{"id":"6ea6be30-40e0-44d3-bedf-11235ffaab31","arxiv_id":"2506.02289","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Full-transport radiation GRMHD simulations show super-Eddington black hole accretion is geometrically thick, drives strong outflows, and radiates with very low efficiency (below about 0.5% at 150 times Eddington).","lead":"This paper uses a new, more accurate computer simulation method to model gas falling onto a stellar-mass black hole at rates from a tenth of the Eddington limit to 150 times it. The simulations show that very fast-eating black holes are surrounded by thick, puffy disks that produce surprisingly little light, which could change how we interpret bright X-ray sources and early-universe galaxies.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The survey's key quantitative claim—super-Eddington disks radiate at <0.5% efficiency—is in direct, unresolved conflict with Fragile et al. (2025), cited in §3.6 (Eddington-capped Mdot, 30–70% efficiency).","rationale":"This is a survey whose central deliverable is the Eddington-ratio dependence of radiative efficiency (Table 1, Fig. 9) and the vertical-flux-driven structural dichotomy. The method has independent support (White et al. 2023 validation), and the paper provides meaningful internal diagnostics: MRI quality factors, resolved thermal scale heights, inflow-equilibrium windows, and run durations exceeding one thermal time in the measurement region. The Table 1 trend is internally coherent, and the factor-of-two measurement-radius sensitivity (the paper notes >45% of luminosity is generated within the inflow-equilibrium region) does not threaten the order-of-magnitude swing between 0.1 and 150 Mdot_Edd. The reader's flagged assumption—the rho0-normalized torus as a proxy for astrophysical feeding—is real but explicitly acknowledged in §5, partially mitigated by the beaming-based inflow argument citing Kaaz et al. (2025), and only testable with a much larger cyclic-zoom calculation. I regard the §3.6 discrepancy with Fragile et al. (2025) as more immediately load-bearing: it is a direct, current, and internally acknowledged contradiction with a comparable calculation reporting the opposite behavior (Eddington-capped Mdot, 30–70% efficiency) in the same trapping-radius regime. The authors' stated reconciliation does not discriminate between the two calculations, and neither does Fig. 9's 'slim disk' comparison, which is a one-parameter normalization to E9-a3 rather than an independent validation of the trend. The claim that super-Eddington flows are inefficient radiators therefore currently rests on an unexplained regime divergence, a correctness risk rather than a consensus objection. This sharpens the condition but does not change the verdict: CONDITIONAL remains appropriate, with the resolved Fragile comparison (or its cross-code reproduction) as the explicit gating item.","tokens_in":32645,"tokens_out":19885,"duration_ms":179122,"concrete_test":"Adopt the Fragile et al. (2025) initial conditions (their generalized analytic thin-disk equilibrium, with the trapping radius small relative to disk size) in the same AthenaK full-transport configuration used here, at the supply level and resolution of model E88-a3, and run to t = 60000 r_g/c. Then compare the time-averaged Mdot at r = 3 r_g (their eq. 5) and eta_rad at r = 20 r_g (their eq. 10). Reproducing super-Eddington Mdot with eta_rad < 1% would show the discrepancy is initial-condition driven and the survey's regime is internally robust; reproducing an Eddington cap with eta_rad ~ 10% would show the headline low-efficiency claim is specific to the torus class, invalidating the §4 observational applications as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and Table 1 assert that super-Eddington accretion sustains Mdot up to 150 Mdot_Edd while radiating at <0.5% efficiency. Section 3.6 records that Fragile et al. (2025), a comparable radiation-GRMHD calculation initialized from generalized thin-disk equilibria in the same trapping-radius regime, instead finds an Eddington cap on Mdot and 30–70% efficiencies. The paper's reconciliation—that its models satisfy the condition that the trapping radius be small compared with the disk size—does not discriminate, since Fragile et al. satisfy that condition by construction. The text concedes 'the origin of the quite different outcomes between these calculations is uncertain and warrants further investigation.' If Fragile et al. are right, radiation feedback self-limits the accretion rate, and the super-Eddington states, their low efficiencies, and the ULX and LRD applications built on them (§4.5, §4.7) would not be realized even given the large external supply whose absence the paper acknowledges in §5 (feeding from the Bondi radius is unmodeled, and outflow feedback could throttle it). Conversely, if this paper is right, a published comparable calculation is wrong in a way not yet identified. Either way, the quantitative headline is not established beyond this paper's torus initial-condition class.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents ten radiation GRMHD simulations of accretion onto 10-solar-mass black holes, computed with the full transport radiation module in AthenaK (White et al. 2023). The models span estimated Eddington ratios from roughly 0.1 to 150, two black hole spins (0.3 and 0.9375), and two initial magnetic field topologies (single-loop with net vertical flux and double-loop without). The paper reports three structural regimes: geometrically thick, radiation-pressure-supported super-Eddington disks; thin, gas-pressure-supported disks with magnetically dominated coronae when net vertical flux is present; and magnetically elevated disks when it is not. It reports low radiative efficiencies at high accretion rates (dropping from about 5% at 0.1 Mdot_Edd to below 0.5% at 150 Mdot_Edd), powerful jets for spinning black holes with net vertical flux, no MAD states, and outflows with mass-loss rates comparable to the accretion rate. The results are applied to X-ray binaries, ULXs, SS 433, and, speculatively, to little red dots.","tokens_in":32869,"tokens_out":4720,"duration_ms":50365,"significance":"If the central results hold, this is a substantial contribution: it is one of the first parameter surveys of black hole accretion using direct (non-M1, non-FLD) radiation transport in full general relativity, and it comes with unusually strong technical support, including MRI quality factors (Q_z >= 15, Q_phi >= 60), at least 20 cells per scale height, resolution studies, and long integration times (60000-70000 rg/c). The finding that super-Eddington flows radiate at efficiencies below about 0.5% would have immediate consequences for interpreting ULX luminosities, X-ray binary spectral states, and high-redshift AGN candidates. The paper also credibly shows that initial net vertical magnetic flux, rather than accretion rate alone, controls the vertical structure of near-Eddington disks. The main caveat is that the headline efficiency result is in direct, unresolved tension with a comparable published calculation (Fragile et al. 2025), and the paper itself concedes that the origin of the difference is uncertain.","major_comments":[{"comment":"The central quantitative claim—that radiative efficiency drops below 0.5% at Mdot ~ 150 Mdot_Edd—is in direct conflict with Fragile et al. (2025), which the paper cites but does not reconcile. The paper's stated response, that its models satisfy the condition that the trapping radius is small compared with the disk size, does not discriminate the two calculations, because Fragile et al. satisfy that condition by construction. The manuscript concedes that 'the origin of the quite different outcomes between these calculations is uncertain and warrants further investigation.' This concession is load-bearing: if Fragile et al.'s Eddington cap and 30-70% efficiencies are correct, then the super-Eddington states reported here, their low efficiencies, and the ULX and LRD applications built on them (§4.5, §4.7) would not be realized even under the large external mass supply whose absence the paper acknowledges in §5. The revision should either identify a concrete cause of the difference (initial-condition class, opacity treatment, radiative-transfer closure, measurement radius, or boundary conditions) or explicitly reframe the results as valid only for this specific initial-condition class.","section":"§3.6, Table 1"},{"comment":"The super-Eddington accretion rates are effectively an input rather than an outcome of a realistic feeding process: the initial torus is confined to 15-58 rg and the peak density rho0 is set arbitrarily, which directly sets the late-time Mdot. The paper correctly states in §5 that it has not modeled how plasma is fed from beyond the Bondi radius and whether radiation and mechanical feedback from outflows can limit this feeding. This limitation is not merely a scope note: if outflow feedback throttles the supply, the low-efficiency super-Eddington states would not be realized in nature even if they are robust within this simulation setup. The authors should either quantify the external mass-supply conditions under which these states are realized (for example, with an estimate of the Bondi-scale inflow rate and a check against the simulated outflow feedback) or restrict the generality claims in the abstract and conclusions accordingly.","section":"§5, §2.2"},{"comment":"The claimed agreement with the slim-disk efficiency trend is not parameter-free. The dashed line in Figure 9 is 'normalized to match our model E9-a3', so the comparison has one free normalization and is evaluated with only ten simulation points spread over a wide range in Mdot. As presented, the agreement cannot be distinguished from the statement that the curve can be shifted vertically to pass near one point. The authors should show the unnormalized slim-disk prediction, or provide a goodness-of-fit metric that accounts for the fitted normalization, before using this agreement as independent support for the low-efficiency trend.","section":"§4.1, Figure 9"}],"minor_comments":[{"comment":"The text repeatedly refers to 'SS 443' (e.g., 'SS 443 (Fabrika 2004)'); this should be SS 433 throughout.","section":"§4.6"},{"comment":"The naming convention described in the text is clear, but Table 1 would benefit from a column explicitly listing the initial magnetic-field topology rather than encoding it only in the name and in the 'DL' suffix, since the topology is a central result driver.","section":"§2.4"},{"comment":"The statement that 'models configured with the single-loop magnetic field do not reach the MAD state' is based on ten simulations and on unpublished work cited as 'Wong et al., in preparation'; the claim would be stronger if the paper either included the non-radiative thin-disk comparison or explicitly flagged it as preliminary.","section":"§3.1"},{"comment":"The bullet list contains a grammatical error ('mass accretion rate that vary'); this should be corrected in the final version.","section":"§5"}],"recommendation":"major_revision","confidential_remarks":"This is a technically impressive survey with strong numerical backing. The unresolved conflict with Fragile et al. (2025) is the main obstacle: the paper's own §3.6 concedes the outcome difference is unexplained, but the abstract and conclusions still present the low-efficiency super-Eddington result as a general finding. I would ask the authors to either supply a discriminating diagnostic or substantially narrow the generality of the claims. The initial-torus feeding limitation is real but, in my view, addressable in revision; the Figure 9 normalization issue is smaller but should be fixed. I do not see grounds for rejection, because the simulations themselves appear internally consistent and well resolved."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing is the survey itself: ten radiation GRMHD runs using full angular-resolved transport in full GR, spanning four Eddington ratios, two spins, and two initial field geometries, on top of the White et al. 2023 method. The headline results—super-Eddington disks radiating below half a percent efficiency, no MAD without added vertical flux, and the structural split between net-flux and no-net-flux disks—are clearly presented and internally consistent with the diagnostics. The resolution and MRI quality factors are reported carefully (Q_z ≥ 15, Q_phi ≥ 60, 20 cells per scale height), and the convergence tests are summarized rather than deferred. The paper does not oversell its own scope; the §5 acknowledgment that Bondi-radius feeding and feedback throttling are unmodeled is honest and directly relevant.\n\nSoft spots, in proportion. First, the Fragile et al. 2025 disagreement is real and not resolved. The paper cites it in §3.6 and says “the origin of the quite different outcomes between these calculations is uncertain and warrants further investigation.” That is the right sentence, but it means the main quantitative claim—efficiency below 0.5% at 150 Mdot_Edd—is currently only established for this particular torus initial-condition class. The stress-test note asks whether the trapping-radius condition discriminates; it does not, and the paper does not claim it does. This is an unresolved discrepancy between comparable calculations, not a fatal flaw, but it should be flagged in any review. Second, Figure 9’s “slim disk trend” is normalized to match model E9-a3; that is a consistency check, not an independent prediction, and the text does not quite say so. Minor, but worth a caveat. Third, there are no error bars on the Table 1 quantities, and no data or code release; for a survey meant to anchor follow-up papers, the missing reproducibility artifacts are a real weakness. Fourth, the SS433 and LRD sections are speculative—but clearly marked as such.\n\nOverall: a serious computational paper, honestly written, with a genuine conflict in the literature that needs referee attention. Who benefits: anyone working on radiation-dominated accretion, ULXs, or thin-disk MAD formation. I would bring it to reading group and cite it, while noting the Fragile tension. Recommendation: definitely send to a serious referee; the Fragile comparison and the missing error bars should be central in the review.","headline":"A credible first full-transport radiation GRMHD survey with a real, unresolved conflict against Fragile et al. 2025 on super-Eddington efficiency—worth refereeing seriously.","tokens_in":33491,"tokens_out":1859,"would_cite":true,"duration_ms":19189,"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":"Full-transport radiation GRMHD survey shows super-Eddington accretion is radiatively inefficient, with efficiency falling from ~5% at 0.1 Eddington to <0.5% at 150 Eddington, and that near-Eddington structure hinges on net vertical…","keywords":["radiative magnetohydrodynamics","general relativity","black hole physics","accretion","super-Eddington accretion","ultraluminous X-ray sources","X-ray binary stars","magnetically arrested disks"],"falsifier":"Measure the intrinsic (de-beamed) radiative efficiency of a super-Eddington accretor with a known black hole mass and accretion rate, such as a ULX with optical/radio constraints; finding an efficiency much above the predicted few percent at these rates would contradict the paper, as would a thin-disk radiation GRMHD run with a single-loop field and no injected vertical flux that reaches horizon flux $\\varphi_3 \\approx 15$.","tokens_in":32385,"feed_emoji":"🕳️","tokens_out":12541,"duration_ms":118846,"temperature":0.7,"pith_summary":"This paper presents a survey of ten radiation-dominated accretion flows around a ten-solar-mass black hole, covering accretion rates from roughly one-tenth to 150 times Eddington, with two spin values and two initial magnetic field geometries. It sets out to establish how the structure and radiative output of black hole accretion change as the feeding rate is pushed far above Eddington, using a direct solution of the angle-dependent radiation transport equation in full general relativity. The headline result is that radiative efficiency collapses as the Eddington ratio rises, from about 5% at one-tenth Eddington to less than 0.5% at 150 Eddington, because super-Eddington flows become geometrically thick, radiation-pressure-supported disks whose narrow funnel-shaped photosphere traps most of the photons. Near and below Eddington, the survey finds two distinct structures set by net vertical magnetic flux: a thin dense layer with a magnetically dominated corona when flux is present, or a magnetically elevated disk when it is not. If the paper is right, super-Eddington sources should not be intrinsically super-Eddington in bolometric luminosity, which directly affects interpretations of ultraluminous X-ray sources, the soft state of X-ray binaries, and the X-ray-faint, red-compact sources recently found by JWST.","feed_headline":"Super-Eddington disks radiate at under 0.5 percent efficiency","feed_subtitle":"Photon trapping and outflows, not the Eddington limit, set how bright super-Eddington sources look.","key_machinery":"The load-bearing object is a full-transport radiation GRMHD algorithm: it evolves the angle-dependent, frequency-integrated specific intensity $\\hat{I}$ through the time-dependent transfer equation in curved spacetime, using a fixed geodesic angular grid of 42 directions per cell, and couples radiation to the fluid through emission, absorption, and a Compton energy-exchange term. This replaces the approximate closures (flux-limited diffusion or the M1 two-moment method) that earlier radiation-GRMHD surveys relied on. The second piece of machinery is the initial magnetic configuration: a single-loop field supplies net vertical poloidal flux at the midplane, while a double-loop field has none, and the survey shows this distinction decides whether a near-Eddington disk becomes a thin gas-pressure-supported layer with a magnetically dominated corona or a magnetically elevated disk. The efficiency diagnostic $\\eta_{\\rm rad} = L/(\\dot{M}c^2)$, evaluated at $r = 20r_g$ against an accretion rate measured at $r = 3r_g$, carries the headline claim.","core_discovery":"The survey's central discovery is that the radiative efficiency of black hole accretion is a strong, monotonically decreasing function of the Eddington ratio: $\\eta_{\\rm rad}$ drops from roughly $5\\%$ at $\\dot{M} \\simeq 0.1\\,\\dot{M}_{\\rm Edd}$ to below $0.5\\%$ at $\\dot{M} \\simeq 150\\,\\dot{M}_{\\rm Edd}$. Super-Eddington flows organize into a geometrically thick, radiation-pressure-supported disk ($H/r \\sim 0.25$) in which photon trapping is set by turbulent advection rather than radiative diffusion, and a narrow funnel-shaped photosphere restricts the escaping radiation, while optically thick equatorial outflows carry an energy comparable to the radiation. Near and below the Eddington rate, the flow structure bifurcates according to net vertical magnetic flux: with flux, a thin ($H/r \\sim 0.02$) dense gas layer sits beneath a magnetically dominated corona, and without flux the flow is magnetically dominated everywhere. None of the ten models reaches the magnetically arrested state (horizon flux $\\varphi_3 \\leq 3.5$, against the MAD threshold $\\varphi_3 \\approx 15$), yet the single-loop models with a rapidly spinning black hole still launch relativistic jets. The paper attributes these results to the direct treatment of radiation transport, free of the closure assumptions used in previous approximate methods, and notes that they align closely with the earlier non-relativistic full-transport models.","pith_inferences":["A corollary the paper leaves implicit: if the efficiency drop is generic, population and feedback models that take a fixed ~10% radiative efficiency will systematically overestimate the radiative output of super-Eddington phases of black hole growth, while mechanical feedback from winds and jets will dominate.","The magnetic-topology dichotomy suggests a possible observational test through radio-X-ray correlation: sources on the net-flux branch should be radio-louder at fixed X-ray luminosity than zero-net-flux sources, which the current paper does not quantify.","A natural extension would be to post-process these snapshots with frequency-dependent Monte Carlo transport to predict spectra and color corrections, which would directly test the spectral-hardening prediction at high Eddington ratios.","The no-MAD conclusion may be resolution- and domain-size dependent; extending the same full-transport algorithm to larger domains with a cyclic-zoom approach could reveal whether flux recycling from the outer disk changes the conclusion."],"forward_implications":["Super-Eddington accretion should not produce intrinsically super-Eddington bolometric luminosities; apparent high luminosities in ULXs require favorable beaming or a lower true accretion rate.","Observations of X-ray binaries in the soft state can be mapped onto the two magnetic-topology branches: radio-quiet, higher-variability sources correspond to zero-net-flux disks, while net-flux disks produce jets and lower variability.","The absence of MAD states in these thin-disk runs implies that magnetically arrested accretion in real thin disks demands a continuing external supply of vertical flux; without it, jets can still form but through a non-MAD mechanism.","Super-Eddington accretion should be accompanied by very low luminosity variability (≲10%) and by significant spectral hardening wherever the photosphere sits close to a magnetically active funnel.","The geometric and outflow properties of the super-Eddington models offer a concrete mechanism for the X-ray underluminosity and the ~3600 Å continuum break reported in little red dots."],"supporting_citations":[{"why":"Supplies the generally relativistic, angle-dependent radiation transport algorithm that all survey models use.","marker":"(White et al. 2023)"},{"why":"Earlier non-relativistic full-transport models of super-Eddington accretion onto stellar-mass black holes whose efficiency and turbulent-cooling trends this survey extends and confirms.","marker":"(Jiang et al. 2014b)"},{"why":"Reports near-100% radiative efficiencies in magnetically arrested super-Eddington flows computed with the M1 closure, the main quantitative contrast for the low efficiencies measured here.","marker":"(McKinney et al. 2017)"},{"why":"Reports high near-Eddington radiative efficiencies from other radiation GRMHD approaches against which the present few-percent efficiencies are compared.","marker":"(Fragile et al. 2023)"},{"why":"Establishes the normalized magnetic flux value φ≈15 that defines the magnetically arrested state used to classify the surveyed models as non-MAD.","marker":"(Tchekhovskoy et al. 2011)"},{"why":"Provides the beaming fitting function against which the angular distribution of the simulated energy outflow is compared.","marker":"(S ˛ adowski & Narayan 2015b)"}],"fun_headline_variants":["Super-Eddington disk efficiency plummets to under 0.5%","Photon trapping and outflows slash black hole radiative yield","Magnetic flux decides: thin disk plus corona or magnetically dominated flow","Jets launch from rapid spin without need for MAD state","Funnel photosphere and equatorial winds dim super-Eddington sources"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a carefully chosen, self-contained gas torus with a hand-picked density represents how gas actually reaches a stellar-mass black hole from outside, so that the disk's own radiation and outflows do not limit the supply; if real feeding is throttled at large radii, the very fast, very dim states found here would not form.","fun_headline_variants_meta":{"raw":{"variants":["Super-Eddington disk efficiency plummets to under 0.5%","Photon trapping and outflows slash black hole radiative yield","Magnetic flux decides: thin disk plus corona or magnetically dominated flow","Jets launch from rapid spin without need for MAD state","Funnel photosphere and equatorial winds dim super-Eddington sources"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000883,"raw_usage":{"total_tokens":3907,"prompt_tokens":1127,"completion_tokens":2780,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":743,"completion_tokens_details":{"reasoning_tokens":2689}},"tokens_in":743,"tokens_out":2780,"duration_ms":22728,"temperature":1.0,"reasoning_tokens":2689,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:27:01.689711+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the intrinsic (de-beamed) radiative efficiency of a super-Eddington accretor with a known black hole mass and accretion rate, such as a ULX with optical/radio constraints; finding an efficiency much above the predicted few percent at these rates would contradict the paper, as would a thin-disk radiation GRMHD run with a single-loop field and no injected vertical flux that reaches horizon flux $\\varphi_3 \\approx 15$.","supporting_citations":[],"review_version":1}