{"id":"fd84431d-48f0-45f9-ac06-b2e99f725ec8","arxiv_id":"2411.11212","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Stacking ALMA 1.2 mm images of 4,103 lensed galaxies yields average dust masses that decline with redshift from z=1 to z=5 and scale with stellar mass and SFR.","lead":"Using ALMA observations of 33 galaxy clusters, the authors measured the average dust content of about 4,100 distant galaxies by stacking their faint millimeter signals. They found that galaxies at earlier cosmic times carried less dust, and that dust mass grows with stellar mass and star-formation rate.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"High-redshift dust masses may be biased low by missing ALMA-detected dusty galaxies absent from the HST/IRAC catalog; the claimed decline and model tension could be partly a selection effect.","rationale":"The strongest_claim has two parts: a redshift decline and positive scaling relations with M* and SFR. The scaling relations are supported by multiple detections and by median-stack and quiescent-exclusion checks; I do not see a serious challenge to the qualitative correlation with M* and SFR. The fragile part is the redshift decline and the 'lower than models at high z' comparison. Both can be produced by the acknowledged incompleteness of the input catalog for mm-bright galaxies. Section 5 quantifies this at least partially: 35 of 180 ALMA SNR>4 sources are absent from the HST/IRAC catalog, and the effect is expected to grow at high z where dust-obscured galaxies drop out of optical/IR selection. This is a selection effect on the very quantity being averaged. The fixed-T SED issue is a genuine systematic but is less decisive: plausible mass-weighted temperatures (15-45 K) shift the highest-z points by roughly 50%, and the commonly assumed evolution toward hotter dust would lower Mdust further, making the observed decline stronger, not creating it. Incompleteness, by contrast, biases the high-z averages in the direction of the headline result. Therefore the single check that would settle the matter is a re-stack that includes the missing ALMA detections, or at least a bounding calculation of their contribution. If that test leaves the z=3-5 averages unchanged, the paper's conclusions stand; if it moves them substantially, the conclusions need to be weakened. The paper's own transparency and robustness tests (median stacks, quiescent exclusion, bootstrap) are real strengths, but none of them addresses this selection effect. Hence I keep the reader's CONDITIONAL verdict: not a rejection, but the quantitative redshift trend should be presented as conditional on completeness.","tokens_in":24866,"tokens_out":7892,"duration_ms":84552,"concrete_test":"Cross-match the full ALCS SNR>4 continuum catalog against Kokorev et al. (2022) and Guerrero et al. (2023); take the ~35 missing sources, assign redshifts from any available ALMA/other photometry or bracket them with z=1-5 priors, and re-run the mean stacks with these sources included in the corresponding M*/SFR/redshift bins (with magnification assigned from cluster models). If adding them raises any z=3-5 bin's average Mdust by ~0.3 dex or converts a 3-sigma upper limit into a detection, the central decline is not robust to catalog incompleteness; if the bins move by <0.1 dex, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing weakness is not the fixed SED but the sample selection. The stacks are built from the Kokorev et al. (2022) HST/IRAC catalog, and Section 5 states that of the 180 ALCS sources detected at SNR>4, only 145 are in that catalog. The 35 missing sources are exactly the mm-bright, dust-obscured galaxies that a UV/optical selection tends to miss, and the authors note the bias should be strongest at high z. Because these sources are absent, the mean 1.2 mm flux in the z=3-5 bins is biased downward; the same bins are mostly 3-sigma upper limits. This is a one-directional bias that acts precisely in the direction of the paper's headline result: lower average dust mass at high z and values below model predictions. The fixed-T=25 K assumption is a real normalization uncertainty, but with mass-weighted temperatures in the 15-45 K range it changes high-z masses by roughly 50%, and if T increases with redshift it strengthens rather than explains the decline. The incompleteness bias is unquantified and could be several times larger. The paper acknowledges it (Section 5) but does not correct for it, so the quantitative redshift trend and the model comparison are conditional on an untested completeness correction.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses ALMA Band 6 continuum imaging from the ALCS survey to stack approximately 4100 lensed galaxies from the Kokorev et al. (2022) HST/IRAC catalog, binning them in redshift, stellar mass, and star-formation rate. From the stacked 1.2 mm fluxes it derives average dust masses using Eq. (1), which assumes a single optically thin modified blackbody with a fixed 25 K mass-weighted dust temperature, and it reports a decline of average dust mass with redshift, positive scaling relations with stellar mass and SFR, and broad agreement with models at z ~ 1-3 but lower-than-predicted dust masses at higher redshift. The analysis includes mean and median stacking, bootstrap and RMS uncertainties, magnification corrections, and a comparison to semi-analytic models using both model dust masses and model-predicted band-6 fluxes.","tokens_in":25121,"tokens_out":7602,"duration_ms":71618,"significance":"If the reported trends are robust, this paper provides a valuable statistical census of dust in typical (not just individually bright) galaxies from z = 1 to z = 5, extending stacking measurements to lower stellar masses and SFRs than most previous work. The study's strengths include the use of a large, homogeneously observed lensed sample; explicit median-stack and bootstrap consistency checks, which support the qualitative trends; CMB correction following da Cunha et al. (2013); and a careful model comparison that distinguishes direct model dust masses from masses derived from predicted band-6 fluxes. The qualitative relations โ€” dust increasing with stellar mass and SFR and decreasing with redshift โ€” are plausible and likely to hold. However, the quantitative redshift decline and the claimed tension with models at high redshift rest on selection and SED assumptions that the paper acknowledges but does not correct, so the result is best regarded as conditional until those biases are bounded.","major_comments":[{"comment":"The sample is drawn from an HST/IRAC catalog that misses 35 of the 180 ALCS sources detected at SNR > 4, as the authors acknowledge in Section 5. Because the missing sources are preferentially the most dusty, and because the effect is expected to be strongest at high redshift, the stacked fluxes in several high-redshift bins (Tables 3 and 4, many of which are 3-sigma upper limits) are biased downward in exactly the direction of the headline result โ€” the decline in average dust mass with redshift and the low values compared to models. The paper does not quantify or correct this incompleteness. The authors should attempt to bound this bias, for example by adding the 35 missing sources using their ALMA positions in the stacks, by reweighting the sample with a completeness model, or by presenting the maximum plausible correction; without this, the redshift trend and model comparison remain conditional.","section":"Section 5 and Section 2.2"},{"comment":"The dust mass conversion assumes a single optically thin modified blackbody with T_rest = 25 K and beta = 1.8 at all redshifts. Since the observed band is fixed, the rest-frame wavelength shifts from about 0.5 mm at z = 1 to about 0.2 mm at z = 5, so Eq. (1) progressively moves off the Rayleigh-Jeans tail and becomes sensitive to the assumed temperature and SED shape. The authors note the ~T^-1 dependence and quote a ~50% change at T = 40 K, but this does not bound the effect of a plausible mass-weighted temperature evolution: a change of ~10 K across the probed redshift range can alter high-z masses by factors comparable to the claimed decline, and the adopted temperature also sets the absolute values used for the model comparison. The paper should present the redshift trend under at least two plausible T(z) prescriptions (e.g., constant 25 K and a mild increase with redshift) to demonstrate that the qualitative decline is robust.","section":"Section 3.3, Eq. (1)"},{"comment":"The scaling relations are fitted using detections only, ignoring the 3-sigma upper limits, as stated in Section 4.1. At high redshift most points are upper limits, so the fitted slopes for Mdust(z) (e.g., b = -1.8 +/- 0.7 and -1.0 +/- 0.3 in Table 5) are likely biased by the detection threshold, which preferentially selects bright sources. A censored likelihood or survival analysis should be used, or the fits should be repeated with upper limits included to bound the effect. In addition, the fitted normalization for the Mdust-M* relation at z = 1-2 (a = 5.2 +/- 13.0 x 10^3 in Table 5) is effectively unconstrained, so the paper's claim of having derived a scaling relation with stellar mass is not supported by this fit; this should be stated explicitly.","section":"Section 4.1 and Table 5"},{"comment":"The sample's stellar mass function shows an overdensity at z > 4 and an under-density at 2 < z < 3 relative to COSMOS2020 (Figure 3), which the authors attribute to photometric redshift misclassification of lower-redshift sources. As they note, this would overestimate dust content at z > 3, working in the opposite direction of the missing dusty galaxies. The net redshift trend therefore depends on two unquantified and competing biases. The authors should attempt to quantify the photo-z contamination, for example by propagating the redshift uncertainties into the stacked fluxes or by re-fitting with the z > 3 bins adjusted to the COSMOS2020 stellar mass function.","section":"Section 5 and Figure 3"}],"minor_comments":[{"comment":"The summary states that the sources were binned in 'five different redshift bins,' but Section 2.3 defines only four redshift bins (13 (plausible, and the paper says the bias could be strongest there), the stacked 1.2 mm flux in the high-z bins is biased low in exactly the direction of the paper's headline decline and the model tension. The authors flag this in Section 5 but don't quantify it. That's the load-bearing caveat. Second, the fixed T=25 K dust SED: a real normalization uncertainty, but at the ~50% level, and if temperature rises with redshift it would make the decline steeper, not explain it. Third, the scaling-relation fits ignore upper limits; minor, but worth noting since several of the high-z points are upper limits.\n\nNone of this is a hidden error; the paper is candid about the SED and the missing sources. But the abstract's 'steady decline' and the model comparison at z>3 should be read as conditional on an untested completeness correction. The low-z results and the low-mass/low-SFR measurements stand on their own.\n\nThis paper deserves a serious referee. I'd send it to review, and I'd ask for a quantitative treatment of the missing detections -- even a simple stacking of the 35 missing sources or a number-counts-based correction would help. Useful for anyone working on dust scaling relations; I'd cite the low-z results with confidence.","headline":"Solid, transparent stacking analysis reaching new low-mass/low-SFR regimes; the low-z scaling relations are likely robust, but the high-z decline is conditional on an unquantified catalog-incompleteness bias.","tokens_in":25747,"tokens_out":6425,"would_cite":true,"duration_ms":58233,"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":"By stacking 4,103 lensed galaxies, this analysis shows a steady decline in average dust mass with redshift from z=1 to z=5, a rise with stellar mass and star formation rate, and lower high-redshift dust masses than current models predict.","keywords":["dust mass","stacking","ALMA","gravitational lensing","galaxy evolution","star formation rate","stellar mass","high-redshift galaxies"],"falsifier":"Stack the same lensed sources in a second ALMA band, for example 3 mm, so that the same rest-frame Rayleigh-Jeans wavelength is probed at all redshifts; if the decline in derived dust mass persists, the central claim survives, while if the decline flattens or vanishes, the fixed-temperature SED conversion was responsible for part of it.","tokens_in":24677,"feed_emoji":"๐ŸŒŒ","tokens_out":10901,"duration_ms":168345,"temperature":0.7,"pith_summary":"The paper sets out to measure how the dust content of ordinary galaxies evolves from $z=1$ to $z=5$, and how that dust tracks stellar mass and star formation rate. Dust matters because it catalyses molecular hydrogen formation, obscures starlight, and records the metal build-up of galaxies. The authors stack ALMA 1.2 mm continuum images of 4,103 galaxies behind 33 lensing clusters, turning many individually undetected galaxies into measurable average fluxes. The central result is a steady decline in average dust mass with redshift, with dust mass rising toward higher stellar mass and higher star formation rate, from which scaling relations are derived. If correct, dust accumulates gradually over cosmic time and current models overproduce dust at $z\\sim4$โ€“5.","feed_headline":"Galaxy dust mass falls steadily from z=1 to z=5","feed_subtitle":"Stacking 4,100 lensed galaxies shows dust rises with stellar mass and star formation, and models overpredict it at high z.","key_machinery":"The engine of the analysis is image-domain continuum stacking with the LineStacker tool: uv-tapered ALMA band-6 maps of 33 clusters are cut into 9.76-arcsecond stamps at catalog positions, averaged, and the central flux is integrated in a 2-arcsecond aperture, reaching stacked noise levels near $4\\times10^{-3}$ mJy per beam. That flux is converted to dust mass with a single optically thin modified blackbody, $M_{\\rm dust}=5.03\\times10^{-31}(S_{\\nu_{\\rm obs}}/f_{\\rm CMB})D_L^2 / [(1+z)^4 B_{\\nu_{\\rm obs}}(T_{\\rm obs})\\kappa_{\\nu_0}(\\nu_0/\\nu_{\\rm rest})^\\beta]$, using $T_{\\rm rest}=25$ K, $\\beta=1.8$, and $\\kappa_{\\nu_0}=0.0431$ m$^2$ kg$^{-1}$ at 352.6 GHz, with lensing-magnification and CMB corrections. Uncertainties combine empty-stack RMS, bootstrap resampling, redshift scatter, and a 20% magnification error; non-detections are reported as $3\\sigma$ upper limits.","core_discovery":"On the paper's own terms, the discovery is that average galaxy dust mass declines steadily from $z=1$ to $z=5$ while rising with both stellar mass and star formation rate at fixed redshift. The authors fit scaling relations of the form $M_{\\rm dust}(x)=a x^b$, finding for example $M_{\\rm dust}\\propto {\\rm SFR}^{0.78\\pm0.07}$ at $1\\le z<2$ and $M_{\\rm dust}\\propto z^{-1.0\\pm0.3}$ at fixed stellar mass $10^{10}$โ€“$10^{11}\\,M_\\odot$. Stacked detections span roughly $3\\times10^6$ to $2.6\\times10^8\\,M_\\odot$, while the highest-redshift bins are mostly $3\\sigma$ upper limits. Comparison with galaxy-formation models shows broad agreement at $z\\sim1$โ€“3 but predicted dust masses that are higher than observed at $z>3$.","pith_inferences":["I infer that the strongest systematic to test is the rest-wavelength shift: since the fixed 1.2 mm band probes about 0.5 mm at $z=1$ and about 0.2 mm at $z=5$, a true evolution of the mass-weighted dust temperature would change the high-redshift masses by roughly 50% and could flatten the reported decline.","I infer that a multi-band stacking test, for example with ALMA band 3 near 3 mm, would sample the same rest-frame Rayleigh-Jeans tail at all redshifts and cleanly separate a real dust-mass decline from an SED assumption.","I infer that if JWST-based source catalogs recover the dusty galaxies missed by the HST-IRAC selection, the high-redshift average dust masses are likely to move upward, shrinking the gap with model predictions."],"forward_implications":["Average dust mass drops by roughly an order of magnitude from $z\\sim1$ to $z\\sim5$, so dust accumulation in typical galaxies is a gradual process tied to cosmic time.","At fixed redshift, dust mass increases with both stellar mass and star formation rate, with the SFR relation closer to linear, making star formation rate a useful dust-mass tracer.","At $z>3$ the measured dust masses fall below the predictions of current galaxy-formation models, implying those models overproduce dust in early galaxies if the measurements hold.","Excluding quiescent galaxies raises the average dust masses in stellar-mass-selected stacks, so quiescent galaxies carry less dust for their stellar mass and lower the population averages."],"supporting_citations":[{"why":"Supplies the modified blackbody formula used to convert 1.2 mm flux into dust mass.","marker":"Kovรกcs et al. (2010)"},{"why":"Sets the adopted dust-mass normalisation and provides a reference for comparing dust-mass evolution.","marker":"Magnelli et al. (2020)"},{"why":"Gives the CMB heating and background corrections applied to the measured fluxes.","marker":"da Cunha et al. (2013)"},{"why":"Supports treating long-wavelength continuum as a dust-mass tracer in the optically thin Rayleigh-Jeans regime.","marker":"Scoville et al. (2016)"},{"why":"Provides the LineStacker tool and its empty-stack and bootstrap routines behind all stacked fluxes and uncertainties.","marker":"Jolly et al. (2020)"},{"why":"Supplies the HST-IRAC catalog with redshifts, SFRs, stellar masses, and magnifications that define every stack.","marker":"Kokorev et al. (2022)"},{"why":"Describes the ALCS survey maps and data that the stacking analysis uses.","marker":"Kohno et al. (2023)"},{"why":"Defines the main sequence used to identify and exclude quiescent galaxies in the robustness check.","marker":"Speagle et al. (2014)"},{"why":"Provides the SHARK model predictions, both direct dust mass and band-6 flux, used for the high-redshift comparison.","marker":"Lagos et al. (2019)"}],"fun_headline_variants":["Galaxy dust mass falls from z=1 to z=5","Dust mass rises with stellar mass and star formation rate","Models overpredict galaxy dust at high redshift","ALMA stacking shows dust mass declining across cosmic time","From z=1 to z=5, galaxy dust declines and models miss high z"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise, discussed but not corrected in Section 5, is that the 1.2 mm continuum traces dust mass through a single optically thin 25 K blackbody at every redshift, so if the true mass-weighted dust temperature or emissivity evolves, part of the apparent redshift decline would be an artifact of that assumption.","fun_headline_variants_meta":{"raw":{"variants":["Galaxy dust mass falls from z=1 to z=5","Dust mass rises with stellar mass and star formation rate","Models overpredict galaxy dust at high redshift","ALMA stacking shows dust mass declining across cosmic time","From z=1 to z=5, galaxy dust declines and models miss high z"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000341,"raw_usage":{"total_tokens":1969,"prompt_tokens":1127,"completion_tokens":842,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":743,"completion_tokens_details":{"reasoning_tokens":757}},"tokens_in":743,"tokens_out":842,"duration_ms":8234,"temperature":1.0,"reasoning_tokens":757,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T18:47:49.769240+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Stack the same lensed sources in a second ALMA band, for example 3 mm, so that the same rest-frame Rayleigh-Jeans wavelength is probed at all redshifts; if the decline in derived dust mass persists, the central claim survives, while if the decline flattens or vanishes, the fixed-temperature SED conversion was responsible for part of it.","supporting_citations":[{"cited_title":"2020, , 892, 66","cited_arxiv_id":null,"evidence_quote":"Sets the adopted dust-mass normalisation and provides a reference for comparing dust-mass evolution."},{"cited_title":"K., & Stanley , F","cited_arxiv_id":null,"evidence_quote":"Provides the LineStacker tool and its empty-stack and bootstrap routines behind all stacked fluxes and uncertainties."},{"cited_title":"ALMA Lensing Cluster Survey: $HST$ and $Spitzer$ Photometry of 33 Lensed Fields Built with CHArGE","cited_arxiv_id":"2207.07125","evidence_quote":"Supplies the HST-IRAC catalog with redshifts, SFRs, stellar masses, and magnifications that define every stack."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the SHARK model predictions, both direct dust mass and band-6 flux, used for the high-redshift comparison."}],"review_version":1}