{"id":"da2e302c-7c28-4af6-9326-bae3619a9558","arxiv_id":"2607.21712","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Non-resonant gg→aa→4γ production could constrain the dimension-6 ALP-gluon coupling down to ~10^-3 TeV^-2 at 300 fb^-1, but the allowed parameter space remains unbounded along multiple flat directions.","lead":"This paper studies LHC collisions that produce pairs of axion-like particles decaying into four photons, and estimates how well such a search could probe ALP interactions, including a dimension-6 gluon contact term not studied before. It also reinterprets an existing ATLAS search for Higgs decays into ALP pairs in a three-dimensional parameter space.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Projected C_G2 reach assumes maximally large Br(a→γγ); non-minimal C_WW or fermionic couplings weaken the quoted 10^-3 TeV^-2 bound, possibly by orders of magnitude.","rationale":"I have read the paper in good faith. The analytic decomposition of the cross section (Eq. 4.5) is consistent with the amplitude structure (Eq. 4.3), and the numerical ratios in Table 1 match the analytic expectations in Sec. 4.1.1. The recasting of the ATLAS bound in Sec. 5.2 is internally coherent. The only load-bearing soft spot is the best-case treatment of the ALP decay width: the projected C_G2 reach is controlled by [Br(a→γγ)]^2, and Br is maximized by construction when C_WW is minimized and C_ψ = 0. This is disclosed, but it means the headline sensitivity is an upper limit on the search's power rather than a robust prediction. My assessment agrees with the reader's weakest_assumption; no additional fatal flaw is present. The CONDITIONAL verdict is appropriate.","tokens_in":58178,"tokens_out":17161,"duration_ms":154722,"concrete_test":"Recompute the N_signal = 3 contour for C_G2 in Fig. 12 at ma = 30 and 200 GeV, replacing C̄_WW with a non-minimal benchmark (e.g., C_WW = 0, corresponding to C_B = C_γγ) and adding a representative fermionic coupling, e.g., C_t/Λ = 0.1 TeV^-1, to Γ_tot in Eq. (4.31). If the resulting bound on C_G2/Λ^2 changes by more than a factor of 2, the quoted 10^-3 TeV^-2 must be presented as a best-case reach.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central sensitivity claim (Sec. 6: C_G2/Λ_a^2 down to 10^-3 TeV^-2 at 300 fb^-1) is derived from N_signal ∝ σ(pp→aa) × [Br(a→γγ)]^2 (Eqs. 4.1, 5.1). The branching ratio is computed in Sec. 4.2 in a deliberately best-case setup: C_WW is fixed to C̄_WW that minimizes Γ_a→EW (Eqs. 4.30–4.32, App. A) and all fermionic couplings C_ψ are set to zero, maximizing Br. In a generic ALP EFT, C_WW and C_ψ are independent parameters; non-minimal values increase Γ_tot^a and reduce Br. Since the bound on C_G2 for fixed signal yield scales as 1/Br (production ∝ C_G2^2, signal ∝ Br^2), a factor-10 reduction in Br weakens the quoted reach by a factor ~10. The authors themselves list the dependence on C_W and on fermionic ALP couplings as a future extension in Sec. 6, so the headline number is conditional on this optimistic decay assumption. This is the same load-bearing caveat identified by the reader.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes and analyzes pp→aa→4γ searches at the LHC as a probe of the ALP EFT, focusing on the first study of non-resonant gg→aa production with dimension-5 (O_G1) and dimension-6 (O_G2) gluon couplings. The production amplitude is decomposed into σ11, σ22, σ12, computed with MadGraph and reweighting; ALP decays are treated in the NWA with Br(a→γγ) normalized to a 'best-case' total width (C_WW minimized, C_ψ=0). Detector effects are modeled with selection cuts, flat 95% photon efficiency, and finite-size detector effects with Ldet=1.08 m calibrated on the ATLAS f_aa. Under a background-free Poisson assumption, 95% CL projections are given for 300 fb^-1 at 13 TeV in two-dimensional slices, and the ATLAS h→aa→4γ search [46] is recast in (Cγγ, CG1, Cah). The main results are a projected reach on C_G2/Λ_a^2 down to ~10^-3 TeV^-2 and the observation that a pp→aa→4γ measurement alone leaves the parameter space unbounded in certain directions.","tokens_in":58505,"tokens_out":11619,"duration_ms":122064,"significance":"This is the first quantitative treatment of non-resonant ALP pair production through O_G2 and provides a clear analytic parameterization (Eqs. 4.3–4.8), a transparent Monte-Carlo reweighting procedure, and an explicit handling of the three components σ11/σ22/σ12 and their interference. The multidimensional presentation of allowed regions, including the sign dependence of C_G2 and finite-size detector effects, is a useful contribution. The central caveat — that the quoted reach assumes the maximal Br(a→γγ) by construction — is acknowledged internally but is not carried into the headline numbers in the abstract and conclusions.","major_comments":[{"comment":"The headline reach C_G2/Λ_a^2 ≲ 10^-3 TeV^-2 is computed in the 'best-case' decay scenario: C_WW is fixed to the value minimizing Γ_a→EW (Appendix A) and all fermionic couplings C_ψ are set to zero, maximizing Br(a→γγ). These are independent ALP EFT parameters, not measured inputs. Since N_signal ∝ σ(pp→aa) Br(a→γγ)^2 and, in the C_G2-dominated region, σ(pp→aa) ∝ C_G2^2, a reduction of Br by a factor x weakens the projected bound on C_G2 by the same factor x. For m_a ≥ 2m_W, even the minimized EW width is nonzero (F(m_a) > 0, Eq. 4.29), and for a generic C_WW or non-vanishing fermionic couplings the maximum Br is smaller. The authors list this dependence as a future extension in Sec. 6, but the abstract and the conclusion quote the 10^-3 TeV^-2 value without this qualifier. Please quantify the degradation for representative non-minimal choices (e.g., C_WW=0 and a fermionic benchmark) and","section":"§4.2, Eqs. (4.30)–(4.32); §5.1, Eq. (5.1); §6"},{"comment":"The extreme bounds on C_γγ in the FSDE-dominated corners (values down to ~10^-9 TeV^-1 at m_a=1 TeV in the lower-right panel of Fig. 9) are sensitive to the modeling of the detector size. The parameter L_det=1.08 m is calibrated to reproduce the ATLAS f_aa, which is a reasonable choice, but the paper does not provide an uncertainty band. Since the bound in this regime is set by the exponential factor in P_aa and by Γ_tot^a, a factor-of-order-unity change in L_det can shift the contour. This does not affect the main C_G2 reach, but it is load-bearing for the statement that C_γγ can be constrained down to 10^-9 TeV^-1; a short scan over L_det (e.g., 0.9–1.5 m) would make the claim robust.","section":"§4.4, Eq. (4.48); lower panels of Fig. 9"}],"minor_comments":[{"comment":"The symbol C_γγ is used both as the parameter being constrained and as the upper-limit value taken from Ref. [46]. Rename the latter, e.g. C_γγ^lim, to avoid ambiguity in the recast inequality.","section":"Eq. (5.7)"},{"comment":"The three efficiency curves ε_cuts,11, ε_cuts,22, ε_cuts,12 are stated to completely overlap. The caption should state this explicitly; if the few-percent differences matter at m_a=30 GeV, they could be shown in a small inset.","section":"Fig. 7"},{"comment":"The flat 95% photon reconstruction efficiency is a useful approximation, but it neglects p_T/η dependence and possible merging of photons from boosted ALPs. The restriction m_a ≥ 30 GeV mitigates the latter; a sentence quantifying the residual uncertainty would be helpful.","section":"§4.3, Eq. (4.41)"},{"comment":"Typographical issue: 'can also be extracted from by di-photon resonance searches' should be 'can also be extracted from di-photon resonance searches'.","section":"§3.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically solid and the authors are transparent about the best-case decay assumption. The revision should focus on making the conditional nature of the headline sensitivity explicit and on quantifying how much the reach degrades when C_WW and fermionic couplings are non-minimal. I do not see a need for rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is the non-resonant gg→aa channel and the dimension-6 operator OG2. Nobody has studied that process before, and the paper makes a good case that it could be competitive with Higgs-resonant searches. The reinterpretation of the ATLAS h→aa→4γ limits in a three-parameter space is also a legitimate piece of work, not a fitted result.\n\nThe analytic amplitude parameterization in Sec. 4.1 is clear and internally consistent, and the Monte Carlo reweighting for σ11, σ22, σ12 is a sound way to keep the interference terms under control. The finite-size detector effects are treated carefully, and the calibration of Ldet against the ATLAS f_aa is a reasonable check. I also give credit for the explicit demonstration that the allowed parameter space remains unbounded in certain directions—that is a non-obvious and useful result about the interplay between ALP production and decay.\n\nThe soft spots are real but disclosed. The headline number—CG2/Λa² down to 10⁻³ TeV⁻² at 300 fb⁻¹—is derived in a deliberately best-case setup: C_WW is fixed to the value that minimizes Γ_EW and all fermionic couplings are set to zero, maximizing Br(a→γγ). That is a modeling choice, not a measured input. If the real ALP has non-minimal electroweak or fermionic couplings, the branching ratio drops and the projected bound weakens, in some regions by orders of magnitude. The paper explicitly acknowledges this dependence in Sec. 6, but it does make the central sensitivity claim conditional. The background-free Poisson assumption, the flat 95% photon efficiency, and the neglect of the quoted +50%/−30% scale and PDF systematics are also simplifications. They are reasonable for a first projection, but they should be propagated before the quoted reach is treated as a firm target.\n\nThe citation pattern looks fine. The paper cites the relevant h→aa searches and ALP EFT literature, and I see no circularity: the limits come from first-principles matrix elements, with only the detector radius calibrated to reproduce ATLAS. No code is shipped, which is a minor reproducibility negative but not disqualifying.\n\nThis paper deserves a serious referee. It is a well-executed first study of an unexplored process and operator, with transparent assumptions and no load-bearing error. A referee should push on the branching-ratio conditionality and the statistical model, but the core analysis holds up. I would send it to peer review.","headline":"First solid projection for non-resonant gg→aa→4γ at the LHC; the headline CG2 reach is conditional on a best-case ALP branching ratio, but the paper is transparent about that.","tokens_in":59071,"tokens_out":1926,"would_cite":true,"duration_ms":22094,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["12.60.-i","13.85.Rm","14.80.Va"],"model":"deepseek-v4-flash","headline":"This paper establishes that a four-photon search for non-resonant axion-like-particle pair production at the LHC could already at 300 fb^-1 probe the dimension-6 ALP-gluon coupling down to 10^-3 TeV^-2, while no single such search can ever","keywords":["axion-like particles","ALP pair production","four-photon final state","dimension-6 ALP operators","LHC phenomenology","long-lived ALPs","effective field theory","Higgs exotic decays"],"falsifier":"Measure the ALP's photon branching ratio or any additional decay width: since the projected reach scales as Br(a→γγ)², detecting a non-negligible a→Zγ, a→W⁺W⁻, or fermionic channel (allowed by gauge invariance above 2m_W) would push the 10^-3 TeV^-2 sensitivity out of reach, directly testing the best-case assumption. Running the four-photon search itself at 300 fb^-1 also settles it: the predicted SM background is ~0.2 events, so a null result places the bound and a 4γ excess whose kinematic distributions disagree with the gg→aa prediction would falsify the EFT interpretation.","tokens_in":58008,"feed_emoji":"⚛️","tokens_out":11233,"duration_ms":106994,"temperature":0.7,"pith_summary":"This paper establishes that the LHC can probe axion-like particles (ALPs) in pairs, not just singly, and gives the first analysis of the non-resonant process gg→aa, where a dimension-6 contact interaction between two ALPs and two gluons competes with the usual dimension-5 gluon coupling. Its central quantitative claim is that with 300 fb^-1 of 13 TeV data, a nearly background-free four-photon search would be sensitive to the dimension-6 coefficient C_G2/Λ²a down to 10^-3 TeV^-2 — roughly an order of magnitude stronger than the Higgs-resonant channel's reach on the analogous h→aa coupling. Equally central is the structural result that no single pp→aa→4γ measurement can ever close the ALP parameter space: cancellations between production and decay, saturation of the photon branching ratio, and long-lived-ALP detector-escape effects leave the allowed region unbounded along specific directions. The authors argue this is generic to collider ALP searches in which the same couplings enter both production and decay, so the same-shaped open regions should appear in other final states. They further recast an LHC search for h→aa→4γ into three-dimensional coupling space, showing that the commonly quoted bounds on C_γγ shift by orders of magnitude once a small gluon coupling C_G1 is admitted.","feed_headline":"Four-photon LHC search could reach ALP couplings of 10^-3 TeV^-2","feed_subtitle":"First study of non-resonant axion pairs shows HL-LHC reach — and why one search can't close the parameter space.","key_machinery":"The engine of the analysis is an amplitude identity and a cancellation: the non-resonant gg→aa cross section factorizes as σ = C_G1⁴σ11 + C_G2²σ22 + C_G1²C_G2σ12, where the interference term σ12 is comparable to the pure terms and changes sign with C_G2; meanwhile the branching ratio Br(a→γγ) = [1 + F(m_a) + k²(m_a) r_γ²]^-1 depends only on the ratio r_γ = C_G1/C_γγ. Combining them, C_G1⁴ production growth cancels against the C_G1^-4 suppression of the branching ratio, creating a flat direction, and Br saturates for C_γγ ≳ 250 C_G1, creating another. Finite-size detector effects — the probability that both ALPs decay within the 1.08 m detector radius, which scales as (Γ_tot^a)² in the low-wi","core_discovery":"Double-ALP production at the LHC is a viable and competitive probe of ALP interactions beyond dimension-5. Focusing on gg→aa→4γ, the paper analyzes for the first time the non-resonant mode, whose rate is a polynomial in the couplings — pure C_G1⁴, pure C_G2², and an interference term C_G1²C_G2 — so different regimes test different operators, including the sign of C_G2. Because the same C_G1 that boosts production also suppresses Br(a→γγ) by opening gluonic decays, the signal is not monotonic: it contains regions where the coupling dependence cancels exactly, a valley of destructive interference for C_G2<0, and a long-lived-ALP regime where the pair escapes the detector before converting to p","pith_inferences":["The strongest apparent C_γγ exclusions (down to ~10^-9 TeV^-1 non-resonant, 10^-5–10^-7 TeV^-1 Higgs-resonant) live where a dimension-6 coefficient vastly exceeds dimension-5 ones; a global fit with EFT-power-counting priors would likely discard those points, reframing these as lifetime exclusions of long-lived ALPs rather than coupling bounds.","Because the cancellation mechanism is structural, a quick cross-check of existing single-ALP limits for the same saturation direction (C_γγ → large, Br saturating) would reveal which published bounds are slices of open regions.","A testable extension: push below ma = 30 GeV with merged-photon topologies for boosted ALPs; the paper's own efficiency curve (ϵ_cuts ≈ 0.002 at 10 GeV) predicts exactly where the non-resonant channel loses sensitivity.","The LO-only gg→aa prediction leaves the σ11/σ12/σ22 ratios vulnerable to NLO QCD corrections; given K-factors of order 5 in the Higgs channel, the 10^-3 TeV^-2 number could move by a comparable factor once the calculation exists."],"forward_implications":["A 300 fb^-1 four-photon search would set the first direct bound on the dimension-6 ALP-gluon contact operator, C_G2/Λ²a ≲ 10^-3 TeV^-2 for ma ≲ 300 GeV (10^-2 at 1 TeV), about an order of magnitude tighter than the Higgs-resonant channel's C_ah/Λ²a ≲ 10^-2 TeV^-2.","No 4γ search alone closes the parameter space: allowed regions stay unbounded along four directions — C_G1 cancellation, C_γγ saturation, the C_G2<0 interference valley, and small-coupling long-lived escape.","The same-coupling cancellation is generic, so cusps, thin allowed bands, and open regions should appear in any ALP search where one coupling controls both production and decay; interpreting such bounds requires a global fit.","In the Higgs-resonant recast, a non-zero C_G1 weakens the inferred limits on C_γγ by 3–5 orders of magnitude (e.g. for C_G1/Λa = 10^-3 TeV^-1), and meaningful C_γγ bounds exist only for C_ah inside a narrow, mass-dependent window.","Sensitivity to the sign of C_G2 is possible only where interference matters (large C_G1); elsewhere the search probes |C_G2| alone, and the best-case Br assumption caps the photon branching ratio at 1/(1+F(ma)) ≈ 0.6 for ma ≫ 2m_W."],"fun_headline_variants":["First non-resonant ALP pair search targets 4γ at LHC","gg→aa→4γ: LHC probes ALP interactions","ALP pair production: 4γ constraints dim-5 and dim-6","LHC 4γ search: non-resonant ALP pairs debut","Non-resonant ALP pairs: 4γ signature at LHC"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The projected reach assumes a best-case ALP whose electroweak decay couplings are tuned to maximize the photon branching ratio; if the real ALP decays more readily into W/Z pairs or fermions, the four-photon signal shrinks as Br(a→γγ)² and the quoted bounds weaken, in some regions by orders of magnitude.","fun_headline_variants_meta":{"raw":{"variants":["First non-resonant ALP pair search targets 4γ at LHC","gg→aa→4γ: LHC probes ALP interactions","ALP pair production: 4γ constraints dim-5 and dim-6","LHC 4γ search: non-resonant ALP pairs debut","Non-resonant ALP pairs: 4γ signature at LHC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001299,"raw_usage":{"total_tokens":5126,"prompt_tokens":725,"completion_tokens":4401,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":469,"completion_tokens_details":{"reasoning_tokens":4300}},"tokens_in":469,"tokens_out":4401,"duration_ms":29057,"temperature":1.0,"reasoning_tokens":4300,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T06:53:28.691806+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the ALP's photon branching ratio or any additional decay width: since the projected reach scales as Br(a→γγ)², detecting a non-negligible a→Zγ, a→W⁺W⁻, or fermionic channel (allowed by gauge invariance above 2m_W) would push the 10^-3 TeV^-2 sensitivity out of reach, directly testing the best-case assumption. Running the four-photon search itself at 300 fb^-1 also settles it: the predicted SM background is ~0.2 events, so a null result places the bound and a 4γ excess whose kinematic distributions disagree with the gg→aa prediction would falsify the EFT interpretation.","supporting_citations":[],"review_version":1}