{"id":"39604ca1-4f4a-4bfe-9656-44c454afbf66","arxiv_id":"1908.08805","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Simulations show gamma-ray telescopes could distinguish preshower-induced air showers from cosmic-ray background using boosted decision trees, but expected event rates are tiny.","lead":"The paper proposes using the Cherenkov Telescope Array (CTA) to search for air showers caused by ultra-high-energy photons that split in Earth's magnetic field. This could open a new observing window for the most energetic photons in the universe, though the predicted event rates are very low.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"BDT separation may be an energy artifact: preshower sample at 40 EeV is compared to protons capped at 3 PeV, so sigma = 29.85 does not establish photon/hadron discrimination.","rationale":"The reader's weakest assumption was the fixed geometry and narrow parameter space. The present review identifies a more specific and potentially more damaging issue: the classifier background is energy-mismatched, so the reported separation and significance do not yet establish preshower-specific discrimination. The paper's own exclusion of protons above 3 PeV is internally reasonable for event-rate estimation, because such protons are rare in 30 hours at this small aperture, but using that exclusion to define the classifier's background conflates 'rare in 30 hours' with 'physically distinguishable from a 40 EeV photon shower.' The proposed same-energy proton simulation directly tests whether the BDT separates by primary type rather than by total energy. The fixed-geometry limitation remains secondary: the aperture and event rates are computed for one arrival direction, so the diffuse-source numbers in Table 1 are not an all-sky estimate. The required corrections are well-defined and the qualitative feasibility could still be rescued, so a conditional verdict rather than accept or reject is appropriate.","tokens_in":6504,"tokens_out":9632,"duration_ms":113949,"concrete_test":"Simulate a proton background sample at E = 40 EeV, with additional samples at 10 EeV and 80 EeV if feasible, for the same zenith angle of 80 degrees, azimuth toward geomagnetic north, and CTA-La Palma detector settings; train and test the same BDT on this energy-matched sample together with the existing preshower sample. Report the CR contamination at the preshower efficiency epsilon = 0.926 and the BDT score separation. If the contamination is much larger than 3.8%, the claimed discrimination is primarily an energy-mismatch artifact and the central claim fails; if contamination remains about 4 percent, the energy-matching concern is refuted.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section 2.1 fixes the preshower primary energy at E = 40 EeV, while the CR background is explicitly restricted to proton energies between 10^4 GeV and 3 x 10^6 GeV, i.e. a factor greater than 1.3 x 10^4 below the signal energy. Because the Hillas size and related image parameters scale steeply with primary energy, the clean BDT separation in Section 3.1 and the headline sigma = 29.85 from Eq. (3.1) are largely a high-energy selection effect; they do not demonstrate that a preshower-induced EAS can be told apart from a proton-initiated EAS of the same energy. The paper's justification for the 3 PeV cap is that higher-energy protons give fewer than one expected event in 30 hours, but that only addresses the expected background rate, not the physical claim of CRE/hadron discrimination. In addition, Eq. (3.1) defines S and B using efficiencies multiplied by counts in the simulated test sample, so the quoted significance grows with the arbitrary number of simulated events and is not an expected detection significance. The flux-weighted yields in Table 1 are below 1.5 x 10^-6 events even for the most optimistic model, so the paper demonstrates classifier separation on synthetic samples rather than observability.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents a simulation-based feasibility study of detecting ultra-high-energy (UHE) photons through the preshower effect with the future CTA-La Palma array. It uses the PRESHOWER code to generate electron-positron cascades from a 40 EeV photon converting in the geomagnetic field, CORSIKA for air-shower development, and sim_telarray for the CTA camera response, with a diffuse source simulated by a 5-degree viewing cone and a nearly-horizontal arrival direction (theta = 80 deg, phi = 180 deg). A TMVA boosted-decision-tree classifier is trained on four Hillas parameters to separate preshower-induced events from a proton background simulated in the 10^4 to 3x10^6 GeV range. The paper reports a maximum classification significance of sigma = 29.85 (epsilon_presh = 0.926, epsilon_CR = 0.038), an effective aperture of 0.05 km^2 for the diffuse source, and expected event numbers below 1.5x10^-6 in 30 hours for SHDM, Auger-limit, and GZK flux models. The authors conclude that preshower events may be distinguished from the cosmic-ray background but that very few events are expected.","tokens_in":6812,"tokens_out":6413,"duration_ms":66704,"significance":"If the discrimination claim held, this would be a useful first step for a novel UHE-photon search channel: the simulation chain is built from established tools (PRESHOWER, CORSIKA, sim_telarray, TMVA), the observation geometry is concrete, and the paper is honest about the very low predicted event rates and about the large flux required for a single event. The central limitation is that the headline separation does not currently support the physical claim of preshower/hadron discrimination, because the signal and background samples differ by about 1.3x10^4 in primary energy and because the quoted significance depends on an arbitrary simulated sample size. With energy-matched samples and a properly defined detection significance, the classification result could be made meaningful; as it stands, the paper demonstrates classifier separation on synthetic samples rather than observability.","major_comments":[{"comment":"The preshower primary is fixed at E = 40 EeV, while the proton background is restricted to 10^4 to 3x10^6 GeV, a factor of about 1.3x10^4 lower in energy. Because the Hillas size and related image parameters scale steeply with primary energy, the separation shown in Figs. 2 and 3 and the quoted sigma = 29.85 are largely an energy selection effect and do not demonstrate that preshower-induced air showers can be distinguished from proton-initiated showers of the same energy. The 3 PeV cap is justified by the expected background count falling below one event in 30 hours, but that addresses the expected rate, not the classifier's physical discrimination capability. The authors should either simulate a same-energy hadronic background or clearly state that the classifier is intended for a background population with a different energy distribution and quantify the resulting selection bias.","section":"Section 2.1 and Section 3.1"},{"comment":"The definitions of S and B in Eq. (3.1) are circular or ambiguous: S is described as the preshower efficiency multiplied by the number of true positives, and B as the background efficiency multiplied by the number of false negatives. True positives already encode the efficiency, and false negatives are not the background events that pass the cut. More importantly, because S and B are proportional to the total number of simulated test events, sigma = 29.85 depends on an arbitrary simulation sample size and is not an expected detection significance for a 30-hour observation. The authors should define S and B as counts after the cut (or use a standard Poisson/Li-Ma significance), report the test-sample sizes, and, if a detection significance is intended, convert the classifier efficiencies into expected signal and background counts for the stated exposure.","section":"Section 3.1, Eq. (3.1)"},{"comment":"All aperture and event-rate numbers are derived for a single arrival geometry (E = 40 EeV, theta = 80 deg, phi = 180 deg, fixed cone and impact radius). The preshower conversion probability depends strongly on the geomagnetic field along the trajectory, so the quoted A_eff and N_preshw values do not generalize to an all-sky diffuse flux without an explicit integration over arrival directions and energies, or at least a demonstration that the dependence is weak. In addition, the expected numbers in Table 1 are far below one even for the most optimistic SHDM model (1.03x10^-4 for the point-source case and 1.49x10^-6 for the diffuse case in 30 hours). The conclusions should therefore state unambiguously that no event is expected in 30 hours under current flux models and that sigma = 29.85 is a classification significance on simulated samples, not an observability claim.","section":"Section 3.2 and Table 1"}],"minor_comments":[{"comment":"The aperture is quoted in km^2, but Eq. (3.2) includes the solid angle Omega; for the diffuse case Omega = 2pi(1-cos 5 deg) is approximately 0.024 sr, so A_eff has units of km^2 sr. Please clarify whether the quoted 0.05 km^2 and 3.40 km^2 include this solid-angle factor and give Omega explicitly for the point-source value.","section":"Section 3.2"},{"comment":"The phrase 'false negatives' for CR background events classified as preshower is nonstandard; these events are usually called false positives or the background efficiency. Please correct the terminology to avoid confusion with the definitions in Eq. (3.1).","section":"Section 3.1"},{"comment":"The symbol d_int is used before its meaning is fully specified; please define it explicitly in the text where it first appears.","section":"Introduction"},{"comment":"The claim that the 3x10^6 GeV cap follows from an expected CR count below one event in 30 hours should be accompanied by a formula or reference for the aperture and spectrum used in that estimate, since it is not derivable from the information given.","section":"Section 2.1"}],"recommendation":"major_revision","confidential_remarks":"This is a proceedings-style feasibility study with an interesting idea and an honest discussion of the very low predicted event rates. The main concerns are fixable: matching the energy of the hadronic background to the signal, redefining the significance statistic in terms of expected counts, and integrating over the relevant geometry. I would not reject the paper, but the headline discrimination claim needs substantial revision before it can be accepted. The use of the PRESHOWER code, which is authored in part by members of this collaboration, is not by itself problematic because the code is a published simulation tool rather than a fitted result."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a genuine attempt to open a new observational channel: using CTA-La Palma in high-zenith mode to catch air showers induced by the geomagnetic preshower effect from UHE photons. That idea is worth taking seriously, and the authors use an established simulation chain (PRESHOWER, CORSIKA, sim_telarray) and a standard TMVA BDT. They also deserve credit for being upfront about the tiny expected event rates under realistic flux models, which essentially makes this a proof-of-concept rather than a detection claim.\n\nThe trouble is the main quantitative claim. The preshower sample is at 40 EeV, while the proton background is explicitly capped at 3x10^6 GeV (3 PeV). Given that Hillas parameters scale steeply with primary energy, the clean BDT separation and the reported sigma = 29.85 mostly tell you that 40 EeV events look different from few-PeV events, not that preshower-induced cascades can be told apart from hadronic cascades of the same energy. The cap is justified on expected background rate grounds, but that does not address the physical discrimination question. If the actual search region is around 40 EeV, then same-energy protons are the relevant background for the classifier.\n\nThere is also a real problem with Eq. (3.1). As written, S and B are efficiencies multiplied by counts in the simulated test sample, so the significance grows with the arbitrary number of simulated events. The reported 29.85 is therefore not an expected detection significance; it is a classifier-quality metric that depends on sample size. The paper does not give the test sample sizes, so the number is essentially uninterpretable. This is fixable, but it should be flagged clearly before the result is used.\n\nThe study also covers a single geometry (40 EeV, zenith 80°, geomagnetic north, one viewing cone) with no systematic uncertainties. That is acceptable for a proof-of-concept, but it limits the generality of the aperture and event-rate numbers.\n\nMy overall reading: the idea is new and the simulation work is honest, but the central claim that preshowers can be effectively distinguished is not supported as stated. The flaws are addressable — rerun the background at matched energy, replace the significance formula with a proper expected-significance calculation, add a few energy/geometry points — and in that sense the paper deserves serious referee attention rather than a desk reject. I would not cite the current version for its quantitative results, but I would keep it on the radar as a promising direction. For a reading group, it is a useful example of how easy it is to overstate classifier separation when signal and background are not energy-matched.\n\nRecommendation: send it to peer review, but with a clear request for same-energy background simulations and a corrected significance calculation.","headline":"Novel but flawed proof-of-concept: the CTA preshower search is promising, but the headline BDT separation compares 40 EeV photons against <=3 PeV protons, so the significance is largely an energy artifact and the quoted sigma is not interpretable.","tokens_in":7407,"tokens_out":2923,"would_cite":false,"duration_ms":31012,"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":"A simulation shows that preshower air showers from 40 EeV photons can be separated from the cosmic-ray background at a significance of 29.85 using next-generation gamma-ray telescopes.","keywords":["preshower effect","ultra-high-energy photons","gamma-ray astronomy","Cherenkov telescopes","boosted decision trees","cosmic-ray background","geomagnetic field","air showers"],"falsifier":"Train the same classifier on a background sample weighted to the true cosmic-ray intensity (the $E^{-2.7}$ proton spectrum at the chosen geometry and observation time) rather than on equal-sized Monte Carlo samples; if the signal efficiency at a fixed background rate drops below about 5σ significance, the claimed discrimination is not robust enough for real observations.","tokens_in":6332,"feed_emoji":"🔭","tokens_out":8352,"duration_ms":83769,"temperature":0.7,"pith_summary":"This paper argues that a screening effect in Earth's magnetic field could turn ultra-high-energy photons into detectable air showers, and that a next-generation gamma-ray telescope array could pick these out from the overwhelming cosmic-ray background. The authors simulate 40 EeV photons arriving from the geomagnetic north at an 80-degree zenith angle, where the conversion probability is highest. A boosted decision tree trained on Cherenkov image shapes separates simulated preshower showers from proton-initiated background showers with a maximum significance of 29.85, identifying 92.6% of preshower events while admitting 3.8% of background events. A point-source version performs even better. The catch is that known production models predict so few ultra-high-energy photons that a 30-hour run would most likely see zero events.","feed_headline":"Preshower photon events survive a cosmic-ray filter at sigma 30","feed_subtitle":"Simulated telescope images separate preshower air showers from hadrons with 92.6% efficiency and 3.8% contamination.","key_machinery":"The load-bearing mechanism is the preshower effect: a 40 EeV photon crossing the geomagnetic field has a conversion probability (here 0.67) of turning into an electron-positron pair far above the atmosphere, and the pair's bremsstrahlung emission produces a collimated cosmic-ray ensemble that later strikes the atmosphere as a single air shower. The classification is carried by a boosted-decision-tree classifier applied to four Hillas image parameters (size, width, length, distance). The high-zenith, north-pointing geometry is what makes the separation work, because the Cherenkov images at 80° zenith are dominated by muons, and the muon content differs between photon-initiated and hadron-initiated showers.","core_discovery":"The central claim is that preshower-induced air showers, produced when an ultra-high-energy photon converts to an electron-positron pair in the geomagnetic field and the pair radiates a cascade of bremsstrahlung photons, leave a distinguishable imprint in the cameras of CTA-La Palma. Using air-shower simulations combined with a detector-response simulation, the authors find that high-zenith (80°) observations pointed toward geomagnetic north, where the field is stronger, produce images dominated by muon Cherenkov light, and that the image-shape parameters (size, width, length, distance) separate preshower events from proton background. The multivariate classifier reaches $\\sigma = 29.85$, $\\varepsilon_{\\mathrm{preshw}} = 0.926$, $\\varepsilon_{\\mathrm{CR}} = 0.038$ for a diffuse source and $\\varepsilon_{\\mathrm{preshw}} = 0.982$, $\\varepsilon_{\\mathrm{CR}} = 0.022$ for a point source. The effective aperture is $0.05\\,\\mathrm{km}^2$ (diffuse) and $3.40\\,\\mathrm{km}^2$ (point source); with the super-heavy dark matter, GZK, and current upper-limit photon fluxes, the expected number of events in 30 hours ranges from $10^{-8}$ to $10^{-4}$, and a flux of $8.73 \\times 10^{3}\\,\\mathrm{km}^{-2}\\,\\mathrm{yr}^{-1}\\,\\mathrm{sr}^{-1}$ would be needed to obtain a single event.","pith_inferences":["An extension not explored in the paper would scan photon energies and arrival directions other than 40 EeV and geomagnetic north; conversion probability and Cherenkov image morphology both vary strongly with geometry, so the claimed separation could map an acceptance-weighted region of the sky.","Because the reported significance comes from Monte Carlo samples with balanced signal and background counts, real-data application would need re-optimisation against a vastly more numerous cosmic-ray background; the quoted efficiencies and significance are best-case values.","If the muon-isolation mechanism is correct, the same high-zenith strategy may transfer to other imaging atmospheric Cherenkov arrays and to lower photon energies, where the preshower conversion probability is smaller but the background environment also differs."],"forward_implications":["If preshower events are real and pass the boosted-decision-tree selection, gamma-ray telescopes would provide a new observational channel for GZK photons and super-heavy dark-matter decay products.","A 30-hour observation run with the specified pointing (80° zenith, toward geomagnetic north) would, under current flux models, expect far fewer than one event, so the practical outcome would be an improved upper limit rather than a detection.","The point-source scenario has an effective area roughly 70 times larger than the diffuse scenario, making targeted searches of candidate sources a more promising route.","Choosing a different cut on the classifier score can eliminate cosmic-ray background entirely at the cost of signal efficiency, enabling background-free searches if the photon flux is sufficiently high."],"supporting_citations":[{"why":"Introduces the preshower effect that screens ultra-high-energy photons in electromagnetic fields.","marker":"[1]"},{"why":"Supplies the preshower simulation algorithm used to compute photon conversion and pair bremsstrahlung in the geomagnetic field.","marker":"[4]"},{"why":"Provides the air-shower simulation engine used to generate particle cascades in the atmosphere.","marker":"[6]"},{"why":"Describes the detector simulation used to model the Cherenkov telescope camera response and trigger.","marker":"[9]"},{"why":"Shows that high-zenith observations isolate the muon component and enable gamma-hadron separation, the basis of the chosen observing mode.","marker":"[11]"},{"why":"Provides the boosted-decision-tree classifier used for the multivariate separation.","marker":"[13]"},{"why":"Gives a super-heavy dark-matter photon flux model used to estimate expected event counts.","marker":"[16]"},{"why":"Gives a GZK photon flux model used to estimate expected event counts.","marker":"[17]"},{"why":"Provides the current experimental upper limits on the 40 EeV photon flux used as a benchmark.","marker":"[18]"}],"fun_headline_variants":["Preshower air showers leave distinct imprints in CTA cameras","Simulated preshower events separate from protons at sigma 30","Gamma-ray preshower filter reaches 92.6% efficiency in simulations","CTA-La Palma simulations spot preshower showers vs proton background","Preshower effect yields distinguishable air showers for gamma-ray telescopes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The results rest on one simulated geometry — a 40 EeV photon arriving at 80° zenith from the geomagnetic north — and on the assumption that the simulated Cherenkov image shapes match what the telescope array will actually record in that configuration.","fun_headline_variants_meta":{"raw":{"variants":["Preshower air showers leave distinct imprints in CTA cameras","Simulated preshower events separate from protons at sigma 30","Gamma-ray preshower filter reaches 92.6% efficiency in simulations","CTA-La Palma simulations spot preshower showers vs proton background","Preshower effect yields distinguishable air showers for gamma-ray telescopes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000233,"raw_usage":{"total_tokens":1558,"prompt_tokens":1072,"completion_tokens":486,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":688,"completion_tokens_details":{"reasoning_tokens":398}},"tokens_in":688,"tokens_out":486,"duration_ms":5834,"temperature":1.0,"reasoning_tokens":398,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:28:19.976078+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Train the same classifier on a background sample weighted to the true cosmic-ray intensity (the $E^{-2.7}$ proton spectrum at the chosen geometry and observation time) rather than on equal-sized Monte Carlo samples; if the signal efficiency at a fixed background rate drops below about 5σ significance, the claimed discrimination is not robust enough for real observations.","supporting_citations":[{"cited_title":"McBreen, C","cited_arxiv_id":null,"evidence_quote":"Introduces the preshower effect that screens ultra-high-energy photons in electromagnetic fields."},{"cited_title":"Heck, et al., FZKA Report 6019 (1998)","cited_arxiv_id":null,"evidence_quote":"Provides the air-shower simulation engine used to generate particle cascades in the atmosphere."},{"cited_title":"Cosmic ray composition measurements and cosmic ray background free gamma-ray observations with Cherenkov telescopes","cited_arxiv_id":"1610.01794","evidence_quote":"Shows that high-zenith observations isolate the muon component and enable gamma-hadron separation, the basis of the chosen observing mode."},{"cited_title":"Ellis, and others, Phys","cited_arxiv_id":null,"evidence_quote":"Gives a super-heavy dark-matter photon flux model used to estimate expected event counts."},{"cited_title":"Gelmini, O","cited_arxiv_id":null,"evidence_quote":"Gives a GZK photon flux model used to estimate expected event counts."},{"cited_title":"The Pierre Auger Observatory: Contributions to the 34th International Cosmic Ray Conference (ICRC 2015)","cited_arxiv_id":"1509.03732","evidence_quote":"Provides the current experimental upper limits on the 40 EeV photon flux used as a benchmark."}],"review_version":1}