{"id":"68e7d3cb-eff0-4c04-adf3-4961e44959dc","arxiv_id":"1908.08423","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Bar size in local spiral galaxies is mainly set by the galaxy's disc scale length, with an extra dependence on stellar mass that turns on only above about 10^10.1 solar masses.","lead":"Using infrared images of nearby spiral galaxies, this paper finds that bar size depends on galaxy mass in two very different ways below and above a mass of about 10^10 solar masses. Because bars are harder to see in distant galaxies, these new scaling relations help correct how astronomers measure bar frequencies across cosmic time.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The novel high-mass residual dependence of bar size on M* after controlling for h or Re (Eq. 4, Table 4) is vulnerable to correlated measurement errors, because M*, Re, h, and a_bar are all measured from the same S4G 3.6 um images and predictor uncertainties are not propagated.","rationale":"Good-faith reading: this is a careful, largely reproducible empirical study with a clear sample definition, public code and data, and model comparison via AIC and bootstrap MSE. The bimodality in Fig. 1/Table 2 is well supported internally and is not the main point of doubt. The truly novel claim is that after removing galaxy or disc size, an extra M* dependence appears only above log M* ~ 10.1. That claim rests entirely on regressions where the predictors and the response all derive from the same 3.6 um images. The reader's weakest-assumption identification is exactly right and is load-bearing because the shared-measurement-error channel has a plausible sign to create a spurious positive residual coefficient: a systematic fluctuation in total 3.6 um flux could raise both measured M* and measured h, while a longer bar also boosts 3.6 um flux and can affect structural fits. A related but secondary issue is that h comes from multi-component decompositions that include a fitted bar component; degeneracy between bar length and disc scale length could strengthen the a-h relation, but this mainly affects the size channel rather than the residual mass channel. The AIC and MSEpred comparisons cannot distinguish these artifacts because all models use the same noisy predictors. Therefore the CONDITIONAL verdict is appropriate: the analysis is sound in its internal logic, but the central new coefficient needs an independent-measurement or Monte-Carlo covariance check before the physical interpretation is accepted.","tokens_in":17885,"tokens_out":6630,"duration_ms":81739,"concrete_test":"Using the public repository data, refit Eq. 4 while drawing M*, Re, and h from their reported uncertainties with an assumed covariance based on shared 3.6 um photometry (e.g., perturb total flux and scale length together). If the high-mass beta2 drops below about 0.1, the residual mass dependence is a correlated-error artifact. Better still: for the S4G spirals overlapping SDSS or GAMA, replace M* with independent SED-based stellar masses and Re/h with structural fits from optical images; refit Table 4 and check whether beta2 remains near 0.35 rather than collapsing toward zero.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central new result is beta2 in Eq. 4/Table 4: at log M* > 10.1, bar size depends on stellar mass even after removing h (beta2 = 0.35 +/- 0.06) or Re (beta2 = 0.47 +/- 0.09). This is the claim that galaxy size alone does not explain bar size at high masses. The paper's evidence for it is the residual trends in Fig. 3 and the AIC/MSE comparisons in Table 5. The problem is that M*, Re, and h are not independent predictors: stellar masses (Munoz-Mateos et al. 2015), half-light radii, and disc scale lengths (Salo et al. 2015) all come from the same S4G 3.6 um images, and bar sizes are measured from those same images. A correlated error between measured h and measured M* -- for example, a luminosity/scale-length fluctuation that raises both -- will produce a spurious positive beta2 when M* is added after h. The 2000 bootstrap resamples resample galaxies, not measurement-noise realizations, so they cannot detect this. The assumed 10% bar-size uncertainty enters only the AIC likelihood and is uniform; predictor errors are effectively set to zero. Until this covariance is bounded, the high-mass residual mass dependence is not established as a physical scaling.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents an analysis of bar sizes (deprojected semi-major axes) in the S4G sample of nearby spiral galaxies, using volume- and mass-limited subsamples to measure how bar size depends on stellar mass, galaxy half-light radius, disc scale length, gas fraction, and Hubble type. The central claims are: (i) the bar-size--stellar-mass relation is bimodal, being nearly flat (slope ~0.1) below log M* ~ 10.1 and steep (slope ~0.6) above; (ii) bar size correlates more strongly with disc scale length (slope ~0.8) and half-light radius (slope ~0.45) than with stellar mass; (iii) after accounting for galaxy size, an additional residual dependence on stellar mass persists only for galaxies above the break mass, with slopes 0.35 (with h) and 0.47 (with Re) from the multivariate fits in Eq. (4) and Table 4; (iv) bar size shows no residual dependence on gas mass fraction or Hubble type after mass/size corrections; and (v) galaxy size can be modeled as a function of stellar mass, bar presence, and bar size. The analysis uses careful sample selection, bootstrap uncertainties, AIC comparison, and out-of-sample prediction errors, and the code and data are publicly available.","tokens_in":18267,"tokens_out":3102,"duration_ms":34714,"significance":"If the central claims hold, this is a valuable empirical contribution to understanding bar formation and growth: it identifies disc scale length as the dominant predictor of bar size and reveals a high-mass residual mass dependence that is not explained by galaxy size alone. The bimodality of the bar-size--mass relation, if real, is a new scaling relation that any model of bar evolution must reproduce, and the observed absence of a present-day gas-fraction dependence constrains theoretical expectations. The paper is careful in its sample selection (volume/mass-limited, inclination cuts), uses bootstrap resampling for parameter uncertainties, compares models with AIC and out-of-sample prediction errors, and ships reproducible code and data. These are genuine strengths. However, the novelty and importance of the result rest heavily on the high-mass residual mass dependence, which is the most vulnerable to systematic uncertainties in the measured predictors.","major_comments":[{"comment":"The central new result -- that bar size depends on stellar mass even after controlling for Re or h at log M* > 10.1 (beta2 = 0.35 +/- 0.06 with h, 0.47 +/- 0.09 with Re) -- is not robustly established because the predictor variables M*, Re, and h are all measured from the same S4G 3.6 micron images (Muse et al. 2015; Salo et al. 2015) and are treated as error-free. Correlated measurement errors among M*, Re, and h (e.g., a fluctuation that raises both the inferred stellar mass and the inferred disc scale length) can generate a spurious positive residual slope of bar size versus M* after removing h or Re. The 2000 bootstrap resamples resample galaxies, not measurement-noise realizations, so they cannot detect this effect. I request an explicit test: either propagate the published uncertainties on M*, Re, and h (including their covariance), or run a Monte Carlo simulation that adds correlated noise to these predictors and show that the recovered beta2 remains consistent with the reported values. Until this is done, the high-mass residual mass dependence should be treated as tentative rather than established, and this caveat should be stated in the abstract and summary.","section":"Section 3.3, Eq. (4), Table 4"},{"comment":"The bimodality of the bar-size--stellar-mass relation is quantified with a broken-linear fit that has an abrupt change in slope, and the model comparison is only between a single straight line and this broken linear form. A smooth transition (e.g., a bent power law with a finite transition width, or a low-order spline) might fit the data as well, which would change the physical interpretation from a distinct break to a gradual steepening. Moreover, the reported break mass (log M* ~ 10.1-10.2) is close to the break seen in the size--mass relation itself (Appendix A), so part of the apparent bimodality could be a projection of that underlying relation. The multivariate fits in Section 3.3 partially address this concern, but the paper should either test alternative functional forms or explicitly justify why an abrupt broken line is the physically preferred model.","section":"Section 3.1, Eq. (1), Table 2"}],"minor_comments":[{"comment":"The assumption of a constant 10% fractional uncertainty on bar sizes is used to compute all AIC values. Since the same uncertainty is used for all fits, it does not affect relative AIC comparisons among the models, but the absolute AIC values are not meaningful if the true uncertainty differs substantially. It would be useful to state this explicitly and perhaps quote qualitative conclusions based on AIC deltas rather than absolute values.","section":"Section 2.1.1"},{"comment":"The statement that 'there is very little correlation' (Spearman r = 0.08, P = 0.11) is slightly undercut by the weak turn-up for very gas-rich galaxies that is mentioned in the same paragraph; the paper correctly notes this is in the opposite sense from the models, but the wording should be tightened to avoid appearing inconsistent.","section":"Section 4.1"},{"comment":"In the fits of galaxy size versus stellar mass and bar size, the binary variable B is defined as 0 for unbarred and 1 for barred galaxies. The model is fitted to all galaxies in the Main Spiral Sample, but for unbarred galaxies the bar size is undefined; it would be helpful to clarify that the product B * log(avis) is set to zero for unbarred galaxies, as the mathematical notation in Eq. (5-6) alone does not make this explicit.","section":"Section 5, Eq. (5-6)"},{"comment":"The paper is generally well written, but there are minor typos and formatting issues (e.g., 'Insitut' in the author affiliation, and some figure captions that use 'fi' ligatures). These do not affect the science.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid empirical study with a clear and useful result, but the novel high-mass residual dependence of bar size on stellar mass is the key claim and it is not yet protected against correlated measurement-error scenarios. The author has a strong track record in this area and the reproducibility package is a plus. I would be comfortable accepting after the error-propagation or Monte-Carlo test is added and the interpretation is adjusted accordingly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is a careful, well-documented empirical study of bar sizes in ~370 S4G spirals. What's genuinely new: the bar size–stellar mass relation is not a single power law. It flattens below log M*≈10.1 (slope ~0.1) and steepens above (slope ~0.6). The paper shows bar size tracks disc scale length more tightly (a∝h^0.8) and that, once size is accounted for, there's a residual mass dependence only at high masses. That residual term is the novel and interesting claim.\n\nThe analysis is solid in the ways that matter: volume- and mass-limited subsamples, bootstrap uncertainties, AIC and out-of-sample prediction errors, and a public GitHub repository with data and notebooks. The bimodality itself is visually clear and survives model comparison; I'd be surprised if that part is wrong.\n\nThe soft spot is the residual high-mass mass dependence (beta2 in Eq. 4/Table 4). M*, Re, h, and a_bar all come from the same 3.6 μm images. The predictors are treated as error-free, and the bootstrap resamples galaxies, not measurement noise. A correlated error between h and M*—say a fluctuation that raises both—will mimic a positive beta2 when M* is added after h. The paper doesn't bound that covariance. The 10% bar-size uncertainty is uniform and only enters the AIC. This doesn't kill the result, but it means the most novel claim is not yet established as physical. Notably, the break mass (~10.1) coincides with the size–mass break in the same data, so the residual could be a refraction of that.\n\nThe rest—no gas-fraction dependence, Hubble-type as side-effect—is fine, if less exciting. The causality discussion is appropriately speculative.\n\nI'd send this to a serious referee. The paper deserves time; the central findings are likely real, and the covariance issue is fixable with a sensitivity analysis (e.g., noise-injection or instrumental-variable style test). A referee should ask for that, not reject the paper.\n\nFor who: anyone working on bar scaling relations, galaxy size–mass relations, or survey completeness. Good reading-group material for a stats-meets-galaxies discussion.\n\nMy recommendation: engage it, but push on the correlated-error front before quoting the high-mass residual.","headline":"New bimodal bar–mass scaling from S4G with a real but covariance-vulnerable residual mass term; worth refereeing, but the high-mass claim needs a correlated-error test.","tokens_in":18768,"tokens_out":2325,"would_cite":false,"duration_ms":24714,"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":"Bar size in spiral galaxies is bimodal in stellar mass and is set mostly by the disc scale length, with an extra mass dependence only in the most massive galaxies.","keywords":["galaxy bars","bar sizes","spiral galaxies","stellar mass","disc scale length","galaxy scaling relations","bar detectability","S4G"],"falsifier":"Re-fit the bar-size relations using independent size and mass estimates for the same galaxies (for example, effective radii from optical imaging or masses from dynamics rather than photometry) and check whether the residual slope of bar size versus stellar mass at fixed size remains about $0.35$--$0.47$ for $\\log (M_\\star/M_\\odot) > 10.1$; if it drops to zero, the extra mass dependence is a correlated-error artifact.","tokens_in":17682,"feed_emoji":"🌌","tokens_out":10452,"duration_ms":93498,"temperature":0.7,"pith_summary":"The paper argues that bar size in spiral galaxies is not a single smooth function of stellar mass. It is nearly constant below about $10^{10.1}\\,M_\\odot$ ($a \\propto M_\\star^{0.1}$) and steepens sharply above it ($a \\propto M_\\star^{0.6}$). Bar size correlates most tightly with the exponential disc scale length ($a \\propto h^{0.8}$), and the correlation with stellar mass at low masses is just a side effect of bigger galaxies having bigger discs. Above the break mass, however, stellar mass matters in its own right: at fixed galaxy size, the more massive galaxy has the longer bar. Because bar size controls whether distant surveys can detect bars at all, this bimodal scaling affects measurements of bar frequency at high redshift as well as theoretical models of bar growth.","feed_headline":"Spiral bar size splits at the 10.1 solar-mass break","feed_subtitle":"Bar length scales as disc size to the 0.8 power, with extra growth only in galaxies above roughly 10 billion solar masses.","key_machinery":"The load-bearing element is a broken-linear (bimodal) power-law fit in the $\\log a_{\\rm vis}$--$\\log M_\\star$ plane, extended to multi-variable fits of $\\log a_{\\rm vis}$ against $\\log M_\\star$ plus $\\log R_e$ or $\\log h$ (Eqns. 1 and 4). The break mass $M_{\\rm brk}$ is a free parameter, so the bimodality is not imposed by hand; it emerges from the data and is checked by bootstrap resampling, Akaike information criterion comparisons, and bootstrap estimates of prediction error. Residual plots of the size-only fits against stellar mass give the decisive visual evidence that an extra mass term appears only above the break.","core_discovery":"The central discovery, stated on the paper's own terms, is that the bar-size--stellar-mass relation in local disc galaxies is bimodal, with a break at $\\log (M_\\star/M_\\odot) \\simeq 10.1$--$10.2$: below the break $a_{\\rm vis} \\propto M_\\star^{0.1}$, above it $a_{\\rm vis} \\propto M_\\star^{0.6}$. Bar size is an even stronger function of galaxy size, $a_{\\rm vis} \\propto R_e^{0.45}$ and $a_{\\rm vis} \\propto h^{0.8}$, and the multi-variable fits of Table 4 show that once both size and mass are included, the residual mass slope below the break is consistent with zero while above the break it remains significant ($\\beta_2 \\simeq 0.35$--$0.47$). The author reads this as evidence that disc scale length is the primary structural determinant of bar size, that a separate mass-driven growth channel operates only in galaxies above roughly $10^{10.1}\\,M_\\odot$, and that the classical correlations with gas fraction and Hubble type are not independent of mass and size.","pith_inferences":["If the bimodal relation is real, then measurements of bar fraction versus redshift need a mass-dependent visibility correction that itself may evolve with the break mass; a constant-size detection threshold will bias high-redshift samples toward the most massive bars.","The $a \\propto h^{0.8}$ scaling invites a direct test in simulations: bar length should be a predictable multiple of the initial disc scale length, and mergers or disc heating that change $h$ should change bar size in the same proportion.","A concrete check that goes beyond the paper's fits is to redo the multi-variable fits with independent sizes and masses (for example, optical effective radii and dynamical masses); this would separate a physical mass effect from correlated photometric errors.","The null gas-fraction result does not necessarily mean gas never slows bar growth; it may mean present-day atomic gas fraction is a poor proxy for the gas content at the epoch of bar formation, a distinction simulations with time-evolving gas fractions could settle."],"forward_implications":["Bar-detectability corrections for large or high-redshift surveys cannot assume a single bar-size--mass scaling; the break at $\\log (M_\\star/M_\\odot) \\simeq 10.1$ means low-mass and high-mass bars fade out at different rates with distance.","Disc scale length, not total stellar mass, is the primary structural control on bar size, so models of bar formation and growth should aim to reproduce $a \\propto h^{0.8}$.","Above the break, a more massive galaxy of the same size will host a longer bar, pointing to a mass-dependent growth process that operates only in the most massive discs.","The absence of any residual bar-size dependence on present-day gas fraction or Hubble type means those classical correlations are not separate physical drivers.","Barred galaxies are more extended than unbarred galaxies of the same mass, and larger bars mark larger discs, so bar presence and bar size can improve predictions of galaxy size from stellar mass."],"supporting_citations":[{"why":"Defines the Spitzer Survey of Stellar Structure in Galaxies that supplies the imaging for every galaxy and bar measurement.","marker":"Sheth et al. 2010"},{"why":"Supplies the visual bar semi-major axes and bar position angles from which deprojected bar sizes are computed.","marker":"Herrera-Endoqui et al. (2015)"},{"why":"Supplies the half-light radii and exponential disc scale lengths used as galaxy-size predictors.","marker":"Salo et al. (2015)"},{"why":"Supplies the distances and stellar masses that define the parent sample and mass bins.","marker":"Muñoz-Mateos et al. (2015)"},{"why":"Provides the morphological classifications identifying barred galaxies and Hubble types.","marker":"Buta et al. (2015)"},{"why":"Established the earlier bar-size--stellar-mass correlation in the same survey that this paper shows is bimodal.","marker":"Díaz-García et al. (2016a)"},{"why":"The companion paper whose bar-detectability argument makes bar-size scaling important for survey comparisons.","marker":"Erwin 2018"},{"why":"Found barred galaxies have larger disc scale lengths at fixed mass, supporting the size-prediction analysis.","marker":"Sánchez-Janssen & Gadotti (2013)"},{"why":"Simulations predicting higher gas fractions produce shorter bars, the baseline for the gas-fraction null result.","marker":"Athanassoula et al. (2013)"}],"fun_headline_variants":["Bar size vs stellar mass is bimodal, break at 10^10.1 Msun","Disc scale length sets bar size; mass adds only above 10^10.1 Msun","Spiral bar length: disc scale dominates, mass matters above 10^10.1","Bar size bimodal: disc scale key, extra mass effect above 10^10.1"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis treats stellar mass, half-light radius, and disc scale length as error-free predictors, even though all three are measured from the same near-infrared images of the same galaxies; if correlated measurement errors generate the residual high-mass dependence of bar size on mass, the most novel conclusion would not survive.","fun_headline_variants_meta":{"raw":{"variants":["Bar size vs stellar mass is bimodal, break at 10^10.1 Msun","Disc scale length sets bar size; mass adds only above 10^10.1 Msun","Spiral bar length: disc scale dominates, mass matters above 10^10.1","Bar size bimodal: disc scale key, extra mass effect above 10^10.1"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001062,"raw_usage":{"total_tokens":4546,"prompt_tokens":1128,"completion_tokens":3418,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":744,"completion_tokens_details":{"reasoning_tokens":3320}},"tokens_in":744,"tokens_out":3418,"duration_ms":24098,"temperature":1.0,"reasoning_tokens":3320,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:42:42.669989+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-fit the bar-size relations using independent size and mass estimates for the same galaxies (for example, effective radii from optical imaging or masses from dynamics rather than photometry) and check whether the residual slope of bar size versus stellar mass at fixed size remains about $0.35$--$0.47$ for $\\log (M_\\star/M_\\odot) > 10.1$; if it drops to zero, the extra mass dependence is a correlated-error artifact.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the Spitzer Survey of Stellar Structure in Galaxies that supplies the imaging for every galaxy and bar measurement."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the visual bar semi-major axes and bar position angles from which deprojected bar sizes are computed."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the half-light radii and exponential disc scale lengths used as galaxy-size predictors."},{"cited_title":"C., et al., 2015, , 219, 3","cited_arxiv_id":null,"evidence_quote":"Supplies the distances and stellar masses that define the parent sample and mass bins."},{"cited_title":"J., et al., 2015, ApJS, 217, 32","cited_arxiv_id":null,"evidence_quote":"Provides the morphological classifications identifying barred galaxies and Hubble types."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The companion paper whose bar-detectability argument makes bar-size scaling important for survey comparisons."},{"cited_title":"A., 2013, , 432, L56","cited_arxiv_id":null,"evidence_quote":"Found barred galaxies have larger disc scale lengths at fixed mass, supporting the size-prediction analysis."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Simulations predicting higher gas fractions produce shorter bars, the baseline for the gas-fraction null result."}],"review_version":1}