{"id":"a8f81089-c25f-45f1-8a64-d6c3388e6b16","arxiv_id":"2411.15297","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Across nine cosmological simulations, dual AGN number densities range from 1e-8 to 1e-3 cMpc^-3 at z = 0-7, with fractions of 0-6% and a redshift peak at z = 1-3.","lead":"This paper measures how often pairs of actively accreting supermassive black holes, called dual AGN, appear in nine different computer simulations of the universe. It finds that their predicted abundance varies by orders of magnitude between simulations, peaks at redshift 1 to 3, and may be too low today to match observations, with direct implications for the AXIS and Athena X-ray telescopes and the LISA gravitational wave observatory.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The z=0 'too few DAGN' claim compares 3D-separated, Lbol>=1e43 simulation pairs to projected, optically/X-ray-selected observed samples without matching selection; the deficit may be a selection artifact rather than DAGN physics.","rationale":"The paper is a careful, transparent comparison, and its main contribution—a uniform DAGN census across nine simulations with a robust z=1-3 peak—does not depend on the specific z=0 comparison. The reader's stated weakest assumption (uncalibrated AGN LFs) is genuine, but it primarily threatens the absolute DAGN number densities and high-z evolution, and the paper partially mitigates it by reporting fractions and by showing results for high stellar-mass and high-luminosity cuts. The z=0 deficit is different: it is the paper's sharpest observational conclusion, and it depends entirely on the equivalence between the simulated and observed samples. Because the paper explicitly worries about projection for forecasts (Sect. 5.1d) yet does not apply the same care to the Liu/Koss comparison, the unresolved selection mismatch is the single most load-bearing gap. My proposed test is a direct, cheap reanalysis of already-existing catalogs. I do not see a reason to change the reader's CONDITIONAL verdict: the paper deserves acceptance conditional on this selection-matched reanalysis (and ideally release of the DAGN catalogs), but the central qualitative findings are likely to survive.","tokens_in":38523,"tokens_out":14333,"duration_ms":133871,"concrete_test":"Recompute the z=0 panels of Fig. 5 with observed-matched selection: project each simulated pair along a random line of sight; keep pairs with projected separation 3.5-30 kpc (Liu) or resolved companion with L2-10keV>1e42 erg/s from Eq. 3 (Koss); restrict to the corresponding redshift/volume and to the three fraction definitions in Fig. 5. If the simulated fractions move above the Liu/Koss values, the deficit is a selection artifact; if they remain below, the deficit claim survives.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline inference that simulations make too few z=0 DAGN (Abstract; Sect. 3.2.1; Fig. 5) rests on a comparison that does not match the observed selection. Simulated DAGN are defined by 3D separation d<=30 pkpc and Lbol>=1e43 erg/s. Liu et al. (2011) use projected separations 3.5-30 kpc and optical emission-line AGN selection; Koss et al. (2012) use L2-10keV>1e42 erg/s at z<0.05. Through the adopted Lbol-Lx relation (Eq. 3), Lbol=1e43 erg/s corresponds to Lx~7.8e41 erg/s, so the simulation sample is fainter than Koss's. Moreover, a projected separation cut admits pairs with 3D separations larger than 30 pkpc, biasing the observed fraction upward relative to the 3D cut. The Fig. 6 caption states that using 3D instead of projected separation 'does not impact the results here,' but no supporting calculation is shown for the z=0 constraints of Fig. 5. Until the simulated fractions are recomputed with the observed projected/luminosity selection, the claimed z=0 deficit is not a secure result. This is more load-bearing than the AGN-LF normalization issue, because the fraction definition largely cancels LF normalization, while the selection mismatch enters directly into the comparison that produces the headline deficit.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper uses nine published large-scale cosmological hydrodynamical simulations (Illustris, TNG50/100/300, Horizon-AGN, EAGLE, SIMBA, BlueTides, and Astrid) to compute the number density, fraction, host-galaxy and MBH properties, and X-ray detectability of dual AGN, defined as two AGN with Lbol >= 1e43 erg/s in distinct galaxies with three-dimensional separation <= 30 pkpc. The main quantitative results are a DAGN number density spanning roughly 1e-8 to 1e-3 cMpc^-3 over z = 0-7, a peak in DAGN number density and fraction at z = 1-3, a claimed deficit of simulated DAGN at z = 0 relative to the constraints of Liu et al. (2011) and Koss et al. (2012), consistency with the Sandoval et al. (2023) upper limit at z ~ 3, and forecasts that future X-ray observatories, especially AXIS, should detect from a few to more than a hundred DAGN depending on survey and simulation.","tokens_in":38823,"tokens_out":10358,"duration_ms":98033,"significance":"The paper is a useful census that applies a uniform DAGN definition across nine simulations and systematically explores the sensitivity of the results to selection cuts in stellar mass, MBH mass, luminosity, and separation. Its qualitative conclusions, such as a DAGN peak at z = 1-3 and fractions of order a few percent, are likely robust to several modeling choices, and the AXIS/Athena detection forecasts with cosmic-variance estimates are a valuable forward-looking element. The authors are transparent about the uncalibrated nature of the simulated AGN luminosity functions and about the dependence of the results on definitions. However, the headline comparison at z = 0 is currently weakened by a mismatch between the simulated selection (3D separation, Lbol >= 1e43 erg/s) and the observed selections (projected separation, X-ray or emission-line thresholds), and the absolute number densities remain conditioned on the AGN luminosity-function normalizations of the host simulations. These issues need to be addressed before the quantitative claims, especially the claimed z = 0 deficit, can be taken at face value.","major_comments":[{"comment":"The claim that all simulations produce too few DAGN at z = 0 is not yet securely established, because the simulated and observed samples are selected differently. The simulation selection uses a 3D separation d <= 30 pkpc and Lbol >= 1e43 erg/s, whereas Liu et al. (2011) use projected separations of 3.5-30 kpc with optical emission-line AGN selection and Koss et al. (2012) use L2-10keV > 1e42 erg/s at z < 0.05. Using Eq. (3), Lbol = 1e43 erg/s corresponds to Lx ~ 7.8e41 erg/s, so the simulated sample contains AGN fainter than the Koss threshold; conversely, a projected-separation cut can include pairs whose true 3D separation exceeds 30 pkpc. These two effects act in opposite directions, so the net bias on the simulated DAGN fraction is unknown. The analogous statement in the Fig. 6 caption that using 3D rather than projected separation 'does not impact the results here' is not supported by a calculation for the z = 0 constraints. Please recompute the DAGN fractions with projected separations and with the luminosity thresholds of the comparison samples, accounting for the redshift window of each survey, and report both the matched and unmatched values.","section":"Sec. 3.2.1 and Fig. 5 (also Abstract and Sec. 6)"},{"comment":"The absolute DAGN number densities are outputs of AGN populations whose luminosity functions are not calibrated and, as the paper notes in Sec. 2.2, overproduce faint AGN at z >= 4 relative to observational constraints (Habouzit et al. 2022). The 1e-8 to 1e-3 cMpc^-3 range quoted in the Abstract therefore mixes DAGN physics with AGN normalization differences between the simulations. The DAGN fraction partly cancels this normalization, but the denominator (total number of AGN) is itself affected by the overproduction of low-luminosity AGN in low-mass galaxies, so the fraction is not fully immune. Please add a robustness test, for example by reweighting or restricting the simulated AGN to the luminosity and host-stellar-mass range where each simulation's AGN luminosity function is in better agreement with observations, and state which of the paper's quantitative conclusions survive. At minimum, the high-redshift (z >= 4) DAGN densities should be explicitly labeled as conditional on the uncalibrated AGN luminosity function.","section":"Sec. 3.1, Sec. 2.2, and Table A1"},{"comment":"The three fraction definitions shown in the rows of Fig. 5 are not all matched to the definitions used by the plotted observational constraints, yet the same observational points are repeated in every panel. Liu et al.'s 1.85% includes a correction for SDSS spectroscopic incompleteness and a projected-separation window of 3.5-30 kpc, while the simulation curves use 3D separations and no incompleteness correction; Koss et al.'s fraction is based on an X-ray-selected parent sample. Repeating the observational values in all panels makes the comparison look more exact than it is. Please place each observational constraint only in the panel(s) whose definition it matches, or state explicitly in the caption and text which rows are not directly comparable.","section":"Sec. 3.2.1 and Fig. 5"}],"minor_comments":[{"comment":"The minimum DAGN number density is quoted as 1e-8 cMpc^-3 in the Abstract and Conclusions but as 1e-7 cMpc^-3 in Sec. 3.1.1; unify these values after deciding whether the zero-DAGN cases are included in the quoted range.","section":"Abstract and Sec. 3.1.1"},{"comment":"The label 'Koos et al. 2012' should read 'Koss et al. 2012'.","section":"Fig. B1"},{"comment":"There are several LaTeX and formatting artifacts, including 'MN L ATEX style file' on the title page and 'T able 1' and 'T able A1' in the text; these should be cleaned before final submission.","section":"Title page and running text"},{"comment":"In the paragraph discussing EAGLE, 'EAGLE MHBs' should be 'EAGLE MBHs'.","section":"Sec. 3.1.1"},{"comment":"The observed points from Liu et al. and Koss et al. are plotted without error bars, while the text notes that the observational error bars can be large and that the values are sensitive to definitions; please either add error bars or clearly state that the plotted values are central estimates.","section":"Sec. 3.2.1 and Fig. 5"},{"comment":"The detectability forecast discusses DAGN with separations larger than 30 pkpc for Athena, but the fiducial DAGN definition in the rest of the paper is d <= 30 pkpc; please clarify that this part intentionally extends the separation range.","section":"Sec. 5.1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a useful multi-simulation census rather than a new simulation campaign, and it fits the scope of MNRAS. The main risk is the z = 0 comparison in Fig. 5, which I expect the authors can fix with a matched-selection recomputation; if that recomputation removes the deficit, the abstract and conclusions must be revised accordingly. I do not see concerns about citation practice or novelty disclosure."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a genuinely useful census—the first uniform DAGN selection applied across nine large-scale simulations. The central numbers (n_DAGN ~ 10^-8 to 10^-3 cMpc^-3, fractions 0–6%, peak at z ~ 1–3) follow directly from the pipeline, and the authors are admirably transparent about how definitions change the results.\n\nWhat is new and good: the uniform selection (L_bol >= 10^43 erg/s, d <= 30 pkpc, distinct galaxies, M* >= 10^9 Msun) across Illustris, TNG50/100/300, EAGLE, SIMBA, Horizon-AGN, BlueTides, and Astrid; the systematic exploration of M* and L_bol cuts; three fraction definitions; and concrete forecasts for AXIS and Athena, including XRB contamination and cosmic variance. The paper also gives a clear picture of where subgrid models diverge—TNG and Horizon produce many DAGN, EAGLE and SIMBA few—and that spread is a useful diagnostic.\n\nSoft spots. The one that matters is the z=0 comparison with Liu+11 and Koss+12. The stress-test concern lands: the simulated sample is 3D-separated with L_bol >= 10^43, while the observed samples are projected-separated and Koss uses Lx > 10^42. The paper notes the luminosity difference but never addresses the 3D-vs-projected issue for this figure. The bias inflates observed fractions, so the claimed 'too few DAGN at z=0' is not yet secure. The authors hedge with 'could', and the deficit is large enough that it may survive a matched comparison, but they need to do the calculation. This seems more important than the AGN-LF normalization issue, since the fraction definition cancels much of that.\n\nThe uncalibrated AGN LFs are still worth flagging: number densities inherit known overproduction of faint AGN at z >= 4, and the 2-dex spread is driven partly by AGN normalization differences. The paper acknowledges this clearly, so it is a caveat rather than a flaw. Minor: no code or catalogs released, though they say 'upon request'.\n\nWho it is for: anyone planning surveys with AXIS or Athena, and simulators who want a target range for DAGN fractions. It is not a paradigm shift, but it is a solid reference. Worth a serious referee: the comparison issue should be fixed in revision, but the paper deserves review. I would send it out.","headline":"A solid, transparent multi-simulation census of dual AGN, but the z=0 deficit claim needs a matched selection comparison before it becomes a secure result.","tokens_in":39496,"tokens_out":5779,"would_cite":true,"duration_ms":49447,"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":"Nine cosmological simulations place the dual-AGN peak at cosmic noon and find too few such systems in the local universe.","keywords":["dual active galactic nuclei","massive black holes","cosmological hydrodynamic simulations","AGN luminosity function","galaxy mergers","X-ray surveys","gravitational wave precursors","cosmic noon"],"falsifier":"A volume-complete, arcsecond-resolution X-ray survey at z ≈ 2–3 that counts AGN pairs with $L_{\\rm bol}\\ge 10^{43}\\,\\mathrm{erg\\,s^{-1}}$ and separation $\\le 30\\,\\mathrm{pkpc}$ could settle it: the simulations predict about $10^{-5}$ to $10^{-3}\\,\\mathrm{cMpc^{-3}}$, so a survey that should contain dozens but finds none would falsify the peak-density prediction, and a detection rate an order of magnitude higher would falsify the current upper limits.","tokens_in":38287,"feed_emoji":"🔭","tokens_out":9686,"duration_ms":81144,"temperature":0.7,"pith_summary":"This paper asks how many dual active galactic nuclei—two accreting massive black holes in separate galaxies, each shining above $10^{43}$ erg/s and separated by no more than 30 proper kiloparsecs—exist across cosmic time. Pulling together nine large cosmological hydrodynamic simulations with different black-hole seeding and feedback recipes, it argues that these systems are rare, making up 0 to 6 percent of all AGN, and that their number density peaks at z ≈ 1–3, just before the cosmic peaks of star formation and AGN activity. At a fixed redshift the simulated DAGN density spans two orders of magnitude, but every simulation agrees on the timing of the peak. It would matter because DAGN are the electromagnetic counterparts that could be caught before massive black holes coalesce, so a reliable census tells future X-ray and gravitational-wave observatories where and when to look; the same simulations produce too few DAGN at z = 0 compared with local surveys, suggesting either overly fast merging or missing physics.","feed_headline":"Simulations agree: dual AGN peak at cosmic noon","feed_subtitle":"DAGN densities span 10^-8 to 10^-3 cMpc^-3, and the local universe may hold too few.","key_machinery":"The carrying definition is a three-part selection rule: a pair of massive black holes each with bolometric luminosity $L_{\\rm bol}\\ge 10^{43}\\,\\mathrm{erg\\,s^{-1}}$, three-dimensional separation $d\\le 30\\,\\mathrm{pkpc}$, and members residing in distinct galaxies. The distinct-galaxy clause is what lets the nine simulations—which treat black-hole dynamics very differently, from repositioning black holes to galactic centres to modelling dynamical friction—be compared on equal footing. Bolometric luminosities are computed from accretion rates via $L_{\\rm bol}=0.1\\,\\dot{M}_{\\rm MBH} c^2$, with a fixed bolometric correction to the 2–10 keV band for the X-ray detectability forecasts.","core_discovery":"On the simulations' own terms, the paper establishes three things. First, DAGN with Lbol ≥ $10^{43}$ erg/s in distinct galaxies and d ≤ 30 pkpc have number densities from about $10^{-8}$ (or zero) to $10^{-3}$ $cMpc^{-3}$ over z = 0–7, with fractions of AGN in pairs of 0–6 percent. Second, all nine simulations place the peak of that population at z = 1–3, slightly earlier than the peak of AGN activity, and the typical DAGN separation grows with time while the median black-hole mass grows with time. Third, the simulated z = 0 DAGN fractions fall below the values reported by local AGN-pair surveys, while the z ≈ 3 predictions are consistent with current upper limits; the paper interprets this as simulations either producing too few pairs or merging them too quickly. The paper also translates the simulated populations into predicted detections for upcoming X-ray telescopes and argues that a large fraction of the DAGN it finds in EAGLE would fall in the LISA gravitational-wave detection region.","pith_inferences":["Beyond the paper: if the z = 0 deficit is physical rather than a luminosity-function artefact, it implies the subgrid models merge massive black holes too promptly; a direct test would be to rerun one simulation with redshift-dependent dynamical-friction delays and see whether the local DAGN fraction rises to the observed value.","Beyond the paper: the two-order-of-magnitude spread in predicted densities makes DAGN a sharper diagnostic of subgrid AGN physics than the overall AGN luminosity function, so matching a single upcoming DAGN survey could break degeneracies between seeding mass, accretion efficiency, and feedback.","Beyond the paper: because the simulations only treat pairs in distinct galaxies, the consensus peak at z = 1–3 may underestimate the true DAGN abundance once same-galaxy pairs at separations below about 5 pkpc, which current surveys cannot resolve, are included; high-resolution zoom-in simulations could test this."],"forward_implications":["The predicted DAGN peak at z ≈ 1–3 means deep surveys should target cosmic noon: the AXIS Deep field is expected to recover about 10^-5 to 10^-3 cMpc^-3 of these systems at z = 2–3.","If the simulated z = 0 shortfall is correct, local AGN-pair surveys are detecting systems that current feedback and merging recipes cannot reproduce, pointing to missing merger delays or overly prompt numerical merging.","Restricting to bright, massive-MBH DAGN, as many observational samples do, changes both the fraction and its redshift evolution, so survey selection must be matched before comparing theory with data.","At least 75 percent of the EAGLE DAGN identified by the paper fall in the LISA signal-to-noise region, meaning a dedicated electromagnetic survey of such systems could flag gravitational-wave progenitor candidates."],"supporting_citations":[{"why":"Provides the AGN luminosity functions of the nine simulations, including the known overproduction of faint AGN that the paper relies on to frame the DAGN predictions.","marker":"Habouzit et al. (2022)"},{"why":"Supplies the low-redshift SDSS DAGN fraction of about 1.85 percent that the simulated z = 0 fractions are compared against and underproduce.","marker":"Liu et al. (2011)"},{"why":"Supplies the local hard-X-ray BAT DAGN fraction used in the z = 0 comparison.","marker":"Koss et al. (2012)"},{"why":"Provides the z ≈ 3 upper limit on the DAGN fraction with which the simulations are consistent at that redshift.","marker":"Sandoval et al. (2023)"},{"why":"Provides the luminous, massive-MBH DAGN fraction selection and constraint used for the z ≤ 3.5 comparison.","marker":"Silverman et al. (2020)"},{"why":"Earlier EAGLE DAGN analysis whose selection criteria and fractions are compared for consistency.","marker":"Rosas-Guevara et al. (2019)"},{"why":"Earlier Horizon-AGN DAGN analysis that supplies the method of keeping one MBH per galaxy for that simulation.","marker":"Volonteri et al. (2022)"},{"why":"Astrid DAGN analysis with massive MBHs whose nearly constant fraction at z = 2–5 is reproduced and discussed.","marker":"Chen et al. (2023)"},{"why":"Supplies the bolometric-to-hard-X-ray luminosity conversion used for the detectability forecasts.","marker":"Shen et al. (2020)"}],"fun_headline_variants":["Dual AGN peak at z=1–3 in all nine simulations","Simulated dual AGN: peak at cosmic noon, scarce locally","Dual AGN densities vary by 100x across simulations","Local dual AGN too few? Simulations say maybe","Future X-ray telescopes key to dual AGN hunt"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the simulated AGN populations are reliable enough for pair counting even though their luminosity functions are not calibrated and are known to overproduce faint AGN in low-mass galaxies at high redshift; if that overproduction dominates the DAGN counts, the predicted densities, the z = 1–3 peak, and the local deficit would be artefacts of the subgrid physics rather than real astrophysics.","fun_headline_variants_meta":{"raw":{"variants":["Dual AGN peak at z=1–3 in all nine simulations","Simulated dual AGN: peak at cosmic noon, scarce locally","Dual AGN densities vary by 100x across simulations","Local dual AGN too few? Simulations say maybe","Future X-ray telescopes key to dual AGN hunt"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000358,"raw_usage":{"total_tokens":2021,"prompt_tokens":1111,"completion_tokens":910,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":727,"completion_tokens_details":{"reasoning_tokens":824}},"tokens_in":727,"tokens_out":910,"duration_ms":8597,"temperature":1.0,"reasoning_tokens":824,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:26:43.042074+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A volume-complete, arcsecond-resolution X-ray survey at z ≈ 2–3 that counts AGN pairs with $L_{\\rm bol}\\ge 10^{43}\\,\\mathrm{erg\\,s^{-1}}$ and separation $\\le 30\\,\\mathrm{pkpc}$ could settle it: the simulations predict about $10^{-5}$ to $10^{-3}\\,\\mathrm{cMpc^{-3}}$, so a survey that should contain dozens but finds none would falsify the peak-density prediction, and a detection rate an order of magnitude higher would falsify the current upper limits.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the AGN luminosity functions of the nine simulations, including the known overproduction of faint AGN that the paper relies on to frame the DAGN predictions."},{"cited_title":"A., Hao L., 2011, , 737, 101","cited_arxiv_id":null,"evidence_quote":"Supplies the low-redshift SDSS DAGN fraction of about 1.85 percent that the simulated z = 0 fractions are compared against and underproduce."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the local hard-X-ray BAT DAGN fraction used in the z = 0 comparison."},{"cited_title":"Searching for the Highest-z Dual AGN in the Deepest Chandra Surveys","cited_arxiv_id":"2312.02311","evidence_quote":"Provides the z ≈ 3 upper limit on the DAGN fraction with which the simulations are consistent at that redshift."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the luminous, massive-MBH DAGN fraction selection and constraint used for the z ≤ 3.5 comparison."},{"cited_title":"M., Bower R","cited_arxiv_id":null,"evidence_quote":"Earlier EAGLE DAGN analysis whose selection criteria and fractions are compared for consistency."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier Horizon-AGN DAGN analysis that supplies the method of keeping one MBH per galaxy for that simulation."},{"cited_title":"F., Faucher-Gigu \\`e re C.-A., Alexander D","cited_arxiv_id":null,"evidence_quote":"Supplies the bolometric-to-hard-X-ray luminosity conversion used for the detectability forecasts."}],"review_version":1}