{"id":"236c120f-f808-4778-946a-dd252eae48b0","arxiv_id":"2505.08855","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Using a texture-memory cooling table in GRMHD, the simulations show a hot two-temperature flow at low accretion rates and a truncated thin disk above about 1% Eddington.","lead":"A new GPU cooling table lets black hole accretion simulations run several times faster, revealing how a hot, puffy gas flow collapses into a thin disk once the infall rate becomes high. The collapse happens near one percent of the Eddington rate, but a numerical floor and short runs keep the exact threshold uncertain.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The h_floor=0.1 aspect-ratio floor pins the outer thin disk, so r_in≈50 rg and mdot_crit≈0.01 are numerical-control-limited; the transition claim needs a floor-convergence test.","rationale":"The reader's weakest_assumption and my independent reading converge on the same load-bearing issue: the h_floor=0.1 aspect-ratio floor. The paper itself provides direct evidence that the outer collapsed disk is floor-limited (Fig. 11(b) shows H/R≈0.10 at r=60 rg for both high-rate runs; Section IV.A says the disk would collapse further without the floor), and Appendix B shows the MRI is not adequately resolved in the thin disk (Q≈5 versus the canonical 10–20). These features undermine the quantitative specificity of the headline astrophysical result: the truncation radius r_in≈50 rg, the one-temperature nature of the outer disk, and the critical accretion rate mdot_crit≈0.01 are all set in a regime where a numerical control parameter, not the physics, is active. The critical rate is also inferred from only two widely separated rates, ~0.2 and ~5×10^-4, so the claimed 0.01 threshold is a broad bracket rather than a measured transition. I do not see this as fatal: the qualitative collapse of a MAD flow at high mdot is consistent with previous work, and the authors explicitly list the floor as a caveat. The appropriate response is therefore to retain the CONDITIONAL verdict, since the claim can only be trusted after a floor-convergence study. My concrete test—lowering h_floor and doubling resolution—directly determines whether r_in and the floor-pinned H/R profile survive, which is the single check that would settle the concern.","tokens_in":23889,"tokens_out":3552,"duration_ms":38850,"concrete_test":"Re-run EXP_HIGH and IMP_HIGH with h_floor=0.03 and doubled resolution (e.g., 768×600×128) for at least 20,000 rg/c after cooling is enabled. If H/R at r=60 rg again sits at the new floor and the inferred truncation radius (the location of the steep H/R transition) shifts by more than ~20% from 50 rg, then the quoted truncation radius and critical accretion rate are floor-limited and the central claim requires substantial qualification.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that the outer, r≳50 rg flow is a physically cold thin disk whose truncation radius and one-temperature status emerge from radiative cooling. The simulations impose a temperature floor via Eq. (19) with h_floor=(H/R)_floor=0.1, and Section IV.A states that without this floor the r>50 rg regions \"would collapse even further.\" Fig. 11(b) confirms both high-accretion-rate runs cool to H/R≈0.10 at r=60 rg, i.e., the outer disk is pinned to the floor, not to a converged thermal equilibrium. Because the floor sets the minimum temperature, it sets the density and Coulomb-coupling rate in the outer disk, and hence the very \"one-temperature thin disk\" and r_in≈50 rg signatures. Appendix B shows MRI quality factors Q≈5 in the thin part of EXP_HIGH/IMP_HIGH, below the canonical 10–20 threshold, so the collapsed disk is underresolved. The critical rate mdot_crit≈0.01 is also only a bracket between runs at ~0.2 and ~5×10^-4, with no run in between. The qualitative collapse is plausible and the caveats are acknowledged, but the quantitative central claims (r_in≈50 rg, mdot_crit≈0.01) are numerical-control-limited.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a GPU-accelerated cooling toolkit for two-temperature GRMHD simulations, using precomputed lookup tables stored in texture memory to speed up radiative cooling (bremsstrahlung, synchrotron, inverse Compton, Coulomb coupling). The toolkit is implemented in H-AMR and applied to MAD simulations around a Kerr black hole with a*=0.9375, spanning Eddington ratios from ~3e-6 to ~0.26. The central scientific claim is that a critical accretion rate near mdot ~ 0.01 separates two regimes: below it, the flow is a geometrically thick, two-temperature hot flow; above it, an outer single-temperature thin disk forms, truncated at rin ~ 50 rg, with an inner two-temperature hot flow and thin filaments. The paper also reports substantial speedups of the cooling prescription relative to M1 radiation transport and global-memory lookup tables.","tokens_in":24109,"tokens_out":1990,"duration_ms":20455,"significance":"If the central claim holds, this is a significant step: it would demonstrate that spectral-state-like transitions can emerge from first-principles GRMHD simulations with local radiative cooling, rather than being imposed by a model. The technical contribution is solid and reproducible: the texture-memory benchmark speedups (Sec. IV, Appendix A), the comparison between implicit and explicit cooling, and the explicit statement of caveats are all strengths. The claim is not circular: the cooling table is built from standard radiative formulas and no parameter is tuned to produce a particular state. However, the quantitative predictions (mdot_crit ~ 0.01, rin ~ 50 rg) are controlled by numerical choices, particularly the aspect-ratio floor h_floor = 0.1, and the simulations are not in inflow equilibrium at the relevant radii. The qualitative collapse is plausible and consistent with prior work, but the quantitative claims are not yet converged.","major_comments":[{"comment":"The imposed temperature floor, Eq. (19) with h_floor=(H/R)_floor=0.1, directly sets the minimum thickness of the outer thin disk. Section IV.A states that without this floor, the r>50 rg regions 'would collapse even further,' and Fig. 11(b) shows both high-rate runs saturating at H/R ~ 0.10 at r=60 rg. Because the floor pins the outer disk temperature, it also controls the density and Coulomb-coupling rate there, and hence the one-temperature plasma condition and the derived truncation radius rin ~ 50 rg. The paper does not provide a floor-convergence test (e.g., varying h_floor with matched resolution and runtime), so the quantitative central claim is numerical-control-limited rather than a converged physical prediction. This point is load-bearing for conclusions (iii) and (iv).","section":"Section III.B.2 and Section IV.A"},{"comment":"The simulations reach inflow equilibrium out to at most ~80 rg (Sec. IV.A, Fig. 2), while the viscous time of the thin disk at the relevant radii is ~10^7 rg/c, as the authors state in the Introduction and in the Conclusions. The truncation radius rin ~ 50 rg and the outer-disk properties are therefore computed from an evolving, non-equilibrium state. The time-averaged profiles in Figs. 4, 5, and 8 may not reflect the steady-state structure. A demonstration that the truncation radius and the outer one-temperature region are stable over a longer timescale, or a quantitative estimate of the expected drift, is needed to support the quantitative claims.","section":"Section IV.A and Conclusions, caveat 2"},{"comment":"The MRI quality factors in the thin part of EXP_HIGH and IMP_HIGH approach Q_r,theta ~ 5, well below the canonical 10–20 range cited by the authors. This means the collapsed thin disk is underresolved, so the turbulent angular-momentum transport and dissipation in the region where the one-temperature condition develops are not reliably captured. The paper presents this as a resolution 'floor' but does not quantify the systematic uncertainty this introduces into rin, mdot_crit, or the temperature profile. At minimum, a resolution study of one high-rate run (e.g., comparing 384x300x64 to a higher resolution) would establish whether the collapse and truncation are robust.","section":"Appendix B and Fig. 17"},{"comment":"The critical accretion rate mdot_crit ~ 0.01 is inferred from a bracket between EXP_HIGH/IMP_HIGH at ~0.17–0.26 and EXP_MID at ~5e-4, with no simulation between. The paper appropriately calls it 'roughly estimated,' but the abstract and conclusions present it without the same caveat ('for accretion rates above ~0.01'). Since the transition could be sharp or gradual, a run at ~0.01–0.05 would materially change the quoted critical value. The claim as stated is not falsified by the current set, but it is weaker than 'revealed.'","section":"Section V and Fig. 12; Conclusions item (iii)"}],"minor_comments":[{"comment":"The optically thick expression in Eq. (11) is attributed to Hubeny 1990; please specify the precise form and the range of validity of the approximation, particularly how the factor 1/(3τ/2 + sqrt(3) + 1/τ_abs) behaves when τ_sca >> τ_abs, since the simulations reach τ_sca ~ 8.","section":"Section II.D and Eq. (11)"},{"comment":"The quality-factor definition in Eq. (B1) uses the local Alfvén speed and rotation frequency; please state explicitly how these are averaged when computing the radial profiles in Fig. 17, since the figure reports 'b^2ρ-weighted' values but the text does not define the weighting.","section":"Section III.C and Fig. 17"},{"comment":"The radiative efficiency in Eq. (22) is computed from the last snapshot only, as noted in the text. Because the cooling output was only saved for the final snapshot, the efficiency values in Table I and Fig. 3 should be labeled as snapshot-based rather than time-averaged; the text does this for the table but Fig. 3 could be misinterpreted.","section":"Section IV.B and Eq. (22)"},{"comment":"The caption for Fig. 6 says the panels are ordered 'from right to left' in accretion rate; the text also uses 'left to right' later. Please make the ordering consistent and unambiguous, and define which panel corresponds to which simulation.","section":"Figure 6 caption and text"},{"comment":"The two panels in Fig. 11 are both labeled '(a)' and '(b)' redundantly below the figure; the text refers to 'Fig. 11(a,b)' but the internal labels should be cleaned up.","section":"Section IV.C and Fig. 11"}],"recommendation":"major_revision","confidential_remarks":"The paper is a strong technical contribution (the cooling-table speedups and the two-temperature implementation are useful and reproducible), and the qualitative emergent collapse is interesting. However, the central quantitative claims are currently pinned by a numerical floor and lack inflow equilibrium. The authors should either add the floor-convergence test and a longer-timescale run, or significantly soften the quantitative claims in the abstract and conclusions. I lean toward major revision rather than rejection because the core physics is plausible and the method is sound, so the fix is within scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth a serious referee, but the headline numbers are softer than the abstract implies. What is genuinely new: a GPU texture-memory lookup table for two-temperature cooling (brems, synchrotron, local Compton, Coulomb) implemented in H-AMR, with clean, reproducible benchmarks. In-code it runs ~2.3x faster than evaluating the cooling analytically, ~3-5x faster than H-AMR's M1 module, and ~5x faster than the same table in global memory. That is a practical contribution. The authors also run five MAD GRMHD simulations spanning mdot ~ 3e-6 to 0.3, which is a wider Eddington-ratio range than any previous single suite, and show a qualitative split: low rates give a thick two-temperature flow; high rates give a cooler thin outer disk with a hot inner region. This is an emergent result, not a fit, and it is consistent with the truncated-disk picture.\n\nThe soft spots are where the stress-test puts them, and they are load-bearing. The imposed aspect-ratio floor h_floor=0.1 sets the thickness of the outer, collapsed disk; the paper says that without the floor the r>50 rg region would collapse further. So r_in ~ 50 rg and the one-temperature outer disk are pinned by a numerical control parameter, not by a converged thermal equilibrium. The MRI quality factor in the thin part of the high-rate runs is Q~5, below the canonical 10-20 threshold, so the collapsed disk is underresolved. The runs only reach ~8e4 rg/c against a viscous time ~1e7 rg/c, so there is no inflow equilibrium. And mdot_crit ~ 0.01 is a bracket between runs at 0.2 and 5e-4, with no intermediate run. The temperature in the outer disk has to be corrected by two orders of magnitude in post-processing because radiation pressure is absent from the simulation. Those are acknowledged caveats, and the qualitative collapse is probably real; the exact transition rate and truncation radius are not.\n\nWhat the paper does well: the methods section is clear, the benchmarks are measured, and the caveats are stated rather than buried. The cooling table construction is standard physics applied carefully, and the implicit/explicit comparison is a useful check.\n\nWho should read it: anyone running radiative GRMHD for stellar-mass black holes, and XRB theorists who want to know where the 'truncated disk + hot inner flow' picture stands in simulation. I would send this to peer review. The right referee report asks for a floor-convergence test (h_floor = 0.05 or lower, or runs showing insensitivity), longer durations at least to ~1e5 rg/c to get closer to inflow equilibrium, and a run at mdot between the bracketing values. Without those, the spectral-state claim should be reported as qualitative.","headline":"A genuinely useful GPU cooling toolkit with an honest but floor-limited first GRMHD survey; the transition numbers need a floor-convergence test before they are quoted.","tokens_in":24764,"tokens_out":3009,"would_cite":true,"duration_ms":30541,"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":"This paper claims that magnetically arrested black hole accretion undergoes a sharp state transition near 1% of the Eddington rate, collapsing from a hot two-temperature inflow into a truncated thin disk.","keywords":["accretion disk","black hole X-ray binaries","GRMHD simulations","spectral state transition","magnetically arrested disk","radiative cooling","two-temperature plasma","GPU texture memory"],"falsifier":"Run the two high-accretion MAD cases again with $h_{\\rm floor}=0.05$ and $0.02$, raise the resolution so the MRI quality factors stay above 10, and extend the runs beyond roughly $10^7\\,r_g/c$ to reach inflow equilibrium; if the outer disk still truncates near $50\\,r_g$ and the collapse boundary stays near $0.01\\,\\dot{M}_{\\rm Edd}$, the claims hold, while a significantly different or missing truncation would show the transition is an artifact of the floor and the short runtime.","tokens_in":23633,"feed_emoji":"🕳️","tokens_out":10398,"duration_ms":100036,"temperature":0.7,"pith_summary":"The paper argues that the observed hard-to-soft spectral state transition of black hole X-ray binaries can emerge from general-relativistic magnetohydrodynamic (GRMHD) simulations once local radiative cooling is included. In a magnetically arrested disk (MAD) around a spinning stellar-mass black hole, accretion rates below about $\\dot{M}\\sim 0.01\\,\\dot{M}_{\\rm Edd}$ settle into a geometrically thick, two-temperature hot flow, while rates above that critical value collapse the outer flow into a cold, single-temperature thin disk truncated near $r_{\\rm in}\\approx 50\\,r_g$. The inner region stays a hot two-temperature flow, giving the truncated-disk-plus-corona geometry long invoked for the hard state. The paper's practical contribution is a GPU texture-memory cooling table, including bremsstrahlung, synchrotron, inverse Compton, and Coulomb processes, that runs 3--5 times faster than standard radiation-M1 closures and about 5 times faster than the same table in global memory.","feed_headline":"Disk collapse at 1% Eddington seen in black hole simulations","feed_subtitle":"Hot, thick accretion flows flatten into a thin disk truncated at 50 gravitational radii above the critical rate.","key_machinery":"The load-bearing mechanism is a precomputed cooling function $Q(H_{\\rm th}, B, n_e, T_e)$ that returns the total radiative loss per unit volume from bremsstrahlung, synchrotron, and inverse-Compton-enhanced synchrotron emission, with an optically thick bridge to blackbody cooling, together with a Coulomb-collision rate table $Q_{cc}(n_e,T_i,T_e)$. Both tables are stored in GPU texture memory and called inside each cell of a two-temperature GRMHD solver, with the cooling added as an external four-force and evolved either explicitly, capped at a 30% internal-energy update, or with an implicit solver. A numerical temperature floor enforcing $(H/R)_{\\rm floor}=0.1$ (Eq. 19) sets the minimum thickness of the collapsed thin disk and is the main control on the quoted truncation radius.","core_discovery":"The paper's central discovery, stated on its own terms, is a critical accretion-rate boundary near $\\dot{m}_{\\rm crit}\\sim 0.01$, where $\\dot{m}=\\dot{M}/\\dot{M}_{\\rm Edd}$, that separates two accretion regimes in MADs. Below the boundary the flow is a radiatively inefficient, geometrically thick ($H/R\\approx 0.2$--$0.35$), two-temperature plasma with near-virial temperatures and $T_e\\ll T_i$. Above the boundary, radiative cooling wins beyond $r\\approx 50\\,r_g$: the outer disk collapses to a thin, optically thick, single-temperature disk with $T_i\\approx T_e\\sim 10^9$ K in the raw simulations, while the region inside $50\\,r_g$ stays puffed up and two-temperature; adding radiation pressure in post-processing lowers the outer temperature to about $10^7$ K. The paper identifies these density and temperature maps with the truncated-disk plus inner-corona configuration invoked for the low/hard state.","pith_inferences":["A decisive follow-up that the paper does not carry out is lowering $h_{\\rm floor}$ below 0.1 at higher resolution; if the truncation radius and critical rate survive that test, they are physical, and if they move, the published values are floor-limited estimates.","The texture-memory table approach is portable: the same trick should accelerate other expensive microphysical cooling integrals, such as neutrino cooling in neutron-star merger simulations, wherever the tabulated parameter space is smooth enough for interpolation.","If the $\\sim 0.01\\,\\dot{M}_{\\rm Edd}$ boundary holds, it makes a testable observational prediction: an X-ray binary crossing that accretion rate should show a changing inner reflection edge and a hard-to-soft spectral pivot at the radius where the disk collapses."],"forward_implications":["If the critical rate is real, hard-state X-ray binaries below $\\dot{m}_{\\rm crit}\\sim 0.01$ should naturally present as hot two-temperature flows, and the soft state above it as a thin outer disk with an inner hot flow, so the disk geometry does not have to be imposed by hand.","A truncation radius of about $50\\,r_g$ in the high-accretion runs gives a specific inner edge for reflection-spectrum and quasi-periodic-oscillation models of black hole X-ray binaries, with the inner hot region also being the natural jet-launching zone.","The transition from one-temperature to two-temperature plasma at $r_{\\rm in}$ implies that spectral models of the hard and intermediate states should couple electrons and ions in the thin disk but decouple them in the inner corona.","The reported computational gains (3--5 times faster than M1-closure radiation schemes, about 5 times faster than a global-memory lookup table) make three-dimensional surveys across the $10^{-6}$--$0.3\\,\\dot{M}_{\\rm Edd}$ range practical at this resolution."],"supporting_citations":[{"why":"Supplies the hot accretion flow cooling prescription and the truncated-disk/RIAF framework that the simulations are designed to test.","marker":"[8]"},{"why":"Provides the optically thin/thick bridging cooling formula and the synchrotron Compton enhancement approximations used in the lookup table.","marker":"[37]"},{"why":"Establishes the two-temperature GRMHD setup and prior evidence that MAD flows can form magnetically truncated disks.","marker":"[34]"},{"why":"Provides a prior GRRMHD simulation at $\\dot{m}=0.3$ whose flow structure is compared with the high-accretion runs.","marker":"[33]"},{"why":"Reports a prior MAD radiation GRMHD simulation that saw disk collapse without a clear truncated disk, framing the new result.","marker":"[35]"},{"why":"Gives a prior scaling $r_{\\rm in}=42\\,r_g(\\dot{m}/0.1)^{-0.60}$ for a truncated cold disk in a hot corona, against which the new truncation radius is compared.","marker":"[25]"},{"why":"Defines the geometrically thin, optically thick accretion disk model that the collapsed outer region is identified with.","marker":"[10]"}],"fun_headline_variants":["1% Eddington rate triggers black hole disk collapse","Thin disk emerges outside 50 r_g above 1% Eddington","Critical accretion rate flips black hole disk shape","Two-temperature flow collapses to thin disk at 1% Eddington","Disk collapse at 1% Eddington reveals truncated inner region"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the numerical floor pinning the disk's thickness at 10% of its radius (the temperature floor of Eq. 19) does not decide where the thin disk forms; lower the floor and the outer disk would collapse further, so the 50-$r_g$ truncation and the 0.01 critical rate are predictions set partly by that floor.","fun_headline_variants_meta":{"raw":{"variants":["1% Eddington rate triggers black hole disk collapse","Thin disk emerges outside 50 r_g above 1% Eddington","Critical accretion rate flips black hole disk shape","Two-temperature flow collapses to thin disk at 1% Eddington","Disk collapse at 1% Eddington reveals truncated inner region"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001788,"raw_usage":{"total_tokens":7100,"prompt_tokens":1050,"completion_tokens":6050,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":666,"completion_tokens_details":{"reasoning_tokens":5961}},"tokens_in":666,"tokens_out":6050,"duration_ms":43271,"temperature":1.0,"reasoning_tokens":5961,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:46:27.662504+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the two high-accretion MAD cases again with $h_{\\rm floor}=0.05$ and $0.02$, raise the resolution so the MRI quality factors stay above 10, and extend the runs beyond roughly $10^7\\,r_g/c$ to reach inflow equilibrium; if the outer disk still truncates near $50\\,r_g$ and the collapse boundary stays near $0.01\\,\\dot{M}_{\\rm Edd}$, the claims hold, while a significantly different or missing truncation would show the transition is an artifact of the floor and the short runtime.","supporting_citations":[{"cited_title":"Jacquemin-Ide, F","cited_arxiv_id":null,"evidence_quote":"Provides the optically thin/thick bridging cooling formula and the synchrotron Compton enhancement approximations used in the lookup table."},{"cited_title":"Tchekhovskoy and D","cited_arxiv_id":null,"evidence_quote":"Establishes the two-temperature GRMHD setup and prior evidence that MAD flows can form magnetically truncated disks."},{"cited_title":"Tchekhovskoy, R","cited_arxiv_id":null,"evidence_quote":"Provides a prior GRRMHD simulation at $\\dot{m}=0.3$ whose flow structure is compared with the high-accretion runs."},{"cited_title":"Jacquemin-Ide, G","cited_arxiv_id":null,"evidence_quote":"Reports a prior MAD radiation GRMHD simulation that saw disk collapse without a clear truncated disk, framing the new result."},{"cited_title":"Their results showed a truncated cold disk embedded in a hot corona, with the inner radius of the cold disk following rin = 42rg ( ˙m/0.1)−0.60","cited_arxiv_id":null,"evidence_quote":"Gives a prior scaling $r_{\\rm in}=42\\,r_g(\\dot{m}/0.1)^{-0.60}$ for a truncated cold disk in a hot corona, against which the new truncation radius is compared."}],"review_version":1}