bearing because the sharpness of the 100-500 Myr timescale claim depends on the effective sample size.","section":"Section 3.4 and Section 5.4"},{"comment":"The paper concludes that 'the clearest correlation timescale is between the SF 100-500 Myr ago and the current HI energy surface density,' but it does not formally test whether rho at 100-200 Myr and 200-500 Myr is significantly larger than rho at the younger time bins. The reported P-values are each against the null of no correlation, not against the null of equal correlation across bins. Since the time bins share the same gas measurements and adjacent SFH bins are correlated, a bootstrap or permutation test of the difference in Spearman coefficients should be added. Without such a test, the claimed preferred timescale is a descriptive statement rather than a statistically supported inference.","section":"Section 5.1 and Section 8"}],"minor_comments":[{"comment":"The phrase 'To investigate the timescales of turbulence low-mass galaxies' is missing the word 'in'; the abstract should also clarify that the 'atomic hydrogen turbulence measures' include both velocity dispersion and energy surface density, which the main text distinguishes.","section":"Abstract"},{"comment":"The sentence 'Sections 5 an present the results' contains a typo and should read 'Sections 5 and 6 present the results.'","section":"Section 4"},{"comment":"The sentence 'As a results larger galaxies, more distant galaxies, and those with multiple HST pointings contribute more regions' should be split or reworded; 'As a results' is a typo.","section":"Section 5.4"},{"comment":"'There effect becomes more significant when the number of regions per galaxy is equal' should read 'This effect becomes more significant...'.","section":"Section 7.1"},{"comment":"The note ends with 'years ag and the error up and down' and should be 'years ago'; also the column-number descriptions appear off by one in several places and should be checked.","section":"Table 8 note"},{"comment":"Several captions contain 'Veloctiy' instead of 'Velocity' (e.g., Figures 5, 6, 8, 11) and '1σbootstrapping' is missing a space; these should be corrected.","section":"Figure captions"},{"comment":"The bootstrap in Section 3.4 is described as 3000 resamples, while Section 5.4 uses 2000 resamples for the equal-region test; the text should clarify which resampling is shown in which figure.","section":"Section 3.4 vs Section 5.4"},{"comment":"All appendix maps are labeled 'Figure 13' with different galaxy names; this is conventional for figure sets, but the captions should explicitly say 'Figure Set 13' so readers are not confused by the repeated number.","section":"Appendix figure set"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid observational study with a clear question and a large, well-characterized sample. The main concern is not the data quality or the absence of robustness checks, but whether the headline correlation can be attributed to turbulence rather than gas column density. I would recommend sending the revised manuscript back to a statistical referee or asking the authors to report partial correlations. The equal-region re-sampling is a good start, but it does not address within-galaxy autocorrelation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the 100-500 Myr correlation between past star formation and HI energy surface density is probably real, but the paper overreaches when the abstract calls it a correlation with \"atomic hydrogen turbulence measures.\" The velocity-dispersion correlations are much weaker than the energy-surface-density correlations, and since energy surface density is proportional to HI column density times sigma-squared, the headline result may be a gas-density correlation rather than a turbulence result.\n\nWhat is genuinely new: the sample is 26 low-mass galaxies instead of the 5 in Papers I and II, with CMD-based SFHs and 400 pc HI kinematics. The H-alpha null is a useful negative result, honestly caveated by fiber coverage. The equal-region resampling test shows IC 2574 and other large galaxies are not carrying the signal by themselves. The mass and sSFR splits are sensible, and the authors are transparent about the influence of NGC 3738 and NGC 5253. No circular reasoning: the SFHs and turbulence measures come from independent data.\n\nThe soft spots are real but manageable. First, the Spearman tests treat roughly 961 regions as independent. Regions inside one galaxy share gas, stars, and large-scale kinematics, and the per-galaxy region counts range from 4 to 235. The P-values therefore overstate significance. The equal-region resampling fixes the galaxy-weighting problem, not spatial autocorrelation. A per-galaxy blocking or mixed model would be the right check.\n\nSecond, and more load-bearing: Sigma_E is proportional to M_HI/A_HI times sigma^2. The full-sample velocity-dispersion correlation is weak (rho = 0.26 at t > 500 Myr), and only after the young-star cut does sigma_central reach rho = 0.44 at 200-500 Myr. So the paper has not isolated the turbulent-velocity contribution from the column-density contribution. The conclusion that stellar feedback sustains HI turbulence on ~100 Myr timescales is supported only if the correlation survives when sigma is separated from column density. The abstract should say \"HI energy surface density,\" not \"turbulence measures.\"\n\nWho this is for: observers and simulators working on feedback and HI turbulence in dwarfs. It deserves a serious referee, but the quantitative timescale claim needs to be softened and the autocorrelation issue addressed. I would send it to review with a request for those revisions, not desk-reject and not accept as is.","headline":"Good sample and a real effect, but the central correlation rides on HI column density more than the paper lets on; the velocity-dispersion evidence is weaker.","tokens_in":54603,"tokens_out":1839,"would_cite":true,"duration_ms":25039,"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":"This paper finds that atomic gas turbulence in low-mass galaxies is tied to star formation that happened 100-500 million years ago, not to recent star formation.","keywords":["star formation histories","HI turbulence","dwarf galaxies","stellar feedback","interstellar medium turbulence","turbulence dissipation timescale","400 pc regions","Spearman correlation"],"falsifier":"Measure the SFH-to-HI-energy correlation separately in inner and outer annuli of a face-on dwarf with a known disk scale-height gradient: the lag should increase with radius if it tracks $\\tau_d = L_D / v_{\\rm rms}$. If the peak lag does not track that prediction, or disappears when the Spearman test is redone with galaxy-level clustering taken into account, the central claim would be falsified.","tokens_in":1872,"feed_emoji":"🌌","tokens_out":2484,"duration_ms":89566,"temperature":0.7,"pith_summary":"This paper asks how long stellar feedback takes to stir the atomic gas in dwarf galaxies, and whether that timescale can explain how the gas keeps churning. The authors compare time-resolved star formation histories with current gas turbulence in 400-parsec patches across 26 low-mass galaxies. They find no correlation between ionized gas velocity dispersion and star formation over the past 5-500 Myr, but a consistent, significant correlation between the atomic hydrogen (HI) energy surface density and star formation activity 100-500 Myr ago. If this is right, stellar feedback from hundreds of millions of years ago is a primary driver of present-day HI turbulence, implying turbulence dissipates over roughly 100 Myr rather than the 5-15 Myr often assumed. That longer timescale would mean supernovae can maintain the observed turbulence with only a few percent energy transfer efficiency.","feed_headline":"Dwarf galaxy gas turbulence lags star formation by 100+ Myr","feed_subtitle":"Across 26 low-mass galaxies, 400-pc patches show atomic gas churn tracks star formation 100-500 Myr ago.","key_machinery":"The analysis is built on 400x400 parsec regions, each with an independently derived star formation history from color-magnitude diagram fitting, and each with HI turbulence measures from two techniques: second-moment maps and Gaussian superprofiles. A superprofile is a co-added, bulk-motion-corrected HI line profile whose core is fit by a Gaussian and whose high-velocity wings trace low-density, turbulent gas; from these come the central velocity dispersion, wing velocity dispersion, wing flux fraction, and the HI energy surface densities $\\Sigma_{E}$. Correlations between each SFR time bin and each turbulence measure are quantified with the Spearman rank coefficient $\\rho$, with bootstrap resampling providing uncertainties, and the physical interpretation uses the dissipation timescale identity $\\tau_d = L_D / v_{\\rm rms}$.","core_discovery":"The central claim, stated most directly in the conclusions, is that the clearest correlation timescale is between star formation 100-500 Myr ago and the current HI energy surface density. At the 400 pc scale, Spearman rank correlations between HI energy surface density and SFR in the 100-200 Myr and 200-500 Myr bins reach $\\rho \\gtrsim 0.4$, with low $P$-values; restricting to regions with recent star formation sharpens the signal, with $\\rho \\simeq 0.52$ in the 100-200 Myr bin. The HI velocity dispersion correlates more weakly, and the correlation in the oldest, unresolved time bin is largely driven by a few high-mass, disturbed galaxies. The ionized gas, measured from H$\\alpha$ line widths, shows no correlation at any probed timescale. A global analysis also shows HI turbulence correlated with star formation 100-300 Myr ago, alongside a milder 25-40 Myr signal consistent with an earlier study. The paper interprets this lag as the dissipation timescale of turbulence in the atomic interstellar medium, with $\\tau_d = L_D / v_{\\rm rms}$ linking the observed delay to the driving scale and velocity dispersion.","pith_inferences":["If the 100-500 Myr lag really is a dissipation timescale, then the same regional analysis applied to a single face-on dwarf with a measured disk scale-height gradient should show the lag increasing with radius, since $\\tau_d = L_D / v_{\\rm rms}$ grows with disk thickness; this radial prediction is left implicit in the paper.","The Spearman tests treat 400 pc regions as independent, but patches inside one galaxy share gas, stars, and large-scale kinematics, and galaxies contribute very unequal numbers of regions; a hierarchical model that accounts for within-galaxy clustering would test whether the 100-500 Myr signal survives at full significance.","If turbulence decays on ~100 Myr timescales, then stochastic variations in a galaxy's recent star formation history should imprint measurable scatter in HI energy density, and outliers such as the disturbed high-mass galaxies in this sample may be systems whose current turbulence still reflects an ancient burst.","The absence of an ionized-gas correlation may partly reflect the paper's insensitivity to the shortest timescales rather than a true physical decoupling, since H$\\alpha$-based SFRs trace star formation within the last ~10 Myr, which the CMD-based SFHs cannot resolve."],"forward_implications":["Previous searches for a link between HI turbulence and recent star formation may have been looking at the wrong timescale; the relevant star formation happened roughly 100-500 Myr ago.","Supernovae can plausibly sustain HI turbulence with energy injection efficiencies of a few percent, resolving the apparent need for efficiencies near or above 100 percent found under short dissipation timescales.","The feedback timescale should vary with galaxy properties: higher specific star formation rates show shorter correlation lags, while lower-mass and lower-sSFR systems trend toward longer lags within the 100-500 Myr range.","Local and global turbulence timescales differ, so galaxy-wide averages mix regions with different driving scales and dissipation rates, which may explain the broader correlation window seen in the full sample."],"supporting_citations":[{"why":"Established the 400 pc regional methodology and the 100-200 Myr HI turbulence correlation in four galaxies that this 26-galaxy sample extends.","marker":"Hunter et al. (2022)"},{"why":"Extended the regional analysis to Holmberg II with a 70-140 Myr timescale; the superprofile and H-alpha stacking methods are inherited from this paper.","marker":"Hunter et al. (2023)"},{"why":"Supplies the global superprofile construction and the 30-40 Myr global feedback timescale that the paper's global analysis directly compares to.","marker":"Stilp et al. (2013c)"},{"why":"Provides the dissipation timescale model $\\tau_d = L_D / v_{\\rm rms}$ and the argument that few-percent supernova efficiencies can sustain HI turbulence.","marker":"Bacchini et al. (2020a)"},{"why":"The MATCH code used to fit region-by-region star formation histories to color-magnitude diagrams.","marker":"Dolphin (2002)"},{"why":"Defines the SFH derivation, artificial-star completeness, and uncertainty treatment applied to every region.","marker":"McQuinn et al. (2010)"},{"why":"Introduced the superprofile decomposition whose central Gaussian and wings define the HI velocity dispersion and energy measures.","marker":"Ianjamasimanana et al. (2012)"},{"why":"Gives the $\\tau_d = L_D / v_{\\rm rms}$ dissipation timescale formulation used to translate the correlation lag into a physical timescale.","marker":"Mac Low (1999)"}],"fun_headline_variants":["Star formation's turbulence echo arrives 100 Myr later","Atomic gas turbulence trails star birth by 100-500 Myr","Dwarf gas turbulence maps to star formation 100-500 Myr ago","Star formation's 100-Myr echo in dwarf galaxy hydrogen"],"cache_read_input_tokens":56576,"weakest_assumption_plain":"The statistical tests treat each 400 pc region as an independent measurement, even though regions within a galaxy share gas, stars, and large-scale kinematics, and galaxies contribute very different numbers of regions, from 4 for Sextans B to 235 for IC 2574.","fun_headline_variants_meta":{"raw":{"variants":["Star formation's turbulence echo arrives 100 Myr later","Atomic gas turbulence trails star birth by 100-500 Myr","Dwarf gas turbulence maps to star formation 100-500 Myr ago","Star formation's 100-Myr echo in dwarf galaxy hydrogen"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000945,"raw_usage":{"total_tokens":4085,"prompt_tokens":1046,"completion_tokens":3039,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":662,"completion_tokens_details":{"reasoning_tokens":2965}},"tokens_in":662,"tokens_out":3039,"duration_ms":22672,"temperature":1.0,"reasoning_tokens":2965,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T14:17:26.456724+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the SFH-to-HI-energy correlation separately in inner and outer annuli of a face-on dwarf with a known disk scale-height gradient: the lag should increase with radius if it tracks $\\tau_d = L_D / v_{\\rm rms}$. If the peak lag does not track that prediction, or disappears when the Spearman test is redone with galaxy-level clustering taken into account, the central claim would be falsified.","supporting_citations":[],"review_version":1}