{"id":"95929892-aded-4c86-951d-1076f52f5b70","arxiv_id":"2501.10295","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":14,"one_line_summary":"GWTC-3 data are consistent with a binary black hole mass distribution that does not evolve with redshift, with the peak location and power-law slope nearly constant below z ~ 1.","lead":"Gravitational wave data from dozens of black hole mergers show no evidence that the masses of merging black holes change with cosmic time. The authors place some of the tightest limits yet on how much the 35-solar-mass peak and the power-law mass slope can vary below redshift 1.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed constancy below z≈1 is partly built into the model: Eq. (2) evolves hyperparameters via a sigmoid, and Table A1 restricts the transition midpoint to z≤0.8, so late or sharp evolution beyond that redshift cannot be represented; the conditional-prior comparison cannot reveal this.","rationale":"I have read the paper as a careful null-result population analysis: the authors survey a plausible parametric family of redshift-dependent mass spectra, report the regions where data are informative, and use conditional priors to separate data-driven constraints from prior extrapolation. The code and data are released, and the caveats in the text are mostly honest. The single most load-bearing weakness is the scope mismatch between the strongest claim and the model family. Every evolving hyperparameter is a sigmoid with transition midpoint capped at 0.8 (Eq. 2 and Table A1), while the headline claim extends to z≈1. The appendix discloses the cap, but the main text does not flag its consequence: within this model, evolution that begins after z≈0.8 is a priori impossible, so the upper edge of the constant below z≈1 statement is not a data-driven exclusion. The conditional-prior technique is a good idea, but it compares two quantities that share the same functional restriction, so it cannot detect a missing family. The concrete test I propose is directly feasible: extend the transition prior and, better, adopt a less parametric redshift dependence to see whether a late or sharp transition is allowed. This does not overturn the paper's modest null conclusion; it determines how broad that conclusion may honestly be stated. I therefore keep the reader's CONDITIONAL verdict, with the condition being exactly this robustness check and a corresponding re-scoping of the claim if needed.","tokens_in":19717,"tokens_out":5017,"duration_ms":54350,"concrete_test":"Re-run the Section 3.1 and 3.2 analyses after replacing the zbar ∼ U(0, 0.8) prior with zbar ∼ U(0, 2), using additional posterior samples or better Monte Carlo variance control to avoid the sampling difficulties reported in Appendix A. In the same run, or as a companion check, use a more flexible redshift dependence, for example piecewise-constant values of μm(z) and α(z) in redshift bins [0, 0.3, 0.6, 1.0] or a linear-in-log(1+z) drift with no upper cap on the transition, and recompute posteriors and a Bayes factor for evolution. If the extended model still keeps μm(z) and α(z) consistent with constant below z≈1, the concern is resolved; if it admits a transition above z≈0.8, the abstract and Section 5 should be re-scoped to no evidence for smooth evolution with midpoint below z≈0.8.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The headline constraint, that the 35 M⊙ peak location μm(z) and power-law slope α(z) remain approximately constant below z≈1, is obtained within a model where every evolving hyperparameter is a sigmoid (Eq. 2) whose transition midpoint has prior zbar ∼ U(0, 0.8) (Table A1). Appendix A explicitly states that wider zbar led to extreme sampling difficulties and was excluded. Consequently, the model has zero prior support for a transition centered above z=0.8. A sharp or late-onset change between z=0.8 and z=1, inside the region covered by the paper's strongest claim, cannot be expressed at all, and even a broad transition centered at z>0.8 is represented only approximately by pushing zbar to the boundary. The conditional-prior comparisons in Figs. 3 and 5 do not resolve this: both the posterior and the conditional prior inherit the same restricted sigmoid family, so they can only show that the data constrain the parameters within that family, not that the family is rich enough to test the claim at z≈1. The data at z>0.8 are sparse, which makes this a genuine coverage gap rather than a harmless technicality. The null result for smooth early-transition evolution is fine; the statement approximately constant below z≈1 is broader than the model family supports.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the binary black hole (BBH) population in GWTC-3 to test whether the primary mass distribution evolves with redshift. The authors model the primary mass distribution as a power-law continuum plus a Gaussian peak near 35 Msun, promote selected hyperparameters to sigmoid functions of redshift (Eq. 2), and perform hierarchical population inference with selection effects. They report no evidence that the Gaussian peak or the power-law continuum evolves, with the strongest constraints being that the peak location and power-law slope are approximately constant below z ~ 1. A companion analysis inverts the question and finds no mass dependence in the redshift evolution of the merger rate. The central conclusion is that current data do not require a redshift-dependent mass spectrum, although a redshift dependence remains possible.","tokens_in":20111,"tokens_out":5983,"duration_ms":64347,"significance":"If the no-evolution result holds, it is an important constraint on formation scenarios that predict strong metallicity-driven mass evolution or hierarchical-merger growth with redshift, and it directly contradicts the recent claims of Karathanasis et al. and Rinaldi et al. The analysis is generally careful: the hierarchical likelihood and Monte Carlo selection-function treatment are standard and well executed, and the public code and data are a strength. The conditional-prior diagnostic is a thoughtful way to separate low-redshift measurements from prior extrapolation. The main weakness is that the headline claim of constancy below z ~ 1 is partly inherited from the restricted sigmoid family and the prior zbar <= 0.8, so the paper overstates the coverage of its strongest constraint.","major_comments":[{"comment":"The model family cannot represent the late-onset or sharp evolution that the abstract's strongest claim is supposed to exclude. Equation (2) restricts every evolving hyperparameter to a single sigmoid, and Table A1 imposes zbar_Lambda ~ U(0, 0.8); Appendix A states that larger transition midpoints were excluded due to 'extreme sampling difficulties.' Consequently, the prior has zero support for a transition centered above z = 0.8, and a sharp change between z = 0.8 and z = 1 cannot be expressed at all, even approximately. The conditional-prior comparisons in Figs. 3 and 5 do not resolve this, because both the posterior and the conditional prior live in the same restricted sigmoid family; they can only show that the data constrain parameters within that family, not that the family is rich enough to test constancy at z ~ 1. I recommend either reframing the claim as 'no evidence within the sigmoid family with zbar <= 0.8' or adding a robustness test with a more flexible redshift dependence, such as binned or Gaussian-process models, that can represent late-onset evolution.","section":"Sec. 2.1, Table A1, Appendix A"},{"comment":"The claim that the 35 Msun peak location and power-law slope are 'constrained to remain approximately constant below z ~ 1' is broader than the analysis supports. The paper itself notes in Sec. 3.1 that 'the results at z ~ 1 are probably extrapolations from intermediate redshift.' Because the prior restricts transition midpoints to z <= 0.8 and the data at z > 0.8 are sparse, the credible intervals near z = 1 are not a direct measurement of constancy; they are an extrapolation within a family that cannot represent a transition beginning above 0.8. The conditional-prior comparison is not a substitute for a model check with an alternative family. The conclusion should be stated as 'current data show no evidence for evolution and exclude large smooth evolution with early onset,' with the z ~ 1 wording softened or explicitly qualified.","section":"Sec. 3.1, Abstract, Sec. 5"},{"comment":"The posterior for the mass-dependent merger-rate slope alpha_z(m1) shows a marked transition feature near m1 ~ 33 Msun in Fig. A4, which the paper interprets as a selection effect tied to the large number of events near 35 Msun. This interpretation is plausible, but it is not demonstrated. A simple injection-recovery check, or a comparison of the alpha_z(m1) posterior with the posterior obtained from a redshift-independent mass model, would strengthen the claim that the feature is not evidence for mass-dependent rate evolution. As written, the statement in Sec. 4 that data are consistent with universal alpha_z and zp is supported, but the paper should be more explicit that the 33 Msun feature is not yet interpretable.","section":"Sec. 4, Fig. A4"}],"minor_comments":[{"comment":"The heading 'DOES THE BLACK HOLE MASS SPECTRUM EVOL VE WITH REDSHIFT?' contains a typo: 'EVOL VE' should be 'EVOLVE'.","section":"Section 3 heading"},{"comment":"The sentence ending 'avoided by subsequent generations of exhibited (Liu & Bromm 2020)' appears garbled; 'exhibited' should likely be 'stars' or similar, and the sentence should be rewritten for clarity.","section":"Introduction, paragraph 2"},{"comment":"The caption contains the typo 'dotted magneta curves'; this should be 'dotted magenta curves'.","section":"Fig. 6 caption"},{"comment":"The exclusion of GW190814 and GW190917 as 'known population outliers' would be more persuasive with a brief sensitivity check showing that including them does not change the main conclusions, especially for the low-mass truncation Mmin.","section":"Appendix B"},{"comment":"The notation dN/dlambda(lambda_i) is slightly confusing because lambda is used for individual-event parameters while Lambda denotes hyperparameters; a distinct symbol for the per-event parameters in Eq. (9) would improve readability.","section":"Eq. (9) and surrounding text"}],"recommendation":"major_revision","confidential_remarks":"The paper is well within the scope of the journal and the central inference is sound within its chosen model family. The reason for major revision is the gap between the strongest claim in the abstract and what the sigmoid family with zbar <= 0.8 can actually test. This is fixable by softening the claim or adding a robustness analysis; I do not see a circularity problem, as the model does not assume the null outcome. The public code and data availability are commendable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid, honest null result on redshift evolution of the BBH mass distribution, worth refereeing, but the abstract's 'approximately constant below z≈1' is a bit stronger than the model actually supports.\n\nThe genuinely new bit is the systematic survey: they let each hyperparameter of the Gaussian peak and the power-law continuum vary as a sigmoid in redshift, and they also flip the question and let the merger-rate evolution vary with primary mass. Previous null results (Fishbach, van Son, Ray, Heinzel) used other parameterizations, so this is a useful cross-check. The paper also does conditional-prior comparisons to separate what the data actually constrain from what is prior extrapolation. That is a good practice, and the posteriors show that the peak location and power-law slope really are constrained by data, not just by the prior. Code and data are on GitHub and Zenodo, and the prose is appropriately cautious about not overclaiming.\n\nThe soft spot is the z̄ ≤ 0.8 prior on the sigmoid transition midpoint, noted in Appendix A. Because of that truncation, the model cannot represent a transition centered above z ≈ 0.8 at all. So the statement that µm(z) and α(z) are 'approximately constant below z ≈ 1' is partly built into the model family. The conditional-prior comparison doesn't fix this: both the posterior and the conditioned prior live in the same restricted family. The data at z > 0.8 are sparse, so this is a real coverage gap, not a technicality. I would not call the central null result wrong — within the sigmoid family, the data really do disfavor early smooth evolution — but the 'below z ≈ 1' phrasing should be softened or backed by robustness tests with, say, linear or broken-power-law evolution shapes.\n\nMinor: GW190814 and GW190917 are excluded as outliers; that is standard but worth keeping in mind when interpreting the bounds. And the paper says it 'disfavors' a shrinking Gaussian peak, while the abstract is more agnostic about peak height; those two phrasings should be reconciled.\n\nBottom line: this is a careful, reproducible analysis that deserves peer review. The authors should be asked to address the transition-prior coverage gap before publication, but I don't think it invalidates the main no-evolution conclusion within the modeled class.","headline":"A careful null result on BBH mass-redshift evolution whose headline claim slightly overstates what the sigmoid model can actually test.","tokens_in":20591,"tokens_out":2710,"would_cite":true,"duration_ms":26263,"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":"This paper claims that the binary black hole mass distribution shows no evidence of evolving with redshift in the GWTC-3 catalog, with the 35-solar-mass peak and power-law slope constrained to stay nearly constant below z≈1.","keywords":["gravitational waves","binary black holes","mass distribution","redshift evolution","GWTC-3","population inference","hierarchical Bayesian","35 solar-mass peak"],"falsifier":"A reanalysis of the same GWTC-3 events with the sigmoid midpoint prior extended to $z \\gtrsim 2$, or with a non-parametric binning of the mass spectrum in redshift, that recovers a $>3\\sigma$ shift in the $35\\,M_\\odot$ peak location or in the power-law slope between $z=0$ and $z=1$ would falsify the claim that these features are constrained to be approximately constant.","tokens_in":19532,"feed_emoji":"🔭","tokens_out":6560,"duration_ms":56459,"temperature":0.7,"pith_summary":"The paper asks whether the masses of merging black holes change as we look deeper into cosmic history, using the 90 binary black hole mergers in the GWTC-3 catalog. It focuses on two well-established features of the primary mass spectrum — a Gaussian excess of approximately 35 solar-mass black holes and a broad power-law continuum from about 10 to above 80 solar masses — and allows each feature's parameters to drift smoothly with redshift. The central finding is that neither feature shows statistically significant evolution: the mean location of the 35 solar-mass peak and the slope of the power-law continuum are constrained to stay approximately constant below redshift z≈1. The data remain consistent with a stationary mass spectrum, although they do not rule out evolution in the peak's height or in the minimum and maximum black hole masses. If correct, this result undercuts recent claims of strong mass-redshift evolution and sharpens the astrophysical question of why the mass spectrum looks so stable.","feed_headline":"No redshift evolution found in black hole merger masses","feed_subtitle":"The 35-solar-mass peak and power-law slope stay put below z≈1, contradicting recent claims of strong evolution.","key_machinery":"The analysis is built on a hierarchical Bayesian population model in which the primary mass distribution is a sum of a power law and a Gaussian peak (Eq. 1), and every hyperparameter of interest is promoted to a smooth sigmoid function of redshift (Eq. 2), with low- and high-redshift asymptotes, a transition midpoint, and a transition width. Selection effects are handled through injection-recovery Monte Carlo averages in the detection expectation term. The load-bearing comparison is the 'conditional prior': for each redshift-varying parameter, the authors compare the full posterior to a prior distribution conditioned on the measured posterior at $z=0$, which shows how much high-redshift behavior is actually informed by high-redshift events rather than extrapolated from local measurements. It is this device that lets them claim the constancy of the peak location and power-law slope is a data-driven result, while the apparent freedom in peak height is a prior effect.","core_discovery":"On its own terms, the paper establishes a null result with quantitative teeth: in the LIGO-Virgo-KAGRA GWTC-3 catalog, there is no evidence that the binary black hole primary mass distribution varies with redshift. The two most precisely measured features — the location of the $35\\,M_\\odot$ peak and the slope of the power-law continuum — are bounded to remain approximately constant below $z\\approx 1$, and the authors show by comparing posteriors to priors conditioned on low-redshift measurements that this constraint is driven by the data rather than by the prior. At the same time, the analysis is careful not to overclaim: evolution in the height of the peak, the minimum mass, or the maximum mass remains possible, and a redshift-dependent mass spectrum is neither ruled out nor required. The paper further inverts the question and finds that the merger rate's redshift evolution shows no mass dependence, consistent with all mass ranges merging in lockstep.","pith_inferences":["The sigmoid prior restricting transition midpoints to $z\\leq 0.8$ means the null conclusion is conditional on evolution being a single smooth step at low redshift; a more flexible non-parametric redshift model could still find structure that this parameterization smooths over.","If the mass spectrum is truly stationary, the astrophysical implication extends beyond metallicity: it would constrain the delay-time distribution of binary black hole mergers to be long enough to homogenize formation epochs across a Hubble time.","A testable extension is to apply the same conditional-prior diagnostic to the next catalog and check whether the allowed drift in peak height narrows, since the current upper bound on peak growth is set by the prior, not the data.","The authors' trick of conditioning priors on the best-measured redshift (or mass) could be applied to other population questions, such as spin evolution with redshift, to distinguish true measurements from prior extrapolations."],"forward_implications":["If the null result holds, theoretical models predicting large shifts in black hole masses between $z=0$ and $z=1$ (e.g., from metallicity evolution or hierarchical mergers) must be reconciled with the data.","A stationary mass spectrum out to $z\\approx 1$ suggests either long delay times between formation and merger that wash out progenitor metallicity trends, or that metal-poor star formation remains significant at late cosmic times.","The constraints on the peak location and power-law slope can inform dark-siren measurements of the Hubble constant, where the mass-redshift relation is a key degeneracy.","Future catalogs from the O4 observing run and beyond, which push the detection horizon deeper, will determine whether the mass spectrum truly remains constant or evolves at $z>1$.","The mass-independent merger-rate evolution found in Section 4 indicates that if multiple formation channels exist, their combined redshift histories synchronize across mass scales."],"supporting_citations":[{"why":"Supplies the GWTC-3 event catalog and the injection sets used to compute selection effects.","marker":"Abbott et al. 2023a"},{"why":"Provides the LVK population-inference results and catalog selection criteria that the analysis builds on.","marker":"Abbott et al. 2023b"},{"why":"Supplies the power-law plus Gaussian baseline parameterization of the primary mass distribution.","marker":"Talbot & Thrane 2018"},{"why":"Provides the redshift-dependent merger-rate model adopted in the likelihood.","marker":"Fishbach et al. 2018"},{"why":"The claim of a large mass shift with redshift that this paper's null result directly contradicts.","marker":"Karathanasis et al. 2023"},{"why":"A recent analysis reporting even stronger redshift evolution, used as the key counterexample to the null conclusion.","marker":"Rinaldi et al. 2024"},{"why":"Earlier search for redshift-dependent maximum mass that also found a null result, providing methodological precedent.","marker":"Fishbach et al. 2021"},{"why":"Flexible non-parametric search for mass-redshift correlation that also found none, supporting the consistency check.","marker":"Ray et al. 2023"}],"fun_headline_variants":["No redshift evolution in black hole merger masses","GWTC-3: no mass evolution in merging black holes","No redshift trend in 35-solar-mass peak and slope","No evidence for redshift evolution in black hole masses"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The null result rests on the assumption that any redshift evolution in a mass-spectrum hyperparameter takes the form of a single sigmoid step whose midpoint is forced to lie at $z \\leq 0.8$, a restriction imposed because wider ranges caused extreme sampling difficulties; if the true evolution is not smooth-sigmoid or starts beyond $z \\approx 0.8$, the analysis could miss it.","fun_headline_variants_meta":{"raw":{"variants":["No redshift evolution in black hole merger masses","GWTC-3: no mass evolution in merging black holes","No redshift trend in 35-solar-mass peak and slope","No evidence for redshift evolution in black hole masses"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000951,"raw_usage":{"total_tokens":4105,"prompt_tokens":1041,"completion_tokens":3064,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":657,"completion_tokens_details":{"reasoning_tokens":3000}},"tokens_in":657,"tokens_out":3064,"duration_ms":23186,"temperature":1.0,"reasoning_tokens":3000,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T19:15:55.292502+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A reanalysis of the same GWTC-3 events with the sigmoid midpoint prior extended to $z \\gtrsim 2$, or with a non-parametric binning of the mass spectrum in redshift, that recovers a $>3\\sigma$ shift in the $35\\,M_\\odot$ peak location or in the power-law slope between $z=0$ and $z=1$ would falsify the claim that these features are constrained to be approximately constant.","supporting_citations":[],"review_version":1}