{"id":"3966637f-787e-407b-a775-f8e6636739dd","arxiv_id":"1908.11665","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Comparing twin cosmological simulations, AGN feedback leaves galaxy alignment with the matter density field unchanged but strengthens position-shape alignments, especially for massive ellipsoids.","lead":"This paper compares galaxy shape and spin alignments in two large cosmological simulations that are identical except for whether black hole feedback is switched on. It finds the feedback does not change how galaxy shapes align with the matter density field, but does change how shapes align around galaxies themselves.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline wδ+ robustness claim is only demonstrated for the simple inertia tensor; the luminosity-weighted reduced tensor changes alignment amplitudes and is not tested for wδ+.","rationale":"Reviewing in good faith, the paper is a clean twin-simulation comparison and the main claims are appropriately scoped. The reader's flagged weakest assumption (transferability of the ε_q=0.15 sub-grid AGN model) is real but explicitly acknowledged in §7 and would require a new simulation suite to test. The more actionable gap is internal: the headline robustness uses one of two shape estimators, and the other is known to change alignment amplitudes and to better mimic luminosity weighting. Since the reduced tensor is explicitly tied to observed shapes in §3.1, the observational relevance of the headline depends on it. Appendix A only checks the e/d and d/e cross-correlations, not the projected matter-shape statistic that carries the abstract claim. A quick recomputation settles whether the concern lands, so the appropriate verdict is conditional acceptance rather than rejection or unconditional acceptance.","tokens_in":24245,"tokens_out":4014,"duration_ms":40283,"concrete_test":"Recompute rp×wδ+ (and rp×wg+ for completeness) for both runs at z=0 and z=1 using the reduced inertia tensor (Eq. 4) for the full N>300, level=1 galaxy sample, with identical binning and the same 8-subbox error estimation as Fig. 4. If the AGN/noAGN difference exceeds the error bars at any rp, the robustness headline is estimator-dependent; if the signals remain consistent within errors, the concern is retired. Publish the simple-versus-reduced comparison in a table.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim (abstract; §6.1.1) that the projected shape–density correlation wδ+ is robust to AGN feedback is demonstrated only for the simple inertia tensor (Eq. 3). Section 3.1 defines the reduced inertia tensor (Eq. 4) as better mimicking luminosity-weighted observed shapes, and Appendix A shows that switching to it lowers tangential and increases radial alignments, with effects large enough to change which population dominates cross-correlations. Horizon-AGN and Horizon-noAGN have very different morphology mixtures (§3.2, Fig. 1), so the observed agreement in wδ+ could be a cancellation that does not survive the reduced-tensor weighting. Appendix A verifies the AGN/noAGN ordering only for the e/d and d/e cross-correlations, not for wδ+. Without a reduced-tensor version of Fig. 4 bottom-right, the abstract's strongest claim about weak-lensing contamination is not tied to the shape definition actually used by lensing surveys.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper investigates whether AGN feedback alters the intrinsic alignments of galaxies by comparing two Horizon simulations that are identical except for the presence of the AGN feedback implementation. The authors measure 3D orientation–separation and spin–separation correlations, projected position–shape correlations (wg+) and shape–density correlations (wδ+), and spin alignments with filaments, at z=0 and z=1, for the full galaxy population, for high-mass ellipsoids, for discs, and for a matched 'twin' sample across the two runs. The main reported findings are that wδ+ for the full population is robust to AGN feedback, that wg+ and ηe around galaxy positions are enhanced by AGN, and that stronger spin–filament alignments in Horizon-AGN arise from the increased abundance of massive pressure-supported galaxies rather than from a change in how the same galaxies orient.","tokens_in":24459,"tokens_out":6648,"duration_ms":61149,"significance":"If the conclusions hold, the paper provides a useful step toward understanding how baryonic feedback processes affect intrinsic alignments, a key contaminant for weak lensing surveys. The analysis design is a genuine strength: the twin-run setup controls for initial conditions and numerical scheme, the matched-galaxy comparison isolates population changes from orientation changes, and the error estimates from eight sub-boxes are a reasonable internal consistency check. The paper is also honest about its limitations, noting the dependence on the sub-grid AGN prescription and the need for survey-like galaxy selections. The main weakness is that the headline wδ+ robustness claim, which is the basis for the weak-lensing contamination statement, is currently tied to a single shape definition and needs an additional test before the abstract's claim is fully supported.","major_comments":[{"comment":"","section":"Sec. 6.1.1, Fig. 4 (bottom-right); Sec. 3.1; Appendix A"}],"minor_comments":[{"comment":"The reference list contains apparent duplicates: Tenneti et al. 2015a and 2015b are both MNRAS 448, 3522, and Hirata et al. 2007a and 2007b are both MNRAS 381, 1197. These should be merged or clearly distinguished.","section":"References"},{"comment":"The V/σ thresholds used for the ellipsoid and disc samples are defined at z=0 and then applied at z=1, which strongly affects sample sizes at z=1 (Table 1). The authors should explicitly justify this choice and discuss its impact on the z=1 comparisons, especially for high-mass ellipsoids where the noAGN sample is extremely sparse.","section":"Sec. 3.2"},{"comment":"The y-axis label '1⋆ξ' is unclear; the caption mentions a probability distribution function re-scaled by the total number of objects times bin size, so the label should state the exact normalized quantity.","section":"Fig. 11"},{"comment":"The density-field tracers are described as unbiased, but they consist of a sub-sample of stars, black holes, gas and dark matter particles. This mixture is not automatically an unbiased tracer of the total matter field; the paper should state more precisely in what sense the 10% convergence test of Chisari et al. (2015) establishes unbiasedness for the alignment statistics used here.","section":"Sec. 5.2"}],"recommendation":"major_revision","confidential_remarks":"The paper is well designed and the core comparison is valuable. The main risk is that the weak-lensing-relevant claim (wδ+ robustness) rests on a shape definition that is not the one the paper itself identifies as observationally motivated. This is fixable by adding a reduced-tensor version of the wδ+ measurement, so I recommend major revision rather than rejection. The authors should also ensure the response to the major comment addresses the possibility that the simple-tensor agreement is a population-weighted cancellation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: this is a clean twin-simulation study of AGN feedback on intrinsic alignments. The headline result, that the density-shape correlation wδ+ for the full sample is robust to AGN while the position-shape wg+ is not, is useful for weak lensing forecasts. The matched twin-galaxy analysis is the real step forward—it separates population selection effects from actual changes in orientation, and it gives the paper a causal structure that simulation comparisons often lack.\n\nThe paper deserves credit for being careful. The 3D and projected statistics are both used, error bars are jackknife over eight sub-boxes, the matching procedure is described and its efficiency quantified, and the authors repeatedly flag that their conclusions depend on one AGN sub-grid model. The comparison with Tenneti et al. (2017) makes the new result clear: previous work said AGN had little effect on alignments; here wg+ shows a significant effect for high-mass ellipsoids, and the filament spin alignments are stronger in the AGN run because of the abundance of pressure-supported galaxies, not because twins orient differently.\n\nWhere the soft spots are. The wδ+ robustness is a null result from one simulation volume with the simple inertia tensor only. The stress-test note points out that the reduced inertia tensor—the one that better mimics luminosity-weighted observed shapes—changes alignment amplitudes, and no reduced-tensor version of the wδ+ panel is shown. That is a legitimate gap. Appendix A shows the AGN/noAGN ordering survives the tensor choice for the e/d and d/e cross-correlations, which is reassuring, but it does not close the gap for wδ+. So the abstract's phrasing is a bit broader than what is demonstrated. I'd call this a minor-to-moderate caveat rather than a fatal flaw, and it is easily addressed in a revision by adding one panel or softening the wording.\n\nThe z=1 high-mass ellipsoid statistics are very sparse, as the paper admits. And the whole analysis leans on one simulation suite; the conclusion about transferability to Euclid/LSST-era calibration is necessarily provisional. The authors say this themselves in Section 7, so the paper is honest about its limits.\n\nIf I were refereeing this, I would recommend acceptance with minor revisions. The central claims are supported within the stated error bars, the matching analysis is a real contribution, and the tensor-choice gap is fixable. I would want to see the reduced-tensor wδ+ test or an explicit boundary on the claim. This paper is for people building intrinsic alignment templates for weak lensing, and for simulators comparing sub-grid physics; they will get value from it.\n\nRecommendation: engage with it. Send to a serious referee.","headline":"A careful twin-simulation comparison showing AGN feedback changes galaxy position-shape alignments but not the density-shape signal used for lensing contamination; the main claim rests on one shape definition and one simulation volume, but the matched-galaxy analysis is a real advance.","tokens_in":25002,"tokens_out":2325,"would_cite":true,"duration_ms":22039,"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":"AGN feedback leaves the weak-lensing alignment signal intact","keywords":["intrinsic alignments","AGN feedback","weak gravitational lensing","galaxy shapes","galaxy spins","cosmic filaments","hydrodynamical simulations","Horizon simulations"],"falsifier":"Take the same twin-simulation setup, raise the quasar-mode efficiency to a substantially higher value, and re-measure $r_p \\times w_{\\delta+}(r_p)$ for all galaxies with more than 300 stellar particles at $z=0$; if the projected shape-density correlation shifts by more than the error bars at $r_p \\approx 1$ to $10$ Mpc, the claim that AGN feedback leaves the weak-lensing contamination unchanged fails beyond this one calibration.","tokens_in":1667,"feed_emoji":"🌌","tokens_out":5717,"duration_ms":86403,"temperature":0.7,"pith_summary":"Does the energy injected by active galactic nuclei into their surroundings change how galaxies' shapes and spins align with the cosmic web? This paper compares two otherwise identical cosmological simulations, one with AGN feedback and one without, and measures shape-density, shape-position, spin-position, and spin-filament correlations. The central result is that the projected shape-density correlation, the quantity that contaminates weak lensing, is the same in both runs, while the shape-position correlation around galaxies is stronger in the AGN run, especially for massive ellipsoids. Using galaxies matched across the two runs, the paper separates population changes from changes in how individual galaxies orient. The practical conclusion is that AGN feedback can be ignored when modelling the GI contamination to cosmic shear, but not when interpreting galaxy-galaxy alignment measurements.","feed_headline":"AGN feedback leaves the weak-lensing alignment signal intact","feed_subtitle":"Twin simulations show density-shape correlations hold while galaxy-galaxy alignments grow stronger.","key_machinery":"The comparison rests on twin cosmological hydrodynamical runs of the Horizon simulation that share identical initial conditions and sub-grid recipes, differing only in the presence of AGN feedback, implemented as a thermal quasar mode and a kinetic radio mode. Galaxy shapes are measured from inertia tensors of stellar particles, with reliable shapes requiring more than 300 particles, and alignments are quantified by the three-dimensional orientation-direction correlation $\\eta_e(r)$ and by the projected correlations $w_{g+}$ and $w_{\\delta+}$. A galaxy-matching procedure pairs objects across the two runs to separate selection effects from genuine orientation changes, while filament extraction with a persistence-based skeleton provides the structure used for spin-filament alignments. Applying the same statistics to matched and unmatched samples is what lets the paper attribute differences either to population composition or to alignment changes.","core_discovery":"Within the Horizon simulation suite, the projected alignment of all resolved galaxy shapes with respect to the matter density field ($w_{\\delta+}$) is unaffected by AGN feedback at $z=0$ and $z=1$, so the predicted contamination to weak lensing is unchanged. By contrast, the projected correlation of shapes around galaxy positions ($w_{g+}$) is significantly enhanced in the AGN run, with the difference driven by high-mass ellipsoids that exist in Horizon-AGN but are absent or more rotation-supported in the noAGN run. When galaxies are matched across the two simulations, the shape-position correlation is still stronger in the AGN run, showing that AGN feedback changes galaxy orientations themselves, not only the sample selection. Spin alignments around filaments are stronger in the AGN run, but matched twins show the same orientation statistics, indicating that the stronger signal comes from a larger abundance of massive pressure-supported galaxies rather than from a change in how those galaxies orient.","pith_inferences":["A stronger AGN prescription than the one calibrated in Horizon-AGN could plausibly alter $w_{\\delta+}$; rerunning the twin comparison with varied feedback efficiency would show whether the GI robustness is general or specific to this calibration.","Because AGN feedback changes $w_{g+}$ at small scales, future lensing surveys measuring galaxy-galaxy alignments could in principle use the small-scale signal to discriminate between feedback models.","The matched noAGN twins of massive AGN galaxies also lie above the same mass threshold, so the perpendicular spin-filament orientation of massive galaxies appears to be inherited from sample selection; an observational test would measure spin-filament alignment among massive quenched galaxies selected by their star-formation state.","The choice of inertia tensor changes the amplitude of radial and tangential alignments but not which run dominates, so the qualitative AGN comparison may hold for other shape-weighting schemes, though quantitative predictions for lensing should be checked with luminosity-weighted shapes."],"forward_implications":["The projected shape-density correlation $w_{\\delta+}$ can be treated as insensitive to AGN feedback in models of weak-lensing contamination, at least for this simulation family.","Modelling of galaxy position-shape correlations $w_{g+}$ from hydrodynamical simulations should account for AGN feedback, particularly at small scales and for high-mass ellipsoids.","Observational position-shape alignment measurements may indirectly constrain AGN feedback strength once population selection effects are separated.","The mass-dependent spin-filament transition is weaker without AGN feedback, suggesting the spin flip of massive galaxies is partly a population-driven feedback effect.","Cross-correlations of ellipsoids around discs show that AGN feedback dampens alignments at $r<0.8$ Mpc and amplifies them at $r>0.8$ Mpc, a pattern reproduced in projection."],"supporting_citations":[{"why":"establishes the inertia-tensor shape definitions, the N>300 reliability cut, and the $w_{\\delta+}$ and $w_{g+}$ estimators used in this comparison.","marker":"Chisari et al. 2015"},{"why":"presents the Horizon-AGN simulation and its AGN feedback implementation, which serves as the reference run.","marker":"Dubois et al. 2014"},{"why":"presents the Horizon-noAGN twin run with identical initial conditions and sub-grid recipes except for AGN feedback.","marker":"Peirani et al. 2017"},{"why":"shows that AGN feedback drives galaxies from rotation-dominated to dispersion-dominated, explaining why high-mass ellipsoids populate Horizon-AGN.","marker":"Dubois et al. 2016"},{"why":"describes the galaxy-matching procedure that produces the twin samples used to separate selection effects from orientation changes.","marker":"Beckmann et al. 2017"},{"why":"provides the earlier, smaller-box result that AGN feedback induces only small changes in galaxy alignments, which this larger comparison extends.","marker":"Tenneti et al. 2017"},{"why":"introduces the persistence-based skeleton extraction used to define the filaments for spin-filament alignments.","marker":"Sousbie 2011"},{"why":"establishes the mass-dependent spin-filament alignment signal in Horizon-AGN that the noAGN comparison shows to be weaker without feedback.","marker":"Codis et al. 2018"}],"fun_headline_variants":["AGN feedback alters galaxy alignments but not lensing signal","Horizon twins show AGN reshapes galaxy-galaxy alignments","Weak-lensing contamination robust to AGN feedback","AGN boosts galaxy-galaxy alignment, lensing signal safe"],"cache_read_input_tokens":27264,"weakest_assumption_plain":"The conclusions depend on the Horizon-AGN sub-grid AGN feedback model, with quasar-mode efficiency $\\epsilon_q=0.15$, being representative of real AGN feedback; if true AGN feedback is stronger or differently calibrated, the alignment signals and the apparent robustness of $w_{\\delta+}$ could change.","fun_headline_variants_meta":{"raw":{"variants":["AGN feedback alters galaxy alignments but not lensing signal","Horizon twins show AGN reshapes galaxy-galaxy alignments","Weak-lensing contamination robust to AGN feedback","AGN boosts galaxy-galaxy alignment, lensing signal safe"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000965,"raw_usage":{"total_tokens":4126,"prompt_tokens":981,"completion_tokens":3145,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":3076}},"tokens_in":597,"tokens_out":3145,"duration_ms":21613,"temperature":1.0,"reasoning_tokens":3076,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:09:14.489183+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the same twin-simulation setup, raise the quasar-mode efficiency to a substantially higher value, and re-measure $r_p \\times w_{\\delta+}(r_p)$ for all galaxies with more than 300 stellar particles at $z=0$; if the projected shape-density correlation shifts by more than the error bars at $r_p \\approx 1$ to $10$ Mpc, the claim that AGN feedback leaves the weak-lensing contamination unchanged fails beyond this one calibration.","supporting_citations":[],"review_version":1}