{"id":"1eb358ff-3271-4201-bd74-b32bbdc9c2bd","arxiv_id":"1908.10936","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"EAGLE simulated galaxies broadly match observed GAMA galaxy shapes in Gini and M20, except for an excess of asymmetry, and image resolution systematically lowers Gini and asymmetry values.","lead":"This paper checks whether simulated galaxies look like real ones by making fake telescope images of EAGLE galaxies and measuring their shapes. It finds broad agreement except that simulated galaxies are too lopsided, and it shows how image blur changes the shape measurements.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The A-excess claim is not yet separated from the mock-image smoothing choice for young stars; the paper itself proposes but does not run the decisive experiment.","rationale":"The reader's weakest assumption is that the mock-image pipeline preserves the true light distribution, with the high simulated A arising from image-generation choices rather than a genuine overabundance of disturbed galaxies. My read agrees: this is the most load-bearing point because A is the only statistic that fails, and the paper's own interpretation routes the failure through the young-stellar source smoothing. The test proposed here directly settles whether the A excess is an artifact: regenerate a subsample with an increased young-star smoothing length and recompare with GAMA. If the excess vanishes, the central claim should be reframed; if it persists, the claim is strengthened. The paper has real independent support: it uses the statmorph code, public EAGLE images, and multi-simulation comparisons, and it includes KS tests for numerical convergence in Appendix B. No machine-checked proof is involved, but the analysis is reproducible in principle. The reader's conditional verdict is appropriate, so I do not change it. A secondary concern is that the GAMA and EAGLE stellar mass distributions differ (Figure 1, medians 10^10.45 vs 10^10.36), and a mass-matched 2D KS or energy-distance test on G-M20 would also be a worthwhile check, but the source-smoothing question is the one that determines the meaning of the single discrepancy reported in the abstract.","tokens_in":26191,"tokens_out":4582,"duration_ms":48577,"concrete_test":"Regenerate SKIRT mock images for a stellar-mass-stratified subsample of Ref-100 (e.g., 200 galaxies spanning 10^10 to 10^11.5 Msun) with the young stellar population (age < 10 Myr) photon-source smoothing length increased from the current 64th-neighbor-of-young-stars value to the 64th-neighbor smoothing length computed using all stellar particles, or to a fixed 1 kpc kernel, holding dust treatment, PSF (FWHM = 1 kpc), noise (S/N = 25), and statmorph settings (including the modified A0 centering) identical; recompute A and compare the resulting A distribution to GAMA. If the median A excess (roughly 0.05) disappears, the 'except A' claim is an image-post-processing artifact; if it persists, the EAGLE asymmetry excess is robust to source smoothing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim has two parts: EAGLE G-M20 morphologies agree with GAMA, while EAGLE asymmetry A is larger. The load-bearing condition for the second part is that the A excess is a property of the simulated galaxies rather than an artifact of the image-generation pipeline. Section 4.2 and Figure 6 show that the excess is driven by 'the light distribution of young star-forming regions... scattered and point-like appearance.' The paper attributes this to the way photon sources are spatially distributed in the mock images (Section 3.1.1: photon emission points are drawn from truncated Gaussians with a 64th-neighbor smoothing length). Section 5(v) explicitly suggests a mitigation: 'assign young stellar particles an increased smoothing length in the mock image generation procedure.' That experiment is never run; Section 4.3 varies only the PSF FWHM, not the source smoothing length. Because A is precisely the statistic most sensitive to pixel-level clumpiness, the headline 'reproduces observations except A' conflates two possibilities: (a) EAGLE galaxies genuinely contain more asymmetric light distributions than SDSS galaxies, or (b) the chosen smoothing kernel makes young stellar sources artificially point-like and clumpy, inflating A. If (b) holds, the correct summary is not that EAGLE reproduces observations except A, but that EAGLE reproduces observations when imaged with this particular smoothing choice except A. The authors flag this limitation themselves, but the missing test leaves the one discrepancy in the paper unsecured.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents non-parametric optical morphologies (Gini, M20, Concentration, Asymmetry) for galaxies in the EAGLE Ref-100 simulation at z=0.1, computed from SKIRT mock images that are post-processed to approximate SDSS imaging. These are compared with morphologies measured from SDSS images of GAMA galaxies at z~0.05 with M*>10^10 M_sun, and with published morphologies for Illustris and IllustrisTNG. The authors report that the Gini-M20 distribution of EAGLE galaxies agrees well with GAMA observations, while the asymmetry statistic A is systematically larger in the simulations. They also study the effect of spatial resolution, finding that Gini and Asymmetry decrease with decreasing resolution, and examine trends between optical morphology and star formation rate, galaxy size, and kinematic morphology indicators. The paper concludes that EAGLE reproduces observed optical morphologies except for the asymmetry excess, which they tentatively attribute to the treatment of young stellar photon sources in the mock-image generation.","tokens_in":26492,"tokens_out":2981,"duration_ms":31329,"significance":"If the conclusions hold, this is a valuable benchmark for hydrodynamic simulations: it provides a like-for-like comparison of simulated and observed non-parametric morphologies using the same measurement code, and it demonstrates that EAGLE and IllustrisTNG produce broadly similar optical morphologies that match GAMA in G-M20 space. The paper also gives useful quantitative results on how non-parametric statistics depend on spatial resolution and viewing angle, which are relevant for future surveys such as LSST. Strengths include the use of publicly available EAGLE mock images, the consistent application of statmorph to both simulated and observed images, the explicit use of external GAMA/SDSS data as a validation target, and the inclusion of convergence and orientation tests in the appendices. The main weakness is that the central exception to the reproduction of observations, the asymmetry excess, is not conclusively separated from a plausible imaging artifact, and the paper itself flags but does not execute the decisive test.","major_comments":[{"comment":"","section":"Sections 4.2 and 5(v)"},{"comment":"","section":"Section 5(i) and Figure 2"},{"comment":"","section":"Sections 3.3.4 and 4.2, Figure 5"}],"minor_comments":[{"comment":"","section":"Section 3.3.4"},{"comment":"","section":"Section 3.3.2"},{"comment":"","section":"Section 4.2"},{"comment":"","section":"Section 4.3"},{"comment":"","section":"Section 3.1.3"},{"comment":"","section":"Section 5(v)"}],"recommendation":"major_revision","confidential_remarks":"The paper is a reasonable fit for MNRAS and the core results are likely to be of interest. The main concern is the unresolved ambiguity in the asymmetry excess, which the authors themselves acknowledge. I would encourage the editor to request the smoothing-length experiment or an equivalent quantitative sensitivity test before publication, as the current wording of the abstract and Section 5(i) overstates the conclusiveness of the 'except A' finding."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a careful, useful paper that gives EAGLE a fair shot at reproducing observed optical morphologies. The main positive result—G-M20 agreement with GAMA—looks credible, but the one discrepancy in the abstract, the asymmetry excess, is not yet separated from the mock-image smoothing recipe for young stars. The authors flag this themselves and even propose the fix, but they don't run it.\n\nWhat's new and good: it is the first non-parametric morphology comparison of EAGLE against GAMA using dust-corrected SKIRT mock images. The resolution study uses thousands of galaxies, going well beyond the 8-galaxy basis of the standard Lotz et al. result, and it directly matters for LSST-era surveys. The cross-simulation comparison with Illustris and IllustrisTNG is informative, and the optical-kinematic morphology correlation, stronger for centrals than satellites, is a genuinely nice added result. The paper is also honest about the comparison being fair: EAGLE's calibration did not use morphology, and the same statmorph code is applied to simulated and observed images, with a documented modification to background subtraction.\n\nSoft spots, in proportion:\n\n- The central G-M20 agreement is asserted from contour overlap; there is no two-sample test for the GAMA-EAGLE comparison, even though the appendix uses KS tests for the convergence check. A 2D KS test would let the reader judge 'agree well' quantitatively. Minor to moderate.\n\n- The asymmetry excess is the only discrepancy, and the paper attributes it to the point-like spatial distribution of young stellar photon sources. That is plausible, and the authors suggest increasing the smoothing length as a test. But they never run it. Section 4.3 varies only the PSF FWHM. Because A is the statistic most sensitive to pixel-scale clumpiness, 'reproduces observations except A' remains ambiguous: it could be the galaxies or the imaging recipe. This is an acknowledged limitation, not a hidden one, but it is the load-bearing discrepancy and the decisive experiment is cheap. Moderate.\n\n- The cross-simulation A comparison in Fig. 5 mixes two different background-subtraction implementations (the authors note this), so the TNG/EAGLE A offset is not a clean comparison. Side point.\n\nBottom line: this paper is for galaxy formation simulators and anyone building mock-image pipelines for large surveys. It deserves a serious referee. I would recommend major revision, not rejection: add a formal comparison test for G-M20 and run the smoothing-length experiment for A. If the asymmetry excess survives, the claim stands; if not, the paper becomes 'EAGLE reproduces observations including A under the right imaging recipe'—still a useful result, just a different message.","headline":"Solid validation of EAGLE optical morphologies with one honest but untested caveat: the asymmetry excess may be a mock-image smoothing artifact.","tokens_in":27046,"tokens_out":4043,"would_cite":true,"duration_ms":39068,"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":"Simulated EAGLE galaxies reproduce observed optical morphologies, except their asymmetry values are systematically too high.","keywords":["galaxy morphology","non-parametric statistics","Gini coefficient","M20","asymmetry","EAGLE simulation","mock images","GAMA survey"],"falsifier":"Regenerate the EAGLE mock images with a larger smoothing length for young stellar photon sources (for example, using the 128th nearest neighbours) and remeasure the Asymmetry statistic: if the high-$A$ tail disappears while $G$-$M_{20}$ and Concentration remain in agreement with GAMA, the discrepancy is an image-generation artifact; if it persists, EAGLE galaxies are intrinsically more asymmetric than observed.","tokens_in":1764,"feed_emoji":"🔭","tokens_out":6135,"duration_ms":128689,"temperature":0.7,"pith_summary":"This paper asks whether the optical shapes of galaxies produced by the EAGLE cosmological simulation look like real galaxies when rendered as realistic telescope images. Using non-parametric statistics (Gini, M20, Concentration, and Asymmetry) on g-band light distributions, the authors find that simulated galaxies occupy nearly the same Gini-M20 and Concentration regions as GAMA galaxies at z~0.05, but have systematically larger Asymmetry values. The paper also establishes that lowering image spatial resolution systematically reduces Gini and Asymmetry, with the Gini of low-mass galaxies most affected. It argues that the asymmetry excess is likely an artifact of how young star-forming regions are rendered as point-like photon sources, rather than evidence that EAGLE produces an overabundance of genuinely disturbed galaxies.","feed_headline":"Mock galaxies match real shapes, but look too lopsided","feed_subtitle":"Gini and concentration match GAMA galaxies; asymmetry gap traces to how young stars are rendered.","key_machinery":"The argument runs through four non-parametric statistics computed from the light distribution of each galaxy image: Gini (how unevenly light is distributed among pixels), M20 (the logarithm of the ratio of the second-order moment of the brightest 20 per cent of light to the total second-order moment), Concentration (five times the base-10 logarithm of the ratio of radii containing 80 and 20 per cent of the light), and Asymmetry (the fractional difference between an image and its 180-degree rotation, with background asymmetry subtracted). The mock images are produced by a three-dimensional Monte Carlo radiative-transfer code that models absorption and scattering by dust, with young star-forming regions rendered as smoothed point sources, then convolved with a Gaussian point-spread function and rebinned to SDSS-like pixel scale with signal-to-noise ratio 25; the same measurement routine is applied to SDSS images of GAMA galaxies. Together with the bulge-strength statistic F, defined as five times the signed distance of a galaxy's (G, M20) point from the early/late-type separation line, these statistics carry the comparison: they translate particle data into quantities a survey would actually measure.","core_discovery":"The paper's central claim is that the optical morphologies of EAGLE galaxies, once rendered as realistic g-band images, reproduce the observed morphologies of low-redshift galaxies in the GAMA survey, with one systematic exception: simulated galaxies are more asymmetric. On those terms, the Gini and M20 distributions of EAGLE galaxies at z=0.1 with stellar masses above $10^{10}$ solar masses nearly overlap those of GAMA galaxies with matched mass selection at z~0.05, and the Concentration distribution is also a close match. The Asymmetry distribution, however, develops a large tail of highly asymmetric galaxies with no observed counterpart, which the paper attributes to the point-like spatial distribution of young stellar photon sources in the mock-image generation rather than an intrinsic excess of disturbed galaxies. The paper also establishes that lowering image spatial resolution systematically reduces Gini and Asymmetry values, with the Gini of low-mass galaxies most affected, and that optical morphology correlates with kinematic morphology more strongly for central than for satellite galaxies.","pith_inferences":["A direct test the paper implies but does not run: regenerate the mock images with a larger smoothing length for young stellar particles and check whether the asymmetry tail disappears while Gini-M20 and Concentration stay matched; if it does, the A excess is an image-rendering artifact.","The resolution effects documented here could be converted into a practical calibration: apply the simulated mass- and FWHM-dependent shifts to observed Gini and Asymmetry measurements before comparing surveys with different seeing. The paper stops at reporting the effect.","The excess of actively star-forming EAGLE galaxies above about $10^{11}$ solar masses, if it reflects weak feedback in the simulation, implies that morphology statistics could serve as an indirect constraint on feedback physics, an extension the paper does not make.","The same mock-image pipeline could be extended to higher redshift or to LSST-like resolution to predict how observed morphology distributions evolve, testing whether the agreement holds beyond the z~0.05 comparison presented here."],"forward_implications":["Because galaxy morphology was not used to calibrate EAGLE, the agreement in Gini-M20 and Concentration counts as an independent prediction of the simulation rather than a retuning of parameters.","The asymmetry excess should be treated as a known bias when using mock EAGLE images to train or test morphological classifiers, merger identifications, or neural networks.","Resolution corrections derived from the simulated sample could be applied to observed surveys with different seeing, though only after checking that the simulated light profiles are realistic at sub-kiloparsec scales.","The correspondence of the kinematic threshold $\\kappa_{co} = 0.4$ means kinematic morphology can serve as a proxy for optical morphology in EAGLE, even though the two criteria select different galaxy populations.","The weaker optical-kinematic correlation in satellites indicates environmental quenching changes a galaxy's light distribution less than its internal dynamics, motivating separate evolutionary tracks for central and satellite galaxies."],"supporting_citations":[{"why":"Defines the EAGLE simulation and its reference model (Ref-100) from which the simulated galaxy sample is drawn.","marker":"Schaye et al. (2015)"},{"why":"Provides the mock images used for the simulated galaxies, including the image-generation choices analyzed in Section 3.1.","marker":"Trayford et al. (2017)"},{"why":"Supplies the T-Type and Galaxy Zoo 2 morphologies used to anchor the GAMA comparison and to test the F statistic.","marker":"Domínguez Sánchez et al. (2018)"},{"why":"Defines the Gini and M20 statistics and the original resolution-dependence study that this work extends.","marker":"Lotz et al. (2004)"},{"why":"Defines the Concentration and Asymmetry statistics used throughout the comparison.","marker":"Conselice et al. (2000)"},{"why":"Supplies the implementation used to compute all non-parametric morphologies and the IllustrisTNG values compared in Figures 2 and 5.","marker":"Rodriguez-Gomez et al. (2019)"},{"why":"Supplies the Illustris non-parametric morphologies and the F bulge-strength statistic used throughout the paper.","marker":"Snyder et al. (2015b)"},{"why":"Supplies the Illustris asymmetry measurements and the background-asymmetry centroid prescription adopted in Section 3.3.4.","marker":"Bignone et al. (2017)"},{"why":"Defines the kinematic metric $\\kappa_{co}$ and the threshold $\\kappa_{co} = 0.4$ used for comparing optical and kinematic morphology.","marker":"Correa et al. (2017)"}],"fun_headline_variants":["EAGLE galaxies match observed shapes, but appear too asymmetric","Simulated galaxies match observed shapes, but asymmetry stands out","Young stars make simulated galaxies look lopsided in EAGLE","Galaxy shapes in EAGLE match reality, except for asymmetry","Galaxy shapes match, but asymmetry is off in simulations"],"cache_read_input_tokens":29184,"weakest_assumption_plain":"The load-bearing premise is that the mock-image pipeline faithfully represents the true light distribution of simulated galaxies, so the excess asymmetry is caused by how young star-forming regions are rendered rather than by an intrinsic overabundance of disturbed galaxies in EAGLE.","fun_headline_variants_meta":{"raw":{"variants":["EAGLE galaxies match observed shapes, but appear too asymmetric","Simulated galaxies match observed shapes, but asymmetry stands out","Young stars make simulated galaxies look lopsided in EAGLE","Galaxy shapes in EAGLE match reality, except for asymmetry","Galaxy shapes match, but asymmetry is off in simulations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001549,"raw_usage":{"total_tokens":6225,"prompt_tokens":1012,"completion_tokens":5213,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":628,"completion_tokens_details":{"reasoning_tokens":5127}},"tokens_in":628,"tokens_out":5213,"duration_ms":40054,"temperature":1.0,"reasoning_tokens":5127,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:29:46.094120+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Regenerate the EAGLE mock images with a larger smoothing length for young stellar photon sources (for example, using the 128th nearest neighbours) and remeasure the Asymmetry statistic: if the high-$A$ tail disappears while $G$-$M_{20}$ and Concentration remain in agreement with GAMA, the discrepancy is an image-generation artifact; if it persists, EAGLE galaxies are intrinsically more asymmetric than observed.","supporting_citations":[{"cited_title":"A., Tissera P","cited_arxiv_id":null,"evidence_quote":"Supplies the Illustris asymmetry measurements and the background-asymmetry centroid prescription adopted in Section 3.3.4."}],"review_version":1}