{"id":"e3eee681-90b6-42b2-9a37-e445b9b5fdd8","arxiv_id":"1908.02372","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A sample of 39 ALMA-detected, optically invisible massive galaxies at z>3 shows that dark dusty galaxies dominate the massive galaxy population in the early universe.","lead":"Astronomers used ALMA to detect 39 massive, dusty galaxies at redshift above 3 that are invisible in even the deepest Hubble and near-infrared images. These galaxies appear to be more numerous than ultraviolet-bright galaxies of the same mass, suggesting that most massive early galaxies were previously missed.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Photometric-redshift systematics are the load-bearing weakness: the z>3 assignment for most ALMA-detected H-dropouts is weakly constrained, and the headline density, SFRD, and halo mass all scale with that assumption.","rationale":"The reader's weakest-assumption analysis identifies exactly the point that carries the paper's quantitative conclusions. My stress pass looked for other internal problems: ALMA flux extraction, selection completeness, stacked-SED interpretation, and clustering. These hold up reasonably well. The ALMA detections are clean, with positional agreement with IRAC; the incompleteness correction is independently supported by the blind GOODS-ALMA/ASAGAO comparison; and the stacked far-infrared SED peak does support a dusty high-z population on average. The real weak point is the sample-level redshift distribution. Because H-dropouts lack H-band detections, the SED fits have weak leverage; the full-template EAzY fits produce broad and sometimes low-z solutions. All headline quantities—space density, star-formation-rate density, and halo mass—are integrals over the z_phot distribution, and a non-negligible z≈2 contaminating population would degrade all of them in the same direction. The paper is honest about this limitation, and the reader's CONDITIONAL verdict is the right calibration. I do not see a reason to move the verdict; instead, the proposed spectroscopic test is the decisive next step.","tokens_in":23215,"tokens_out":9209,"duration_ms":112398,"concrete_test":"Run ALMA CO(4→3) or [CII] 158 μm line scans on a flux- and mass-stratified random subset of at least 10 of the 39 ALMA-detected H-dropouts to obtain spectroscopic redshifts independent of dust templates. Decide the concern by two numbers: the fraction with z_spec<3 and the median Δz/(1+z_spec) between z_spec and z_phot. If more than ~25% of the subset falls at z<3, or if the median offset exceeds ~0.2, recompute the z>3 space density, SFRD, and clustering bias with the measured redshift distribution; the central 'dominant population' claim stands only if the corrected density stays within a factor of ~2 of 2×10^-5 Mpc^-3 and the H-dropout SFRD still exceeds that of massive LBGs at z>3.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim—that ALMA-detected H-dropouts are the bulk population of massive galaxies missed at z>3—requires that essentially all 39 sources really are at z>3. The evidence for this is the photometric-redshift distribution, and the paper itself labels those redshifts 'admittedly uncertain' and notes that spectroscopic confirmation is limited to two Lyα sources (z=3.097 and z=5.113) plus one literature source at z=3.709. For the remaining sources, EAzY fits use photometry in which H-band is only an upper limit; the code was run without a magnitude prior and with the full template set, including old-and-dusty templates. Extended Data Table 2 shows the fragility: several ALMA-detected objects have best-fit z_phot below 3, including GDS-43215 at z=2.91, UDS-34637 at z=2.84, UDS-37649 at z=2.82, and COS-31483 at z=2.97, and many individual 68% intervals span Δz≈1–3. The far-infrared and radio color diagnostics (S870/S450, S1.4/S870, S870/S24) are applied only to subsets or as lower limits and are template dependent; they can flag a dusty high-z population on average, but they do not directly measure the contamination fraction. If a non-negligible fraction of the 39 are actually z≈2 dusty galaxies—a population with similarly red H−[4.5] colors is known at those redshifts—the sources leave the z>3 volume, the quoted space density of 2×10^-5 Mpc^-3 shrinks by the same fraction, the SFRD 'ten times larger than massive LBGs' is reduced by both the removal and the lower 870-μm K-correction, and the halo-mass and progenitor interpretation loses its anchor. The conclusion may be right, but the quantitative headline is currently supported by a redshift assumption, not by measured redshifts.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The Letter reports ALMA 870 micron continuum observations of 63 H-dropouts (IRAC [4.5]<24, H-band undetected to ~27 mag) in three CANDELS fields, detecting 39 sources at S/N>4 with fluxes 0.6-8 mJy. The authors interpret these ALMA-detected H-dropouts as massive dusty star-forming galaxies at z>3 on the basis of photometric redshifts (median z~4), two new Ly-alpha spectroscopic redshifts (z=3.097 and z=5.113), one literature source at z=3.709, and far-infrared/radio color arguments. They derive a space density of ~2e-5 Mpc^-3, a star-formation-rate density about ten times that of equally massive LBGs, and a clustering bias implying Mh~1e13 Msun, concluding that H-dropouts represent the bulk population of massive galaxies missed by LBG selection at z>3. The ALMA data are clean and the analysis is careful, but the z>3 assignment for most of the sample rests on photometric redshifts that the authors themselves call 'admittedly uncertain.'","tokens_in":23627,"tokens_out":5742,"duration_ms":63169,"significance":"If the z>3 interpretation is correct, this is a significant result: it identifies a sizable population of massive, dusty galaxies missed by UV selection, raises the inferred cosmic SFR density at z~4-6, and provides a quantitative target for galaxy formation models. The paper's strengths include the clean ALMA detections with positional agreement to IRAC, the stacked far-infrared SED giving L_IR=2.2e12 Lsun and T_dust=36.7 K, the direct comparison with external samples (ZFOURGE, ALESS, UV luminosity functions), and public code/data availability. However, the headline abundance, SFRD, and halo mass are all conditional on the photometric-redshift distribution, and the authors' own text and Extended Data Table 2 show that individual redshifts are highly uncertain, with some best-fit values below z=3. The significance is therefore not yet fully established.","major_comments":[{"comment":"The z>3 assignment for most of the 39 ALMA-detected H-dropouts is the load-bearing assumption for the paper's central claims, yet only three spectroscopic redshifts are available (two new Ly-alpha detections and one literature source at z=3.709). EAzY was run with the full template set including old-and-dusty templates and without a magnitude prior, and Extended Data Table 2 lists several ALMA-detected objects with best-fit z_phot below 3, including GDS-43215 at z=2.91, UDS-34637 at z=2.84, UDS-37649 at z=2.82, and COS-31483 at z=2.97, with many individual 68% intervals spanning dz~1-3. Because the space density, star-formation-rate density, and halo mass estimates all scale with the assumption that essentially all sources are at z>3, the authors should quantify the expected contamination from z~2 dusty galaxies with similar H-[4.5] colors and show the main results with those sources removed or marginalized over the full photometric-redshift probability distributions. The statement in the main text that 'all the available data points to the ALMA-detected H-dropouts being massive, dusty star-forming galaxies at z>3' is too strong given the large individual redshift uncertainties.","section":"Methods §2.2 and Extended Data Table 2"},{"comment":"The far-infrared and radio color diagnostics invoked in support of z>3 do not directly measure the contamination fraction of the full sample. The S870/S450 and S1.4/S870 colors are available only for the CANDELS-COSMOS subset, are template dependent (the model lines in Extended Data Fig. 4 depend on the assumed infrared SED), and for non-detections only upper or lower limits enter the comparison. The comparison between z_FIR and z_opt in Extended Data Fig. 4c shows substantial dispersion, so these colors can indicate that the population is dusty and high-redshift on average but cannot certify individual z>3 membership for the 39 sources. The manuscript should state this limitation explicitly and treat the far-infrared colors as supporting rather than confirming evidence.","section":"Extended Data Fig. 4 and Methods §1.4"},{"comment":"The clustering measurement assumes that the 39 ALMA-detected H-dropouts occupy the same cosmic volume as the H-selected galaxy sample at 3.5<z<5.5. If a non-negligible fraction of the sample is actually at z~2, the cross-correlation amplitude, and hence the inferred bias b=8.4±1.5 and halo mass Mh~1e13 h^-1 Msun, would be diluted and would not describe z>3 host halos. The authors should either restrict the clustering analysis to sources with more secure z>3 membership or present the halo-mass result as conditional on the photometric-redshift distribution rather than as an independent confirmation of the z>3 interpretation.","section":"Methods §2.3 and Fig. 4"}],"minor_comments":[{"comment":"There are typos in the caption: 'Red filed circles' should be 'Red filled circles', and 'Purple fileld pentagons' should be 'Purple filled pentagons'.","section":"Fig. 3 caption"},{"comment":"The phrase 'more than than 80%' should be 'more than 80%'.","section":"Main text, paragraph 6"},{"comment":"The text refers to 'CANDLES-COSMOS'; the survey name should be 'CANDELS-COSMOS'.","section":"Methods §1.3"},{"comment":"The main text cites 'Extended Data Figure 2' for the redshift and stellar mass distributions of ALMA-detected versus undetected H-dropouts; the relevant panel is Extended Data Fig. 2b, so the callout should be made panel-specific.","section":"Extended Data Fig. 2 callout"},{"comment":"The sentence 'The prior on the observed magnitudes was not used' is a nonstandard modeling choice for EAzY and deserves a brief justification, since using or not using the prior can shift the photometric redshift distribution.","section":"Methods §2.2"}],"recommendation":"major_revision","confidential_remarks":"The paper is suitable for the journal and the ALMA data are valuable. My concern is the gap between the strength of the headline claim ('dominant population', 'majority of the most massive galaxies ... missed') and the redshift evidence, which the authors themselves describe as 'admittedly uncertain'. I would ask the editor to require that either the photometric-redshift contamination be quantified and the main numbers recomputed, or that the abstract and title be scaled back to reflect the conditional nature of the z>3 assignment."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things before reading it: (1) this is the paper that turned H-dropouts from a curiosity into a statistical population, with 39 clean ALMA detections and a publicly documented selection; (2) the entire quantitative punchline—space density, SFRD ten times LBGs, halo mass—rests on photometric redshifts for 37 of the 39 sources, which the authors themselves call 'admittedly uncertain.' Read it for the data, not for the headline numbers.\n\nWhat's genuinely new: 1.8-min ALMA integrations at 870 µm reaching 0.6 mJy detect 62% of [4.5]-bright, H-faint sources. That's a real step beyond the handful of extreme SMGs previously known. The stacked IR SED gives L_IR=2.2e12, T_dust=36.7 K, consistent with normal massive star-forming galaxies at z~4, not just starbursts. The selection, photometry, and completeness correction (factor 2–2.4) are described carefully. The authors also connected to external surveys—ZFOURGE LBGs, ALESS SMGs, UV LFs—so the comparison is reproducible. Credit where due: this is a solid observational advance.\n\nThe soft spot is exactly where the stress-test puts it: redshift. Only two sources have spec-z from their own X-Shooter follow-up (z=3.097, z=5.113), plus one similar-flux source in the literature at z=3.709. For the rest, EAzY fits with no magnitude prior and the full template set; several best-fit z_phot fall at z<3 (GDS-43215 at 2.91, UDS-34637 at 2.84, UDS-37649 at 2.82, COS-31483 at 2.97), and the 68% intervals often span 1–3 in redshift. The far-IR/radio colors (S870/S450, S1.4/S870, S870/S24) are suggestive but template-dependent and applied to subsets. If even 30–40% of the sample is actually at z~2–3, the claimed space density and SFRD drop by a comparable factor, and the 'dominant population' claim weakens to 'a significant population.' The authors do flag this; the paper is honest. But the abstract reads stronger than the data support.\n\nThe clustering bias b=8.4±1.5 is a first attempt but with 39 tracers the error bars are likely underestimated; treat that as indicative.\n\nBottom line: this is a paper to engage with and cite—it defines the H-dropout population and provides the target list for JWST and ALMA follow-up. It deserves a serious referee and, with more spec-z, would be fully convincing. I'd recommend publishing after the redshift caveat is made more prominent, or with a revised abstract that states the density is measured under the photo-z assumption.","headline":"A solid observational advance that defines the H-dropout population, but the z>3 verdict rests on photometric redshifts the authors themselves call uncertain; cite it, read it, and wait for spectroscopy.","tokens_in":24229,"tokens_out":3492,"would_cite":true,"duration_ms":33057,"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":"A population of massive dusty galaxies at $z>3$, invisible in optical and near-infrared light, represents the bulk of massive galaxies in the early Universe and dominates their star formation.","keywords":["dusty star-forming galaxies","H-dropouts","massive galaxies at high redshift","cosmic star-formation history","submillimeter observations","Lyman-break galaxies","galaxy formation","z > 3 galaxies"],"falsifier":"Measure spectroscopic redshifts for a random, statistically meaningful subset of the 39 submillimeter-detected H-dropouts, using submillimeter lines such as [C II] 158-micron or CO, or near-infrared spectroscopy. If more than about a quarter of the sample turns out to lie at $z<3$, the $z>3$ space density and star-formation-rate density would fall below those of equally massive ultraviolet-bright galaxies, and the claim that H-dropouts dominate the massive end at $z>3$ would be falsified.","tokens_in":23038,"feed_emoji":"🌌","tokens_out":12718,"duration_ms":123865,"temperature":0.7,"pith_summary":"The paper argues that the most massive galaxies in the first two billion years after the Big Bang (redshifts $z>3$) have been systematically missed by surveys that select galaxies in rest-frame ultraviolet light. It identifies a population of 39 galaxies that are bright at 4.5 microns but invisible in the deepest near-infrared H-band imaging, and detects them in 870-micron submillimeter continuum. These objects have a space density of about $2\\times10^{-5}$ per cubic megaparsec—two orders of magnitude higher than extreme starbursts—and star-formation rates near 200 solar masses per year, making them plausible 'bulk' massive galaxies rather than rare outliers. The paper concludes that they contribute ten times more star-formation-rate density than equally massive ultraviolet-bright galaxies at $z>3$, that they live in the most massive dark-matter halos of their epoch, and that their abundance challenges current models of massive-galaxy formation.","feed_headline":"Most massive early galaxies are optically invisible","feed_subtitle":"39 dusty galaxies at z>3 outshine equally massive UV-bright galaxies in star formation ten to one","key_machinery":"The load-bearing object is the H-dropout: a galaxy detected at 4.5 microns but not in deep H-band imaging, which selects red, potentially high-redshift massive galaxies. The argument is carried by 870-micron submillimeter detections of 39 such objects; dust continuum at this wavelength directly traces obscured star formation, and ratios such as 870-to-450 micron and radio-to-870 micron provide redshift constraints that support $z>3$. Stacking the far-infrared photometry fixes the mean dust temperature and infrared luminosity, converting individual faint detections into a reliable population-level spectral energy distribution. Finally, the angular cross-correlation of the H-dropouts with surrounding H-band-selected galaxies provides the galaxy bias from which the dark-matter halo mass is inferred.","core_discovery":"Using short 870-micron observations of H-dropouts—galaxies selected to be bright in 4.5-micron light and undetected in near-infrared H-band imaging deeper than $H\\approx27$ mag—the paper detects 39 of 63 candidates. Stacking their infrared spectral energy distributions yields a median redshift of $z\\approx4$, a median stellar mass of $10^{10.6}\\,M_\\odot$, a dust temperature of $36.7\\pm2.1$ K, and an infrared luminosity $L_{\\mathrm{IR}}=2.2\\pm0.3\\times10^{12}\\,L_\\odot$, typical of massive star-forming galaxies at that redshift. The 870-micron fluxes (0.6–8 mJy, median 1.6 mJy) imply star-formation rates of roughly $200\\,M_\\odot\\,\\mathrm{yr}^{-1}$. After correcting for incompleteness, the space density is about $2\\times10^{-5}\\,\\mathrm{Mpc}^{-3}$, and the star-formation-rate density of these galaxies is about 10 times that of equally massive Lyman-break galaxies. Cross-correlating the sample with H-band-selected neighbors gives a bias $b=8.4\\pm1.5$, corresponding to halos of mass $10^{13\\pm0.3}\\,h^{-1}M_\\odot$ at $z=4$, so the paper identifies them as the likely ancestors of the largest ellipticals in today's groups and clusters. Its central claim is that the majority of the most massive galaxies at $z>3$ have been missed by Lyman-break selection and are optically dark.","pith_inferences":["Editorial inference: if the photometric redshifts are systematically overestimated, the derived space density and star-formation-rate density at $z>3$ would shrink; measuring [C II] 158-micron or CO redshifts for a random subset of the 39 detections would test this directly.","Editorial inference: the completeness correction for chance superpositions reaches a factor of about two, so the true space density could be even higher; deeper submillimeter imaging would reveal whether fainter H-dropouts add to the census.","Editorial inference: extending the same H-dropout color selection to wider, shallower surveys would show whether the claimed dominance holds beyond the three deep fields analyzed here.","Editorial inference: a substantial contamination by $z\\sim2$ dusty galaxies would move the population out of the 'early universe' window and weaken the challenge to early galaxy formation; the comparison with equally massive ultraviolet-bright galaxies at $z>3$ would then need to be renormalized."],"forward_implications":["The cosmic star-formation-rate density at $z>3$ includes a large dusty component invisible to ultraviolet surveys; massive H-dropouts alone contribute roughly ten times the star-formation-rate density of equally massive Lyman-break galaxies.","Their space density is comparable to that of the most massive quiescent galaxies at $z\\sim3$, giving the early Universe enough massive progenitors to explain the rapid appearance of quiescent ellipticals.","With halo masses around $10^{13}\\,M_\\odot$ at $z\\approx4$, these galaxies are the natural ancestors of the giant ellipticals that now anchor groups and clusters.","Current semi-analytic models underpredict the abundance of such galaxies by one to two orders of magnitude, so if the measurement stands, galaxy formation models need revision.","Spectroscopic follow-up of the entire H-dropout population—particularly at mid-infrared wavelengths—would confirm the redshifts and sharpen all derived densities."],"supporting_citations":[{"why":"Supplies the ultraviolet-based cosmic star-formation history that the paper shows to be incomplete.","marker":"1"},{"why":"Provides a spectroscopically confirmed extreme H-dropout starburst at $z=5.18$, demonstrating that such objects exist.","marker":"2"},{"why":"Gives the extreme-starburst sample whose space density is compared to the H-dropouts.","marker":"5"},{"why":"Establishes the infrared color selection of massive $z>3$ galaxies that the H-dropout criterion refines.","marker":"11"},{"why":"Supplies the bright submillimeter-galaxy comparison sample with measured 870-micron fluxes.","marker":"12"},{"why":"Provides the empirical dust-temperature and infrared SED calibration used to derive $L_{\\mathrm{IR}}$ and star-formation rates.","marker":"13"},{"why":"Confirms a similar-flux H-dropout at $z=3.709$, direct evidence that such objects exist at $z>3$.","marker":"15"},{"why":"Gives the semi-analytic galaxy formation model whose predicted H-dropout density is compared and found too low.","marker":"19"},{"why":"Frames the tension between observed massive galaxies and predicted halo abundances in the cosmological model.","marker":"21"}],"fun_headline_variants":["Massive early galaxies hide in plain sight","Optically dark galaxies dominate early massive population","Hidden giants: most massive early galaxies are UV-invisible","Survey misses bulk of massive early galaxies","Dusty behemoths outnumber UV-bright peers at z>3"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the photometric redshifts place the 39 submillimeter-detected H-dropouts at $z>3$—a premise the paper itself calls 'admittedly uncertain'—so if many of these galaxies are actually at $z\\approx2$, the derived space density, star-formation-rate density, and halo masses at $z>3$ would drop and the central claim would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Massive early galaxies hide in plain sight","Optically dark galaxies dominate early massive population","Hidden giants: most massive early galaxies are UV-invisible","Survey misses bulk of massive early galaxies","Dusty behemoths outnumber UV-bright peers at z>3"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000792,"raw_usage":{"total_tokens":3605,"prompt_tokens":1180,"completion_tokens":2425,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":796,"completion_tokens_details":{"reasoning_tokens":2349}},"tokens_in":796,"tokens_out":2425,"duration_ms":18807,"temperature":1.0,"reasoning_tokens":2349,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:46:50.735372+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure spectroscopic redshifts for a random, statistically meaningful subset of the 39 submillimeter-detected H-dropouts, using submillimeter lines such as [C II] 158-micron or CO, or near-infrared spectroscopy. If more than about a quarter of the sample turns out to lie at $z<3$, the $z>3$ space density and star-formation-rate density would fall below those of equally massive ultraviolet-bright galaxies, and the claim that H-dropouts dominate the massive end at $z>3$ would be falsified.","supporting_citations":[{"cited_title":"& Dickinson, M","cited_arxiv_id":null,"evidence_quote":"Supplies the ultraviolet-based cosmic star-formation history that the paper shows to be incomplete."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides a spectroscopically confirmed extreme H-dropout starburst at $z=5.18$, demonstrating that such objects exist."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the extreme-starburst sample whose space density is compared to the H-dropouts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the infrared color selection of massive $z>3$ galaxies that the H-dropout criterion refines."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the bright submillimeter-galaxy comparison sample with measured 870-micron fluxes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the empirical dust-temperature and infrared SED calibration used to derive $L_{\\mathrm{IR}}$ and star-formation rates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Confirms a similar-flux H-dropout at $z=3.709$, direct evidence that such objects exist at $z>3$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the semi-analytic galaxy formation model whose predicted H-dropout density is compared and found too low."},{"cited_title":"L., Capak, P., Masters, D","cited_arxiv_id":null,"evidence_quote":"Frames the tension between observed massive galaxies and predicted halo abundances in the cosmological model."}],"review_version":1}