{"id":"dab95c15-18ef-434f-8cf4-a7f022ec342b","arxiv_id":"2505.05909","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":2,"one_line_summary":"A 160-hour solar tracking campaign with a 7-GridPix detector at CAST found no axion signal and set a new best helioscope limit g_ae*g_aγ < 7.35e-23 GeV^-1 at 95% confidence.","lead":"A new low-background GridPix detector installed at the CAST solar axion telescope collected about 160 hours of solar data and set a new upper limit on the combined axion-electron and axion-photon couplings. The limit, g_ae*g_aγ < 7.35e-23 GeV^-1 at 95% confidence, is the best helioscope bound to date and demonstrates the detector's potential for rare-event searches.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline limit is computed with the massless conversion probability of Eq. 5.2, but is quoted up to m_a = 10 meV; the coherence form factor suppresses the low-energy g_ae signal by tens of percent near that mass, so the single quoted value may not be valid across the stated mass range.","rationale":"The reader's weakest assumption is background transferability and veto efficiency. That concern is legitimate but secondary: the claimed improvement over the previous limit is about 10%, and a plausible few-percent tracking-correlated background shift would not erase it, especially since the analysis already assigns systematics to signal, background, and position. The more load-bearing issue is that the central result is a single number quoted for 'axion masses below 10 meV', while the conversion probability used in Eq. 5.2 is the m_a = 0 expression. At m_a = 10 meV, the axion-photon coherence length is not negligible at the ~1 keV energies where the g_ae flux peaks. The resulting signal suppression directly changes the conversion of observed counts into a coupling limit, so the headline limit may be too strong near the upper end of the quoted mass range. If the paper intended the massless limit, the abstract overstates the mass range; if it intended a mass-averaged limit, the missing qL dependence is an internal inconsistency. Either way, a concrete check can settle the matter: recompute the limit with the full coherence form factor and see whether the curve stays below the previous CAST bound up to 10 meV. This is an analytical check requiring no new data, and if the curve remains below 8.1×10^{-23} GeV^{-1} throughout the range, the concern is resolved and the acceptance stands. Until then, acceptance should be conditional on that verification or on an explicit restatement of the mass domain.","tokens_in":13984,"tokens_out":23712,"duration_ms":273578,"concrete_test":"Recompute the Section 6 expected and observed limits as a function of m_a by replacing P in Eq. 5.2 with P(q) = P(q=0) · [sin(qL/2)/(qL/2)]^2, q = m_a^2/(2E), and integrating over the g_ae solar flux, the energy-dependent detection efficiency, and the same unbinned likelihood and systematics. Produce a curve of the g_ae·g_aγ limit from m_a = 0 to 10 meV. If the curve crosses 8.1×10^{-23} GeV^{-1} before 10 meV, the headline claim must be restricted to the mass range where the limit remains below the previous bound, or the mass-dependent conversion probability must be folded into the analysis.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section 5.1 defines the vacuum conversion probability as P_{a→γ,vacuum} = ε0 ħ c^3 (g_aγ B L / 2)^2, with no dependence on axion mass. For m_a > 0, the conversion probability should carry the coherence factor [sin(qL/2)/(qL/2)]^2, where q = m_a^2/(2E). Taking L = 9.26 m and E ≈ 1 keV, the energy at which the g_ae solar flux peaks, gives qL ≈ 2.35 for m_a = 10 meV and a suppression factor of about 0.62; at E = 0.5 keV the suppression is near 0.09. Because the expected signal is proportional to P, an observed no-excess upper limit on the coupling product scales as P^{-1/2}. Ignoring the form factor therefore makes the quoted 7.35×10^{-23} GeV^{-1} limit too strong near m_a = 10 meV by an amount that may be large enough to erase the claimed improvement over the previous CAST 2013 limit of 8.1×10^{-23} GeV^{-1}. The paper quotes a single limit value 'for axion masses below 10 meV' and does not present a mass-dependent exclusion curve or state that the value refers only to the m_a → 0 limit. Thus the domain of validity of the central claim is not established by the text as written.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports a search for solar axions produced through the axion-electron coupling using a 7-GridPix detector at CAST, with about 160 hours of active solar tracking data from 2017-2018. The analysis uses an MLP-based cluster classifier, scintillator and outer-chip vetoes, a FADC rise-time cut, and a raytracing-based signal spatial template. A Bayesian unbinned likelihood with nuisance parameters yields the central result: g_ae·g_aγ < 7.35×10^-23 GeV^-1 at 95% CL, quoted for axion masses below 10 meV, claimed to improve the previous CAST 2013 bound by about 10%. An ancillary limit g_aγ < 9.0×10^-11 GeV^-1 is also reported.","tokens_in":14303,"tokens_out":10024,"duration_ms":108468,"significance":"If correct, the central limit would be the best helioscope bound on the g_ae–g_aγ product and a valuable demonstration of GridPix technology for future experiments such as IAXO. The analysis has notable strengths: the limit-setting procedure was blinded (expected limits were computed before unblinding), the signal efficiency is calibrated with 55Fe and eight X-ray energies, the telescope response is validated against PANTER measurements, and the analysis framework is open source. However, the quoted limit is stated for a mass range in which the conversion probability used in the signal model is not valid, so the numerical claim as written is not established. The paper is sound in design but needs a substantive revision of the mass dependence before acceptance.","major_comments":[{"comment":"The signal model in Eq. (5.2) uses the massless conversion probability P_{a→γ,vacuum} without a coherence form factor, yet the abstract and Section 6 quote the limit \"for axion masses below 10 meV.\" For m_a = 10 meV and E ≈ 1 keV, qL = m_a^2 L/(2E) ≈ 2.35, and the factor [sin(qL/2)/(qL/2)]^2 is about 0.62, dropping to about 0.09 at E = 0.5 keV. Because the upper limit on g_ae·g_aγ scales roughly as (signal-weighted P)^{-1/2}, the quoted 7.35×10^-23 GeV^-1 is too strong near the upper end of the claimed mass range; the correction can plausibly exceed the 10% improvement over the CAST 2013 limit. Please provide a mass-dependent exclusion curve, or explicitly restrict the claim to the m_a → 0 limit and state the correspondingly narrow mass range for which the quoted number is valid.","section":"§5.1, Eq. (5.2); §6 and Abstract"},{"comment":"The displayed simplification L = e^{-s_tot} ∏_i (1 + s_i/b_i) is not correct as written if the product runs over all bins, because for bins with zero candidates the ratio of Poisson likelihoods contributes e^{-s_i}, not (1 + s_i/b_i). The product should run only over bins containing a candidate. Please correct the derivation or clarify the notation; if the numerical implementation uses the candidate-only product, state this explicitly so the manuscript is self-consistent.","section":"§5, likelihood definition"}],"minor_comments":[{"comment":"The limit is quoted on the product g_ae·g_aγ, but the prior and posterior in Eq. (5.1) are defined on g_ae with no discussion of how g_aγ is held fixed or marginalized. Please state explicitly how the product limit is obtained from L(g_ae).","section":"§5.1, Eq. (5.1)"},{"comment":"The systematic budget lists σ_s, σ_b, and σ_xy, but does not mention an uncertainty on the solar axion flux model from Ref. [15] or on the absolute normalization of the raytracing signal image. Please state whether these are included in σ_s or justify their neglect.","section":"§5.2"},{"comment":"The FADC rise-time cut removes events outside the 1st and 99th percentiles, corresponding to an efficiency of about 98%, but Eq. (5.3) lists only ε_veto and ε_S. The total efficiencies in Table 3 suggest this factor is included; please add it to the formula or explain where it enters.","section":"§4 and Eq. (5.3)"},{"comment":"The color scale in Fig. 9 is labeled \"Axion flux [a.u.]\" but the colorbar ticks are shown in scientific notation with a numeric scale; please clarify the units or remove the numeric ticks to avoid confusion.","section":"§6, Fig. 9"},{"comment":"The observed limit is stronger than the median expected limit; reporting the probability of obtaining a limit this strong under the background-only hypothesis would help quantify the significance of the downward fluctuation.","section":"§6"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid experimental contribution and the analysis appears careful, but the mass-coherence issue directly affects the headline number and must be fixed before publication. The likelihood formula error is also in a central equation and should be corrected. I do not see indications of an integrity problem; the revision is straightforward in principle but essential."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Readable, honest experimental paper from CAST: the first physics result of the 7-GridPix detector. The detector work is real and well documented: 160 h of tracking, background at 1.06e-5 keV^-1 cm^-2 s^-1, MLP plus veto chain, a blind expected-limit protocol, 55Fe and multi-energy calibrations, and raytracing checked at PANTER within 2.6%. No red flag in the candidate set: no events land near the axion image. The new bound g_ae*g_aγ < 7.35e-23 GeV^-1 improves the 2013 CAST g_ae limit by about 10%. That part is credible.\n\nThe soft spot is the mass range. Eq. 5.2 and the vacuum conversion probability in Sec. 5.1 have no m_a dependence, yet the abstract and Sec. 7 quote the limit \"for axion masses below 10 meV.\" At 10 meV, qL at the solar g_ae spectrum peak (~1 keV) is O(2), so the coherence form factor suppresses the signal by about 0.6, and far more at lower energies. Ignoring that makes the quoted limit too strong by roughly the square-root inverse, enough to erase the claimed improvement over CAST 2013 near 10 meV. The paper needs a mass-dependent exclusion curve, or at minimum a clear statement that the quoted number is the m_a -> 0 limit with the mass range chosen where the form factor is negligible. As written, the central claim's domain is not established.\n\nThe other assumptions — background taken outside tracking, 86% outer-chip veto from random-coincidence bootstrap — are standard and reasonably handled with nuisance parameters; I treat those as minor. The rest of the analysis is clean.\n\nWho gets value: the axion community, especially IAXO detector development. It deserves a serious referee and should be publishable after the mass-dependent conversion is fixed or the claim is restricted. Recommend major revision, not desk reject.","headline":"A careful blinded helioscope search with a new GridPix detector gives a modest best limit, but the quoted mass range is too broad because the massless conversion probability is used for axion masses up to 10 meV.","tokens_in":15213,"tokens_out":3899,"would_cite":true,"duration_ms":41666,"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 reports a search for solar axions at CAST with a seven-chip GridPix detector and finds no signal excess, setting a new upper limit on the product of the axion-electron and axion-photon couplings of $g_{ae}\\,g_{a\\gamma} <…","keywords":["axion","axion-electron coupling","helioscope","GridPix","Timepix","solar axion","background suppression","rare event search"],"falsifier":"Reanalyze the published tracking candidates with the raytraced axion-image region masked out; if the resulting limit moves by more than the quoted uncertainties, the background interpolation is driving the claimed improvement. Conversely, an excess of low-energy clusters whose positions follow the expected axion image would directly falsify the no-signal conclusion.","tokens_in":13824,"feed_emoji":"🔭","tokens_out":7143,"duration_ms":75541,"temperature":0.7,"pith_summary":"This paper reports a search for solar axions using a new seven-chip GridPix detector at the CAST helioscope. Axions produced in the Sun through their coupling to electrons could reconvert into X-rays in the experiment's magnet, and the detector was built to catch those X-rays with very low background. After about 160 hours of solar tracking, no signal excess was seen. The collaboration therefore sets a new upper limit of $g_{ae}\\,g_{a\\gamma} < 7.35\\times10^{-23}\\,\\mathrm{GeV}^{-1}$ at 95% confidence for axion masses below 10 meV, improving the previous best helioscope limit by about 10%. This demonstrates that GridPix technology, with its single-electron sensitivity and integrated vetoes, is competitive for rare-event searches.","feed_headline":"CAST's new GridPix detector cuts axion-electron limit by 10%","feed_subtitle":"Machine-learning cuts and vetoes make this the strongest helioscope bound on axion couplings.","key_machinery":"The central object is the 7-GridPix detector: a gas ionization chamber read out by seven Timepix pixel chips with integrated micromesh amplification grids, giving single-electron sensitivity and per-pixel spatial resolution. The central chip sits at the focal plane of the X-ray telescope, while the six surrounding chips act as an active veto. Background suppression combines a multilayer perceptron classifier trained on simulated X-ray events and real background clusters, a scintillator veto, a longitudinal pulse-shape cut based on the flash ADC, and a pointing veto using the outer chips. The expected axion image comes from a dedicated raytracing simulation, and the final limit is set with an unbinned Bayesian likelihood that is marginalized over nuisance parameters using a Metropolis-Hastings Markov chain Monte Carlo.","core_discovery":"The central claim is that a search for solar axions produced through the axion-electron coupling, using a 7-GridPix detector at CAST, finds no excess above background and therefore sets a new upper limit $g_{ae}\\,g_{a\\gamma} < 7.35\\times10^{-23}\\,\\mathrm{GeV}^{-1}$ at 95% confidence for axion masses below 10 meV. This improves on the previous CAST bound of $8.1\\times10^{-23}\\,\\mathrm{GeV}^{-1}$ from 2013 by about 10%. The same tracking data also yield a complementary limit on the axion-photon coupling alone, $g_{a\\gamma} < 9.0\\times10^{-11}\\,\\mathrm{GeV}^{-1}$, which does not surpass CAST's best axion-photon limit but adds independent model constraints.","pith_inferences":["A direct consistency check would be to re-run the limit calculation with the raytraced axion-image region masked out; if the limit changes by much more than the quoted uncertainties, the reported improvement depends on the background interpolation in the signal region.","The 86% outer-chip veto efficiency, estimated by bootstrapping random coincidences, suggests that engineering a veto with higher efficiency or a lower random-coincidence rate would directly translate into a stronger $g_{ae}\\,g_{a\\gamma}$ limit with the same exposure.","The ultra-thin silicon-nitride window gives improved transmission below 3 keV, which is exactly where electron-coupled solar axions peak; the same detector concept could be pointed at other low-energy solar or dark-matter signals that are invisible to detectors with thicker windows."],"forward_implications":["The best helioscope constraint on the product $g_{ae}\\,g_{a\\gamma}$ is now this $7.35\\times10^{-23}\\,\\mathrm{GeV}^{-1}$ limit, replacing the 2013 CAST value as the reference for this coupling product.","The analysis pipeline is deliberately generic in the assumed signal spectrum, so the same detector data and background model can be reused for other solar axion production channels without redesigning the limit calculation.","The detector's background rate of about $1.06\\times10^{-5}\\,\\mathrm{keV}^{-1}\\,\\mathrm{cm}^{-2}\\,\\mathrm{s}^{-1}$ at roughly 80% signal efficiency in the 0.2-8 keV range puts GridPix on par with the best CAST readout systems for low-background X-ray searches.","The derived $g_{a\\gamma}$ limit, while not competitive alone with CAST's best axion-photon bound, is already combined with newer Micromegas data in the collaboration's ongoing effort to strengthen helioscope constraints.","With only 160 hours of tracking time, the improvement over the previous limit indicates that most of the gain comes from lower background and better signal efficiency rather than longer exposure."],"supporting_citations":[{"why":"Supplies the previous best helioscope limit on $g_{ae}\\,g_{a\\gamma}$ that this analysis improves.","marker":"[21]"},{"why":"Defines the solar axion flux from the axion-electron coupling used to model the expected signal.","marker":"[15]"},{"why":"Establishes the CAST helioscope and its data-taking framework for this search.","marker":"[9]"},{"why":"Introduces the pixel readout chip technology underlying the GridPix detector.","marker":"[22]"},{"why":"Provides the raytracing simulation that gives the expected axion image used in the likelihood.","marker":"[32]"},{"why":"Supplies the Metropolis-Hastings algorithm used to marginalize the Bayesian likelihood.","marker":"[45]"}],"fun_headline_variants":["GridPix lowers CAST's axion-electron limit by 10%","New CAST limit on axion-electron coupling from GridPix","GridPix detector improves CAST axion-electron limit","CAST's GridPix tightens axion-electron bound","GridPix at CAST shaves 10% off axion-electron limit"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result assumes that the background model built from data taken while the telescope was not tracking the Sun, together with the 86% veto efficiency estimated from chance coincidences, remains valid during the hours the telescope points at the Sun.","fun_headline_variants_meta":{"raw":{"variants":["GridPix lowers CAST's axion-electron limit by 10%","New CAST limit on axion-electron coupling from GridPix","GridPix detector improves CAST axion-electron limit","CAST's GridPix tightens axion-electron bound","GridPix at CAST shaves 10% off axion-electron limit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000666,"raw_usage":{"total_tokens":3087,"prompt_tokens":1044,"completion_tokens":2043,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":660,"completion_tokens_details":{"reasoning_tokens":1955}},"tokens_in":660,"tokens_out":2043,"duration_ms":15959,"temperature":1.0,"reasoning_tokens":1955,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:53:10.834662+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Reanalyze the published tracking candidates with the raytraced axion-image region masked out; if the resulting limit moves by more than the quoted uncertainties, the background interpolation is driving the claimed improvement. Conversely, an excess of low-energy clusters whose positions follow the expected axion image would directly falsify the no-signal conclusion.","supporting_citations":[{"cited_title":"Barth, A","cited_arxiv_id":null,"evidence_quote":"Supplies the previous best helioscope limit on $g_{ae}\\,g_{a\\gamma}$ that this analysis improves."},{"cited_title":"Llopart, M","cited_arxiv_id":null,"evidence_quote":"Introduces the pixel readout chip technology underlying the GridPix detector."},{"cited_title":"TrAXer - an interactive real-time x-ray raytracer","cited_arxiv_id":null,"evidence_quote":"Provides the raytracing simulation that gives the expected axion image used in the likelihood."},{"cited_title":"Metropolis, A.W","cited_arxiv_id":null,"evidence_quote":"Supplies the Metropolis-Hastings algorithm used to marginalize the Bayesian likelihood."}],"review_version":1}