{"id":"9b2e61f8-9b89-4e86-8d04-6bce8b5d92ce","arxiv_id":"2505.21492","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"The dust removal timescale in galaxies decreases from about 1.8 Gyr at z~0.05 to about 0.4 Gyr at z>3, indicating more efficient dust clearing in the early universe.","lead":"Using infrared-selected galaxies from the Herschel Space Observatory, the authors fit how each galaxy's dust-to-stellar mass ratio declines with stellar age, and measure the e-folding ('dust removal') timescale in nine low-redshift bins and three high-redshift bins up to z~5. They report that this timescale drops from about 1.8 billion years in the local universe to under 500 million years at z>3, implying faster dust clearing in the early universe.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The high-z <500 Myr timescale depends on CIGALE/MAGPHYS age comparability and on an age cutoff at the cosmic age boundary; a plausible 0.2 dex age offset, acknowledged in Sec. 2.1, changes tau by roughly the full claimed signal.","rationale":"The reader's weakest_assumption correctly identifies the exponential model and the comparability of ages from the two SED codes as critical. My stress-test sharpens this into a specific, load-bearing vulnerability: the high-redshift tau is not only conceptually dependent on the age axis, but numerically fragile because a plausible 0.2 dex age systematic, acknowledged in the paper's Sec. 2.1, moves tau from 0.43 Gyr to below 0.3 Gyr. In addition, the z>2.57 bin's age range is truncated by the age of the Universe, so the fit's dynamic range is small and the slope is likely prior-dominated. The reader's verdict of CONDITIONAL is appropriate: the work is transparent and the data are mostly public, but the headline evolutionary claim should not be accepted as quantitatively solid until the age-axis systematics are propagated into tau and the z~0.9 code-switch is explicitly tested. I do not see evidence of internal inconsistency; the issue is an unquantified systematic rather than a logical error. The concrete tests above would settle whether the high-z result survives.","tokens_in":13556,"tokens_out":3691,"duration_ms":41914,"concrete_test":"Re-fit Eq. (1) for the DD20 z=2.57-5.234 bin (N=124) after (a) restricting to galaxies with mass-weighted age < 0.8 times the age of the Universe at their individual redshift, and (b) applying a uniform +0.2 dex shift to all ages, the inter-code dispersion reported by Pacifici et al. (2023). If tau rises above ~0.6 Gyr or the fit becomes unconstrained in either case, the <500 Myr result is not robust. As a cross-code check, fit the same bin using MAGPHYS-derived ages for the GAMA z~0.85-0.9 galaxies to test whether the apparent z~0.9 discontinuity in tau is a code effect.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that the fitted e-folding timescale tau in Eq. (1) decreases from ~1.8 Gyr at z~0.05 to 0.43 +/- 0.33 Gyr at z>3. The highest-redshift point is the load-bearing end of the trend, and it rests entirely on the stellar ages from the DD20/CIGALE sample. Two compounding issues make this point insecure. First, the paper corrects the DD20 dust masses by -0.3 dex and SFRs by -0.1 dex to match MAGPHYS-based GAMA values, but applies no correction to stellar ages. Pacifici et al. (2023) report mass-weighted age differences of at least 0.2 dex across codes and SFH assumptions, and the paper itself states in Sec. 2.1 that a -0.2 dex age shift would reduce the inferred tau below 300 Myr. Thus the quoted systematic uncertainty in the age axis is comparable to the whole claimed evolutionary signal. Second, in the z=2.57-5.234 bin the stellar ages are hard against the age of the Universe at the median redshift (~2.2 Gyr). The exponential fit then has a very limited dynamic range; if the ages cluster near the cosmic-age ceiling, the fitted slope is primarily set by the SED prior rather than by a measured decline. This is not a niche worry: the paper's own discussion of high-z ages is framed entirely around the constraint that galaxies cannot be much older than the Universe, which is exactly the truncation that makes tau poorly constrained. Because the trend from z~0.8 (tau ~1.6 Gyr) to z~0.9-1.21 (tau ~1.07 Gyr) coincides with the switch from MAGPHYS/GAMA to CIGALE/DD20, the apparent break may be a code-age offset rather than a physical transition. The conservative conclusion is that the claim 'tau <500 Myr at z>3' is not yet robust to plausible SED age systematics.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper measures the dust removal timescale tau defined by the exponential decline of the dust-to-stellar mass ratio with stellar age (Eq. 1) for galaxies over 0 < z < 5.3. The analysis combines 120,061 GAMA galaxies (z < 0.9) with 300 dusty star-forming galaxies from Donevski et al. (2020) at 0.9 < z < 5.234, using published dust masses, stellar masses, and mass-weighted ages from MAGPHYS and CIGALE SED fitting. Fitting the relation in nine redshift bins yields tau decreasing from 1.77 +/- 0.04 Gyr at z < 0.1 to 0.43 +/- 0.33 Gyr at z > 2.57, with a normalization A that rises from 10^-2.24 to 10^-1.93. The authors interpret this as more efficient dust removal at high redshift, driven by AGN feedback, supernova shocks, and astration.","tokens_in":13886,"tokens_out":10967,"duration_ms":113422,"significance":"The low-redshift GAMA analysis is based on a very large sample and yields well-constrained tau values, with internal robustness checks in Appendix B. If the high-redshift trend were secure, the paper would provide the first broad-redshift measurement of dust removal timescales and directly connect to theoretical expectations of faster dust processing at early epochs. The analysis uses exclusively public catalog data, which is a strength. However, the high-redshift conclusion rests almost entirely on a single bin of 124 galaxies, and the paper itself identifies a 0.2 dex age systematic that is not propagated into the reported uncertainties. The central claim is therefore not yet established at the level claimed in the abstract.","major_comments":[{"comment":"The analysis applies -0.3 dex and -0.1 dex corrections to the DD20 dust masses and SFRs to bring them into line with the GAMA/MAGPHYS values, but applies no correction to the stellar ages. The paper itself states (Sec. 2.1) that a 0.2 dex decrease in stellar age would reduce the inferred tau to below 300 Myr, and it cites Pacifici et al. (2023) showing that mass-weighted ages vary by at least 0.2 dex among SED fitting codes. Because age is the independent variable in Eq. (1), this systematic uncertainty propagates directly into tau. For the highest-redshift bin, a plausible +/-0.2 dex age offset changes tau by a large fraction of the claimed evolutionary signal. The authors should refit the DD20 bins with ages shifted by +/-0.2 dex and add the resulting systematic uncertainty to the quoted tau values before the abstract claim can be considered supported.","section":"Sec. 2.1 / Sec. 3, Table 1"},{"comment":"The highest-redshift bin gives tau = 0.43 +/- 0.33 Gyr from 124 galaxies, which is only 1.3 sigma away from zero. The two lower-redshift DD20 bins (1.07 +/- 0.93 Gyr from 20 galaxies and 1.51 +/- 0.87 Gyr from 153 galaxies) are statistically consistent with the GAMA values at z < 0.9. Thus the claimed monotonic decrease from z ~ 0.05 to z > 3 is driven almost entirely by this one bin. The paper should present a formal test of whether a constant tau across all redshifts is rejected (for example, a chi-square or Bayesian comparison), and should discuss the possibility that the single low bin reflects selection effects or systematic errors rather than a genuine evolutionary trend.","section":"Sec. 3, Table 1, z=2.57-5.234 row"},{"comment":"In the z=2.57-5.234 bin, the inferred stellar ages lie close to the age of the Universe at the median redshift, as indicated by the vertical dashed lines in Fig. A.1. The exponential fit in this bin therefore has limited dynamic range, and the fitted slope may be controlled by the SED-fitting prior that prevents ages from exceeding the cosmic age, rather than by a measured decline in the dust-to-stellar ratio. The paper should quantify how many galaxies in this bin have ages within, say, 0.1 dex of the cosmic-age ceiling and should test the stability of tau when those objects are removed from the fit.","section":"Sec. 3 / Appendix A, Fig. A.1"},{"comment":"The exponential relation Mdust/Mstellar = A exp(-age/tau) is adopted from Michalowski et al. (2019) and applied to star-forming DSFGs without independent validation. In galaxies with ongoing star formation, the dust-to-stellar mass ratio can decline with stellar age because the stellar mass grows while dust production continues, so the fitted tau is not uniquely a dust removal timescale. The authors should justify the physical interpretation by checking whether the dust mass itself declines with age in the DD20 sample, or by explicitly including a dust production term when modeling the observed ratio.","section":"Sec. 3, Eq. (1)"},{"comment":"The GAMA sample is Herschel-selected and includes galaxies across the full mass and morphology range, while the DD20 sample is restricted to massive (M > 10^10 solar masses) dusty star-forming galaxies. The paper demonstrates overlapping SFR-stellar mass distributions in Fig. 1, but the GAMA bins at z = 0.7-0.9 contain only 155 and 98 galaxies, and the DD20 bins are coarse (20, 153, and 124 galaxies). The apparent continuity of tau across z ~ 0.9 could therefore reflect the switch between samples and selection functions rather than a physical continuity. The authors should either match the selection criteria more closely in the GAMA subsamples (as attempted in Appendix B.4, although the uncertainties are large) or explicitly model a sample-calibration offset in the joint fit.","section":"Sec. 2.2 / Fig. 1"}],"minor_comments":[{"comment":"The abstract states 'less than 450 Myr at z>3', based on tau = 0.43 +/- 0.33 Gyr. Since the 1-sigma upper bound is about 0.76 Gyr and the 2-sigma upper bound exceeds 1 Gyr, the abstract should either quote the uncertainty or soften the claim to 'consistent with tau below about 0.8 Gyr'.","section":"Abstract"},{"comment":"The paper states that the DD20 sample contains 300 galaxies, but Table 1 lists 20 + 153 + 124 = 297 galaxies across the three DD20 redshift bins. Please clarify whether three galaxies were excluded from the fits and why.","section":"Sec. 2"},{"comment":"The sentence 'A decrease by 0.2 dex in stellar age would reduce the inferred dust removal timescale to below 300 Myr' appears without derivation. Please show the calculation or provide a reference that supports this quantitative statement.","section":"Sec. 2.1"},{"comment":"The caption of Fig. 2 refers to vertical dashed lines with colors matching the redshift bins, whereas Fig. A.1 describes 'vertical dashed black lines' in the last row. Please make the descriptions consistent.","section":"Fig. 2 / Fig. A.1"},{"comment":"The abbreviation 'MS' is used for the main sequence without expansion at first use. Please write 'main sequence (MS)' and then use 'MS' thereafter.","section":"Sec. 2.2"}],"recommendation":"major_revision","confidential_remarks":"The paper's central claim is fragile because the high-redshift bin is the only point that drives the trend and that point is vulnerable to the age-systematics issue that the authors themselves identify. I recommend major revision rather than rejection because the requested tests (age-offset fits, significance tests, dynamic-range checks) are straightforward with the public data and could materially change the conclusions. The authors should also consider whether the abstract overstates the result given that the GAMA-only data show a non-monotonic pattern, with tau decreasing to z ~ 0.5 and then increasing again by z ~ 0.8."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThe genuinely new thing here is the first attempt to measure the dust removal timescale out to z~5, using the Michalowski et al. (2019) exponential decay relation on a large Herschel-selected sample. That is worth something, and the low-redshift GAMA fits are clean and carefully described. But the headline claim—that tau drops below 500 Myr at z>3—is not yet supported. The whole high-z signal rests on one bin (z=2.57-5.234, N=124, tau = 0.43 ± 0.33 Gyr). Formally that is about 4 sigma from the 1.77 Gyr low-z value, but the systematic floor is far larger. The authors themselves note in Sec. 2.1 that a 0.2 dex shift in stellar age would give tau below 300 Myr. A 0.2 dex age difference between MAGPHYS and CIGALE is exactly what Pacifici et al. (2023) report, and the high-z ages are also pushed against the cosmic age ceiling at z~3.3. The observed break conveniently coincides with the switch from the GAMA/MAGPHYS sample to the DD20/CIGALE sample. So the claimed evolution could be largely a code-dependent age offset.\n\nThe paper does several things well. It is explicit about the SED-fitting caveats, applies a -0.3 dex correction to the DD20 dust masses and -0.1 dex to the SFRs to place them on the MAGPHYS scale, and tests five GAMA subsamples as robustness checks. The appendix honestly admits the exponential model fails for the youngest GAMA galaxies. The discussion of astration, SN shocks, and AGN feedback is sensible, even if it cannot distinguish the mechanisms.\n\nThe weaker spots are not hard to find. The low-z GAMA trend is not monotonic—tau dips at z~0.5 and then rises again at z~0.7-0.8. The intermediate DD20 bin (z=1.21-2.57, tau = 1.51 ± 0.87 Gyr) is consistent with no evolution. The normalisation A increases with redshift, which the authors themselves attribute partly to selection; if selection can bias A, it can also bias tau. And the DD20 catalog is not public, so the crucial high-z data cannot be checked.\n\nWho should read this? Dust evolution and quenching modellers will want the table of tau values, but they should treat the z>2 points as upper limits or provisional until the age systematics are quantified. I would send this to peer review rather than desk reject, with the clear expectation that the authors propagate a 0.2 dex age systematic into the tau uncertainties, release the high-z fits, and soften the abstract. The physical idea is plausible; the present analysis does not nail it down.","headline":"Plausible but fragile: the high-z dust removal timescale claim hinges on a single bin and is within the systematic age offset the authors themselves quote.","tokens_in":14571,"tokens_out":5687,"would_cite":false,"duration_ms":57988,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Dust vanishes faster in high-redshift galaxies, with removal timescales falling from 1.8 Gyr to about 0.4 Gyr by z>3.","keywords":["dust removal timescale","dust-to-stellar mass ratio","high-redshift galaxies","Herschel galaxies","SED fitting","galaxy evolution","AGN feedback","interstellar medium dust"],"falsifier":"Fit the same exponential relation to dust-to-stellar ratios of a redshift-complete, mass-selected sample at $z = 0$–$5$ using a single SED-fitting code; if $\\tau$ stops declining with redshift, the trend is a selection or calibration artifact. A cheaper check follows from the paper's own sensitivity note: if independent stellar-age indicators (for example rest-frame near-infrared colours or spectral indices) placed the high-redshift ages 0.2 dex lower, the inferred removal timescale would drop below 300 Myr, so any systematic age bias of that size would falsify the quoted values.","tokens_in":13265,"feed_emoji":"🌌","tokens_out":6136,"duration_ms":61262,"temperature":0.7,"pith_summary":"This paper uses more than 120,000 Herschel-detected galaxies spanning redshifts 0 to about 5 to measure how long dust survives in galactic interstellar media. Fitting an exponential decline of the dust-to-stellar mass ratio with stellar age in nine low-redshift and three high-redshift bins, it finds the e-folding dust removal timescale shrinks from 1.77 Gyr at z~0.05 to 0.43 Gyr at z~3.3. The conclusion is that dust removal is not a fixed process but becomes more efficient at earlier cosmic epochs, likely because AGN-driven outflows, supernova shocks, and rapid star formation remove dust faster. If correct, galaxy evolution models must treat dust destruction as redshift-dependent rather than assuming a constant removal timescale.","feed_headline":"Dust removal accelerates to under 500 Myr by z>3","feed_subtitle":"Analysis of 120,000 Herschel galaxies finds dust-to-star ratios decay four times faster in early cosmic epochs.","key_machinery":"The analytical motor is the single-exponential relation $M_{\\rm dust}/M_{\\rm stellar} = A e^{-\\mathrm{age}/\\tau}$, applied separately to redshift-binned samples. Here $\\tau$ is the e-folding dust removal timescale and $A$ is the maximum dust-to-stellar ratio; fitting this curve to thousands of galaxies per bin converts a scatter plot of dustiness versus stellar age into two redshift-dependent numbers. The data are SED-derived dust masses, stellar masses, and mass-weighted ages from MAGPHYS for the GAMA galaxies and CIGALE for the DD20 dusty star-forming galaxies, with a 0.3 dex downward correction applied to the high-redshift dust masses to align the two codes.","core_discovery":"The central claim is that the timescale on which galaxies shed their dust, defined by the relation $M_{\\rm dust}/M_{\\rm stellar} = A e^{-\\mathrm{age}/\\tau}$, decreases from about 1.8 Gyr at low redshift to less than about 0.5 Gyr at $z > 3$, with the highest-redshift bin fitted at $\\tau = 0.43 \\pm 0.33$ Gyr. The paper also finds that the normalisation constant $A$, which sets the maximum dustiness a galaxy can reach, rises with redshift from $\\log_{10} A = -2.24$ at $z \\sim 0.05$ to $-1.93$ at $z > 3$, indicating that younger galaxies are dustier. The authors interpret the short high-redshift timescale as the signature of AGN activity, supernova shocks, and astration acting on timescales of a few hundred million years.","pith_inferences":["Editorial extension: if the exponential fit is interpreted causally, the dust removal timescale becomes a cosmic chronometer, so measuring the age at which a galaxy's dustiness drops by one e-fold could date the onset of quenching at high redshift.","Editorial extension: applying the same exponential model to the gas-to-stellar mass ratio in these samples would test whether the trend is specific to dust destruction or instead reflects general ISM removal.","Editorial extension: a dedicated survey of several hundred $z > 3$ dusty galaxies processed with a single SED-fitting code would test whether the sub-500 Myr timescale is robust or an artifact of combining two catalogues with different dust models."],"forward_implications":["Galaxy evolution models that assume a single, gigayear-scale dust removal time will have to make the removal rate redshift-dependent, because the fitted e-folding time drops by roughly a factor of four from $z \\sim 0.05$ to $z > 3$.","High-redshift star-forming galaxies should typically show low dust-to-stellar ratios for their age, since dust is cleared within a few hundred million years by AGN outflows, supernova shocks, and rapid star formation.","The rising normalisation constant implies the maximum dust-to-stellar ratio was higher in the early Universe, consistent with rapid dust production occurring before efficient removal begins.","Independent observations of quiescent galaxies at $z = 3$–$4$ with dust removal times of 0.2–0.7 Gyr reinforce the interpretation that AGN feedback sets the dust-removal pace at high redshift."],"supporting_citations":[{"why":"Supplies the exponential decline relation between dust-to-stellar mass ratio and stellar age used as Eq. (1), and the low-redshift baseline for removal timescales.","marker":"Michałowski et al. (2019)"},{"why":"Provides the GAMA DR3 catalogue with MAGPHYS-derived dust masses, stellar masses, and mass-weighted ages for the 120,061 low-redshift galaxies.","marker":"Driver et al. (2016)"},{"why":"Provides the DD20 sample of 300 massive dusty star-forming galaxies up to $z = 5.234$ with CIGALE-derived properties that extend the analysis to high redshift.","marker":"Donevski et al. (2020)"},{"why":"Quantifies MAGPHYS versus CIGALE parameter offsets on the GAMA sample, justifying the 0.3 dex correction applied to the DD20 dust masses.","marker":"Paspaliaris et al. (2023)"},{"why":"Shows stellar masses are consistent across many SED-fitting codes while highlighting the age–dust degeneracy, supporting the comparability assumption across the two samples.","marker":"Pacifici et al. (2023)"},{"why":"Provides the redshift-dependent star-forming main sequence used to verify that the GAMA and DD20 samples trace comparable star-forming populations.","marker":"Speagle et al. (2014)"},{"why":"Reports independent dust removal timescales of 0.2–0.7 Gyr in quiescent galaxies at $z = 3$–$4$, cited as consistency for the short high-redshift timescale.","marker":"Sato et al. (2024)"},{"why":"Supplies theoretical dust formation and removal timescales from supernovae and sputtering, used to interpret the measured $\\tau$ values.","marker":"Gall & Hjorth (2018)"}],"fun_headline_variants":["Dust removal timescale shrinks 4x by redshift 3","Galaxies shed dust faster in early cosmic epochs","High-z galaxies purge dust in under 500 Myr","Dust removal accelerates to 500 Myr by z>3","Sub-Gyr dust removal in high-redshift galaxies"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole trend rests on the assumption that the fall-off of the dust-to-stellar mass ratio with stellar age in each redshift bin is caused by dust removal rather than by stellar mass growth or selection, and that the mass-weighted ages from the two SED-fitting codes are accurate and directly comparable across redshifts.","fun_headline_variants_meta":{"raw":{"variants":["Dust removal timescale shrinks 4x by redshift 3","Galaxies shed dust faster in early cosmic epochs","High-z galaxies purge dust in under 500 Myr","Dust removal accelerates to 500 Myr by z>3","Sub-Gyr dust removal in high-redshift galaxies"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000367,"raw_usage":{"total_tokens":1985,"prompt_tokens":970,"completion_tokens":1015,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":586,"completion_tokens_details":{"reasoning_tokens":932}},"tokens_in":586,"tokens_out":1015,"duration_ms":11145,"temperature":1.0,"reasoning_tokens":932,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:26:38.402017+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fit the same exponential relation to dust-to-stellar ratios of a redshift-complete, mass-selected sample at $z = 0$–$5$ using a single SED-fitting code; if $\\tau$ stops declining with redshift, the trend is a selection or calibration artifact. A cheaper check follows from the paper's own sensitivity note: if independent stellar-age indicators (for example rest-frame near-infrared colours or spectral indices) placed the high-redshift ages 0.2 dex lower, the inferred removal timescale would drop below 300 Myr, so any systematic age bias of that size would falsify the quoted values.","supporting_citations":[],"review_version":1}