{"id":"284c19cd-d45f-4c3c-9165-b6660088a975","arxiv_id":"2412.00455","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Simulations show that a new time-of-flight neutron detector with 121 scintillator cells per layer and two layout options can identify and reconstruct neutrons from Bi+Bi collisions at 3A GeV with about 70% purity and roughly 1e9 events per month.","lead":"This paper describes the design and simulated performance of a new high-granularity neutron detector being built for the BM@N heavy-ion experiment, which should measure neutron flow and yields at energies up to 4A GeV. It matters because such measurements could constrain the symmetry-energy term of the nuclear equation of state, which is poorly known at high baryon density.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Internal inconsistency in the headline yield: 500 n/s acceptance flux cannot produce 1.2e9 reconstructed single-neutron events/month at 50% efficiency; the missing spill-cycle parameter also prevents reproduction.","rationale":"The reader's weakest assumption (DCM-QGSM-SMM and Geant4 fidelity) is legitimate, but it is an external physics-modeling uncertainty that would affect many detector-simulation studies. Even if the generator is off by tens of percent, the detector could still function as designed. The more immediate and decisive problem is that the paper's own quantitative claim about reconstructed neutron yield is not internally consistent: the stated acceptance flux of 500 n/s, when combined with the stated 50% efficiency and 76% single-neutron fraction, cannot produce 1.2e9 reconstructed single-neutron events per month. The calculation also omits the spill repetition rate, so the result is not reproducible from the text. This concern is prior to any modeling question: if the arithmetic of the yield claim is unresolved, the summary's central quantitative promise is unsupported. The conditional verdict remains appropriate because the detector design and simulation studies are otherwise credible; the condition should explicitly include reconciling the yield estimate and reporting the spill-cycle input. Therefore the reader's verdict is unchanged.","tokens_in":7959,"tokens_out":10077,"duration_ms":101101,"concrete_test":"Recompute the monthly yield from the stated inputs: T_month = 2.592e6 s, F_accept = 500 s^-1, efficiency = 0.5, single-neutron fraction = 0.76. Then N_single = F_accept x T_month x efficiency x f_single = 4.9e8, versus the claimed 1.2e9. To settle, request from the authors (a) the exact spill-cycle length and number of spills per month used in Sec. 3.1, and (b) whether the 500 n/s in Sec. 1 refers to the same acceptance and reaction as the yield calculation. If no single consistent set of parameters yields 1.2e9, the headline yield must be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing problem is the internal consistency of the headline yield number, not the generator model. Section 1 states that the estimated neutron flux at BM@N is about 500 neutrons per second in the HGND acceptance. One month of continuous running is 2.592e6 s, so the total number of primary neutrons entering the detector acceptance in a month is at most 1.3e9. Section 3.1 then claims that, with a mean HGND efficiency of 50%, about 1.2e9 single-neutron events will be reconstructed per month, rising to 1.5e9 if multiple-neutron events are included. Both numbers exceed (or nearly equal) the total incoming neutron count: 500 n/s x month x 0.5 = 6.5e8 reconstructed events, and applying the stated 76% single-neutron fraction lowers this to roughly 4.9e8. As written, 1.2e9 is arithmetically impossible unless the acceptance flux is about 1200 n/s, not 500 n/s. In addition, the yield formula gives no spill repetition rate or number of spills per month, so the '1e6 ions per spill, 50% duty factor, 70% efficiency, 2% interaction length' inputs cannot be combined by the reader to reproduce the result. This is not an external modeling uncertainty; it is an internal inconsistency in the paper's central quantitative claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes the design and simulation studies of a highly granular time-of-flight neutron detector (HGND) for the BM@N fixed-target program at JINR. Two geometric options are presented (a single module at 5 m and a two-arm module at 7 m), along with the scintillator/SiPM cell design, the FPGA/TDC readout, and Geant4-based simulations using the DCM-QGSM-SMM event generator for Bi+Bi and Xe+CsI collisions. The authors report a neutron detection efficiency around 50%, an energy resolution of 10-30% from time-of-flight reconstruction with 150 ps time smearing, and compare a cluster method with a graph-neural-network approach, quoting reconstruction efficiency at a fixed purity of 0.7. They estimate a yield of about 1e9 reconstructed neutrons per month of BM@N operation.","tokens_in":8215,"tokens_out":9491,"duration_ms":94719,"significance":"If correct, the paper demonstrates a workable detector concept for neutron measurements in the 300 MeV to 4 GeV range at BM@N, with two independent reconstruction strategies and a transparent simulation setup. The strengths are the clearly specified thresholds, time cuts, time smearing, reconstruction definitions, and the held-out 50/50 train/test split for the machine-learning method. The performance numbers are internally plausible as preliminary simulation results. However, the central yield estimate contains an arithmetic inconsistency with the stated acceptance flux, and several simulation inputs are not fully documented; these issues must be resolved before the quantitative physics claims can be accepted.","major_comments":[{"comment":"The yield estimate is not internally consistent with the quoted acceptance flux. The Introduction states that the neutron flux in the HGND acceptance is about 500 neutrons/s; over 30 days this gives 1.3e9 incoming neutrons, and multiplying by the quoted 50% mean efficiency gives at most 6.5e8 reconstructed neutrons. Since single-neutron events are a subset of all neutron events, the claimed 1.2e9 reconstructed single-neutron events is arithmetically impossible under these inputs. The estimate also cannot be reproduced because the listed inputs ('1e6 ions per spill, 50% duty factor and 70% efficiency of Nuclotron, 2% interaction length of target') omit the spill repetition rate or number of spills per month. Please supply a complete formula with all beam parameters and a corrected yield number, and harmonize it with the Summary's 'about 1e9'.","section":"Section 3.1"},{"comment":"The 'mean efficiency of the HGND of 50%' used in the yield calculation is not defined. Figure 7 shows a strong energy dependence of the detection efficiency, so a single 50% value must be accompanied by the weighting spectrum. The yield should be computed by averaging the energy-dependent efficiency over the simulated primary-neutron spectrum at the chosen HGND position, and the resulting spectrum-weighted mean should be quoted instead of an ad-hoc 50% number.","section":"Section 3.1, Fig. 7"},{"comment":"The paper does not provide any validation of the DCM-QGSM-SMM event generator for neutron production at BM@N energies, nor of the Geant4 hadronic transport models used in bmnroot. The quoted multiplicity, efficiency, purity, and yield all inherit these model assumptions. Please add a comparison with existing data (for example, the Xe+CsI run at 3.8A GeV already collected in 2023, or published neutron spectra from similar reactions) or at least state the generator dependence as a systematic uncertainty with a concrete cross-check; without this, the quantitative claims are conditional on an unvalidated input model.","section":"Section 3"},{"comment":"The beam energy labels are inconsistent between the text and the figure captions. Section 3.1 states the Bi+Bi performance studies are done at 3.0A GeV, but the Fig. 9 caption reads 'Bi+Bi @ 3.8A GeV'; for Xe+CsI the text says 3.8A GeV while the Fig. 10 caption reads 3A GeV. Beam energy directly changes the neutron multiplicity and time-of-flight distributions used to derive the yield, so these inconsistencies must be resolved and the simulation results checked against the actual energy used.","section":"Section 3.1, Figs. 9 and 10"}],"minor_comments":[{"comment":"The velocity cut 'v < c' is not a meaningful restriction for massive particles, all of which satisfy v < c; the text should specify the actual threshold value used for gamma and charged-particle rejection.","section":"Section 4.1"},{"comment":"The sentence 'The difference for 1 GeV neurons' should read '1 GeV neutrons'.","section":"Fig. 7"},{"comment":"The text refers to an energy-resolution comparison in Fig. 13 without stating numerical values or the definition of the systematic shift for the cluster method; a short quantitative summary would improve reproducibility.","section":"Section 4.3, Fig. 13"},{"comment":"There are several typographical errors, including 'syimmetric' (Introduction), 'assemled' and 'matix' (Section 2.1), 'L VDS' (Sections 2.1 and 2.2), and 'yieds' (Summary); these should be corrected.","section":"Throughout"},{"comment":"The right panel of Fig. 6 is described as 'background neutron on all surfaces of the HGND', but the text does not specify which surfaces are included or whether this background includes neutrons from hadronic interactions in the detector material; a sentence clarifying the definition would help.","section":"Section 3.1, Fig. 6"}],"recommendation":"major_revision","confidential_remarks":"The inconsistency in Section 3.1 is the main obstacle; it appears to be an error in the luminosity/yield calculation rather than in the detector simulation itself, so it can be fixed within the scope of the manuscript. I do not see grounds for rejection, but the corrected arithmetic and a clearer statement of model dependence are needed before the quantitative claims can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this is a genuine progress report on a neutron ToF detector for BM@N, with real new content over the group's own earlier papers: the 5m vs 7m two-arm layout comparison, a GNN-based reconstruction pipeline, and monthly neutron yield estimates. Second, the headline yield number is internally inconsistent with the stated neutron flux, and the paper should not be accepted until that is fixed.\n\nThe simulation work is described in enough detail to follow: Geant4/bmnroot, DCM-QGSM-SMM, thresholds, time cuts, 150 ps smearing. The efficiency curves and energy resolution plots are preliminary but plausible, and the cluster-vs-GNN comparison is a legitimate cross-check. The GNN is tested on a held-out 50% split, which is more than many ML-in-physics papers do. All of that deserves credit.\n\nNow the soft spots. Section 1 says the neutron flux in the HGND acceptance is about 500 neutrons per second. A month is 2.592e6 s, so at most 1.3e9 primary neutrons enter the acceptance. With the stated 50% mean efficiency, you cannot reconstruct more than ~6.5e8 neutron hits per month. The paper claims 1.2e9 single-neutron events, and 1.5e9 including multiple-neutron events. That is arithmetically impossible under the stated assumptions. Also, the yield formula lists ions per spill, duty factor, Nuclotron efficiency, and interaction length, but no spill repetition rate, so the reader cannot reproduce the number. This is an internal inconsistency in the central quantitative claim, not an unavoidable external uncertainty. It may be a simple missing parameter or a typo in the flux, but as written it fails. There are also smaller inconsistencies: Fig. 9's caption says Bi+Bi at 3.8A GeV while the text says 3.0A GeV, and Fig. 10's caption says 3A GeV while the text says 3.8A GeV. And the performance plots have no error bars or statistical uncertainties, which is a minor issue for a simulation study but worth noting.\n\nThe physics motivation is sound: the symmetry-energy term at baryon densities above 1 GeV per nucleon is poorly constrained, and a high-statistics neutron detector at BM@N could provide new data. The design choices are reasonable. But the paper is a detector R&D report, not a physics result; it will be useful to the heavy-ion instrumentation community and to BM@N collaborators. It deserves peer review, but the referee should insist on corrected yield arithmetic, error estimates, and consistent beam-energy labels.","headline":"Useful BM@N neutron detector simulations, but the headline yield is internally inconsistent with the stated flux and needs a correction.","tokens_in":8892,"tokens_out":3437,"would_cite":false,"duration_ms":33262,"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":"This paper claims that a newly designed highly granular time-of-flight neutron detector can identify neutrons and reconstruct their energies in heavy-ion collisions at BM@N up to 4A GeV, with a simulated yield of about one billion…","keywords":["time-of-flight neutron detector","heavy-ion collisions","neutron reconstruction","azimuthal neutron flow","plastic scintillator","silicon photomultiplier","graph neural network","neutron yields"],"falsifier":"In the first Xe+CsI data run, compare the measured single-neutron event rate per $10^6$ beam ions with the simulated value of about 12.7% of events at the HGND surface; a sustained rate disagreement beyond a factor of two, after correcting for live time and trigger efficiency, would falsify the monthly yield estimate.","tokens_in":7739,"feed_emoji":"⚛️","tokens_out":8846,"duration_ms":80908,"temperature":0.7,"pith_summary":"The paper reports the design and simulated performance of a new highly granular time-of-flight neutron detector for the BM@N heavy-ion experiment. Its central claim is that the detector can identify primary neutrons and reconstruct their kinetic energies in the range from about 300 MeV to 4 GeV, using plastic scintillator cells read out by silicon photomultipliers and time-of-flight measurement. In simulations of Bi+Bi collisions at 3A GeV, the detector reaches roughly 50% detection efficiency, 10–30% energy resolution, and a reconstruction purity of 0.7 at that efficiency, yielding an estimated $1.2\\times10^9$ reconstructed single-neutron events per month of running (about $1\\times10^9$ in the summary's rounded figure). A first Xe+CsI run at 3.8A GeV is predicted to give about $0.9\\times10^9$ single-neutron events per month. These numbers matter because neutron yields and azimuthal neutron flow are observables sensitive to the symmetry-energy term of the equation of state of dense nuclear matter, which is poorly constrained above 1 GeV per nucleon.","feed_headline":"Neutron detector promises a billion events a month","feed_subtitle":"Time-of-flight cells would measure neutron energies and flow in heavy-ion collisions up to 4A GeV.","key_machinery":"The load-bearing object is the HGND module: an $11\\times11$ matrix of $4\\times4\\times2.5\\,\\mathrm{cm}^3$ polystyrene scintillator cells read out by silicon photomultipliers on one side and LEDs for calibration on the other, with copper absorbers between layers and a veto first layer. Time-of-flight is the working identity: neutron velocity is inferred from the distance to the target divided by the hit time, and kinetic energy follows from $E = m_n(1/\\sqrt{1-(v/c)^2}-1)$, supported by a 100 ps FPGA-based TDC and measured cell resolution of 130–150 ps. The copper absorber and layer stacking create a sampling calorimeter that converts neutron interactions into small clusters of fired cells, which are then grouped either by geometric and time clustering or by a graph neural network whose nodes are hit coordinates and energy deposits. The simulation chain of a heavy-ion event generator plus a detector-response simulation provides the efficiency, background, and yield numbers.","core_discovery":"The central claim is that a compact, highly segmented detector—16 alternating layers of $11\\times11$ arrays of $4\\times4\\times2.5\\,\\mathrm{cm}^3$ plastic scintillator cells interleaved with 3 cm copper absorber plates, with the first layer acting as a charged-particle veto—can turn time-of-flight hits into a clean neutron sample in the fixed-target BM@N environment. Neutron kinetic energy is obtained from the fastest hit in a cluster through $E = m_n(1/\\sqrt{1-(v/c)^2}-1)$, with per-cell time resolution measured at 130–150 ps and a TDC precision near 40 ps. With a 35 ns time cut that suppresses background by about a factor of 6 while retaining 92% of primary neutrons, the simulated detection efficiency is about 50% and the energy resolution is 10–30% over $0.3$–$4$ GeV. Two reconstruction strategies—a cluster method and a graph-neural-network method—both reach a reconstruction efficiency around 0.7 at fixed purity 0.7 in simplified single-neutron tests, with the machine-learning method correcting time-of-flight overestimates at 2–4 GeV.","pith_inferences":["If the simulated performance transfers to the real detector, this would extend neutron-flow constraints from sub-GeV heavy-ion measurements into the multi-GeV regime where the symmetry-energy term of the nuclear equation of state is least constrained.","The graph-neural-network energy-regression model is trained only on events identified as containing neutrons; retraining on events with multiple neutrons would likely be needed before the reported purity holds for high-multiplicity central collisions.","A natural testable extension is to use the two 7 m 'arms' to measure the reaction-plane dependence of neutron azimuthal flow differentially in rapidity, which the current single-neutron performance study does not yet demonstrate."],"forward_implications":["HGND can deliver about $1\\times10^9$ reconstructed single-neutron events per month of BM@N running, giving a dataset large enough for differential neutron yield and azimuthal-flow measurements at beam energies up to 4A GeV.","The 35 ns time cut removes roughly six times more background than signal while keeping 92% of primary neutrons, so primary neutrons with kinetic energy above about 300 MeV can be selected cleanly.","The two independent reconstruction methods—cluster analysis and graph neural networks—can cross-check each other, and the machine-learning route compensates the time-of-flight energy overestimation seen in the cluster method at 2–4 GeV.","If multiple-neutron events are also reconstructed, the monthly statistics rise from about $1.2\\times10^9$ to about $1.5\\times10^9$ reconstructed neutron events for Bi+Bi at 3A GeV."],"supporting_citations":[{"why":"Defines the BM@N spectrometer layout and the available space that fixes the two HGND positions at 5 m and 7 m from the target.","marker":"[15]"},{"why":"Supplies the baseline HGND concept of alternating scintillator layers and copper absorbers that this paper optimizes.","marker":"[16]"},{"why":"Provides the 100 ps FPGA-based TDC whose precision underlies the time-of-flight energy reconstruction.","marker":"[17]"},{"why":"Underlies the detector-response simulation used for the efficiency, resolution, and yield studies.","marker":"[18]"},{"why":"Is the simulation and analysis framework in which the HGND performance studies are carried out.","marker":"[19]"},{"why":"Supplies the Monte-Carlo generator for Bi+Bi and Xe+CsI collisions that produces the neutron multiplicities and spectra driving the yield estimate.","marker":"[20]"},{"why":"Gives the measured 130–150 ps scintillator cell time resolution used as the input to the energy-resolution studies.","marker":"[21]"},{"why":"Provides the graph-convolution architecture on which the machine-learning neutron reconstruction is based.","marker":"[22]"}],"fun_headline_variants":["Neutron time-of-flight detector with ML reconstruction for BM@N","Highly granular neutron detector measures flow in heavy-ion collisions","New ToF neutron detector achieves 50% efficiency in simulations","Graph neural networks improve neutron time-of-flight reconstruction"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The yield and performance numbers assume that the simulated neutron multiplicities and energy spectra from the event generator, and the simulated detector response, match the real Bi+Bi and Xe+CsI collisions at BM@N.","fun_headline_variants_meta":{"raw":{"variants":["Neutron time-of-flight detector with ML reconstruction for BM@N","Highly granular neutron detector measures flow in heavy-ion collisions","New ToF neutron detector achieves 50% efficiency in simulations","Graph neural networks improve neutron time-of-flight reconstruction"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000612,"raw_usage":{"total_tokens":2815,"prompt_tokens":883,"completion_tokens":1932,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":499,"completion_tokens_details":{"reasoning_tokens":1864}},"tokens_in":499,"tokens_out":1932,"duration_ms":16139,"temperature":1.0,"reasoning_tokens":1864,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T05:22:09.637668+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In the first Xe+CsI data run, compare the measured single-neutron event rate per $10^6$ beam ions with the simulated value of about 12.7% of events at the HGND surface; a sustained rate disagreement beyond a factor of two, after correcting for live time and trigger efficiency, would falsify the monthly yield estimate.","supporting_citations":[{"cited_title":"Development of High Granular Neutron Time-of-Flight Detector for the BM@N experiment","cited_arxiv_id":"2309.09610","evidence_quote":"Supplies the baseline HGND concept of alternating scintillator layers and copper absorbers that this paper optimizes."},{"cited_title":"Finogeev, F","cited_arxiv_id":null,"evidence_quote":"Provides the 100 ps FPGA-based TDC whose precision underlies the time-of-flight energy reconstruction."},{"cited_title":"URL https://git.jinr.ru/nica/bmnroot","cited_arxiv_id":null,"evidence_quote":"Is the simulation and analysis framework in which the HGND performance studies are carried out."},{"cited_title":"Baznat, A","cited_arxiv_id":null,"evidence_quote":"Supplies the Monte-Carlo generator for Bi+Bi and Xe+CsI collisions that produces the neutron multiplicities and spectra driving the yield estimate."},{"cited_title":"Measurement of Time Resolution of Scintillation Detectors with EQR-15 Silicon Photodetectors for the Time-of-Flight Neutron Detector of the BM@N Experiment","cited_arxiv_id":"2309.03614","evidence_quote":"Gives the measured 130–150 ps scintillator cell time resolution used as the input to the energy-resolution studies."}],"review_version":1}