{"id":"9d9f9f6f-9e99-4aa7-a22e-65ed1d87b709","arxiv_id":"2010.15834","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The authors obtain analytic two-loop helicity amplitudes for three-photon production at hadron colliders in leading-color QCD and release a public code that evaluates the finite remainders stably across physical phase space.","lead":"Physicists computed the two-loop QCD corrections to the production of three photons in quark-antiquark collisions, the missing ingredient for next-to-next-to-leading-order predictions at the LHC. They ship a public C++ library that evaluates the result fast enough to run inside a full collider simulation.","discovery_kind":"extension","skeptic_critique":{"model":"selfhosted/deepseek-v4.1-flash","headline":"The abstract's 'across the whole physical phase space' and 'stability demonstrated' outrun figure 2, whose tail includes points wrong by orders of magnitude on a single cut set; the body acknowledges this and defers a rescue system.","rationale":"The reader's weakest_assumption identifies the same load-bearing point: the stability claim is a sample statistic over one cut set, the code returns no per-point error estimate, and the paper defers a rescue system. I agree, and I sharpen two details the reader leaves implicit. First, figure 2's tail reaches d = -4, so the failing points are not merely imprecise but wrong by orders of magnitude, which matters more for differential distributions than for an integrated Monte Carlo. Second, the mismatch is specifically between abstract:s3/s4 ('across the whole physical phase space', 'stability across phase-space is demonstrated') and a body that tests one ATLAS 8 TeV sample and explicitly says the rescue system is future work. That is an abstract-versus-body scope issue, not a defect in the amplitude calculation. The rest of the validation chain is solid for what it claims: one-loop checks against OpenLoops and ref. [10], permutation and parity relations verified numerically, pole structure checked, collinear factorization checked at one loop to O(epsilon^2) and at two loops to O(epsilon^0), and reference values in tables 4-6 reproduced by the shipped assembly script. The two-loop collinear check for the like-helicity (q,gamma) case is a consistency check on the normalized splitting amplitude rather than a comparison to an independent two-loop splitting function, and the phenomenological agreement in ref. [16] reuses these same amplitudes, so neither is a fully independent cross-check; but both are disclosed and neither is presented as more than it is. The leading-color truncation in eq. 2.14 is stated in the abstract and body and delegated to ref. [16], so it is scoped rather than hidden. The priority claims in abstract:s5 and section 3.3 are carefully qualified ('publicly available', 'such a decomposition') and are consistent with the cited state of the field, including ref. [15]. None of this changes the scientific verdict. A minor abstract revision narrowing 'across the whole physical phase space' to the tested phase-space region would be appropriate, but the physics claim stands and the reader's ACCEPT is the right call.","tokens_in":23982,"tokens_out":6920,"duration_ms":133639,"concrete_test":"Using the public library, generate a phase-space sample concentrated near the surface s12 + s23 = 0 (the letter the paper itself identifies as a spurious singularity), evaluate H(2) in double and quadruple precision on each point, and measure the fraction with d < 4. If that fraction is substantially above the 0.1% reported for the ATLAS 8 TeV sample, the abstract's 'across the whole physical phase space' is not supported and should be softened to the tested cut set; if it stays near 0.1%, the concern does not land and the abstract wording is defensible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central deliverable is a public code for 'immediate phenomenological applications' (abstract:s5), and its usability rests on the numerical-stability claim. The body is more careful than the abstract. Figure 2 plots d = -log10|H_double/H_quad - 1| over 90000 points with the ATLAS 8 TeV cuts of ref. [16], and its x-axis runs down to d = -4, i.e. some sampled points have relative error of order 10^4, not merely low precision. The paper states 'less than 0.1% of the sampled points have an accuracy of less than four digits' and attributes the failures to proximity to spurious singularities such as the letter s12 + s23, then explicitly defers a precision-rescue system ('We leave this for future work', section 4). Against this, abstract:s3 says the remainders are 'given in a form that can be efficiently evaluated across the whole physical phase space' and abstract:s4 says 'numerical stability across phase-space is demonstrated.' The demonstration is a single cut set; the universal wording is not supported by it. A user running different cuts, or a region denser in spurious-singularity surfaces, receives a number with no in-code error signal. This is a presentation and robustness gap rather than a physics error, and the authors disclose the underlying limitation, but the abstract's scope claim is the one place where the shipped deliverable's stated guarantee is stronger than the evidence shown.","agreement_with_reader":"agree"},"referee_report":{"model":"selfhosted/deepseek-v4.1-flash","summary":"The manuscript presents the first analytic decomposition of a two-loop five-point QCD scattering amplitude into master integrals, for the process q qbar -> gamma gamma gamma in the leading-color (planar) approximation, together with the corresponding two-loop finite remainders expressed in the pentagon-function basis of ref. [23]. The two independent helicity amplitudes A_{+++} and A_{-++} are computed with the two-loop numerical unitarity framework as implemented in Caravel, with the master-integral coefficients reconstructed analytically from finite-field evaluations over rational phase-space points. The authors provide the all-orders-in-epsilon master-integral decomposition, the epsilon^0 remainders, reference values on a specified phase-space point (Tables 4-6), a validation chain (four-point amplitudes against ref. [10], five-point one-loop against OpenLoops, all permutation/parity relations of Table 1, collinear factorization as in Appendix B), and a public C++ library that assembles the helicity- and color-summed squared remainders H^(1) and H^(2). A stability study over 90000 phase-space points with the ATLAS 8 TeV cuts of ref. [16] is presented in Fig. 2.","tokens_in":24200,"tokens_out":3917,"duration_ms":70815,"significance":"If the results hold, this is a genuinely important contribution: it is the first publicly available two-loop five-point amplitude suitable for on-the-fly NNLO phenomenology, and the first analytic master-integral decomposition for a physical two-loop five-point process. The paper earns credit for shipping reproducible deliverables (ancillary files with master-integral definitions, amplitude and remainder expressions, an assembly example reproducing Tables 4-6, and a public code), for a validation chain that is unusually complete for this genre, and for quantifying numerical stability rather than asserting it. The results have already been used in ref. [16], which is a meaningful falsifiable test of practical usability. The leading-color approximation and the omission of the non-planar N_f contribution are stated explicitly and their phenomenological adequacy is deferred to ref. [16], which is appropriate.","major_comments":[],"minor_comments":[{"comment":"The abstract states that the remainders are 'given in a form that can be efficiently evaluated across the whole physical phase space' (abstract:s3) and that 'numerical stability across phase-space is demonstrated' (abstract:s4), whereas the demonstration in §4 and Fig. 2 uses a single cut set (the ATLAS 8 TeV selection of ref. [16]) and the authors themselves note that a precision-rescue system is deferred to future work. The body is more careful than the abstract, and the limitation is disclosed, so this is a wording issue rather than a physics issue; I would nonetheless ask the authors to soften the abstract to match the evidence, e.g. by stating that stability is demonstrated on a representative phenomenological phase-space sample and that a small fraction of points near spurious-singularity surfaces lose accuracy.","section":"Abstract; §4 and Fig. 2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The text states that 'less than 0.1% of the sampled points have an accuracy of less than four digits', but the histogram axis extends to d = -4, i.e. some points are wrong at the level of the leading digit. Please quote the actual number of such points, the worst observed value of d, and the fraction of the 90000 points with d below, say, 1 and 2. This would let a reader judge the practical risk more concretely than the single '0.1%' figure does.","section":"§4, Fig. 2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Since the library returns a number without an error estimate and the problematic regions are, as the authors state, explicitly known (surfaces where letters such as s12 + s23 vanish), it would be cheap and useful to document those letters and the associated phase-space regions in the README, even if a full precision-rescue system is left for future work. I would encourage this as a stopgap.","section":"§4, public code","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The one-loop 'Pentagon' entry reads MI[``Pentagon'', i, s12, s23, s45, s45, s15], listing s45 twice; presumably the fourth argument should be s34. Please check this against the ancillary master-integral definition files, since a mislabelled permutation representative would be easy for a user to propagate.","section":"Table 8","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The sentence enumerating the collinear limits reads 'either (gamma,gamma), (q,q), (q,gamma) or (q,gamma)', with the last pair repeated; the final entry should presumably be (qbar,gamma), consistent with the separate '(qbar,gamma) limit' paragraph that follows.","section":"Appendix B, list of collinear configurations","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The prefactor of the one-loop-squared term is written as 'N_c^2/4 N_c', which is ambiguous as typeset. Please write it unambiguously (e.g. (N_c^2/4) N_c) and state explicitly that the C_F^2 factor of the one-loop amplitude has been replaced by its leading-color value for consistency with the two-loop approximation, as is done in the text for Eq. (4.2).","section":"Eq. (4.3)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The claim that 'this is the first time that such a decomposition has been made available for a two-loop five-point QCD amplitude' would be strengthened by one explicit sentence distinguishing it from refs. [46,48], which provide analytic five-gluon amplitudes but, as I understand, only in the form of finite remainders rather than master-integral decompositions valid to all orders in epsilon. As written, a reader may not immediately see why the two claims are compatible.","section":"§3.3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"A few typographical and typesetting issues should be fixed: 'Aside form the fact' in §3.2; 'heliciy' in the sentence introducing Eq. (B.3); and in §4 the sentence 'the amplitude must stay regular As an example, s12 + s23 is a letter...' is missing a sentence break and a full stop.","section":"§3.2 and Appendix B","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The ancillary path is given as 'anc/example assembly.m' with a space; please confirm the actual filename (presumably example_assembly.m) so that readers can locate it.","section":"§3.5","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"For reproducibility, please state the version of PentagonFunctions++ the library was tested against, and the license under which the code is released.","section":"§4, code availability","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"minor_revision","confidential_remarks":"The novelty claims are appropriately hedged and the overlap with ref. [15] (which computed the same amplitudes numerically and via interpolation) is explicitly credited and distinguished on the grounds of public availability and analytic form; I see no disclosure problem. The paper is a strong fit for the journal and the requested changes are editorial."},"author_rebuttal":null,"desk_editor":{"model":"selfhosted/deepseek-v4.1-flash","letter":"Bottom line: this is the paper that actually hands you the triphoton two-loop remainders, in a public C++ library, and the companion Matrix paper (ref. [16]) has already run them through a Monte Carlo. If you work on five-point two-loop amplitudes or on triphoton phenomenology, this is the one to read rather than the earlier interpolation-based study [15]. It is not a methods paper — the pipeline is Caravel plus finite-field reconstruction plus the simplification of ref. [49], all the authors' own machinery.","headline":"A genuine deliverable: the first analytic all-orders master-integral decomposition of a five-point two-loop QCD amplitude, plus a public code for triphoton NNLO — the physics is solid, the abstract's phase-space wording is the one loose thread.","tokens_in":24922,"tokens_out":1053,"would_cite":true,"duration_ms":22244,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":{"msc":[],"pacs":["12.38.Bx","13.85.-t"],"model":"selfhosted/deepseek-v4.1-flash","headline":"The two-loop QCD corrections to three-photon production are now written down in full analytic form and shipped as runnable code, so a five-point two-loop process can be used in phenomenology without re-deriving it.","keywords":["two-loop scattering amplitudes","perturbative QCD","three-photon production","numerical unitarity","pentagon functions","NNLO corrections","helicity amplitudes","collider phenomenology"],"falsifier":"Evaluate the shipped $H^{(2)}$ in double and quadruple precision on phase-space points deliberately placed near the surfaces where letters such as $s_{12}+s_{23}$ vanish, under cuts different from the ATLAS 8 TeV set used in the paper; a substantial rate of points below four correct digits there would undercut the stability claim, while an independent reimplementation matching tables 4 to 6 at the stated reference point would support it.","tokens_in":25257,"feed_emoji":"⚛️","tokens_out":2452,"duration_ms":46620,"temperature":0.7,"pith_summary":"The paper sets out to make the double-virtual NNLO QCD corrections to $qbar q to gammagammagamma$ usable, rather than merely computable. Its authors obtain the two independent helicity amplitudes at two loops in the leading-color approximation, first as an analytic decomposition into master integrals valid to all orders in the dimensional regulator $epsilon$, and then as finite remainders expanded in a basis of pentagon functions. The remainder coefficients are simplified by multivariate partial fractioning so that they can be evaluated quickly, and the whole thing is packaged in a public C++ library that returns the helicity- and color-summed squared remainders. The practical stake is that triphoton production has shown slow perturbative convergence and a tension with LHC Run 1 data at NLO, so a stable two-loop ingredient is what a full NNLO prediction needs. The authors report that the code evaluates the two-loop quantity in a few seconds per phase-space point and agrees with quadruple precision to better than four digits on more than 99.9 percent of 90000 sampled points.","feed_headline":"Two-loop amplitudes for three photons, ready to run","feed_subtitle":"A public code evaluates the five-point two-loop remainder in seconds, matching quadruple precision to four digits on 99.9% of sampled points","key_machinery":"Two ingredients carry the argument. The first is two-loop numerical unitarity: the integrand is decomposed into master integrands and surface terms, and the coefficients are fixed by solving cut equations built from products of tree amplitudes, evaluated in finite-field arithmetic so that the coefficients are exact rational numbers for rational phase-space points. The second is the pentagon-function basis for five massless particles, which turns the master-integral decomposition into a common function space shared by the one-loop and two-loop amplitudes and by the infrared subtraction operators, so that the finite remainder is again a linear combination of the same functions with rational coefficients. Reconstruction of those rational coefficients proceeds by fixing denominators from a subset of the symbol alphabet letters that are polynomial in the Mandelstam invariants, then fitting the numerators with a multivariate Newton method. The photon's lack of color ordering multiplies the pentagon-box topologies from five to thirty, which is why the non-planar alphabet, closed under all photon permutations, is needed even in the leading-color approximation.","core_discovery":"The central claim is a pair of firsts, stated in the abstract and again in section 3.3: analytic master-integral coefficients for a two-loop five-point QCD amplitude, and a five-point two-loop process released publicly in a form ready for immediate phenomenological use. Concretely, the authors compute $A^{(2,0)}_{-++}$, $A^{(2,N_f)}_{-++}$, $A^{(2,0)}_{+++}$ and $A^{(2,N_f)}_{+++}$ by two-loop numerical unitarity with finite-field arithmetic and functional reconstruction, keeping the full $epsilon$ dependence in the master-integral decomposition. They then convert to finite remainders in the pentagon-function basis of the five-massless-particle case, where the coefficient functions are rational in the Mandelstam invariants and $mathrm{tr}_5$ and have a simpler denominator structure than the master-integral coefficients themselves. The paper quantifies the gain: the largest remainder needs 171 independent rational functions against 1320 for the corresponding all-order coefficients, and the pole in $mathrm{tr}_5^2$ present in the coefficients is absent from the remainders. The released library evaluates the squared remainders $H^{(1)}$ and $H^{(2)}$ and was already used in a full NNLO triphoton calculation.","pith_inferences":["The paper's stability evidence is a population statistic, not a per-point bound: the library returns a number with no error flag, so a user running cuts that concentrate near surfaces where letters like $s_{12}+s_{23}$ vanish could silently receive a wrong value. A wrapper that evaluates at perturbed points and checks the spread would close this gap cheaply.","The claim of a first rests on a comparison the authors cannot make directly, since the earlier interpolation-based triphoton study does not publish explicit two-loop amplitudes; independent agreement is therefore established only indirectly, through the NNLO cross section it feeds. A direct numerical comparison at a handful of published phase-space points would make the claim much harder to disput","The reference values in tables 4 to 6, together with the assembly script, are the natural anchor for anyone reimplementing this: agreement at the stated phase-space point is a cheap and sharp test of an independent pipeline.","If the spurious-singularity explanation of the unstable points holds, then the same diagnostic should apply to the five-parton two-loop remainders already in the literature, which would make it a general tool rather than a fix specific to this process."],"forward_implications":["A complete NNLO QCD prediction for triphoton production becomes feasible on the same footing as diphoton, since the double-virtual ingredient can now be called as a library rather than reconstructed by interpolation.","The all-order-in-$epsilon$ master-integral decomposition gives a starting point for defining the three-loop remainder that N$^3$LO corrections would require.","Because the remainders have fewer spurious denominators than the master-integral coefficients, they are the better object to hand to a Monte Carlo, and the paper's own timing (seconds per point in double precision) supports on-the-fly use.","The same reconstruction pipeline, with photons treated as unordered colorless particles, extends to other five-point processes with more than one colorless final-state particle.","The observed correlation between unstable evaluations and proximity to spurious singularity surfaces gives a concrete diagnostic for building precision-rescue logic around the shipped code."],"supporting_citations":[{"why":"Supplies Caravel, the numerical-unitarity framework the authors modify to handle external photons and on which the whole computation runs.","marker":"[36]"},{"why":"Provides the pentagon-function basis for five massless particles and the evaluation code the released library links against.","marker":"[23]"},{"why":"Supplies the reconstruction and multivariate partial-fractioning procedure the authors adapt from remainders to master-integral coefficients.","marker":"[49]"},{"why":"Establishes the integrand decomposition into master integrands and surface terms that the cut equations are built on.","marker":"[25]"},{"why":"Underlies the finite-field evaluation and functional reconstruction that make exact rational coefficients accessible.","marker":"[26]"},{"why":"Defines the infrared subtraction operators $I^{(1)}$ and $I^{(2)}$ used to convert renormalized amplitudes into finite remainders.","marker":"[54]"},{"why":"Provides the non-planar symbol alphabet whose polynomial subset fixes the denominators of the rational coefficients.","marker":"[65]"},{"why":"The NNLO triphoton calculation that already consumes these amplitudes, and the source of the phase-space distribution used for the stability study.","marker":"[16]"}],"fun_headline_variants":["Three photons at two loops, packaged as public code","First public five-point two-loop process: three photons","Triphoton two-loop remainders cut 1320 functions to 171","Two-loop triphoton amplitudes in analytic master integrals","Five-point two-loop QCD remainders, stable across phase space"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The claim that the released code is trustworthy rests on a sample statistic rather than a guarantee: fewer than 0.1 percent of 90000 points under one specific set of LHC cuts fell below four correct digits, the library returns no per-point error estimate, and the authors explicitly defer a precision-rescue system to future work.","fun_headline_variants_meta":{"raw":{"variants":["Three photons at two loops, packaged as public code","First public five-point two-loop process: three photons","Triphoton two-loop remainders cut 1320 functions to 171","Two-loop triphoton amplitudes in analytic master integrals","Five-point two-loop QCD remainders, stable across phase space"]},"model":"selfhosted/deepseek-v4.1-flash","effort":"low","attempts":1,"cost_usd":0.0,"raw_usage":{"total_tokens":1176,"prompt_tokens":948,"completion_tokens":228,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":157}},"tokens_in":948,"tokens_out":228,"duration_ms":6318,"image_pages":[],"temperature":1.0,"reasoning_budget":50,"reasoning_tokens":157,"transport_provider":"selfhosted","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-10-31T01:00:11.939921+00:00","model_set":{"reader":"selfhosted/deepseek-v4.1-flash"},"falsifier":"Evaluate the shipped $H^{(2)}$ in double and quadruple precision on phase-space points deliberately placed near the surfaces where letters such as $s_{12}+s_{23}$ vanish, under cuts different from the ATLAS 8 TeV set used in the paper; a substantial rate of points below four correct digits there would undercut the stability claim, while an independent reimplementation matching tables 4 to 6 at the stated reference point would support it.","supporting_citations":[{"cited_title":"Caravel: A C++ Framework for the Computation of Multi-Loop Amplitudes with Numerical Unitarity","cited_arxiv_id":"2009.11957","evidence_quote":"Supplies Caravel, the numerical-unitarity framework the authors modify to handle external photons and on which the whole computation runs."},{"cited_title":"Pentagon Functions for Scattering of Five Massless Particles","cited_arxiv_id":"2009.07803","evidence_quote":"Provides the pentagon-function basis for five massless particles and the evaluation code the released library links against."},{"cited_title":"Analytic Form of the Planar Two-Loop Five-Parton Scattering Amplitudes in QCD","cited_arxiv_id":"1904.00945","evidence_quote":"Supplies the reconstruction and multivariate partial-fractioning procedure the authors adapt from remainders to master-integral coefficients."},{"cited_title":"Two-loop integrand decomposition into master integrals and surface terms","cited_arxiv_id":"1510.05626","evidence_quote":"Establishes the integrand decomposition into master integrands and surface terms that the cut equations are built on."},{"cited_title":"Scattering amplitudes over finite fields and multivariate functional reconstruction","cited_arxiv_id":"1608.01902","evidence_quote":"Underlies the finite-field evaluation and functional reconstruction that make exact rational coefficients accessible."},{"cited_title":"The singular behaviour of QCD amplitudes at two-loop order","cited_arxiv_id":"hep-ph/9802439","evidence_quote":"Defines the infrared subtraction operators $I^{(1)}$ and $I^{(2)}$ used to convert renormalized amplitudes into finite remainders."},{"cited_title":"Bootstrapping pentagon functions","cited_arxiv_id":"1712.09610","evidence_quote":"Provides the non-planar symbol alphabet whose polynomial subset fixes the denominators of the rational coefficients."},{"cited_title":"Triphoton production at hadron colliders in NNLO QCD","cited_arxiv_id":"2010.04681","evidence_quote":"The NNLO triphoton calculation that already consumes these amplitudes, and the source of the phase-space distribution used for the stability study."}],"review_version":1}