{"id":"7a2fc800-dd60-4c39-ba1e-75ba1f15b37b","arxiv_id":"2506.00946","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A new reflection model, relxillA, bins the disk by emission angle rather than by radius, matches ray-tracing spectra far better than existing models, and fixes a coding bug in relxill v2.4.","lead":"Astrophysicists built a new X-ray reflection model, relxillA, that tracks how the emission angle of radiation varies across a black hole's accretion disk more accurately than previous models. The model removes a known systematic error that would otherwise bias measurements of black hole spin and tests of gravity with next-generation X-ray observatories.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The accuracy claim rests on a fixed 10-bin emission-angle grid validated at only one spin and three inclinations, with no convergence test; moreover the paper's binning is internally inconsistent (uniform μ vs uniform θ), so the method as written is ambiguous.","rationale":"The paper's central claim is credible in outline: replacing a disk-averaged or annulus-averaged specific intensity with emission-angle-resolved zones is a natural fix, and the comparison to blackray, a public ray-tracing code, is a sensible accuracy test. The timing overhead is modest (Table 1), the EXO 1846–031 test is honestly reported and shows no improvement for current data, and the simulated X-IFU+LAD fit shows a dramatic improvement in residuals. However, the claim is presented as a general property of the model, while the validation is narrow. The strongest support is Fig. 4: a fit of relxillA to data simulated with blackray. That intrinsically validates the model at that one parameter point, but does not establish the 10-bin discretization as sufficient across spin, inclination, ionization, and emissivity. The paper contains no convergence test, and bin-midpoint quadrature can be inaccurate when the integrand varies strongly within a bin; the paper itself shows >10% deviations for v2.4 at 80°. The internal inconsistency between the text's uniform-μ description and the equations' uniform-θ bins is a separate, concrete issue: if the implemented code follows Eq. (10), the model is not sampling μ uniformly, undercutting the stated design; if it follows the text, the equations in the paper are wrong. Either way a reader cannot reproduce the method from the manuscript alone. The v2.5 result (χ²/dof=2.71 vs 1.45 for v2.4) is a red flag that the r_e correction interacts strongly with the averaging scheme; because relxillA uses the same correction, its good fit should be shown not to be accidental. None of this implies the result is wrong; it means the paper needs either a convergence study/public code inspection or a clear acknowledgment that the model is calibrated only for the tested regime. The verdict remains CONDITIONAL, with the condition being the missing validation and the bin-definition ambiguity.","tokens_in":15003,"tokens_out":14358,"duration_ms":150052,"concrete_test":"Inspect the relxillA implementation (source code or FITS table definitions) to determine whether the zone edges are μ_i=i/10 or θ_i=iπ/20, then run a convergence and parameter-scan test: compare relxillA with 10, 20, 40, and 80 bins (in both μ and θ variants) against blackray on a grid covering α=0, 0.5, 0.9, 0.998; θ_obs=15°, 30°, 50°, 70°, 80°; logξ=0, 1, 2, 3; and q=2, 3, 5. If the 10-bin result deviates from the converged (80-bin) result by more than ~1% anywhere in the grid, or if the μ-binned and θ-binned implementations disagree materially, the central accuracy claim is not established as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that relxillA removes the emission-angle approximation error—rests on a single discretization choice: 10 bins in emission angle with the specific intensity evaluated at the bin midpoint. The direct accuracy validation against blackray covers only α=0.998 at θ_obs=20°, 45°, 80° with one set of reflection parameters (Γ=2, q=3, logξ=3.1, AFe=1); the headline simulated fit uses θ_obs=15°, q=5, logξ=1.0, a point not in that validation. No convergence test in the number of bins is presented, so it is unknown whether 10 bins is adequate or merely the value that was tried. A second, concrete problem is internal inconsistency: the text says the 10 zones are uniform in μ=cosθ_e (0 to 1 in steps of 0.1), but the defining equations (5) and (10) use θ_i=iπ/20, i.e. equal-width bins in θ_e, not μ. These two schemes differ substantially, especially at grazing angles, and the paper does not say which one is implemented. If the code follows the equations, the stated rationale of uniform sampling across the full angular distribution is false; if it follows the text, the equations are wrong. Either way the method is not reproducible as written. Finally, the unexplained degradation of relxill v2.5 (χ²/dof=2.71 vs 1.45 for v2.4) in the simulated fit shows that the r_e bug fix can have large, counterintuitive effects when combined with angle averaging; since relxillA uses the corrected r_e, its good performance should be demonstrated not to depend on a coincidental cancellation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents relxillA, a new version of the relxill reflection model that replaces the disk-averaged or annulus-averaged emission-angle treatment with a discretization of the emission angle itself. Following the exact integral formulation in Eq. (1), the model divides the emission-angle range into ten zones and evaluates the specific intensity at a representative angle within each zone, summing the ten resulting convolution integrals (Eq. 10). The authors report that relxillA reproduces ray-traced spectra from their blackray code much better than relxill v2.4 and v2.5, fits simulated NewAthena/X-IFU+LAD spectra with residuals consistent with noise, and gives results comparable to older models on a NuSTAR observation of EXO 1846-031, while adding no new free parameters. The paper also identifies and corrects an r_e^4 versus r_e error in relxill v2.4, issuing this as relxill v2.5.","tokens_in":15350,"tokens_out":4304,"duration_ms":44918,"significance":"If the accuracy claim holds, relxillA addresses a known systematic limitation of the relxill family that becomes critical for high-throughput next-generation X-ray instruments. The paper is methodologically transparent in writing down the exact Eq. (1) and the approximations in Eqs. (4)-(10), and it explicitly reports the code fix behind v2.5. The comparison against blackray, including the reltrans comparisons in Fig. 3, is a useful benchmark, and the modest runtime increase reported in Table 1 makes the model practical. The central claim is plausible, but it rests on a rather narrow validation set and on a binning prescription that is internally inconsistent between the text and the equations; both points need to be resolved before the model can be used as a community standard. The paper does not ship machine-checked proofs, but the model code is intended for public release in the relxill package, which is a strength if the final version includes the corrected equations.","major_comments":[{"comment":"The binning prescription is internally inconsistent. The text states that the cosine of the emission angle, mu = cos theta_e, is divided into 10 equal-width intervals from 0 to 1 in steps of 0.1, but Eqs. (5), (8), and (10) define theta_i = i pi/20, i.e., equal-width bins in theta_e rather than in mu. These two discretizations are not equivalent: equal-width theta bins give non-uniform mu widths that are largest near grazing emission, while equal-width mu bins concentrate zones near theta_e = 90 degrees. Because the paper's stated rationale is uniform sampling across the full angular distribution, the implemented scheme must be specified unambiguously. The authors should correct either the text or the equations and, ideally, report results for both prescriptions to demonstrate that the claimed accuracy is not an artifact of the chosen binning.","section":"Section 1 and Eqs. (5), (10)"},{"comment":"The accuracy validation of relxillA covers only a single spin value (alpha = 0.998), three observer inclinations (20, 45, 80 degrees), and a single set of reflection parameters (Gamma = 2, q = 3, log xi = 3.1, AFe = 1). The headline spectral fit in Section 4 uses theta_obs = 15 degrees, q = 5, and log xi = 1.0, which lies outside this validated grid. Moreover, no convergence test in the number of emission-angle zones is presented, so it is not established that ten zones is sufficient across the parameter space of spin, emissivity index, ionization, and iron abundance. The authors should add a convergence study (e.g., comparing 5, 10, 20, and 50 zones) and at least validate the model at the parameters used in the simulated fit of Fig. 4, including a lower spin value, since the emission-angle distribution changes strongly with spin.","section":"Section 3, Figs. 2-3"},{"comment":"The unexplained degradation of relxill v2.5 relative to v2.4 in the simulated fit needs a quantitative explanation. The paper states that v2.5 'does not do better than' v2.4, but Fig. 4 shows chi^2/dof = 26279/9714 = 2.71 for v2.5 versus 14110/9714 = 1.45 for v2.4, a dramatic worsening. Since relxillA incorporates the same r_e correction as v2.5, the good fit of relxillA could in principle rely on a cancellation between the corrected radial weighting and the new angle binning. The authors should identify which spectral regions drive the v2.5 residuals, explain why the r_e correction has such a large effect for these particular parameters, and demonstrate that relxillA's performance is robust rather than coincidental.","section":"Section 4, Fig. 4"}],"minor_comments":[{"comment":"The color bar and zone boundaries in Fig. 1 would be easier to interpret if the contour between zones were overlaid on the blackray panels, so the reader can directly see how well the piecewise-constant emission-angle zones match the true map.","section":"Section 3, Fig. 1"},{"comment":"The column header 'N ENER CONV' is unclear; it should be written as 'N_ENER_CONV' with a definition in the caption, and the table would benefit from a row for relxilllpA at N_ENER_CONV = 32768 if available.","section":"Table 1"},{"comment":"The notation theta_bar_e is defined only parenthetically; the authors should state explicitly whether the representative angle is the midpoint in theta_e or in mu, and should include a footnote that the choice is tested as part of the convergence study.","section":"Section 3, Eqs. (5)-(10)"},{"comment":"The sentence 'we can use theta_bar_e = (theta_i + theta_{i+1})/2, but its exact value has no significant impact on the final result' is a strong claim that should be supported by a numerical test, especially because the choice of representative angle interacts with the bin-width inconsistency noted above.","section":"Section 3, text after Eq. (10)"},{"comment":"The y-axis label '2/' in the residual panels should read 'chi^2' or 'Delta chi', and the three panels should use a common y-axis scale to make the residual amplitudes directly comparable.","section":"Section 4, Fig. 4"}],"recommendation":"major_revision","confidential_remarks":"The accuracy benchmark blackray is the authors' own code, and the simulated data are from the authors' prior work. This is not by itself disqualifying, but the paper would be materially stronger with an independent cross-check of at least one spectrum, or with the release of the exact input scripts used for the blackray comparisons. The internal inconsistency between the text and the equations describing the binning is the kind of issue that will be caught immediately by any reader trying to reproduce the method, so it must be fixed before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper presents relxillA, a new reflection model in the relxill family that bins the disk by emission angle instead of averaging over annuli, and it fixes an r_e^4 bug in relxill v2.4. The binning idea is not new—reltrans already does it—but the implementation here is clean, introduces no new parameters, and the accuracy comparisons against the authors' ray-tracing code blackray show relxillA tracks the exact calculation much better than standard relxill at high inclination. The bug fix is clearly identified, and the model is only moderately slower than the existing ones.\n\nThe paper does several things well. The exact integral in Eq. (1) and the binning approximation are laid out clearly. The comparison at theta_obs = 20, 45, 80 degrees shows relxillA within a few tenths of a percent of blackray, while relxill v2.4 has up to 10% deviations at 80 degrees. The simulated NewAthena/X-IFU+LAD spectrum fits with no residuals in relxillA, while v2.4 and v2.5 show large residuals. The EXO 1846-031 application honestly shows no improvement for current data, which is a good sign that the model is not overfitting.\n\nThree soft spots. First, the bin definition is internally inconsistent. The text says the 10 zones are uniform in mu = cos theta_e from 0 to 1 in steps of 0.1, but Eqs. (5) and (10) define the bin edges as theta_i = i pi/20, i.e. uniform in theta. These differ significantly at grazing angles, and the paper does not say which is implemented. That must be fixed for the method to be reproducible. Second, the accuracy validation is thin: one spin (a=0.998), three inclinations, one set of reflection parameters, and no convergence test in the number of bins. The headline simulated fit uses a different parameter set (theta=15, q=5, logxi=1.0) that is not in the validation grid, so the accuracy claim there is extrapolated. Third, the unexplained degradation of relxill v2.5 in the simulated fit (chi2/dof=2.71 vs 1.45 for v2.4) is puzzling. If the r_e bug fix makes the angle-averaged model worse for that parameter set, the interactions between the correction and the averaging approximation need to be understood. The authors should demonstrate that relxillA's good performance is not a lucky cancellation.\n\nWho is this for? X-ray spectroscopists who use relxill and are preparing for next-generation missions. The model is a genuine improvement for high-inclination sources. With the bin definition clarified and a convergence test included, this would be a solid contribution. As it stands, the internal inconsistency and the thin validation prevent full acceptance.\n\nSend it to peer review—it deserves referee time—but the referee should insist on the bin definition being pinned down and add a convergence test or a broader validation scan.","headline":"A useful new relxill variant that bins by emission angle and fixes a real bug, but the bin definition is ambiguous and the validation is thin.","tokens_in":15976,"tokens_out":5171,"would_cite":true,"duration_ms":40113,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper presents relxillA, an extension of the relxill reflection model that replaces radial annuli with emission-angle zones, and shows it eliminates the residuals that plague older models on simulated next-generation X-ray spectra.","keywords":["X-ray reflection spectroscopy","accretion disk","emission angle","relxillA","ray tracing","black hole spin","NewAthena/X-IFU","X-ray binaries"],"falsifier":"Fit the same simulated NewAthena/X-IFU+LAD spectra with relxillA using 20 or 50 bins in $\\mu$ instead of 10; if the fit improves substantially or parameter estimates shift beyond the 90% confidence intervals, the 10-bin discretization is not converged and the claimed accuracy does not hold. A complementary check would compare relxillA against an independent ray-tracing or Monte Carlo transfer code at spins such as $\\alpha = 0, 0.5, 0.9$ and inclinations near $60^\\circ-85^\\circ$.","tokens_in":14751,"feed_emoji":"🕳️","tokens_out":4814,"duration_ms":45433,"temperature":0.7,"pith_summary":"The paper claims that the dominant inaccuracy in current relativistic reflection models comes from averaging the non-isotropic reflected intensity over emission angles, and that this can be fixed with a modest computational cost. It presents relxillA, which divides the disk into zones of equal-width bins in $\\mu = \\cos\\theta_e$ rather than radial annuli, so each zone uses a reflection spectrum computed at the appropriate emission angle. Against simulated NewAthena/X-IFU+LAD spectra of bright black hole X-ray binaries, relxillA yields $\\chi^2/\\mathrm{dof} \\approx 1.03$ while relxill v2.4 and v2.5 give about 1.45 and 2.71, and its spectra agree with accurate ray-tracing to well within 1%. The authors also fit a NuSTAR spectrum of EXO 1846-031 and find no significant differences among models, suggesting current data cannot distinguish the improvement even though future high-quality data will require it.","feed_headline":"Emission-angle fix makes reflection model fit next-gen spectra","feed_subtitle":"Dividing the disk by emission angle instead of radius removes the residuals that plague relxill on future X-ray data.","key_machinery":"The central object is the relxillA model, which divides the accretion disk into up to 10 zones defined by intervals of $\\mu = \\cos\\theta_e$ of width 0.1, covering $\\mu$ from 0 to 1. Each zone uses the non-relativistic reflection spectrum computed at a representative emission angle within that zone, and these zone spectra are then combined through the relativistic transfer function and the correct $r_e$ weighting. This replaces the approximation of using an average spectrum over the whole disk or over radial annuli, and it is only moderately slower than the standard relxill calculation.","core_discovery":"The central claim is that replacing the whole-disk or radial-annulus averaging of the reflected intensity with a 10-zone decomposition in emission angle $\\theta_e$ makes relativistic reflection spectra accurate enough to fit high-quality black hole spectra without large residuals. The model also corrects a bug in relxill v2.4 where the integral weighting contained $r_e^4$ instead of $r_e$; relxill v2.5 fixes that bug alone, but the paper shows this is still not enough. With relxillA, the simulated NewAthena/X-IFU+LAD spectra are fitted with no visible residuals, whereas relxill v2.4 and v2.5 leave systematic structure in the residuals. The paper further shows that for the current NuSTAR observation of EXO 1846-031, the improved model gives statistically similar fits to the older models, so past analyses with relxill v2.4 remain valid for available data, while the improvement becomes necessary only with higher-quality data.","pith_inferences":["If the emission-angle decomposition is the main source of error, similar zone-based angular treatments could improve other relativistic models, including non-Kerr extensions and time-dependent reverberation codes.","The fixed 10-bin grid in $\\mu$ may need a convergence check; testing 20 or 50 bins on simulated high-quality spectra would directly test whether the discretization is saturated.","The validation is currently limited to one spin value and three inclinations, so a broader parameter scan could reveal settings where the 10-bin approximation degrades.","Because the model changes predicted spectra only subtly, the practical payoff begins with X-IFU/LAD-like throughput, not with current observatories."],"forward_implications":["Relativistic reflection fits on next-generation X-ray data can use relxillA without introducing any new model parameters or degeneracies.","The r^4 bug in relxill v2.4 is corrected in v2.5, but the angular averaging error remains and is larger than the bug for high-inclination sources.","Current NuSTAR-quality data cannot distinguish relxillA from older models, so archival analyses with relxill v2.4 do not need to be redone.","The speed penalty of relxillA is modest (about 17 ms per spectrum at default resolution versus 14 ms for relxill v2.4), keeping it usable for spectral fitting.","For simulated NewAthena/X-IFU+LAD spectra, relxillA removes the large residuals that make relxill v2.4 and v2.5 statistically unacceptable fits."],"supporting_citations":[{"why":"Supplies the simulated NewAthena/X-IFU+LAD observations and the earlier demonstration that relxill and relxillnk cannot fit them.","marker":"Liu et al. (2025)"},{"why":"Defines the relxill model that this paper extends, including its treatment of the emission angle.","marker":"García et al. (2014)"},{"why":"Provides the blackray ray-tracing code used as the accurate reference for emission-angle maps and spectra.","marker":"Abdikamalov et al. (2024)"},{"why":"Introduces the transfer function formalism that underlies the flux integrals in Eqs. (1)-(10).","marker":"Cunningham & Bardeen (1973)"},{"why":"Gives the pedagogical derivation of the exact relativistic flux calculation that the model approximates.","marker":"Bambi (2024)"},{"why":"Defines reltrans v1.0.1, one of the publicly available reflection models compared against blackray.","marker":"Ingram et al. (2019)"},{"why":"Defines reltrans v2.0, another reflection model compared against blackray in the lamppost setup.","marker":"Mastroserio et al. (2021)"},{"why":"Provides the detailed emissivity-profile study of EXO 1846-031 that contextualizes the NuSTAR fit in Section 5.","marker":"Li et al. (2024)"}],"fun_headline_variants":["Emission-angle zones sharpen black hole reflection spectra","Dividing disk by emission angle fixes residual structure","Reflection model corrected for next-gen X-ray quality","Emission-angle accuracy essential for future X-ray data","Improved emission-angle model beats relxill v2.5"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The accuracy of relxillA is validated only against the authors' ray-tracing code blackray for a single spin value and three viewing angles, and it assumes the fixed 10-bin discretization in emission angle remains accurate for all spins, emissivity profiles, and ionization parameters used in spectral fits.","fun_headline_variants_meta":{"raw":{"variants":["Emission-angle zones sharpen black hole reflection spectra","Dividing disk by emission angle fixes residual structure","Reflection model corrected for next-gen X-ray quality","Emission-angle accuracy essential for future X-ray data","Improved emission-angle model beats relxill v2.5"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000425,"raw_usage":{"total_tokens":2153,"prompt_tokens":891,"completion_tokens":1262,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":1186}},"tokens_in":507,"tokens_out":1262,"duration_ms":11248,"temperature":1.0,"reasoning_tokens":1186,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:55:17.372970+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fit the same simulated NewAthena/X-IFU+LAD spectra with relxillA using 20 or 50 bins in $\\mu$ instead of 10; if the fit improves substantially or parameter estimates shift beyond the 90% confidence intervals, the 10-bin discretization is not converged and the claimed accuracy does not hold. A complementary check would compare relxillA against an independent ray-tracing or Monte Carlo transfer code at spins such as $\\alpha = 0, 0.5, 0.9$ and inclinations near $60^\\circ-85^\\circ$.","supporting_citations":[{"cited_title":"B., Ayzenberg, D., Bambi, C., et al","cited_arxiv_id":null,"evidence_quote":"Provides the blackray ray-tracing code used as the accurate reference for emission-angle maps and spectra."},{"cited_title":"T., & Bardeen, J","cited_arxiv_id":null,"evidence_quote":"Introduces the transfer function formalism that underlies the flux integrals in Eqs. (1)-(10)."}],"review_version":1}