{"id":"42490e29-61c9-4079-b431-efc95e402b5b","arxiv_id":"2607.26045","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":1,"one_line_summary":"Across six pencil-beam benchmarks, the unofficial OpenMC photonuclear branch reproduces MCNPX integrated photoneutron yields within 0.7% under common ENDF7u data, while emitted-neutron spectra vary more between codes and data libraries.","lead":"An unofficial version of the open-source Monte Carlo code OpenMC, which adds photonuclear reactions, is checked against two established codes and a simple formula. It agrees with MCNPX to within 0.7% on total neutron yields, but energy spectra show larger model-dependent differences.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Headline 0.7% agreement is not tied to statistical uncertainties; without error bars the OpenMC–MCNPX 'verification' cannot be distinguished from Monte Carlo noise, and this is the load-bearing gap in the central claim.","rationale":"The reader's verdict CONDITIONAL is supported by the lack of statistical uncertainty quantification and the absence of reproducible artifacts. The reader's stated weakest_assumption, the broomstick single-collision premise, is not the most load-bearing: a quantitative estimate shows the axial leakage fraction is below 1e-16 for the given dimensions, so the lateral-current tally equals first-collision production to extremely high precision. The real soft spot is the unreported Monte Carlo statistical error on the headline 0.7% agreement. This gap is directly testable from the existing tallies and, if material, undermines the verification claim. Since the reader already conditioned on this issue, my independent concern does not move the verdict; it sharpens the rationale. I therefore recommend UNCHANGED, with the concrete test proposed above as the next step.","tokens_in":14679,"tokens_out":14766,"duration_ms":145651,"concrete_test":"Using the stored OpenMC and MCNPX surface-current tallies for the six configurations, compute the variance of the integrated outward current from the reported bin-wise standard errors (assuming independence or using covariances from the tally bins). Form the difference Δ = Y_OpenMC − Y_MCNPX and its standard error σ_Δ. If |Δ|/σ_Δ < 2 for any benchmark, the 0.7% agreement is consistent with statistical noise, and the authors must report the uncertainty or re-run with more particles to claim sub-percent verification; if |Δ|/σ_Δ > 3 for all six, the verification claim is confirmed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim — 'OpenMC and MCNPX agreed within approximately 0.7% in integrated neutron yield for all six benchmark configurations' (Abstract, Sec. III.B) — is presented without any quantification of the statistical uncertainty in those yields. Section II.B states that 'OpenMC and MCNPX uncertainties correspond to the standard errors reported by their respective Monte Carlo tally estimators,' but no standard errors are reported, plotted, or used to bound the difference. For the lower-yield configurations (e.g., 9Be at 5 MeV, where the integrated yield is on the order of 1e-3 per source photon), 10^8 source photons imply Poisson relative errors around 0.3–0.5%; a 0.7% code difference could therefore be within ~2σ of zero. Without error bars, the sub-percent agreement cannot be distinguished from Monte Carlo noise, and the verification 'stamp' is not statistically grounded. This is the most load-bearing unaddressed issue because it directly supports the paper's main conclusion. By contrast, the single-collision premise that concerned the reader is quantitatively safe: for r = 1e-6 cm and L ~ 3000 cm, the solid-angle fraction for axial escape is ~(r/L)^2/4 < 1e-16, so the lateral-current tally equals first-collision production to high precision. The paper's self-acknowledged need to 'establish' the finite-length approximation (Sec. II.C) is a real but minor omission, not a threat to the central comparison.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a verification study of an unofficial photonuclear development branch of OpenMC using six single-collision broomstick benchmarks. Targets are 2H, 9Be, and 238U, each exposed to a monoenergetic photon beam (5, 5, and 15 MeV, respectively) and to a continuous 1–20 MeV Kramers bremsstrahlung spectrum. OpenMC is compared with MCNPX under the same ENDF7u photonuclear data, with FLUKA using its native photonuclear models, and with a first-collision analytical solution for integrated neutron yield. Additional OpenMC runs with IAEA/PD-2019 quantify nuclear-data sensitivity. The principal finding is that OpenMC and MCNPX agree within about 0.7% in integrated neutron yield for all six configurations, while FLUKA and IAEA/PD-2019 deviate by larger amounts. The paper concludes that the branch reproduces integrated neutron production in the tested idealized geometries, while energy-differential spectra remain model- and data-sensitive.","tokens_in":15068,"tokens_out":7803,"duration_ms":78940,"significance":"This is a timely and appropriately scoped verification of a capability that is not yet in the official OpenMC distribution. The use of a common evaluated library for the primary OpenMC–MCNPX comparison is a sound way to separate implementation differences from data differences, and the chosen nuclides exercise distinct photonuclear mechanisms (two-body disintegration, multi-body breakup, actinide multi-channel/fission competition). The work also usefully demonstrates that integral agreement does not imply spectral agreement, and it quantifies sensitivity to the underlying nuclear-data evaluation. The paper is honest about the verification/validation distinction and about the idealized nature of the broomstick geometry. If the central quantitative claim is supported by reported statistical uncertainties, the paper will be a useful reference for the community; as it stands, that support is missing.","major_comments":[{"comment":"The central claim, “OpenMC and MCNPX agreed within approximately 0.7%,” is not accompanied by any statistical uncertainty. Section II.B says that uncertainties correspond to the standard errors of the respective tally estimators, but no standard errors are given, plotted, or used to test whether 0.7% is statistically distinguishable from zero. For a low-yield configuration such as 9Be at 5 MeV (integrated yield on the order of 1e-3 per source photon), 1e8 source photons imply a Poisson relative error of roughly 0.3–0.5%; a 0.7% code difference is then on the order of 1–2 sigma. The same issue affects the energy-resolved comparison in Fig. 4, where low-count bins could be statistically indistinguishable. Please report per-bin and integrated relative errors (or confidence intervals) for all codes and all configurations, and state explicitly whether the quoted 0.7% is the observed differenc","section":"III.B / Fig. 5 / Sec. II.B"},{"comment":"The paper itself states that the axially thick approximation underlying Eqs. (8) and (11) “must therefore be established by evaluating the finite-length interaction factor,” but no such evaluation is reported for any target. This matters for the analytical reference, especially for 238U, where L is only 400 cm and the factor 1−exp(−Σtot L) is not identically unity. The reader is asked to accept that the approximation is exact without a quantitative check. Please report either the minimum value of the finite-length factor over the contributing photon-energy range or the resulting correction to Yn for each case, even if the correction is negligible.","section":"II.C, Eqs. (7)–(11)"}],"minor_comments":[{"comment":"The analytical solution is explicitly constructed from the same ENDF7u data used by OpenMC and MCNPX. Although Sec. II.C acknowledges this, the abstract and title present the analytical solution as a separate verification reference. Please qualify it as a consistency check of the implementation under a shared data library rather than an independent verification of the yield.","section":"Abstract / Sec. II.C"},{"comment":"The photofission multiplicity νf is said to be energy dependent, but no explicit prescription or source for νf(Eγ) is given. Since the 238U cases rely on this quantity, a brief definition or citation would help.","section":"Sec. II.C, Eq. (2)"},{"comment":"For a verification of an unofficial development branch, the exact code state matters. Please provide the specific commit hash (or version) of the Stein photonuclear branch used for all OpenMC calculations, and state whether the input decks are available in a repository.","section":"Sec. II.B / Reproducibility"},{"comment":"Several deviation labels are difficult to read (e.g., “1.87e-02%” and “3.64e-04%”), and the plot mixes different marker styles across panels. A small table of the integrated yields and deviations with uncertainties would be clearer than the crowded figure.","section":"Fig. 5"},{"comment":"The FLUKA comparison combines implementation, model, and data differences. The paper states this correctly, but the conclusion should avoid implying that FLUKA provides an independent verification of OpenMC; it is better described as a cross-model plausibility check.","section":"Sec. IV / Limitations"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing you should know: this is the first systematic verification of the photonuclear branch of OpenMC, and the core claim—near-identical integrated neutron yields between OpenMC and MCNPX under common ENDF7u data—is credible and carefully scoped. The paper cleanly separates a common-data code comparison, a model-based FLUKA comparison, and a nuclear-data sensitivity study, and it is honest about what each can and cannot show. The analytical solution is a consistency check rather than an external benchmark, and the authors say so.\n\nWhat is genuinely new: the continuous-spectrum benchmarks, the FLUKA and analytical references, and the IAEA/PD-2019 sensitivity quantifications. The observation that integral agreement does not guarantee spectral agreement, especially for the monoenergetic 2H case, is worth keeping in mind. The treatment of 238U with competing channels is sensible, and the paper does not overclaim validation.\n\nThe soft spots are real but not fatal. The main one: the 0.7% agreement between OpenMC and MCNPX is presented without numerical statistical uncertainties, despite Section II.B saying the codes report standard errors. For the low-yield cases (e.g., 9Be at 5 MeV, yield around 1e-3 per source photon), 1e8 photons give Poisson relative errors around 0.3–0.5%, so a 0.7% difference could be within about two sigma. The sub-percent claim needs error bars to be meaningful. That is the load-bearing gap.\n\nThe broomstick single-collision premise itself is safe. With radius 1e-6 cm and lengths in the hundreds to thousands of centimeters, the solid-angle fraction for secondary transport is vanishingly small; the lateral-current tally is essentially first-collision production. The paper's caution about the finite-length factor for 238U is a minor omission, not a threat to the comparison.\n\nAlso worth flagging: no input decks, commit hashes, or other reproducible artifacts are provided. That matters for a verification study of an unofficial branch, and it is an easy fix. The same-data analytical reference limits its role, but the MCNPX comparison is independent and carries the verification.\n\nWho this is for: anyone using or considering the unofficial OpenMC photonuclear branch for source-strength or activation work, and the OpenMC development community. It deserves a serious referee, but the referee should insist on reported uncertainties and reproducibility artifacts. I would engage with it; it is a solid, honest piece of work that needs a moderate revision rather than a rewrite.","headline":"Useful, well-scoped verification of the unofficial OpenMC photonuclear branch, but the headline 0.7% agreement is not tied to reported statistical uncertainties and the paper ships no reproducible inputs.","tokens_in":15509,"tokens_out":1541,"would_cite":true,"duration_ms":16689,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["25.20.-x","02.70.Uu","28.20.-v"],"model":"deepseek-v4-flash","headline":"OpenMC's photonuclear branch reproduces MCNPX's integrated neutron yields to within 0.7% across six broomstick benchmarks.","keywords":["photonuclear reactions","OpenMC","Monte Carlo verification","photoneutron production","MCNPX","FLUKA","photonuclear data","broomstick benchmark"],"falsifier":"A decisive check would be to rerun the 238U benchmarks with a shorter target length (so Σtot L is no longer ≫1) and compare against the full Eq. (5) rather than the axially thick limit; if OpenMC and MCNPX stop tracking the analytical prediction in exactly the way the factor 1−exp(−Σtot L) departs from 1, the 'current equals first-collision production' identification is the fragile step. Alternatively, an experimental 2H photodisintegration spectrum at 5 MeV would discriminate between ENDF7u's reported kinematic bug and the other models, since the paper attributes the spectral discrepancy to a","tokens_in":14618,"feed_emoji":"⚛️","tokens_out":5039,"duration_ms":44898,"temperature":0.7,"pith_summary":"The paper verifies the photonuclear capability of an unofficial OpenMC development branch. Using six 'broomstick' benchmarks—2H, 9Be, and 238U targets hit by monoenergetic 5 or 15 MeV photons and by a continuous 1–20 MeV LINAC-like spectrum—it finds that OpenMC and MCNPX agree within about 0.7% on integrated neutron yield when both use the same ENDF7u data. A first-collision analytical solution confirms the yields for the light nuclides, while FLUKA, using its own photonuclear models, lands 6–9% away in the monoenergetic cases and within about 4% for the continuous source. The paper's second point is that integral agreement does not imply spectral agreement: the 2H monoenergetic case shows similar totals but visibly different neutron energy spectra. A reader should care because it clears the integrated-yield barrier to adopting the photonuclear branch while mapping where further work—spectra, kinematics, nuclear data—still sits.","feed_headline":"OpenMC photonuclear branch matches MCNPX within 0.7%","feed_subtitle":"Six broomstick benchmarks on 2H, 9Be, and 238U verify integrated yields; spectra still diverge.","key_machinery":"The broomstick benchmark: an extremely long, infinitesimally thin cylinder (radius 1e-6 cm, lengths 3000 cm for 2H, 1500 cm for 9Be, 400 cm for 238U) aligned with a pencil photon beam. This geometry suppresses secondary transport, so the outward neutron current through the lateral surface can be compared directly with first-collision production. The analytical solution computes Y_n = ∫ (dNγ/dEγ) [1−exp(−Σtot L)] w(Eγ) dEγ, where w(Eγ) = Σn/Σtot is the expected number of neutrons produced per first photon interaction, with Σn the macroscopic neutron-production cross section (including multiplicities) and Σtot the total photon interaction cross section. In the axially thick limit the bracket b","core_discovery":"The central claim is that, under common ENDF7u photonuclear data, the OpenMC photonuclear branch's integrated neutron-production response is statistically indistinguishable from MCNPX's: the two codes agree within 0.7% across all six benchmark configurations. The paper argues this verifies the branch's handling of the common evaluated cross sections, reaction multiplicities, source weighting, and integrated neutron tallies for the tested nuclides and spectra. It also establishes a boundary: agreement in total yield does not certify agreement in the emitted-neutron spectrum, demonstrated most sharply by the monoenergetic 2H case, where OpenMC and MCNPX get the same total but different spectra","pith_inferences":["Beyond the paper: the same six benchmarks could serve as regression tests if the photonuclear branch is ever proposed for integration into the official OpenMC distribution.","Beyond the paper: the 2H spectral mismatch points to a library-level kinematic issue (the paper cites an NJOY processing bug in older evaluations), implying that other codes using ENDF7u may also mis-sample spectra at 5 MeV even when yields are correct.","Beyond the paper: extending the broomstick design to angular-distribution scoring or to additional nuclides (e.g., 197Au, 63Cu) would test whether the 0.7% agreement holds beyond the three selected materials.","Beyond the paper: a quick analytic check of 1−exp(−Σtot L) over the 238U spectra would bound the residual bias in the analytical reference; the paper leaves that check unquantified."],"forward_implications":["OpenMC's photonuclear branch can be trusted for integrated neutron-yield estimates in the tested energy–nuclide combinations, with the same caveats that apply to MCNPX-based results.","Users needing neutron energy or angular distributions should not rely on integral verification: the 2H monoenergetic case shows that identical totals can hide different spectra.","Nuclear-data choice matters more than code choice in several configurations: switching OpenMC from ENDF7u to IAEA/PD-2019 changed yields by up to 11.3%, larger than the code-to-code spread.","FLUKA's native-model results are within about 6–9% (monoenergetic) and about 4% (continuous) of the analytical reference, narrowing under broadband averaging even though local models differ.","The first-collision analytical solution is a practical reference for separating implementation errors from evaluated-data differences in photonuclear code verification."],"fun_headline_variants":["OpenMC, MCNPX neutron yields match within 0.7%","OpenMC matches MCNPX photonuclear yields within 0.7%","0.7% yield agreement: OpenMC vs MCNPX","OpenMC photonuclear passes MCNPX comparison within 0.7%","OpenMC matches MCNPX to 0.7% on neutron yields"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is the 'infinitesimal radius' broomstick condition—radius 10^-6 cm—which guarantees that every neutron created in the first collision exits through the lateral surface without secondary transport, so the tallied outward current equals first-collision production; the paper also assumes for all cases, including 238U, that the axial interaction factor 1−exp(−Σtot L) is effectively unity, yet it does not report the numerical check.","fun_headline_variants_meta":{"raw":{"variants":["OpenMC, MCNPX neutron yields match within 0.7%","OpenMC matches MCNPX photonuclear yields within 0.7%","0.7% yield agreement: OpenMC vs MCNPX","OpenMC photonuclear passes MCNPX comparison within 0.7%","OpenMC matches MCNPX to 0.7% on neutron yields"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000649,"raw_usage":{"total_tokens":2875,"prompt_tokens":861,"completion_tokens":2014,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":605,"completion_tokens_details":{"reasoning_tokens":1915}},"tokens_in":605,"tokens_out":2014,"duration_ms":14714,"temperature":1.0,"reasoning_tokens":1915,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T00:45:00.029663+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive check would be to rerun the 238U benchmarks with a shorter target length (so Σtot L is no longer ≫1) and compare against the full Eq. (5) rather than the axially thick limit; if OpenMC and MCNPX stop tracking the analytical prediction in exactly the way the factor 1−exp(−Σtot L) departs from 1, the 'current equals first-collision production' identification is the fragile step. Alternatively, an experimental 2H photodisintegration spectrum at 5 MeV would discriminate between ENDF7u's reported kinematic bug and the other models, since the paper attributes the spectral discrepancy to a","supporting_citations":[],"review_version":1}