{"id":"42cba051-f609-49ae-918f-be17ae5794e1","arxiv_id":"1909.01565","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":11,"one_line_summary":"Using Bayesian analysis of stellar ages, young stellar object counts, and the Milky Way's star formation rate, the paper concludes that bound star clusters form through sustained gas accretion with low star formation efficiency, not by global collapse or by star formation efficiency increasing…","lead":"This paper tests competing explanations for how dense groups of stars, called bound star clusters, form in our galaxy. It finds that the best match to observations is a 'conveyor belt' picture, where gas flows in steadily while stars form slowly, rather than a sudden global collapse.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"ATLASGAL clump-to-progenitor identification is the load-bearing assumption; if most ATLASGAL clumps are unbound, the global SFR argument against GC/GCD (Eq. 56) loses its target and the conveyor-belt conclusion is unsupported.","rationale":"The reader's weakest assumption identifies exactly the same load-bearing concern: the ATLASGAL clump sample is assumed to be representative of direct bound-cluster progenitors, which underpins both the epsilon_ff comparison (Section 3.2.1) and the Section 3.4.2 global SFR calculation (equation 56) that rules out GC and GCD. I agree that this is the pivot. The paper's own caution in Section 3.4.1 acknowledges the uncertainty. My stress-test confirms that if most ATLASGAL clumps are unbound (as suggested by the 5-10% bound fraction of star formation in spiral galaxies cited in the introduction), the global SFR overproduction argument loses its target, and the conveyor-belt conclusion would not follow from this analysis alone. The Bayesian model-comparison is careful and the code is public, so the conditional verdict is appropriate; the concern does not invalidate the paper but does prevent an unconditional acceptance. Since my reading aligns with the reader's conditional verdict and identified weakness, no verdict change is needed.","tokens_in":38841,"tokens_out":7305,"duration_ms":72216,"concrete_test":"Determine the bound fraction of ATLASGAL clumps by measuring their virial parameters (ratio of kinetic to gravitational energy) and identify the subset whose binding energy exceeds their turbulent kinetic energy. Recompute the GC/GCD SFR (equation 56) using only the total mass of that bound subset, leaving all other analysis unchanged. If the bound mass is less than ~2e6 Msun (so that the predicted SFR drops below ~2 Msun/yr), the global SFR argument no longer rules out GC/GCD, and the conveyor-belt conclusion loses its primary support. Alternatively, compare the epsilon_ff distributions of bound versus unbound ATLASGAL clumps; if they differ significantly, the sample is not representative of bound-cluster progenitors.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim depends on identifying the ATLASGAL clump population (Heyer et al. 2016) as direct progenitors of bound clusters like the ONC and NGC 6530. This identification is load-bearing in two places: (1) Section 3.2.1 states that the ATLASGAL sample 'very likely represents a survey of YSOs in objects that will become clusters like NGC 6530 or the ONC', which makes the observed epsilon_ff distribution a constraint on cluster-formation models; and (2) Section 3.4.2 uses the total mass of ATLASGAL clumps, Mtot ~ 1e7 Msun, in equation (56) to argue that GC and GCD models overproduce the Galactic SFR (11 Msun/yr vs ~2 Msun/yr). If most ATLASGAL clumps are unbound and will never form a bound cluster, then the epsilon_ff distribution for these clumps does not directly constrain cluster progenitors, and the GC/GCD SFR argument is misapplied: the bound-cluster-forming subset might have a smaller total mass, relaxing the overproduction constraint. The paper itself flags this uncertainty in Section 3.4.1: 'we do not in fact know if the density range that is selected by ATLASGAL corresponds well to the conditions that delineate between bound and unbound star formation.' Without a demonstration that ATLASGAL clumps are predominantly bound cluster progenitors, the exclusion of GC/GCD, and hence the unique support for the conveyor-belt mode, is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper investigates how gravitationally bound star clusters assemble and argues that the available Milky Way observations favor a 'conveyor belt' formation mode. The authors construct six zero-dimensional, analytically solvable model scenarios—static (ST), conveyor belt (CB), conveyor belt with rapid dispersal (CBD), global collapse (GC), global collapse with dispersal (GCD), and increasing star formation efficiency (IE)—and test them with MCMC forward modeling against three data sets: the stellar age distributions in the ONC and NGC 6530, the distribution of star formation efficiencies per free-fall time (epsilon_ff) inferred from YSO counts in ATLASGAL clumps, and the global star formation rate of the Milky Way. They find that ST and CB(p=0) cannot reproduce the accelerating-but-extended age distributions; IE needs an efficiency slope delta at the prior boundary (delta=3) for the ages, in conflict with the near-constant efficiency (delta≈0) required by the YSO data; and GC/GCD models that fit the ages and YSO counts imply the ATLASGAL clumps alone form stars at roughly 11 Msun/yr, several times the observed total Galactic SFR. CB and CBD, by contrast, naturally produce accelerating star formation at constant low efficiency (~1%) and predict a contribution of ~0.3 Msun/yr from the ATLASGAL clumps. The paper concludes that cluster-forming regions accrete gas at an increasing rate while forming stars inefficiently and without global collapse.","tokens_in":39211,"tokens_out":19379,"duration_ms":171385,"significance":"The paper is significant for the star formation field and is unusually transparent in its construction: the six models have closed-form solutions, the fitting code is publicly available on Bitbucket, and the treatment of observational uncertainties (biased log-normal age errors, finite-window SFR averaging, selection of gas-dominated systems) is careful and clearly described. The global SFR argument of Section 3.4.2 is a valuable and falsifiable discriminant that ties the cluster-formation question to the Milky Way's total star formation budget, and the paper's prediction that ATLASGAL clumps contribute roughly 10% of the Galactic SFR is testable. The conclusion that bound clusters form via conveyor-belt accretion with epsilon_ff ~ 0.01 and the Goldbaum et al. p=3 scaling directly addresses the long-standing problem that gas clouds as massive as mature clusters are not observed. If the central claim survives scrutiny, it should reshape how the field interprets cluster formation, and the framework will be reusable as Gaia-based ages and YSO surveys improve.","major_comments":[{"comment":"The identification of the ATLASGAL clump population as the direct progenitor population of bound clusters like the ONC and NGC 6530 is load-bearing in two places: Section 3.2.1 uses it to turn the observed epsilon_ff distribution into a constraint on cluster formation models, and Section 3.4.2 applies the GC/GCD models to the full ATLASGAL mass Mtot ≈ 1e7 Msun in Eq. (56) to exclude those scenarios on the grounds that they overproduce the Galactic SFR. The paper asserts that the clumps 'very likely' become ONC-like clusters, but in Section 3.4.1 it concedes that 'we do not in fact know if the density range that is selected by ATLASGAL corresponds well to the conditions that delineate between bound and unbound star formation.' If most ATLASGAL clumps are unbound and will never form bound clusters, the epsilon_ff constraint would not directly apply to cluster progenitors, and the mass entering Eq. (56) should be reduced to the bound-forming subset, which would relax the overproduction argument and remove the unique support for the conveyor-belt conclusion. The authors should quantify the bound fraction of the ATLASGAL sample (e.g., via virial-parameter estimates) and redo the Section 3.4.2 calculation for the bound subset only, or otherwise demonstrate that the exclusion of GC/GCD is insensitive to this fraction.","section":"Sections 3.2.1, 3.4.1, and 3.4.2 (Eq. 56)"},{"comment":"The paper states in Section 3.1.1 that the posterior PDFs for some parameters and the relative goodness-of-fit are sensitive to the assumed stellar-age error parameters (sigma, b), which observationally lie in a range up to about 0.2–0.3 dex, and it limits the analysis to features that are robust. This promise is not fully delivered in the global SFR argument: Eq. (56) is normalized to the age-derived value tau_coll = 0.03, and the exclusion of the IE model relies on the age-derived delta sitting at the prior boundary (delta = 3). Both discriminants are therefore downstream of the age-error assumption. I ask the authors to demonstrate explicitly, over the plausible range of (sigma, b), that the tau_coll posterior for GC/GCD and the delta posterior for IE remain incompatible with the YSO and SFR constraints; if only the qualitative conclusion is claimed, the text should make that limitation clearer in Sections 3.3 and 3.4.","section":"Sections 3.1.1, Table 3, and 3.4.2 (Eq. 56)"},{"comment":"In the epsilon_ff likelihood, clumps with no detected YSOs are treated as detections at their stated 2-sigma upper limits rather than as censored data points. A proper treatment would integrate the model prediction up to the upper limit for these clumps (a survival-analysis likelihood). Because the derived values epsilon_ff ≈ 0.01–0.02 and sigma_log_eps ≈ 0.15 dex drive the combined constraints of Section 3.3 and the SFR estimates of Section 3.4, the authors should either implement the censored likelihood or demonstrate that the posterior PDFs in Table 4, and the conclusions drawn from them, are unchanged when non-detections are handled consistently.","section":"Sections 3.2.1 and 3.2.2 (Eq. 42)"}],"minor_comments":[{"comment":"There is a typo in the sentence 'objects that are will become clusters like NGC 6530 or the ONC'; it should read 'objects that will become clusters.'","section":"Section 3.2.1"},{"comment":"The double-peaked age distribution predicted by the GC and GCD models is mentioned but never displayed or explained in the main text; adding a sentence on the origin of the second peak (stars formed near the collapse singularity) would aid the reader.","section":"Section 3.1.2"},{"comment":"Several of the marginalized posteriors are effectively unconstrained (e.g., CB p=3 for the ONC gives log epsilon_ff = −0.24 with asymmetric uncertainties spanning more than an order of magnitude); the text should flag which parameters the age data do not constrain, to avoid over-reading the 'successful' fits.","section":"Table 3"},{"comment":"The 'combined' constraints are overlaps of independently derived posterior PDFs rather than a joint fit; the paper should state explicitly that this identifies consistency regions rather than joint posterior modes.","section":"Section 3.3"},{"comment":"MCMC convergence is asserted by visual inspection only; reporting a quantitative diagnostic (e.g., Gelman-Rubin R-hat or effective sample size) would be appropriate, particularly since the tails of the chains drive the SFR distributions in Figure 8.","section":"Section 3.1.3"},{"comment":"The GC/GCD analysis imposes the prior t_ff,0 > 0.3 Myr when fitting the YSO data, but Eq. (56) sets t_ff,0 = 0.3 Myr for the SFR calculation; the authors should clarify the relationship between these two choices and confirm explicitly the direction of the resulting bias (the claim that Eq. 56 is a lower limit appears correct but is not argued).","section":"Sections 3.2.3 and 3.4.2"}],"recommendation":"major_revision","confidential_remarks":"This is a serious, well-executed study and the modeling framework will likely become a standard reference. The main risk to the paper's central claim is the load-bearing assumption that ATLASGAL clumps are the direct progenitors of bound clusters; the authors themselves flag the key uncertainty in Section 3.4.1, yet the concluding claim is stated more strongly than that caveat supports. I would request the robustness analyses described in the major comments rather than rejecting; the paper is close to publishable, but the uniqueness of the conveyor-belt conclusion should be tempered or quantitatively defended. The omission of quantitative model-comparison metrics is also worth addressing given the acknowledged sensitivity to age-error parameters."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does actual work: it builds simple analytic cartoons for six cluster-formation scenarios and then confronts them jointly with three independent constraints—Gaia-era stellar age distributions in ONC and NGC 6530, ATLASGAL YSO counts, and the Milky Way's global SFR. The comparative Bayesian treatment is new, the code is public, and the fits are reported honestly. The cleanest result is the exclusion of the increasing-efficiency model: stellar ages want a strongly time-varying epsilon_ff, while YSO counts demand near-constant epsilon_ff, and no parameter choice squares that circle. That is a solid, reproducible finding.\n\nThe conveyor-belt conclusion is plausible but not bulletproof. The authors are upfront that their models are zero-dimensional cartoons, and they note that none reproduce the age distributions in detail and that posteriors are sensitive to the adopted age-error model. Those are honest caveats, not fatal flaws. The softest spot is the one the stress-test flags: treating the ATLASGAL clump population as the direct progenitors of bound clusters like the ONC. That identification underpins both the epsilon_ff distribution comparison and, more importantly, Equation 56, where GC/GCD models overproduce the Galactic SFR by a factor of ~5. If most ATLASGAL clumps are unbound and will never form bound clusters, that SFR argument loses its target—the bound-forming subset could have much less mass. The authors themselves acknowledge in Section 3.4.1 that they do not know if the ATLASGAL density selection matches the bound/unbound boundary. So the conclusion that global collapse is ruled out is not fully established; it is a well-argued conditional statement.\n\nI also want to credit the continuity with older arguments: the GC/GCD overproduction is a direct cousin of Zuckerman & Evans and Krumholz & Tan, and the paper makes that lineage explicit. That is the right kind of cumulative science.\n\nWho should read this: anyone working on star cluster formation, star formation efficiency, or the interpretation of ATLASGAL and Gaia data. It is a good paper for a reading group because the models are transparent and the failure modes are discussable. I would cite it for the model-comparison framework and the IE exclusion. It deserves a serious referee—not a desk reject—and the referee should push on the ATLASGAL identification and maybe on the prior choices, but the core analysis is careful and reproducible.","headline":"A careful Bayesian model comparison that makes a real case for the conveyor-belt picture, with the ATLASGAL-to-progenitor identification as the main load-bearing caveat—worth serious engagement even if the conclusion stays conditional.","tokens_in":39733,"tokens_out":1514,"would_cite":true,"duration_ms":19253,"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":"Bound star clusters form in a conveyor-belt mode: gas streams in while stars form slowly, so the cluster never assembles all its gas at once; global collapse and time-dependent efficiency models fail the data.","keywords":["star cluster formation","conveyor belt accretion","star formation efficiency","free-fall time","young stellar objects","Orion Nebula Cluster","ATLASGAL","Galactic star formation rate"],"falsifier":"A survey that measured star formation efficiency per free-fall time as a function of evolutionary stage in a large sample of dense clumps and found it rising with time, rather than staying near one percent, would falsify the conveyor-belt conclusion. Equivalently, finding a massive, dense, quiescent gas cloud with a free-fall time shorter than the age of its cluster and no embedded stars would resurrect the static-cloud scenario.","tokens_in":38639,"feed_emoji":"⭐","tokens_out":6816,"duration_ms":62014,"temperature":0.7,"pith_summary":"This paper asks why some star-forming regions become gravitationally bound clusters while most stars drift apart, and argues that the answer is a conveyor belt: gas streams into the cluster-forming hub at an accelerating rate while stars form slowly and steadily, so the full cluster mass is never assembled as gas at one time. The authors test this against three observational constraints, stellar age distributions in the Orion Nebula Cluster and NGC 6530, counts of young stellar objects in dense ATLASGAL gas clumps, and the Milky Way's total star formation rate, and find that only conveyor-belt-type models survive. If correct, bound clusters form over many free-fall times at roughly one percent efficiency per free-fall time, and the puzzle of missing massive gas clouds disappears because no such cloud ever needs to exist.","feed_headline":"Star clusters assemble via conveyor-belt gas, not collapse","feed_subtitle":"Age and gas data favor slow star formation fed by streaming gas, not a single collapse.","key_machinery":"The argument rests on a family of one-zone analytic models in which the star formation rate is $\\dot M_*=\\epsilon_{\\rm ff} M_g/t_{\\rm ff}$ and gas evolves as $\\dot M_g = \\dot M_{\\rm acc} - (1+\\eta)\\dot M_*$, with $t_{\\rm ff}$ the free-fall time and $\\eta$ a mass-loading factor. Each scenario, static, conveyor belt, global collapse, and increasing efficiency, is a different prescription for $\\dot M_{\\rm acc}$, $t_{\\rm ff}$, or $\\epsilon_{\\rm ff}$, and the models are compared to data through Bayesian likelihoods for the age distribution and the $\\epsilon_{\\rm ff}$ distribution, plus an analytic calculation of the Galaxy-wide star formation rate implied by the ATLASGAL clump population. The load-bearing identity is the global SFR equation, which shows that global-collapse models overproduce stars unless star formation efficiency is low enough to prevent bound clusters from forming.","core_discovery":"The central claim is that the best available explanation for all three constraints is the conveyor belt mode: gas accretes at an increasing rate, approximately as $\\dot M_{\\rm acc}\\propto t^3$, while the central cluster-forming region has a star formation efficiency per free-fall time $\\epsilon_{\\rm ff}\\simeq 0.01$ that is low and roughly constant. In this picture the observed acceleration of star formation in young clusters reflects the growing gas mass available in the hub, not a global collapse of the cloud or a time-dependent efficiency. Global collapse models fail because, once tuned to produce accelerating star formation, they predict that the dense ATLASGAL clumps should form stars at roughly five to ten times the Milky Way's entire star formation rate; increasing-efficiency models fail because the stellar ages demand $\\epsilon_{\\rm ff}$ that rises steeply with time while the YSO-gas correlation demands it be nearly constant.","pith_inferences":["Going beyond the paper, the conveyor-belt picture suggests that a protocluster's final mass is set more by how long its feeder filaments keep supplying gas than by the initial cloud mass; this could be tested by comparing filament inflow rates to cluster stellar masses.","The same reasoning predicts that in galaxies with more gas accretion, a higher fraction of stars should end up in bound clusters, a correlation that could be measured with resolved young-stellar-object surveys in nearby galaxies.","A direct test would be to measure $\\epsilon_{\\rm ff}$ in a single protocluster as a function of time by dating successive generations of embedded stars; if it rises, the conveyor-belt conclusion would need revision.","The paper's argument treats gas density and star formation as volume-averaged; a more detailed version might distinguish the hub from filaments, and could reveal whether stars formed in filaments also join the bound cluster."],"forward_implications":["Massive gas clouds as massive as the future cluster need not exist at any time; the gas is consumed and replenished simultaneously.","Observed dense gas clumps like ATLASGAL are caught mid-accretion, which is why they look gas-rich and young relative to their embedded stars.","A successful cluster-forming region should show a near-constant $\\epsilon_{\\rm ff}\\sim 0.01$ while its gas mass and star formation rate rise together.","The Milky Way's total star formation rate is consistent with dense clumps producing only about a tenth of all stars, matching the observed bound-cluster fraction.","Acceleration of star formation in a cluster is a sign of accelerating gas supply, not of an imminent global collapse."],"supporting_citations":[{"why":"Supplies the ATLASGAL young-stellar-object counts and the measured distribution of $\\epsilon_{\\rm ff}$ that every model is fitted against.","marker":"Heyer et al. (2016)"},{"why":"Proposed the conveyor-belt picture that the paper adopts as its preferred explanation for cluster formation.","marker":"Longmore et al. (2014)"},{"why":"Derives the $\\dot M_{\\rm acc}\\propto t^3$ accretion law used as the fiducial conveyor-belt accretion history.","marker":"Goldbaum et al. (2011)"},{"why":"Supplies the Gaia-kinematic membership and stellar ages for the Orion Nebula Cluster sample.","marker":"Kounkel et al. (2018)"},{"why":"Supplies the NGC 6530 stellar age sample and the velocity dispersion used to set its free-fall time.","marker":"Prisinzano et al. (2019)"},{"why":"Gives the total mass of ATLASGAL clumps used in the Galactic star-formation-rate calculation.","marker":"Urquhart et al. (2018)"},{"why":"Sets the observed Milky Way star formation rate of about two solar masses per year that disfavours global collapse.","marker":"Chomiuk & Povich (2011)"},{"why":"Formulates the increasing-efficiency model that the paper rules out by combining age and YSO constraints.","marker":"Lee et al. (2016)"},{"why":"Articulates the global-collapse-plus-dispersal model that the paper tests and rejects on star-formation-rate grounds.","marker":"Vázquez-Semadeni et al. (2019)"}],"fun_headline_variants":["Conveyor-belt gas builds star clusters, collapse fails","Star clusters: accretion wins over collapse","Slow gas feed, not collapse, forms bound clusters","Constant low efficiency builds clusters, not collapse"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that the ATLASGAL clumps, the dense dusty gas clouds in which young stellar objects are counted, are the true gas-rich progenitors of bound clusters like the Orion Nebula Cluster and NGC 6530; if most of them will never become bound clusters, the star-formation-rate test that rules out collapse loses its target.","fun_headline_variants_meta":{"raw":{"variants":["Conveyor-belt gas builds star clusters, collapse fails","Star clusters: accretion wins over collapse","Slow gas feed, not collapse, forms bound clusters","Constant low efficiency builds clusters, not collapse"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000572,"raw_usage":{"total_tokens":2681,"prompt_tokens":899,"completion_tokens":1782,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":515,"completion_tokens_details":{"reasoning_tokens":1731}},"tokens_in":515,"tokens_out":1782,"duration_ms":13324,"temperature":1.0,"reasoning_tokens":1731,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:13:49.494546+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A survey that measured star formation efficiency per free-fall time as a function of evolutionary stage in a large sample of dense clumps and found it rising with time, rather than staying near one percent, would falsify the conveyor-belt conclusion. Equivalently, finding a massive, dense, quiescent gas cloud with a free-fall time shorter than the age of its cluster and no embedded stars would resurrect the static-cloud scenario.","supporting_citations":[{"cited_title":"J., Krumholz M","cited_arxiv_id":null,"evidence_quote":"Derives the $\\dot M_{\\rm acc}\\propto t^3$ accretion law used as the fiducial conveyor-belt accretion history."}],"review_version":1}