{"id":"d4f5a3c3-ada3-4c72-96d6-cd11fa6cd956","arxiv_id":"2501.12008","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Conventional dust radiative transfer on TNG100 cannot reproduce the red V−J colors of massive star-forming galaxies at z≈2; only screen-like dust around stellar populations younger than ~1 Gyr can match the JWST data.","lead":"Massive star-forming galaxies in the TNG100 simulation at cosmic noon are too blue in V−J by about 0.9 magnitudes compared to JWST observations, even with state-of-the-art dust radiative transfer. The mismatch indicates that dust must be arranged in dense screens around young stars, a geometry standard models do not produce.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central claim depends on assuming TNG100+BPASS dust-free V−J colors are realistic; this is flagged by the authors (footnote 10) but never directly tested, and a redder true intrinsic V−J would flatten the required attenuation-reddening relation.","rationale":"The reader's weakest-assumption analysis identifies the realism of the intrinsic dust-free TNG100 UVJ fluxes as the key condition for the central claim, and I agree. The paper itself repeatedly flags this assumption (footnote 10, Sect. 5.1, Sect. 4.3) without providing a quantitative test against observed intrinsic colors, so it is a genuine gap rather than a manufactured one. The concern is load-bearing because the required steepness of the attenuation-reddening relation is computed as the difference between observed and simulated dust-free V−J; if that baseline shifts redward, the required slope decreases. The paper's robustness tests (BPASS being the reddest template, SIMBA/EAGLE agreement in V−J) narrow the plausible offset to roughly 0.2–0.4 mag but do not eliminate it, since all simulations draw on similar stellar population synthesis assumptions. The proposed SED-fitting test directly measures the quantity in question and would settle whether the dust-geometry conclusion is an artifact of the intrinsic-color assumption. I do not see a more severe threat: the observed V-band brightness offset (1.6 mag) independently supports the need for stronger attenuation, the CANDELS/3D-HST comparison supports the observed red V−J distribution, and the toy-model result is internally consistent. The paper is honest about its caveats and the analysis is carefully executed; the condition that the intrinsic-color assumption be tested before the geometry conclusion is used to revise dust models is appropriate. Thus the reader's CONDITIONAL verdict stands unchanged.","tokens_in":36555,"tokens_out":7958,"duration_ms":89429,"concrete_test":"Run a Bayesian SED fit (e.g., Prospector or Bagpipes) on the JWST/NIRCam photometry of the 162 massive (M⋆ > 10^11 M⊙) UVJ-selected star-forming galaxies, using a flexible two-component dust attenuation model and a broad prior on star-formation history, to derive the posterior distribution of intrinsic (dust-free) V−J. Compare this posterior to the TNG100 dust-free V−J distribution for the same mass and sSFR selection. Then re-derive the required attenuation-reddening relation (as in Fig. 6) using the observed intrinsic V−J posterior: if the required slope falls below the simulated TNG100 relation, the screen-like geometry conclusion is weakened; if the posterior is consistent with TNG100, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The quantitative conclusion that all conventional dust radiative transfer models fail and that screen-like attenuation is required is derived by subtracting the simulated dust-free V−J from the observed V−J (Sect. 4.1). If the true intrinsic V−J of massive z≈2 star-forming galaxies is redder than TNG100+BPASS predicts, the required reddening (1.11 mag) and V-band attenuation (2.05 mag) are overestimated, softening the steepness requirement. The authors acknowledge this in footnote 10 and Sect. 5.1, and Appendix B shows BPASS yields the reddest V−J among the templates tested, so template choice is conservative. However, the underlying stellar population properties (ages, metallicities) themselves are not validated against observations: the comparison to SIMBA and EAGLE shows V−J varies by only ~0.17 mag between simulations, and all three share similar stellar physics assumptions. An unresolved 0.3–0.5 mag offset in intrinsic V−J would reduce the required attenuation-reddening slope from the screen value of ~0.66 toward ~0.4, potentially bringing it within reach of clumpy or mixed geometries that the paper dismisses. This is the single most load-bearing assumption because the U−V and V-band magnitude discrepancies are less sensitive to it: U−V is robust across simulations, and the 1.6 mag V-band brightness offset directly indicates insufficient AV independent of intrinsic V−J.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the rest-frame UVJ colors of galaxies at z≈2 in the TNG100 cosmological simulation, post-processed with the SKIRT dust radiative transfer code, and compares them to JWST/NIRCam photometry from the DAWN JWST Archive. The authors report that massive (M_star>10^11 M_sun) star-forming galaxies in TNG100 are too blue in V−J by ≈0.9 mag and too bright in the V band by ≈1.6 mag relative to observations, across a broad range of dust models, SED libraries, dust-to-metal ratios, and dust assignment schemes. They infer that reproducing the observed colors would require an attenuation-reddening relation (AV versus AV−AJ) as steep as a dust screen, and they show that a toy model in which dust screens obscure stellar populations younger than ≈1 Gyr reproduces the data. The central claim is that conventional dust radiative transfer with dust tracing metals fails to reproduce the mass-dependent UVJ diagram at cosmic noon.","tokens_in":36883,"tokens_out":6357,"duration_ms":69543,"significance":"If the result holds, it is an important challenge to the standard assumption that diffuse ISM dust tracing metals can reproduce the UVJ colors of massive dusty star-forming galaxies at cosmic noon. The paper merits attention for its extensive robustness testing: multiple SED libraries (BPASS, FSPS-MILES, BC03, TODDLERS), three dust models (THEMIS, Draine & Li, Zubko), fdust variations up to 5, alternative dust assignment criteria, alternative observational catalogs (CANDELS/3D-HST with updated EAZY templates), and cross-simulation comparisons with SIMBA and EAGLE. The authors also make their catalogs and analysis scripts publicly available, which strengthens reproducibility. The main conclusion would have broad implications for SED fitting and for subgrid dust modeling in cosmological simulations, provided the dependence on the assumed intrinsic stellar populations is adequately addressed.","major_comments":[{"comment":"The quantitative conclusion that the attenuation-reddening relation must be 'as steep as a dust screen' is obtained by subtracting the simulated dust-free V−J (TNG100+BPASS) from the observed V−J. The authors correctly flag this assumption in footnote 10 and in Section 5.1, but they do not directly test it. The comparison to SIMBA/EAGLE in Section 5.1 shows only a ~0.17 mag spread in intrinsic V−J among hydrodynamical simulations that share similar stellar physics, and the Appendix B SED-template comparison is conservative (BPASS gives the reddest V−J) without probing the underlying age–metallicity distribution. If the true intrinsic V−J of massive z≈2 star-forming galaxies were ~0.3–0.5 mag redder than TNG100+BPASS, the required V−J reddening would drop from 1.11 mag to roughly 0.6–0.8 mag, lowering the required slope in Fig. 6 from ~0.66 toward ~0.4, which could bring it within reach of the clumpy or mixed star-to-dust geometries the paper argues are excluded. I request a sensitivity analysis that re-derives the required AV versus AV−AJ point under plausible redder intrinsic V−J offsets (e.g., +0.2, +0.3, +0.5 mag), and a direct comparison of the massive TNG100 stellar populations to observational constraints (e.g., rest-frame optical spectral indices or dust-free colors from SED fitting with flexible SFH priors).","section":"Section 4.1; footnote 10; Section 5.1"}],"minor_comments":[{"comment":"The text 'for stellar ages above ≈300 Gyr' should read 'above ≈300 Myr' (or another value within the plotted range 10^7–10^9.5 yr); 300 Gyr exceeds the age of the Universe and is inconsistent with the plotted grid.","section":"Fig. 7 caption"},{"comment":"Eq. (1) uses t⋆>tsplit for the unattenuated component and t⋆<tsplit for the attenuated component; since tsplit is described as a threshold, the boundary t⋆=tsplit should be assigned explicitly (e.g., t⋆≥tsplit for one component) to avoid ambiguity.","section":"Eq. (1)"},{"comment":"The phrase 'we find a good agreement' should be 'we find good agreement' or 'agreement is good'; throughout the text, several words contain spurious spaces (e.g., 'e ffective', 'di fference') that should be cleaned in production.","section":"Section 2.1"},{"comment":"The statement that homogeneous shell models reproduce the observed star-forming sequence at t⋆∼10 Myr is not accompanied by a quantitative closeness measure; reporting the median V−J offset for those models would help the reader judge the match.","section":"Section 4.2"},{"comment":"The median differences quoted for Fig. 5 and the required attenuation point in Fig. 6 would benefit from bootstrap uncertainties, since the histograms are broad and the observational sample size is modest.","section":"Section 4.1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a good fit for A&A and the authors are transparent about the intrinsic-color caveat. The requested sensitivity analysis is achievable within the scope of a revision and would materially strengthen the central claim. I see no concerns about novelty disclosure or citation behavior."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nShort version: this is a convincing negative result with one load-bearing caveat the authors admit but don't close. They show that at z≈2 the JWST/NIRCam data contain a population of massive (M* > 10^11 Msun) star-forming galaxies with very red V−J colors, and that after running SKIRT on TNG100 over a wide grid of post-processing choices (fdust 0–5, four SED libraries, three dust models, alternative dust assignment), no conventional dust-RT model gets them red enough. The discrepancy is ~0.9 mag in V−J and the simulated galaxies are ~1.6 mag too bright in V. This is a real problem for the standard assumption that dust simply follows metals.\n\nWhat is new: earlier work (Donnari 2019, Akins 2022) saw hints of missing red star-forming galaxies, but this paper makes the mass dependence explicit and adds the AV versus AV−AJ diagnostic, which shows that matching the data requires an attenuation-reddening relation as steep as a dust screen. It also shows that SIMBA and EAGLE don't fix it, and that a toy model works only if dust screens obscure stellar populations younger than ~1 Gyr. The paper is well structured, careful, and all code and data are public on GitHub.\n\nThe soft spot: the calculation subtracts the simulated dust-free V−J from the observed V−J. If TNG100+BPASS underestimates the true intrinsic V−J of massive z≈2 star-forming galaxies by even 0.3–0.5 mag, the required steepness drops from the screen value toward values that clumpy or mixed geometries might reach. The authors flag this in footnote 10 and Section 5.1, and they show BPASS gives the reddest V−J of the templates they tried and that SIMBA and TNG100 differ by only 0.17 mag. But that last comparison is not reassuring if all three simulations share the same stellar physics assumptions. A quantitative test of the intrinsic-color assumption—for instance, comparing dust-corrected observed stacks to TNG100 SEDs—would materially strengthen the case.\n\nMinor: they rely on EAZY photo-z and rest-frame colors, but the new-template re-run on CANDELS/3D-HST gives similar results, so this is not a major worry.\n\nBottom line: this is a paper for galaxy formation modelers, dust-RT practitioners, and anyone doing SED fitting of dusty star-forming galaxies. It deserves a serious referee. I'd send it out; the referee should require a sensitivity test on intrinsic colors, but the central claim is likely to survive.\n\nRecommendation: accept for peer review.","headline":"A robust negative result on dust models at cosmic noon, with one untested intrinsic-color assumption the authors flag but don't close.","tokens_in":37509,"tokens_out":4811,"would_cite":true,"duration_ms":45300,"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":"At cosmic noon, dust models make massive galaxies 0.9 mag too blue in V−J","keywords":["mass-resolved UVJ diagram","dust radiative transfer","SKIRT","TNG100","cosmic noon","dust attenuation","massive star-forming galaxies","JWST/NIRCam"],"falsifier":"Measure the dust-free (attenuation-corrected) V−J colors of massive star-forming galaxies at $z\\approx2$, for example from Balmer-decrement-selected samples with spatially resolved spectroscopy; if the intrinsic colors already match the observed colors, the inferred screen reddening disappears. Alternatively, resolved HST/JWST imaging of dusty star-forming galaxies that finds no dust screens with $\\tau_V\\approx3$ covering populations up to ~1 Gyr would falsify the toy model.","tokens_in":36337,"feed_emoji":"🌌","tokens_out":7887,"duration_ms":78212,"temperature":0.7,"pith_summary":"This paper tests whether current galaxy formation simulations can reproduce the observed mass trend in the UVJ color-color diagram at redshift $z\\approx2$. The authors post-process TNG100 galaxies with the SKIRT dust radiative transfer code and find that massive ($M_\\star\\gtrsim10^{11}\\,M_\\odot$) star-forming galaxies come out too blue in $V-J$ by about 0.9 mag and too bright in the V band by about 1.6 mag compared with JWST/NIRCam data. They show the discrepancy is robust to SED templates, dust models, dust-to-metal ratios, and to comparisons with other simulations. To match the data, an attenuation-reddening relation as steep as a dust screen is required; in their toy model this means dust screens obscuring stellar populations with ages below about 1 Gyr. The paper thereby makes the case that the reddening of massive dusty star-forming galaxies at cosmic noon is a star-to-dust geometry problem that conventional dust radiative transfer does not yet capture.","feed_headline":"Dust models make cosmic-noon galaxies 0.9 mag too blue","feed_subtitle":"Massive star-forming galaxies at z≈2 demand screen-like dust reddening no standard model delivers.","key_machinery":"The central diagnostic is the mass-resolved UVJ diagram—rest-frame $U-V$ versus $V-J$ colors split by stellar mass—together with the attenuation-reddening relation $A_V$ versus $A_V-A_J$. The UVJ diagram isolates the red sequence of massive dusty star-forming galaxies, and the $A_V$ versus $A_V-A_J$ plane distinguishes a dust screen (steep, linear, slope 0.664 with the THEMIS extinction curve) from mixed star-dust geometries (flatter, saturating relations). The paper uses SKIRT radiative transfer on TNG100 as the standard machinery, and then replaces it with a two-component toy model in which star particles younger than $t_\\mathrm{split}$ are dimmed through a THEMIS screen of optical depth $\\tau_V$, with screen optical depths at U and J set by the THEMIS extinction ratios. That toy model is what identifies the requirement that screens must cover stellar populations up to $\\approx1$ Gyr.","core_discovery":"The central claim is that no dust radiative transfer setup explored here can produce the observed V−J colors of massive star-forming galaxies at $z\\approx2$, because the attenuation-reddening relation of simulated galaxies is too flat. With TNG100 stellar populations, BPASS SEDs, TODDLERS for young regions, and THEMIS diffuse dust, the simulated galaxies reach median V−J colors still $\\approx0.9$ mag bluer than the JWST/NIRCam distribution, and their V-band magnitudes are $\\approx1.6$ mag too bright. Reverse-engineering the needed attenuation gives 1.11 mag of V−J reddening, 0.41 mag of U−V reddening, and $A_V\\approx2.05$ mag, while the simulated massive star-forming galaxies have median $A_V\\approx0.35$ mag. A thin dust screen has a steep, linear relation between $A_V$ and $A_V-A_J$, with slope 0.664 for THEMIS, and only such a steep relation reaches the observed colors; the simulated relation flattens at high optical depth because stars and dust are mixed. The paper's positive result is a toy model in which each stellar population younger than $t_\\mathrm{split}$ is attenuated by an isolated screen, which matches the data for $t_\\mathrm{split}\\approx1.6$ Gyr and screen optical depth $\\tau_V\\approx3.2$.","pith_inferences":["If the screen-like geometry is correct, dust radiative transfer codes need a subgrid prescription that binds dust to stellar populations up to ~1 Gyr old, not only to <30 Myr birth clouds.","The paper implies the tension should peak at the epoch and mass range where dusty star-forming galaxies dominate, so applying the same mass-resolved comparison at $z<1.5$ and $z>3$ would map where the missing reddening begins and ends.","Part of the observed red V−J sequence may be template-dependent: the paper shows that updating EAZY templates moves CANDELS/3D-HST massive galaxies from a pile-up at $V-J\\approx1.8$ to redder colors, so the same comparison with even newer SED libraries is worth running.","Stacked rest-frame UV-to-mid-IR SEDs of massive $z\\approx2$ star-forming galaxies could test whether their integrated attenuation curves are as steep as a screen, as the toy model predicts."],"forward_implications":["Reproducing the JWST/NIRCam UVJ data requires an attenuation-reddening relation as steep as a dust screen, which none of the explored radiative transfer models delivers.","Massive star-forming TNG100 galaxies would need $A_V \\approx 2.05$ mag and 1.11 mag of V−J reddening, far above the simulated median $A_V \\approx 0.35$ mag.","In the toy model, dust screens must cover stellar populations up to $\\gtrsim1$ Gyr; at $t_\\mathrm{split}=1.6$ Gyr about 73% of stellar mass is obscured.","A constant dust-to-metal ratio cannot reproduce the observed mass trend: low-mass star-forming TNG100 galaxies are too red while massive ones are too blue.","The failure persists across TNG100, SIMBA, and EAGLE and across SED-template and dust-model variations, pointing to a systematic gap in current modeling."],"supporting_citations":[{"why":"Provides the SIMBA dust-attenuated UVJ colors used for comparison and shows that low-mass galaxies dominate the red population in that simulation, unlike observations.","marker":"Akins et al. 2022"},{"why":"Supplies an alternative set of TNG100 dust-attenuated colors ('Dust model C') based on line-of-sight extinction, which also fails to reproduce the massive red sequence.","marker":"Nelson et al. 2018"},{"why":"Provides public EAGLE SKIRT dust-attenuated fluxes used in the comparison to test whether the failure is simulation-specific.","marker":"Camps et al. 2018"},{"why":"The DirtyGrid library of analytical dust radiative transfer models; used to identify the homogeneous shell geometry as the only standard geometry that can match the observed colors.","marker":"Law et al. 2018"},{"why":"Defines the UVJ demarcation line used to classify star-forming versus quiescent galaxies in observations and to quantify quiescent-region contamination.","marker":"Williams et al. 2009"},{"why":"Provides the CANDELS/3D-HST photometric catalogs whose rest-frame colors are compared with JWST/NIRCam and rederived with newer templates.","marker":"Skelton et al. 2014"},{"why":"Establishes the TNG100+SKIRT UVJ post-processing methodology in the local Universe that this paper extends to $z\\approx2$.","marker":"Gebek et al. 2024"},{"why":"The TODDLERS SED library used to model unresolved dust around young (<30 Myr) stellar populations.","marker":"Kapoor et al. 2023"},{"why":"Observed specific dust masses used to validate that the TNG100 dust masses adopted in the radiative transfer are plausible.","marker":"Donevski et al. 2020"}],"fun_headline_variants":["Standard dust models leave z~2 galaxies 0.9 mag too blue","Cosmic-noon galaxy colors defy standard dust physics","Only screen-like dust can redden massive z~2 galaxies","TNG100+SKIRT: no standard dust model matches UVJ data","Massive star-forming galaxies need dust screens, not mixtures"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that the simulated dust-free V−J colors of massive galaxies at $z\\approx2$, set by TNG100 stellar ages and metallicities through the BPASS templates, are realistic; if the intrinsic colors are too blue, the required dust reddening is overestimated and the discrepancy may be a stellar-population problem.","fun_headline_variants_meta":{"raw":{"variants":["Standard dust models leave z~2 galaxies 0.9 mag too blue","Cosmic-noon galaxy colors defy standard dust physics","Only screen-like dust can redden massive z~2 galaxies","TNG100+SKIRT: no standard dust model matches UVJ data","Massive star-forming galaxies need dust screens, not mixtures"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0002,"raw_usage":{"total_tokens":1499,"prompt_tokens":1195,"completion_tokens":304,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":811,"completion_tokens_details":{"reasoning_tokens":215}},"tokens_in":811,"tokens_out":304,"duration_ms":3606,"temperature":1.0,"reasoning_tokens":215,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T17:36:49.117794+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the dust-free (attenuation-corrected) V−J colors of massive star-forming galaxies at $z\\approx2$, for example from Balmer-decrement-selected samples with spatially resolved spectroscopy; if the intrinsic colors already match the observed colors, the inferred screen reddening disappears. Alternatively, resolved HST/JWST imaging of dusty star-forming galaxies that finds no dust screens with $\\tau_V\\approx3$ covering populations up to ~1 Gyr would falsify the toy model.","supporting_citations":[],"review_version":1}