{"id":"e6b8b7d5-cd6b-4c61-803c-86a9b2e21828","arxiv_id":"2412.20242","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"A BDT-based simulation predicts 5 sigma sensitivity to pp to th with h to gamma gamma at the HL-LHC for tan(beta)=1, cos(alpha-beta)=0.1, chi_tc=5 after about 2700 inverse femtobarns.","lead":"This paper simulates a rare collision process at the proposed High-Luminosity LHC: production of a Higgs boson together with a top quark through a flavor-changing interaction allowed in the Two-Higgs Doublet Model type III. It reports that a machine-learning classifier could make the process discoverable (5 sigma) after about 2700 inverse femtobarns for optimistic model parameters.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Table III violates the χ_tc^2 scaling required by Eq. (22): S1/S2 cross-sections rise by only ~2.5–3.5× for χ=1→5 instead of 25×, so the quoted luminosities inherit an unverified MC normalization.","rationale":"The reader's REJECT is well-founded. This is a standard phenomenological projection whose numerical outcome depends entirely on the MC event normalization. The decisive weak point is Table III, which contradicts the paper's own coupling formula in Eq. (22): for a fixed scenario the only χ_tc-dependent ingredient is g_tch, so the amplitude is linear in χ_tc and the inclusive cross-section must be quadratic. The tabulated increments are 2.5× and 3.5× for χ=1→5, i.e., roughly the √χ behavior one would get if the implemented coupling were accidentally proportional to √χ_tc or if event weights were not squared. This also explains the surprising χ=3 result: with honest χ² scaling, χ=3 could not reach 4.4σ at 3000 fb^-1 when χ=5 reaches 5σ at 2700 fb^-1. I therefore agree with the reader's weakest-assumption identification. Additional criticisms, such as the vacuous KS overtraining check, are secondary and do not need to be invoked. The proposed concrete test—re-running the named generator chain at three χ values and comparing inclusive cross-section ratios—will settle whether the error is a typo in Table III (in which case rescaling might partly restore the analysis) or a model-file bug (in which case the forecasts are invalid). Until that check is done, the verdict should remain REJECT.","tokens_in":12005,"tokens_out":17401,"duration_ms":184171,"concrete_test":"Ask the authors to provide the FeynRules/UFO model files and run MadGraph5_aMC@NLO for pp→th+X at √s=13 TeV with fixed tanβ=1, cos(α−β)=0.1, for χ_tc = 1, 3, and 5, reporting inclusive cross-sections before any selection or BDT. If the ratios are 1:9:25, Table III is wrong and all significance curves must be rescaled; if the ratios are instead ~1:?:2.5 or ~1:?:3.5, the implemented model has the wrong χ-dependence and the quoted luminosities are invalid. As a cross-check, insert an arbitrary htc coupling into a minimal UFO and verify that the squared amplitude is proportional to the square of that coupling.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eq. (22) makes g_tch linear in χ_tc, and the quoted signal has a single FCNC vertex, so at fixed tanβ and cos(α−β) the cross-section σ(pp→th+X) must scale as χ_tc^2. Table III gives S1: 0.01 fb (χ=1) → 0.025 fb (χ=5), a factor of 2.5 instead of 25, and S2: 0.004 fb → 0.014 fb, a factor of 3.5. This is not a harmless typo: the χ=3 HL-LHC point inherits the same normalization. If the generator were correct, σ(χ=3)/σ(χ=5) = (3/5)^2 = 0.36; starting from 5σ at 2700 fb^-1 for χ=5, the χ=3 significance at 3000 fb^-1 would be about 2σ, not 4.4σ. The reported χ=5 and χ=3 significances are instead mutually consistent with σ growing roughly as χ^0.5–0.6, which points to a model-file or normalization error rather than a simple table misprint. Because Fig. 6 and the abstract's luminosity projections are built on these event samples, the central quantitative claim is unsupported until the generator output is verified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies single Higgs boson production in association with a top quark, pp -> th + X, within the Two-Higgs-Doublet Model of type III, focusing on the final state t -> l nu b and h -> gamma gamma. The authors extract the flavor-changing coupling g_tch from the model, constrain the parameter space using the CMS upper limit on BR(t -> ch) and the projected HL-LHC limit, and perform a Monte Carlo analysis with FeynRules, MadGraph5, Pythia8, Delphes3, and a Boosted Decision Tree classifier. They claim that for tan(beta)=1, cos(alpha-beta)=0.1, and chi_tc=5, a 5-sigma signal can be reached at L_int >= 2700 fb^-1 with 5% systematic uncertainty, and that for chi_tc=3 the significance is about 4.4 at 3000 fb^-1 under the HL-LHC projection on BR(t -> ch).","tokens_in":12254,"tokens_out":8296,"duration_ms":89245,"significance":"If correct, the paper would provide a concrete and testable HL-LHC prospect for a direct flavor-changing neutral scalar interaction involving the top quark and the 125 GeV Higgs boson. The use of both the current CMS bound and the projected HL-LHC bound on BR(t -> ch) as anchors is appropriate, and the paper makes the BDT-based analysis reproducible in broad outline. However, the central quantitative results are internally inconsistent: the signal cross-sections in Table III do not follow the chi_tc^2 scaling required by Eq. (22), and the reported chi_tc=3 significance cannot be derived from the chi_tc=5 results under the paper's own model. These issues affect the abstract, Fig. 6, and the conclusions, so the central claim is unsupported until the Monte Carlo normalization is verified and the numbers are corrected.","major_comments":[{"comment":"Eq. (22) defines g_tch as linear in the parameter chi_tc, and each diagram contributing to pp -> th + X contains one flavor-changing htc vertex, so for fixed tan(beta) and cos(alpha-beta) the signal cross-section must scale as chi_tc^2. Table III reports sigma(S1, chi_tc=5)/sigma(S1, chi_tc=1) = 0.025/0.01 = 2.5 and sigma(S2, chi_tc=5)/sigma(S2, chi_tc=1) = 0.014/0.004 = 3.5, both far below the required factor of 25. This is not a rounding issue; it indicates that the Monte Carlo normalization or model implementation does not follow Eq. (22). Because the luminosity projections in Fig. 6 and the abstract are built from these event samples, the absolute cross-sections and the derived significances are not credible without a corrected generation and an explicit check of the chi_tc^2 scaling.","section":"Eq. (22) and Table III"},{"comment":"The text introduces chi_tc=3 for scenario S1 to satisfy the HL-LHC projection on BR(t -> ch), but Table III does not provide a cross-section for this value. Under the chi_tc^2 scaling that follows from Eq. (22), one would have sigma(chi_tc=3) = (3/5)^2 sigma(chi_tc=5) = 0.36 sigma(chi_tc=5). Starting from the paper's own chi_tc=5 result of a 5-sigma signal at L_int ~ 2700 fb^-1 with kappa=5%, the chi_tc=3 significance at 3000 fb^-1 should be approximately 2 sigma, not the quoted 4.4 (or 4.2 with systematics). The quoted values are part of the abstract and conclusions, so this inconsistency must be resolved by reporting the actual chi_tc=3 cross-section and recomputing the significance.","section":"Sec. III and Fig. 6"}],"minor_comments":[{"comment":"Equation (21) is typeset incompletely: the line breaks after \"-LY = - g/(2MW)\" and several parentheses are unbalanced, which makes it difficult to verify the extraction of g_tch in Eq. (22).","section":"Eq. (21)"},{"comment":"In Eq. (16), the expression for B_f contains a repeated factor \"(r2 + r2 - 1)\" where one of the factors presumably should involve r1 or r3; please correct this typographical error.","section":"Eq. (16)"},{"comment":"Table III lists only chi_tc=1 and chi_tc=5, although chi_tc=3 is used later in Sec. III and in the abstract; a row for chi_tc=3 with its uncertainty should be added.","section":"Table III"},{"comment":"The caption of Fig. 6 does not state which curves correspond to kappa=0 and kappa=5% for both scenarios; the reader must infer this from the text.","section":"Fig. 6 caption"}],"recommendation":"major_revision","confidential_remarks":"The scaling inconsistency in Table III is severe and should be treated as the primary issue in review. A rerun of the Monte Carlo generation and an independent check of sigma(chi_tc=5)/sigma(chi_tc=1) are necessary before the quantitative claims can be trusted. I do not see a circularity problem in using BR(t -> ch) limits as external constraints; the issue is purely the internal consistency of the reported numbers."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nShort version: the paper is a standard but competent collider-sensitivity study, and the model setup is fine. The problem is the central numbers. Eq. (22) makes g_tch linear in chi_tc, and the signal has one FCNC vertex, so sigma(pp->th+X) must scale as chi_tc^2. Table III gives S1: 0.01 fb (chi=1) to 0.025 fb (chi=5), a factor of 2.5, and S2: 0.004 to 0.014, a factor of 3.5. These are not 25-fold increases. That is not a cosmetic typo: the chi=3 point used for the HL-LHC projection inherits the same normalization. Starting from 5 sigma at 2700/fb for chi=5, chi=3 at 3000/fb should give roughly 2 sigma, not 4.4. The reported significances are mutually consistent with sigma growing like chi^0.5–0.6, which points to a model-file or normalization error. Until the generator output is checked, the abstract's headline results should not be taken at face value.\n\nWhat is genuinely good: the paper is transparent about its scenarios, it uses an external constraint (CMS and projected HL-LHC limits on BR(t->ch)) as an anchor, and it gives a concrete 2HDM-III four-zero-texture benchmark for a process that has been studied before only in EFT or other frameworks. The BDT analysis and the comparison with straight cuts are nicely presented. The KS overtraining check, though, is vacuous—a KS value in [0,1] is not evidence of no overtraining, and the authors shouldn't present it as such.\n\nOne more thing: the abstract's emphasis on 5 sigma at chi=5 sits awkwardly with the HL-LHC projection that forces them to drop to chi=3 and 4.4 sigma. That tension is a consequence of the scaling problem, not an independent flaw.\n\nOverall: the model and approach are legitimate, and the flaw looks fixable if the Monte Carlo normalization is corrected. But as written, the central quantitative claim is not supported. I would send this to peer review—a serious referee can ask the authors to verify the scaling and rerun the projections. If the numbers hold after correction, the paper is a useful, if incremental, contribution. If they don't, the conclusion collapses.\n\nI wouldn't cite it in its current form, but I'd be curious to see the corrected version.","headline":"A legitimate 2HDM-III sensitivity study whose central numbers don't survive internal consistency checks: Table III violates the chi_tc^2 scaling required by Eq. (22), so the luminosity projections are unsupported as written.","tokens_in":12929,"tokens_out":1184,"would_cite":false,"duration_ms":12904,"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 flavor-changing top-charm-Higgs coupling, absent in the Standard Model, can be probed through pp→th+X at the HL-LHC, where a BDT analysis predicts a 5σ reach for χ_tc=5.","keywords":["2HDM type III","flavor-changing neutral scalar interactions","top-Higgs associated production","HL-LHC","boosted decision trees","signal significance","top quark FCNC"],"falsifier":"Generate new signal samples for χ_tc=1 and χ_tc=5 with the same model implementation and event generator used in the paper, holding all other parameters fixed, and compare the cross-sections; if the ratio is not 25, the implementation does not follow the analytic coupling, and the reported 5σ reach at 2700 $fb^{-1}$ for χ_tc=5 would need to be recalculated.","tokens_in":11761,"feed_emoji":"⚛️","tokens_out":12757,"duration_ms":117543,"temperature":0.7,"pith_summary":"In the two-Higgs-doublet model of type III (2HDM-III), the paper argues that a flavor-changing neutral scalar interaction (FCNSI) coupling the top quark, the charm quark, and the Higgs boson—absent in the Standard Model—can be probed through $pp\\to th+X$ production at the High-Luminosity LHC. With the decay chain $t\\to\\ell\\nu_\\ell b$ and $h\\to\\gamma\\gamma$, and separating signal from background using boosted decision trees, they predict a $5\\sigma$ significance for $\\tan\\beta=1$, $\\cos(\\alpha-\\beta)=0.1$, $\\chi_{tc}=5$, at integrated luminosity $\\mathcal{L}_{\\rm int}\\gtrsim 2700~\\text{fb}^{-1}$ including a $5\\%$ systematic uncertainty, and $\\sigma\\approx4.4$ for $\\chi_{tc}=3$ at $\\mathcal{L}_{\\rm int}=3000~\\text{fb}^{-1}$ under the HL-LHC projection $\\text{BR}(t\\to ch)<10^{-4}$. This matters because it offers a direct search channel for a coupling that is otherwise constrained only indirectly through the decay $t\\to ch$, and because the machine-learning approach extends the reach well beyond simple kinematic cuts.","feed_headline":"Flavor-changing top-Higgs coupling hits 5-sigma at HL-LHC","feed_subtitle":"Rare top-charm-Higgs coupling could show up in th(γγ) events within 3000 fb^-1 of HL-LHC data.","key_machinery":"The flavor-violating top-charm-Higgs vertex $g_{tch}$ from Eq. (22) of the 2HDM-III, which is linear in the parameter $\\chi_{tc}$ and scales as $1/\\tan\\beta$, is the object that drives $pp\\to th+X$ production. On the analysis side, the workhorse is a boosted decision tree trained on photon $p_T$, lepton $p_T$, and jet pseudorapidity variables, which separates the signal from the dominant Standard Model backgrounds; the classifier output is then scanned to maximize the significance $S/\\sqrt{S+B+(0.05B)^2}$.","core_discovery":"Within the two-Higgs-doublet model of type III, the authors claim that the flavor-changing neutral scalar coupling $g_{tch}$, derived in Eq. (22), is proportional to $\\chi_{tc}/\\tan\\beta$, and that it controls the $pp\\to th+X$ production cross-section. For their benchmark scenario S1 ($\\tan\\beta=1$, $\\cos(\\alpha-\\beta)=0.1$), after applying a boosted decision tree to the decay chain $t\\to\\ell\\nu_\\ell b$ with $h\\to\\gamma\\gamma$, they predict a signal significance of $5\\sigma$ or more for $\\chi_{tc}=5$ when the integrated luminosity reaches about $2700~\\text{fb}^{-1}$ (with $5\\%$ systematic uncertainty), and about $4.4\\sigma$ ($4.2$ with systematics) for $\\chi_{tc}=3$ at $3000~\\text{fb}^{-1}$, which respects the HL-LHC projection $\\text{BR}(t\\to ch)<10^{-4}$. They conclude that the HL-LHC could find evidence for this new physics process, and that the BDT analysis substantially outperforms simple kinematic cuts.","pith_inferences":["The signal cross-sections in Table III scale more slowly with χ_tc than Eq. (22) dictates (S1 moves from 0.01 fb to 0.025 fb when χ_tc goes from 1 to 5, whereas the square scaling implies a factor 25). If the table reflects the generated samples, the model implementation is inconsistent with the analytic coupling; if the table is a transcription error, the χ_tc=5 significance at 2700 fb^-1 would b","The same BDT pipeline, with the h→γγ resonance as the anchor, could be applied to flavor-violating decays of the heavier neutral scalars H0 and A0 in the same model, where the background composition is similar.","If the upper limit on BR(t→ch) tightens below 10^-4, the allowed χ_tc shrinks; the trend in the paper's parameter scan suggests the reach would degrade roughly in proportion, so the 5σ window for χ_tc=5 and tanβ=1 would close, leaving the process as an evidence-level probe only for the smallest couplings."],"forward_implications":["With 3000 fb^-1, the HL-LHC can search for the FCNSI coupling down to χ_tc≈3, a region consistent with the projected limit BR(t→ch)<10^-4, so pp→th+X becomes a complementary direct probe of the same physics that t→ch searches constrain indirectly.","The BDT-based selection roughly doubles the expected significance relative to the cut-based analysis described in the paper, so the classifier gain is a key part of the reach claim.","For the most favorable benchmark, a 3-sigma hint is predicted by about 1000 fb^-1, while 5-sigma discovery requires around 2700 fb^-1 once 5% background systematics are included.","At tanβ=3, the predicted significance is markedly lower, so the sensitivity is concentrated at small tanβ where g_tch ∝ 1/tanβ is largest."],"supporting_citations":[{"why":"Sets the current upper limit on BR(t→ch) that bounds the parameter χ_tc.","marker":"[16]"},{"why":"Supplies the HL-LHC projection BR(t→ch)<10^-4 used as the conservative benchmark for χ_tc.","marker":"[36]"},{"why":"Proposed pp→th production as a probe of top flavor-changing neutral scalar couplings.","marker":"[17]"},{"why":"Previous study of pp→th with h→γγ at the LHC, the cut-based baseline the BDT analysis improves on.","marker":"[19]"},{"why":"Implements the 2HDM-III Lagrangian with FCNSI in the event-generation chain.","marker":"[40]"},{"why":"Generates the signal and background Monte Carlo samples.","marker":"[41]"},{"why":"Simulates the HL-LHC detector response used for the analysis.","marker":"[43]"},{"why":"Provides the boosted decision tree algorithm used to separate signal from background.","marker":"[37]"},{"why":"Defines the significance formula S/√(S+B+(κB)^2) used for the reach projections.","marker":"[49]"}],"fun_headline_variants":["Flavor-changing top-Higgs: 5σ possible at HL-LHC luminosity","Rare th + X via FCNSI may hit 5σ at 2700 fb^-1","Single Higgs with top quark: BDT finds 5σ reach","Top-charm-Higgs coupling tested by th(γγ) at HL-LHC","FCNSI-driven th production shows 5σ promise at HL-LHC"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reach projections stand on the Monte Carlo simulation implementing the model's coupling exactly as derived, so the signal rate grows with the square of the flavor parameter χ_tc; if the implementation drifts, every significance number changes.","fun_headline_variants_meta":{"raw":{"variants":["Flavor-changing top-Higgs: 5σ possible at HL-LHC luminosity","Rare th + X via FCNSI may hit 5σ at 2700 fb^-1","Single Higgs with top quark: BDT finds 5σ reach","Top-charm-Higgs coupling tested by th(γγ) at HL-LHC","FCNSI-driven th production shows 5σ promise at HL-LHC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000618,"raw_usage":{"total_tokens":2936,"prompt_tokens":1084,"completion_tokens":1852,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":700,"completion_tokens_details":{"reasoning_tokens":1743}},"tokens_in":700,"tokens_out":1852,"duration_ms":18492,"temperature":1.0,"reasoning_tokens":1743,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T23:25:34.780806+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Generate new signal samples for χ_tc=1 and χ_tc=5 with the same model implementation and event generator used in the paper, holding all other parameters fixed, and compare the cross-sections; if the ratio is not 25, the implementation does not follow the analytic coupling, and the reported 5σ reach at 2700 $fb^{-1}$ for χ_tc=5 would need to be recalculated.","supporting_citations":[{"cited_title":"Arroyo-Ure˜ na, J","cited_arxiv_id":null,"evidence_quote":"Sets the current upper limit on BR(t→ch) that bounds the parameter χ_tc."},{"cited_title":"Measurement of the B0 S → µ+µ− decay properties and search for the B0 → µ+µ− decay in proton-proton collisions at √s = 13 TeV","cited_arxiv_id":null,"evidence_quote":"Supplies the HL-LHC projection BR(t→ch)<10^-4 used as the conservative benchmark for χ_tc."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proposed pp→th production as a probe of top flavor-changing neutral scalar couplings."},{"cited_title":"Search for flavor-changing neutral current interactions of the top quark mediated by a Higgs boson in proton-proton collisions at 13 TeV","cited_arxiv_id":null,"evidence_quote":"Previous study of pp→th with h→γγ at the LHC, the cut-based baseline the BDT analysis improves on."},{"cited_title":"Artificial Intelligence for High Energy Physics, Chapter 2: Boosted Decision Trees, pages 9–58","cited_arxiv_id":null,"evidence_quote":"Implements the 2HDM-III Lagrangian with FCNSI in the event-generation chain."},{"cited_title":"The Elements of Statistical Learning","cited_arxiv_id":null,"evidence_quote":"Generates the signal and background Monte Carlo samples."},{"cited_title":"Christensen, C´ eline Degrande, Claude Duhr, and Benjamin Fuks","cited_arxiv_id":null,"evidence_quote":"Simulates the HL-LHC detector response used for the analysis."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the boosted decision tree algorithm used to separate signal from background."},{"cited_title":"Salam, and Gregory Soyez","cited_arxiv_id":null,"evidence_quote":"Defines the significance formula S/√(S+B+(κB)^2) used for the reach projections."}],"review_version":1}