{"id":"f52e0a6a-ac07-465e-91d8-f4e69c863681","arxiv_id":"2412.00599","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Galaxies above the star-forming main sequence form stars faster in their centers (outside-in growth), while those below it form stars faster in their outskirts (inside-out growth), a pattern the authors interpret as self-regulated size evolution.","lead":"This paper maps where stars are forming inside about a thousand distant galaxies and shows that the pattern flips depending on whether a galaxy is above or below the average star-forming main sequence. The pattern suggests galaxies cycle between growing their dense centers and growing their outer disks, which would keep their sizes in check.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Shared pixel-level SED-fit noise can build the ΔMS–sSFR-gradient correlation; the outside-in/inside-out self-regulation claim is not yet separated from this covariance.","rationale":"The paper's strongest claim is that the spatial gradient of sSFR is tied to distance from the MS and that this drives a self-regulating cycle of size growth. That claim would be secure if the correlation were a property of the true sSFR fields rather than of the SED-fitting model. The reader correctly identified the restrictive tau-model SFH as the weak spot. My read sharpens this: the problem is not just that an alternative SFH could shift the absolute values, but that the independent variable ΔMS and the dependent variable y are aggregates of the same per-pixel fits, so even uncorrelated pixel errors contribute a non-zero covariance when the error variances of central and outskirt pixels differ. The age–dust degeneracy in 7-band fits makes those error variances large and spatially varying, so the covariance can be significant. The paper's robustness checks are helpful but do not speak to this: varying the aperture keeps the same q_i maps, and the star-forming-only subsample still uses the same model. I therefore would not accept the paper as is, but I also would not reject it; the interpretation is plausible and consistent with existing compaction/quenching scenarios. A null simulation using posterior draws is a tractable, decisive check. The reader's conditional verdict is the right one; my concern is a sharper reason for the same condition, hence UNCHANGED.","tokens_in":18047,"tokens_out":11801,"duration_ms":130545,"concrete_test":"Construct 1000 mock realizations of the sample under the null that each pixel's true sSFR is independent of ΔMS: draw per-pixel SFHs from the iSEDfit posterior distributions, run the same pixel SED-fitting, MS-fitting, and partial-correlation pipeline, and record the distribution of r. If the 95th percentile of the null r values is clearly below 0.48, the shared-fit-error explanation is rejected and the claim survives; if the null distribution spans 0.48, the reported correlation is a fitting artifact rather than evidence for outside-in/inside-out growth along the MS.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central correlation in Fig. 4 is between two quantities constructed from the same per-pixel SED fits (Secs. 2.2, 3.1): y = log(sSFR_central/sSFR_outskirts) and x = log(sSFR_galaxy/sSFR_MS). Let q_i be the fitted log sSFR of pixel i, w_i its stellar-mass weight, and ε_i the pixel-level fitting error. The error entering y is approximately the mean ε_i in the central region minus the mean ε_i in the outskirts; the error entering x is approximately Σ_i w_i ε_i (plus MS-ridge uncertainty). Consequently Cov(x,y) has a term that is nonzero whenever the central and outskirt per-pixel error variances differ. The age–dust–SFH degeneracy in a single exponentially declining tau model with τ=0.01–1 Gyr and only 7 broad-band filters is severe and spatially varying, so this shared-error term can be substantial. The aperture-robustness check in Sec. 2.3 reuses the same q_i maps and cannot diagnose the effect; the star-forming-only check does not either. The paper's Sec. 2.4 admission that the MS normalization is 'largely dependent on the assumed model of SFH' applies to both x and y. Without a null that preserves the pixel error covariance while breaking the physical relation, the reported partial correlation r=0.48±0.02 is not yet evidence for self-regulated size growth along the MS.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents spatially resolved star formation histories for ~997 galaxies in GOODS-South at 0.5≤z≤2.0, derived from pixel-by-pixel SED fitting with an exponentially declining tau model. Dividing each galaxy into a central region within the half-mass ellipse and an outer annulus, the authors compute the ratio of 100-Myr specific star formation rates in the two regions and correlate it with the galaxy's offset from the star-forming main sequence (ΔMS), finding a partial correlation of 0.48±0.02 after controlling for stellar mass and redshift. They interpret this as evidence for self-regulated size growth: galaxies above the MS grow outside-in (central build-up), while those below grow inside-out (outer star formation), connecting the MS scatter to structural evolution.","tokens_in":18316,"tokens_out":6180,"duration_ms":59355,"significance":"If correct, this is a valuable observational result: it directly links the position of a galaxy relative to the star-forming main sequence to the spatial distribution of recent star formation, providing a possible mechanism for the near-constant scatter of the MS and for the flat size–mass relation. The analysis uses a reasonably large, homogeneous HST/CANDELS sample and includes useful robustness checks (varying the central aperture definition in Sec. 2.3, restricting to star-forming galaxies). The partial-correlation framework is appropriate for separating the gradient–ΔMS relation from mass and redshift trends. However, the central claim rests on pixel-level SED fitting with a single, restrictive SFH model, and no null test is presented that would separate a true physical correlation from one induced by shared measurement errors between the two axes of Fig. 4.","major_comments":[{"comment":"The two variables in Fig. 4, log(sSFR100,central/sSFR100,outskirts) and ΔMS = log(sSFR100,galaxy/sSFRMS), are both constructed from the same pixel-by-pixel SED fits (Sec. 2.2). Pixel-fit errors in the central and outer regions enter the y-axis with opposite signs and also enter the galaxy-integrated sSFR that defines ΔMS, so correlated, spatially varying fit errors can generate a spurious correlation. The aperture checks in Sec. 2.3 reuse the same fitted pixel maps and therefore cannot diagnose this effect, and the star-forming-only subsample does not break the shared-noise covariance. The authors should add a null test that preserves the pixel error covariance while destroying the physical relation—for example, randomly permuting central/outskirt pixel assignments within each galaxy, or simulating mock galaxies with no intrinsic gradient but with the same noise properties—and show that the partial correlation is not reproduced by noise alone.","section":"Sec. 3.1 / Fig. 4, with Secs. 2.2 and 2.4"},{"comment":"The analysis relies on a single exponentially declining tau model with τ in [0.01, 1.0] Gyr, fixed Chabrier IMF and Calzetti dust law, and only seven broad-band filters. This prior strongly restricts the allowed SFH shapes and can introduce spatially varying systematic biases, since the age–dust degeneracy behaves differently in high-S/N central pixels than in low-surface-brightness outskirts. Both the y-axis and the x-axis of Fig. 4 are derived from these fits, so the measured correlation could partly be an artifact of the assumed model. The paper itself states in Sec. 2.4 that the derived MS normalization is 'largely dependent on the assumed model of SFH'; the same caveat applies to the sSFR gradient. The authors should repeat the analysis with alternative SFH models (e.g., delayed-τ or non-parametric SFHs, different dust laws) or quantify how the partial correlation and the inferred growth patterns change under plausible model variations.","section":"Sec. 2.2"},{"comment":"The conclusion that the observed sSFR gradient implies 'outside-in' or 'inside-out' growth assumes that the current 100-Myr sSFR ratio directly predicts the future direction of half-mass-radius evolution. The supporting mass-weighted age ratio in Sec. 3.2 is derived from the same SED fits and is subject to the same systematics; it does not independently confirm the temporal interpretation. To make the central claim load-bearing, the authors should connect the observed gradients to actual size evolution—for example, by comparing with galaxies observed at two epochs, with simulation predictions, or by explicitly deriving the expected ΔR50 from the measured sSFR profile and showing that it matches the observed size trends in Fig. 5. At minimum, the assumption that a single snapshot maps onto a growth direction should be stated and discussed.","section":"Secs. 3.1 and 4"}],"minor_comments":[{"comment":"Report the uncertainties on the fitted MS parameters (sb, β, μ), since these uncertainties propagate into ΔMS and hence into the partial correlation.","section":"Sec. 2.4"},{"comment":"Clarify whether the reported partial correlation coefficients are computed for the full sample of 997 galaxies or for the star-forming subsample, and state the sample size used in the PINGOUIN calculation.","section":"Sec. 3.1"},{"comment":"The color coding by size is hard to read in printed grayscale; consider using a colormap with larger contrast or adding contours.","section":"Fig. 4 / Fig. 5"},{"comment":"The phrase 'observed similar trends' should be quantified; please state how the correlation coefficient and slope change for the 1 kpc aperture case.","section":"Sec. 2.3"},{"comment":"The Limitations subsection does not mention the potential shared-noise/circularity issue or the SFH-model dependence beyond the MS normalization; these should be acknowledged as limitations of the present analysis.","section":"Sec. 4.2"},{"comment":"'997 galaxies' appears in the abstract and in figure captions; please ensure the number is consistent and define the sample selection in one place.","section":"Abstract and Sec. 1"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses an interesting and timely question, and the observed correlation, if physical, would be an important contribution. However, the analysis currently lacks the null tests and SFH-model sensitivity checks needed to rule out the plausible alternative that the correlation is generated by shared pixel-level SED-fit errors. The covariance argument is well grounded in the manuscript: the paper itself concedes in Sec. 2.4 that the MS normalization is model-dependent, and the aperture tests in Sec. 2.3 cannot address the shared-noise issue. I would therefore recommend a major revision rather than rejection, because the central question is important and the necessary tests appear feasible within the scope of the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line first. This paper reports a genuinely new measurement: for ~997 GOODS-S galaxies at 0.5≤z≤2.0, it correlates the ratio of central to outskirt sSFR with distance from the star-forming main sequence, and finds a partial correlation of 0.48±0.02 after controlling for mass and redshift. That is exactly the trend predicted by Tacchella et al. 2016b, so the paper is best read as an observational confirmation rather than a new mechanism. The sample construction, the aperture variation checks, and the star-forming-only subsample are all sensible. The age-ratio panel (Fig. 7) adds independent support for the same direction. I believe the authors are reporting a real signal, not a fabrication.\n\nThe soft spot is load-bearing. The stress-test note is correct: x and y in Fig. 4 are built from the same pixel-level SED fits. y is the difference between central and outskirt pixel sSFR; x is the galaxy-wide sSFR relative to a ridge fitted to those same galaxy sSFR values. With a single exponentially declining tau model, tau in [0.01,1] Gyr, and seven broad-band filters, the age–dust–SFH degeneracy is severe and spatially varying. The covariance between x and y can be nonzero even if no real physical gradient exists. The aperture robustness check reuses the same maps, so it cannot diagnose this. Neither can the star-forming-only split. The paper's own statement in Section 2.4 that the MS normalization is largely dependent on the assumed SFH model applies to both axes. What is missing is a null test that preserves the pixel error covariance while breaking the physical relation—for example, fitting with a more flexible SFH and checking whether the correlation survives, or a bootstrap/permutation test on pixel noise. Without that, the reported r=0.48 is consistent with the self-regulation story but does not establish it.\n\nThe self-regulation interpretation is also a step beyond the data. The authors infer future size change from current sSFR gradients. That is a reasonable inference, but the cyclic compaction-depletion-replenishment language is a narrative fit, not a demonstrated oscillation from these snapshots. The lack of public data products makes it harder for someone else to run the null test themselves. That is a minor issue on its own but matters here because the central number is suspect until tested.\n\nWho gets value? Anyone working on MS scatter, resolved SFHs, or size evolution. It is a useful paper to know, and a serious referee should see it. I would send it to review, not desk reject, but I would ask for a robustness test against alternative SFH models and a clear treatment of the shared-noise covariance before publication.","headline":"A plausible but not-yet-established observational confirmation of the predicted outside-in/inside-out growth pattern across the MS; the key correlation shares SED-fit noise on both axes, so it needs a robustness test before it can anchor the self-regulation picture.","tokens_in":18926,"tokens_out":2995,"would_cite":true,"duration_ms":34031,"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":"For 997 galaxies at z 0.5–2, the sign of the central-to-outskirt star-formation ratio flips at the main-sequence ridge, predicting whether a galaxy shrinks or grows.","keywords":["galaxy evolution","star-forming main sequence","spatially resolved star formation histories","pixel-by-pixel SED fitting","inside-out growth","outside-in growth","galaxy sizes","GOODS-S"],"falsifier":"Observe the same sample with spatially resolved nebular emission maps, such as H-alpha or [O II], to measure the current central-to-outskirt sSFR ratio independently of SED modeling; if the sign of the ratio does not flip from positive to negative as galaxies cross ΔMS = 0, or if galaxies with positive ratios show no subsequent decrease in half-mass radius, the self-regulation claim would be falsified.","tokens_in":1960,"feed_emoji":"🌌","tokens_out":5269,"duration_ms":182366,"temperature":0.7,"pith_summary":"The paper claims that where a galaxy sits relative to the star-forming main sequence (the tight relation between star formation rate and stellar mass) tells you which way its size is evolving. Using about 997 GOODS-S galaxies at redshifts 0.5 to 2.0, the authors reconstruct spatially resolved star formation histories and find that galaxies on the upper envelope of the main sequence have higher specific star formation in their centers than in their outskirts, meaning they grow outside-in and should shrink. Galaxies below the main sequence show the opposite pattern, growing inside-out and should expand. This anticorrelation with galaxy size implies a self-regulating cycle: gas compaction builds the center, central depletion shifts star formation outward, and the galaxy grows again, which would keep the size–mass relation flat. If correct, a galaxy's offset from the main sequence becomes a predictor of its structural growth direction.","feed_headline":"Galaxy growth direction flips across the star-forming main sequence","feed_subtitle":"Above the main sequence galaxies build their centers; below it, their outskirts grow—keeping the size–mass relation flat.","key_machinery":"The load-bearing object is the spatially resolved star formation history reconstructed pixel-by-pixel with an exponentially declining tau model, split into a central region within the half-mass ellipse R50s and an outskirts region from 1 to 3 R50s. These histories yield the 100 Myr specific star formation rates and mass-weighted ages for each region, summarized as sSFR100,central/sSFR100,outskirts and t50,central/t50,outskirts. The comparison quantity is ΔMS = log(sSFR100,galaxy/sSFRMS), where the main sequence ridge is fitted as sSFRMS = sb (M⋆/$10^{10}$ M⊙)^β (1+z)^μ with best-fit values sb = 0.22, β = −0.39, and μ = 0.85.","core_discovery":"The central discovery is a correlation between the spatially resolved specific star formation rate ratio, log(sSFR100,central/sSFR100,outskirts), and the distance from the main sequence, ΔMS, with a partial correlation of 0.48 ± 0.02 after taking stellar mass and redshift into account. Galaxies above the main sequence ridge (ΔMS > 0) show centrally enhanced specific star formation, meaning their stellar mass doubles faster inside the half-mass ellipse than outside, which is outside-in growth and a shrinking half-mass radius. Galaxies below the ridge (ΔMS < 0) show the reverse, with more active star formation in the outskirts, inside-out growth, and an expanding half-mass radius. The trend anticorrelates with galaxy size: larger galaxies above the main sequence are shrinking while smaller galaxies below it are growing, which the authors interpret as self-regulation of size growth along the main sequence. Mass-weighted age ratios independently support the same picture, with younger centers in galaxies above the main sequence and similar central and outskirts ages in galaxies below it.","pith_inferences":["If the correlation is causal, the relatively cheaply measured ΔMS could serve as a proxy for the direction of size evolution in large surveys that lack spatially resolved data.","The paper notes in Section 2.4 that its main sequence normalization depends on the assumed star formation history model, so rerunning the analysis with nonparametric or bursty star formation histories would test whether the sign flip at ΔMS = 0 is an artifact of the tau model.","The natural next test is to follow individual galaxies through repeated compaction and regrowth cycles in simulations, which would quantify how much of the flat size–mass relation arises from this oscillation rather than from steady growth.","Extending this analysis to z > 3 with JWST would reveal whether the same compaction–depletion–regrowth cycle operated near the peak of cosmic star formation or whether early galaxies grew more monotonically."],"forward_implications":["A galaxy's offset from the main sequence ridge predicts the direction of its size evolution: above the ridge galaxies build their centers and shrink, while below the ridge they build their outskirts and grow.","The anticorrelation between growth direction and size keeps the size–mass relation nearly flat, matching the observed shallow slopes for star-forming galaxies.","The pattern holds across stellar masses from about 10^9.8 to 10^11.5 solar masses and redshifts 0.5 to 2.0, with no strong redshift dependence, meaning the self-regulation cycle operates over several billion years of cosmic time.","Massive galaxies below the main sequence show depleted centers with active outskirts, connecting the growth-direction picture to inside-out quenching and the transition toward quiescence.","The observed sequence—central enhancement above the main sequence, central depletion, then outward growth—supports an oscillatory, rather than monotonic, drift around the main sequence, consistent with compaction-driven bulge growth followed by disc regrowth."],"supporting_citations":[{"why":"Supplies the GOODS-S sample, stellar mass maps, PSF-corrected half-mass sizes, and the pixel-by-pixel SED fitting methodology.","marker":"Mosleh et al. 2020"},{"why":"Provides the SFH reconstruction equations and the spatially resolved analysis framework that this paper extends to central and outskirts regions.","marker":"Jain et al. 2024"},{"why":"Provides the redshift-dependent main sequence reference and sSFR threshold context used to separate star-forming and quiescent galaxies.","marker":"Speagle et al. 2014"},{"why":"Supplies the main sequence fitting form used in Equation 1 and the compaction–depletion–replenishment self-regulation scenario.","marker":"Tacchella et al. 2016a"},{"why":"Predicted variations in the sSFR gradient across the main sequence that this paper tests observationally.","marker":"Tacchella et al. 2016b"},{"why":"Provides the four-phase blue-nugget evolutionary picture against which the observations are interpreted.","marker":"Zolotov et al. 2015"},{"why":"Establishes the theoretical basis for self-regulation of galaxies along the main sequence through gas inflow and feedback.","marker":"Dekel et al. 2013"},{"why":"Documents the nearly flat size–mass relation for star-forming galaxies that the self-regulation interpretation is invoked to explain.","marker":"Suess et al. 2019a"},{"why":"Provides observational context for inside-out quenching in galaxies below the main sequence.","marker":"Whitaker et al. 2017"}],"fun_headline_variants":["Main sequence position steers galaxy size growth direction","Galaxies self-regulate size growth along the main sequence","Above MS: outside-in growth, below MS: inside-out","MS ridge flips galaxy growth from outside-in to inside-out","Galaxy growth flips with star-forming main sequence"],"cache_read_input_tokens":20864,"weakest_assumption_plain":"The whole analysis assumes that pixel-by-pixel SED fitting with a single exponentially declining tau model recovers the true spatially resolved star formation histories, so if real star formation histories are more complex, every derived sSFR ratio, the fitted main-sequence ridge, and the ΔMS correlation could change.","fun_headline_variants_meta":{"raw":{"variants":["Main sequence position steers galaxy size growth direction","Galaxies self-regulate size growth along the main sequence","Above MS: outside-in growth, below MS: inside-out","MS ridge flips galaxy growth from outside-in to inside-out","Galaxy growth flips with star-forming main sequence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001052,"raw_usage":{"total_tokens":4487,"prompt_tokens":1085,"completion_tokens":3402,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":701,"completion_tokens_details":{"reasoning_tokens":3322}},"tokens_in":701,"tokens_out":3402,"duration_ms":29898,"temperature":1.0,"reasoning_tokens":3322,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T05:12:02.294237+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Observe the same sample with spatially resolved nebular emission maps, such as H-alpha or [O II], to measure the current central-to-outskirt sSFR ratio independently of SED modeling; if the sign of the ratio does not flip from positive to negative as galaxies cross ΔMS = 0, or if galaxies with positive ratios show no subsequent decrease in half-mass radius, the self-regulation claim would be falsified.","supporting_citations":[],"review_version":1}