{"id":"ab031aed-7ef8-45a3-a82f-b615b61bc2c7","arxiv_id":"2502.02647","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A semi-analytic model seeded with simulation-derived star formation rules reproduces JWST's bright galaxies at z>11 via a top-heavy IMF or boosted star formation, and finds low-mass galaxies drove reionisation.","lead":"The authors combine a semi-analytic galaxy formation model with gas physics taken from a radiation-hydrodynamics simulation to explain why JWST sees more bright galaxies in the early universe than expected. Their models reproduce the observed bright galaxy abundance at redshift 11 to 20 and indicate that low-mass galaxies supplied most of the ionising photons that reionised the universe.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Load-bearing ambiguity in the stochastic sampling: if fcold and f_sphinx are drawn independently from their marginal SPHINX20 PDFs, the product that sets star formation is uncalibrated to SPHINX20's joint distribution, so the bright-end z>11 LF match and the reionization budget could be artifacts…","rationale":"I read the paper as a constructive, transparent model-comparison study: the eIMF route is physically motivated through the Chon et al. IMF prescription and gives a relatively parameter-light match to the JWST bright-end excess, while the paper openly labels the eSFE model as an 'extreme' alternative and discusses its ad hoc fcold=0.5 and f*=0.8 choices at z>=14. The reader's weakest-assumption identification of the SPHINX20-to-DELPHI mass extrapolation is reasonable and related to calibration, but my worry is more specific and cuts across all models, including the nominally independent eIMF path. In DELPHI, the actual star formation is set by fcold*f_sphinx, and the paper never demonstrates that the Monte Carlo sampling preserves the SPHINX20 joint distribution of these two quantities. If it does not, the stochastic bursts that power the bright end at z>=11 are partly numerical, and the reionization source census inherits the same bias. This is a concrete, checkable ambiguity rather than a disagreement with the broad astrophysical scenario. I therefore keep the reader's CONDITIONAL verdict unchanged, but with an additional explicit condition: verify the joint sampling before relying on the bright-end LF or the 85% low-mass reionization fraction. If the authors confirm joint sampling, the concern dissipates and the paper's central claims remain plausible.","tokens_in":34968,"tokens_out":8947,"duration_ms":98617,"concrete_test":"Instrument DELPHI's Monte Carlo assignment and extract the joint distribution p(fcold, f_sphinx) it actually draws for each halo-mass/redshift bin. From SPHINX20, compute the directly measured joint distribution and the derived distribution of y = M_*[<30 Myr]/M_g = fcold*f_sphinx for the same bins (e.g., z~9, M_h~1e8-1e10). Perform a two-sample test (2D KS on the joint PDF, or quantile comparison on y) between SPHINX20 and DELPHI. If they disagree, rerun all four models sampling from the SPHINX20 joint 2D PDF (or directly resampling SPHINX20 galaxies) and regenerate the z=11-14 UV LFs and cumulative ionizing emissivity. If the bright-end LF shifts by more than ~0.3 dex or the M*<1e9 contribution changes by more than ~10%, the central claim is not robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that both eIMF and eSFE explain the z>=11 bright-end excess rests on the stochastic SFE implementation of Sects. 2.2-2.3. Eq. 1 defines f_sphinx = M_*[<30 Myr]/(M_g fcold), so the physical quantity controlling DELPHI star formation is the product fcold*f_sphinx = M_*[<30 Myr]/M_g. The paper displays only marginal PDFs of fcold and f_sphinx (Figs. 2-3) and says that 'to assign fcold and f_sphinx values to delphi halos, we choose the sphinx probability distribution ... using a standard Monte Carlo sampling.' It does not state that these two quantities are drawn jointly, and no validation is shown that the sampled product distribution reproduces the distribution actually measured in SPHINX20. If fcold and f_sphinx are sampled independently from marginals, the code can generate unphysical combinations, e.g., high cold-gas fraction and high recent SFE simultaneously, that are rare or absent in SPHINX20. Such artificial starbursts are exactly the rare objects that populate the bright end of the UV LF at z>=11 and contribute to the reionizing photon budget. The fiducial model's underprediction and the eIMF/eSFE 'success' could therefore be an artifact of the sampling scheme rather than a physical outcome. This is a load-bearing calibration issue, not a cosmetic scatter issue.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper couples the DELPHI semi-analytic galaxy formation model to ISM properties measured in the SPHINX20 radiation-hydrodynamics simulation. Cold gas fractions fcold and star formation efficiencies f_sphinx are Monte Carlo sampled from SPHINX20 PDFs, introducing stochastic star formation into DELPHI. Three models are compared: a fiducial model using the SPHINX20-derived distributions, an eIMF model with a metallicity- and redshift-dependent top-heavy IMF, and an eSFE model with ad hoc increasing star formation efficiency and cold gas fraction for massive halos at z>7. The fiducial model underpredicts the bright end of the observed UV LF at z>=11, while the eIMF and eSFE models reproduce it. The paper additionally compares stellar mass functions, dust masses, mass-metallicity relations, UV slopes, Balmer breaks, and xi_ion against JWST and ALMA data, and uses the models to infer that galaxies with M*<10^9 Msun provide about 85% of the escaping ionizing photons down to z~7.","tokens_in":35341,"tokens_out":5933,"duration_ms":60449,"significance":"If the results hold, the paper provides a useful demonstration that a semi-analytic model can use resolved hydrodynamical simulation results to generate stochastic star formation and make multi-observable predictions for JWST. The eIMF model's agreement with the z>=11 bright-end UV LF is a genuine prediction in the sense that the IMF prescription is imported from prior cluster-formation simulations and not tuned to the JWST LF, and the paper gives a broad set of falsifiable predictions (xi_ion, beta slopes, Balmer breaks, SMF separation, luminosity density evolution) that can discriminate between the models. The reionization source claim is also presented with two escape-fraction prescriptions, which strengthens its qualitative robustness. However, the central claim that both the eIMF and eSFE models explain the bright JWST excess is currently conditional on two load-bearing issues: the stochastic sampling of fcold and f_sphinx is not shown to preserve the joint SPHINX20 distribution, and the eSFE boost is imposed by hand rather than derived or independently calibrated.","major_comments":[{"comment":"The stochastic sampling of fcold and f_sphinx is underspecified in a way that directly affects the central claim. Equation (1) defines f_sphinx as M_*[<30 Myr]/(M_g fcold), so the physical quantity that sets the new stellar mass in Eq. (2) is the product fcold*f_sphinx = M_*[<30 Myr]/M_g. The manuscript displays only marginal PDFs of fcold and f_sphinx (Figs. 2 and 3) and states that DELPHI halos are assigned values by 'standard Monte Carlo sampling' of the SPHINX20 distributions. If fcold and f_sphinx are drawn independently from their marginals, the product distribution will not reproduce the joint distribution realized in SPHINX20, and rare combinations of high cold gas fraction and high recent SFE could be generated artificially. Such artificial bursts are exactly the rare objects that populate the bright end of the z>=11 UV LF and contribute to the reionizing photon budget. The authors should state whether the draws are joint and should validate that the sampled product distribution matches the SPHINX20 distribution of M_*[<30 Myr]/M_g. If the draws are in fact independent, the analysis must be rerun by sampling from the joint fcold-f_sphinx distribution before the fiducial underprediction and the eIMF/eSFE 'success' can be interpreted.","section":"Sects. 2.2-2.3, Eq. (2)"},{"comment":"The eSFE model's high-redshift boost is imposed rather than derived. The model sets f*(Mh,z)=0.8 and fcold=0.5 for massive halos at z>=14, with a linear interpolation between the SPHINX20 values at z=7 and these imposed values at z=14. No independent physical motivation or calibration is provided for these specific numbers beyond the need to increase the bright-end UV LF. Consequently, the eSFE agreement with the z>=11 JWST data in Fig. 7 is partly a consequence of construction, and it cannot be presented on the same footing as the eIMF model in the abstract's claim that both models 'can explain' the abundance of bright galaxies. The eSFE model should be explicitly reframed as a deliberately extreme scenario, or its parameter values should be calibrated against independent high-resolution simulations or observations, before it is used to support the paper's main conclusion.","section":"Sect. 2.6.2 and Table 1"},{"comment":"The extrapolation of SPHINX20 PDFs to the most massive DELPHI halos is a key uncertainty for the bright end. As the text acknowledges, at any redshift DELPHI contains halos up to two orders of magnitude more massive than the most massive SPHINX20 halos, and those massive halos are assigned fcold and f_sphinx from smaller systems. The argument that this induces limited error relies on visual convergence of the PDFs for the few most massive SPHINX20 bins. Because the z>=11 bright-end LF is one of the two central claims, the paper should quantify the sensitivity of the bright-end results to this mass matching, for example by testing an alternative extrapolation or by checking against a larger-volume simulation with more massive halos.","section":"Sect. 2.2"}],"minor_comments":[{"comment":"Equation (10) contains an unresolved citation placeholder '(?)' that should be replaced with the intended reference.","section":"Sect. 4.1, Eq. (10)"},{"comment":"The citation list in the discussion of accreting black holes includes '?;' between references; this placeholder needs to be fixed.","section":"Sect. 1"},{"comment":"The text says the PDFs are built in stellar mass bins, while Figs. 2 and 3 and the assignment procedure at the end of the section use halo mass bins; this inconsistency should be resolved.","section":"Sect. 2.2 and Figs. 2-3"},{"comment":"The right panel does not show the uncertainty ranges from the five random-seed runs; adding them or stating that they are omitted for clarity would make the presentation consistent with the left panel.","section":"Fig. 17"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the scope of A&A and the eIMF prediction is genuinely interesting, but the main claim is currently more conditional than the abstract suggests. I would ask the authors to clarify the joint sampling, validate the product distribution, and reframe the eSFE model before reconsideration. I do not see a novelty disclosure concern; the DELPHI and SPHINX20 components are prior work and the integration is new."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this paper deserves a serious referee. The eIMF half of the argument is a real prediction; the eSFE half is built to reproduce the excess it then claims to explain. And the stochastic-sampling ambiguity in Sect. 2.2 could be load-bearing, so it should be fixed before the bright-end match is treated as physical.\n\nWhat's genuinely new: feeding SPHINX20 cold-gas fractions and SFE distributions into DELPHI as stochastic draws, the eSFE variant itself, and the unusually broad comparison against UV LFs, stellar mass functions, dust masses, beta slopes, Balmer breaks, xi_ion, and reionisation history. The multi-observable sweep is systematic and the paper is clearly written. The eIMF model's form is imported from earlier independent work, but using it here to match the JWST bright end at z>11 is a legitimate, non-tuned success.\n\nNow the soft spots, in proportion. First, eSFE: the high-z boost (f*=0.8, fcold=0.5 at z>=14, with a weighted average between z=7 and 14) is introduced to lift the bright end to the observed level. The paper does call it an 'extreme model', but the abstract still presents both eIMF and eSFE as successful explanations; that circularity should be made much more explicit. Second, the reionisation analysis fits a constant fesc per model to match reionisation history. The claim that low-mass galaxies (M*<1e9 Msun) provide about 85% of ionising photons is robust to that fit because a constant fesc cancels in the cumulative distribution, but the fitted fesc values themselves are not constraints. Third, the stress-test concern is real: the methods text says 'standard Monte Carlo sampling' of 'the sphinx probability distribution', but the figures show marginal PDFs and the text never states that fcold and f_sphinx are drawn jointly. If they are sampled independently, the product fcold*f_sphinx is uncalibrated to SPHINX20's joint behaviour, and the artificial starbursts could sit exactly on the z>11 bright end. That would weaken both the fiducial model's underprediction and the eIMF/eSFE 'success'. The eIMF model is less exposed because its luminosity boost comes from the IMF rather than the sampling, but the bright-end LF in all models depends on this sampling.\n\nWho this is for: anyone working on JWST UV LFs or reionisation sources. The eIMF model produces concrete, testable signatures (xi_ion, Balmer breaks, metal enrichment) that make the paper worth citing. Treat the eSFE model as an upper envelope, not a physical proposal.\n\nRecommendation: send to peer review. Ask the authors to (a) validate that the sampled fcold*f_sphinx product reproduces SPHINX20's joint distribution, (b) separate the tuned eSFE from the predictive eIMF in the abstract and conclusions, and (c) downplay the fitted fesc as an inference.","headline":"Solid, careful model comparison; the eIMF path is a genuine prediction, the eSFE path is tuned, and the stochastic sampling needs a joint-distribution check.","tokens_in":35913,"tokens_out":3456,"would_cite":true,"duration_ms":35010,"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":"Two physical mechanisms explain JWST's bright galaxies at $z \\geq 11$.","keywords":["high-redshift galaxies","reionisation","UV luminosity function","stellar initial mass function","star formation efficiency","JWST","semi-analytic galaxy formation","radiation-hydrodynamics simulation"],"falsifier":"A JWST/NIRSpec measurement of the ionising photon production efficiency $\\xi_{\\rm ion}$ in bright galaxies at $z \\sim 12-17$ would settle the eIMF model, which predicts $\\log_{10}\\xi_{\\rm ion}\\simeq25.55\\,[{\\rm Hz\\,erg^{-1}}]$ against roughly $25.2$ for the other models; likewise, a stellar mass function measurement at $M_\\ast\\sim10^9\\,M_\\odot$ and $z\\sim12$ would test the eSFE model, which predicts a number density about 100 times that of the fiducial model.","tokens_in":34751,"feed_emoji":"🔭","tokens_out":10634,"duration_ms":92938,"temperature":0.7,"pith_summary":"The paper argues that the bright galaxies JWST sees in excess at $z \\geq 11$ can be explained by either of two physical mechanisms: a stellar initial mass function that becomes top-heavy in low-metallicity, high-redshift systems, or a star formation efficiency that rises with redshift. Both prescriptions reproduce the observed ultraviolet luminosity function when inserted into a semi-analytic model whose star-forming gas properties are sampled from a high-resolution radiation-hydrodynamics simulation, so the comparison is anchored in simulated interstellar-medium physics rather than free fitting alone. The paper also claims that star formation in galaxies below about $10^9\\,M_\\odot$ of stars supplies roughly 85% of the ionising photons down to the midpoint of reionisation at $z \\sim 7$, identifying low-mass galaxies as the key reionisation sources. If correct, the JWST bright-end excess is not a crisis for galaxy formation, and the reionising population is pinned to a specific, faint class of galaxies.","feed_headline":"Two mechanisms explain JWST's bright early galaxies","feed_subtitle":"Top-heavy stars or faster star formation both match the UV galaxy excess; low-mass galaxies drive reionisation.","key_machinery":"The load-bearing machinery is the transfer of interstellar-medium statistics from the high-resolution SPHINX20 radiation-hydrodynamics simulation into the DELPHI semi-analytic model: probability distributions of cold gas fraction (gas below 1000 K) and star formation efficiency (stars younger than 30 Myr divided by cold gas mass) are sampled by Monte Carlo for each halo, injecting stochastic, bursty star formation. Two luminosity-boosting prescriptions are layered on top: the eIMF model makes the stellar mass function increasingly top-heavy at low metallicity and high redshift, raising the UV luminosity and ionising-photon output per unit stellar mass; the eSFE model raises cold gas fractions and star formation efficiencies in massive halos above $z\\sim7$, saturating at fixed values at $z\\ge14$. These ingredients turn the semi-analytic model into an ensemble whose bright end can match JWST while its faint end is set by simulated interstellar-medium physics.","core_discovery":"On the paper's own terms, the central discovery is that the overabundance of bright galaxies at $z \\geq 11$ is reproduced by two distinct and physically motivated prescriptions, not by exotic sources: the eIMF model, in which the stellar initial mass function becomes increasingly top-heavy at low metallicity and high redshift, and the eSFE model, in which the cold gas fraction and star formation efficiency of massive halos increase with redshift above $z \\sim 7$. Both match the observed ultraviolet luminosity function from $z \\sim 5$ out to $z \\sim 15-20$ once dust attenuation is included at low redshift, with dust negligible above $z \\sim 12$. The paper further establishes that, in every model consistent with reionisation constraints, galaxies with stellar masses below roughly $10^9\\,M_\\odot$ provide about 85% of the escaping ionising photons down to the reionisation midpoint at $z \\sim 7$. The models also place the mass-metallicity relation in place by $z\\sim17$ and bracket current measurements of UV spectral slopes, Balmer break strengths, and ionising photon production efficiencies, with the eIMF model predicting a factor 2--2.5 higher $\\xi_{\\rm ion}$ at $z \\geq 12$.","pith_inferences":["We infer that the current JWST bright-end data, taken alone, cannot distinguish between the two mechanisms; the degeneracy has to be broken by stellar mass functions or spectral indicators, a point the paper itself makes.","If low-mass galaxies supply most of the ionising photons, then bright-galaxy surveys may be probing the wrong population for understanding reionisation; the relevant sources sit near or below current detection limits.","The eSFE model's assumed saturation values at $z\\ge14$ ($f_\\ast=0.8$, $f_{\\rm cold}=0.5$) could be checked directly with larger-volume radiation-hydrodynamics simulations; if those simulations find lower efficiencies in massive halos, the eSFE bright end would weaken.","A natural testable extension is to combine the eIMF prediction of high $\\xi_{\\rm ion}$ with 21-cm observations of the reionisation midpoint: the model's larger photon output requires a lower escape fraction to match the timing, and future neutral-hydrogen measurements could verify that combination."],"forward_implications":["If the central claim is right, the JWST bright-end excess at $z \\geq 11$ is reproduced within standard structure formation, with no need for additional sources such as faint active galactic nuclei to explain the ultraviolet luminosity function.","The eIMF and eSFE models predict stellar mass functions that differ by more than 2.5 orders of magnitude at $M_\\ast \\sim 10^9\\,M_\\odot$ and $z \\sim 12$, so jointly measured luminosity and stellar mass functions will separate the two mechanisms.","Because sub-$10^9\\,M_\\odot$ galaxies dominate the reionisation budget, deeper surveys should find that the sources completing reionisation are mostly fainter than current JWST detection limits.","The eIMF model predicts $\\xi_{\\rm ion}\\sim10^{25.55}\\,{\\rm Hz\\,erg^{-1}}$ for faint galaxies at $z \\gtrsim 12$, a factor 2--2.5 above the other models, making ionising photon production efficiency a direct observational test of a top-heavy IMF.","Dust attenuation shapes the bright end of the UV luminosity function only at $z \\lesssim 11$; at higher redshifts the intrinsic and observed luminosity functions coincide, so dust cannot mask a model failure there."],"supporting_citations":[{"why":"Supplies the SPHINX20 radiation-hydrodynamics simulation from which the cold gas fraction and star formation efficiency distributions are drawn.","marker":"Rosdahl et al. (2022)"},{"why":"Provides the previous DELPHI model (delphi23) that this work extends with stochastic interstellar-medium prescriptions.","marker":"Mauerhofer & Dayal (2023)"},{"why":"Motivates the metallicity- and redshift-dependent top-heavy IMF used in the eIMF model.","marker":"Chon et al. (2022)"},{"why":"Provides the evolving-IMF prescription that the eIMF model adapts for its stellar mass function.","marker":"Cueto et al. (2024)"},{"why":"Supplies the $\\beta$-dependent escape fraction relation used in one of the two reionisation scenarios.","marker":"Chisholm et al. (2022)"},{"why":"Provides the ionised volume filling fraction evolution equation used to compute reionisation histories.","marker":"Madau (2017)"}],"fun_headline_variants":["JWST's bright galaxies: two models, one fit","Top-heavy stars or fast growth: both explain cosmic dawn","Small galaxies delivered 85% of reionising photons","Cosmic dawn's bright galaxies need no exotic physics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the probability distributions of cold gas fraction and star formation efficiency measured in the simulated low-mass halos carry over unchanged to the much more massive halos that the semi-analytic model produces, and that in the eSFE model these quantities rise to fixed high values at $z\\ge14$; if real massive halos do not follow that extrapolation, the predicted abundance of bright galaxies at $z>10$ changes.","fun_headline_variants_meta":{"raw":{"variants":["JWST's bright galaxies: two models, one fit","Top-heavy stars or fast growth: both explain cosmic dawn","Small galaxies delivered 85% of reionising photons","Cosmic dawn's bright galaxies need no exotic physics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00069,"raw_usage":{"total_tokens":3279,"prompt_tokens":1254,"completion_tokens":2025,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":870,"completion_tokens_details":{"reasoning_tokens":1959}},"tokens_in":870,"tokens_out":2025,"duration_ms":15910,"temperature":1.0,"reasoning_tokens":1959,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T11:35:03.330125+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A JWST/NIRSpec measurement of the ionising photon production efficiency $\\xi_{\\rm ion}$ in bright galaxies at $z \\sim 12-17$ would settle the eIMF model, which predicts $\\log_{10}\\xi_{\\rm ion}\\simeq25.55\\,[{\\rm Hz\\,erg^{-1}}]$ against roughly $25.2$ for the other models; likewise, a stellar mass function measurement at $M_\\ast\\sim10^9\\,M_\\odot$ and $z\\sim12$ would test the eSFE model, which predicts a number density about 100 times that of the fiducial model.","supporting_citations":[{"cited_title":"& Dayal, P","cited_arxiv_id":null,"evidence_quote":"Provides the previous DELPHI model (delphi23) that this work extends with stochastic interstellar-medium prescriptions."}],"review_version":1}