{"id":"07b8e767-abc1-4e3f-aacc-a75735ae135d","arxiv_id":"hep-ph/0012260","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"FeynArts 3 generates Feynman diagrams at three levels of field detail, inserts model-allowed particles into them, and emits both publication-quality LaTeX or PostScript figures and symbolic amplitudes.","lead":"FeynArts 3 is a Mathematica package that enumerates Feynman diagrams for a given process and turns them into symbolic amplitudes, and this paper documents how it works and how to drive it. Physicists who compute higher-order corrections can use it to replace weeks of hand drawing and algebra with a scripted pipeline.","discovery_kind":"unclear","skeptic_critique":{"model":"selfhosted/deepseek-v4.1-flash","headline":"The completeness and symmetry-factor claims of CreateTopologies (Section 3.1) rest on an asserted power-set equivalence test with no independent enumeration; the single quoted count, 2214 two-loop 2 to 2 topologies, is the one place a silent error would surface.","rationale":"Genre matters here. This is a CPC program description; the accepted evidence standard is a clear algorithmic account plus distributed, runnable code, and FeynArts meets that (open source, downloadable, manual included). Under instruction 7 the shipped code is itself evidence, which is why I do not treat the absence of a QGRAF cross-check as a defect that sinks the claim. My concern is narrower: the one place the paper makes a sharp quantitative assertion about completeness, namely 2214 two-loop 2 to 2 topologies, is unverified, and the symmetry-factor sentence in Section 3.1 is compressed enough that a reader cannot confirm the free-action assumption it relies on. That is worth one check, not a rejection. On the reader's three grounds: (i) the SUSY point is weaker than the reader allows. The abstract's 'support for supersymmetric models' concerns the code's ability to handle Majorana fermions and fermion-number-violating couplings, and Section 3.6 demonstrates exactly that via the flipping-rule algorithm and Eq. (2). The missing MSSM model file is a data file, and the program-summary box states plainly that model files other than SM and QCD 'must be set up by the user.' So this is disclosed, not promised and withheld. (ii) The performance figure lacking hardware detail is a minor reporting gap, not load-bearing. (iii) The cross-check concern is the real one, and I agree with it, but I would frame it as a validation gap rather than a correctness risk, since the code is available for anyone to run the check. Hence partial agreement and UNCHANGED: the reader's CONDITIONAL is defensible, and my sharpened version of concern (iii) does not move it.","tokens_in":10560,"tokens_out":5562,"duration_ms":108376,"concrete_test":"Run QGRAF, or an independent topology enumerator, for two-loop 2 to 2 one-particle-irreducible topologies and compare the count to the 2214 quoted in Section 3.1. Separately recompute the symmetry factor of the one-loop 2 to 2 topologies and a sample of two-loop ones directly from |Aut| and compare to the RelativeCF values FeynArts assigns. A mismatch in either the count or the symmetry factors would show the power-set equivalence and symmetry-factor procedure is incomplete; agreement would close the concern.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central technical claim is that CreateTopologies returns exactly the distinct topologies with correct symmetry factors (Section 3.1). The paper is candid that the naive canonical-order comparison can miss equivalent graphs when symmetric vertex groups exist, and it supplies the repair: compare the 'power set' obtained by permuting the tagged symmetric-vertex indices. What it does not supply is an argument that this repair is complete, i.e. that permuting those indices generates the full orbit and that the orbit size equals the ratio of automorphism groups used to divide s ('s is divided by the number of versions found'). The symmetry-factor bookkeeping is stated in a single sentence and, read literally, is delicate: dividing 1/|Aut_old| by the number of equivalent versions yields 1/|Aut_new| only under a free-action assumption the text does not state. If that assumption fails for some topology class, the error is silent and propagates into every RelativeCF and hence every amplitude. The only quantitative anchor is '2214 two-loop 2 to 2 topologies' in about 10 seconds, quoted without hardware, version, or independent enumeration. The mitigating factor is real: the code is open source and downloadable, so the claim is checkable by any reader, but the paper itself performs no such check, and the one worked example (Z to b bbar, 14 classes-level diagrams) is likewise not compared to an independent generator. This is a validation gap rather than a demonstrated error.","agreement_with_reader":"partial"},"referee_report":{"model":"selfhosted/deepseek-v4.1-flash","summary":"This is a program-description paper for FeynArts 3, a Mathematica package that generates Feynman topologies, inserts model-allowed fields into them, applies Feynman rules to produce symbolic amplitudes, and renders publication-quality diagrams in PostScript or LaTeX. The paper is organized as a CPC-style Program Summary followed by a long write-up: Section 2 walks a user through the three-step workflow (CreateTopologies, InsertFields, CreateFeynAmp) on the example Z -> b bbar, and Section 3 explains the underlying concepts, namely recursive leg attachment from pre-supplied starting topologies with a power-set equivalence test and symmetry-factor division (Section 3.1), the three field levels of generic, classes, and particles (Section 3.2), the intersection-of-allowed-lists insertion algorithm including mixing propagators (Section 3.3), the graphics pipeline and LaTeX output format (Section 3.4), the structure of the FeynAmp output with its generic amplitude plus Insertions replacement rules (Section 3.5), and the flipping-rule algorithm that permits Majorana fermions and fermion-number-violating couplings (Section 3.6). No new physics results are claimed; the contribution is the tool, its algorithmic design, and its availability under the GNU library general public license.","tokens_in":10793,"tokens_out":3884,"duration_ms":80502,"significance":"FeynArts is one of the standard front ends of the perturbative QCD and electroweak pipeline, and this write-up documents the design decisions that made version 3 usable at scale. Three elements are genuinely valuable and are stated clearly: the three-level field concept (Section 3.2), which lets the bulk of the algebra be done once at generic level and the particle sums be deferred to a fast numerical stage; the flipping-rule algorithm (Section 3.6, Eq. (2)), which extends automated diagram generation to Majorana fermions and hence to supersymmetric models where the textbook arrow-ordering prescription fails; and the custom PostScript prologue with the feynarts.sty style, which removes external figure files from LaTeX documents. The paper also earns credit for shipping open-source, downloadable code with a comprehensive manual, for quoting concrete performance figures (2214 two-loop 2 -> 2 topologies in about 10 seconds; about a minute for a one-loop 2 -> 2 electroweak amplitude set), and for stating its restrictions plainly in the Program Summary rather than burying them. If the tool behaves as described, the practical value to the community is high and durable.","major_comments":[{"comment":"The symmetry-factor bookkeeping is compressed into a single sentence: after each leg is added, topologically equivalent versions are gathered and replaced by one representative whose s is divided by the number of versions found. Read literally, dividing 1/|Aut_old| by the number of equivalent versions yields the correct 1/|Aut_new| only if the permutation action used to generate those versions is free on the relevant set, and the text does not state or justify that assumption. The same paragraph concedes that the naive canonical-order comparison can miss equivalent graphs when symmetric vertex groups exist, and offers the power-set permutation of tagged symmetric-vertex indices as the repair, but gives no argument that this permutation generates the full orbit rather than a subset of it. Because s propagates directly into RelativeCF and hence into every amplitude, an incompleteness here would be silent. A short orbit-stabilizer argument, or a pointer to the precise statement in Ref. [9], would close this.","section":"Section 3.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper contains no independent validation of the diagram set or of the symmetry factors. The only quantitative anchor is the count of 2214 two-loop 2 -> 2 topologies in about 10 seconds, quoted without machine specification, Mathematica version, or comparison to an independent enumeration, and the single worked example (Z -> b bbar, 14 classes-level diagrams) is not compared to a published result or to a comparable generator such as QGRAF [4] or GRACE [5], both of which the paper itself names as equivalent in scope. Since the central claim is that the returned diagram set is exactly the set the model allows, a short validation subsection reporting at least one cross-check of topology counts and one of a known amplitude would materially strengthen the paper. This is a validation gap rather than a demonstrated error, and the open-source distribution makes the check feasible for any reader.","section":"Section 3.1 and Section 2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The abstract lists 'support for supersymmetric models' among the main features of version 3, but Section 2 states that a model file for the Minimal Supersymmetric Standard Model 'will be available soon', and the Program Summary states under Restrictions on the complexity that model files other than the Standard Model and QCD are not contained in FeynArts and must be set up by the user. What version 3 actually provides is the algorithmic capability (the flipping rule of Section 3.6) plus the absence of a shipped MSSM model file. The abstract phrasing will lead users to expect a file that is not in the distribution. Rewording the abstract to distinguish algorithmic support from shipped model files would resolve this without any change to the technical content.","section":"Abstract and Section 2","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The entries 'Memory required to execute with typical data: 8M words' and 'No. of bits in a word: 8' appear inconsistent with each other and with the platforms listed; if 8M words at 8 bits per word is intended, the resulting figure is small for a Mathematica session, and if a 32- or 64-bit word is intended the memory entry should be restated. Please check these two form entries.","section":"Program Summary","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The timing figures ('about a minute' for a one-loop 2 -> 2 electroweak amplitude set, 'about 10 seconds' for the 2214 two-loop topologies) are useful but are given without processor, clock speed, or Mathematica version. A parenthetical specification would make them reproducible.","section":"Section 2 and Section 3.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The three-loop limit is presented as a restriction of the program, but Section 3.1 makes clear that the loop order is set by which starting topologies are supplied as external input, so the limit is a property of the shipped data rather than of the algorithm. Saying so at the point where the restriction is listed would help users who need higher orders.","section":"Program Summary, Restrictions on the complexity","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The treatment of mixing propagators is dispatched in two sentences, with the gamma-Z mixing field as the only example. Since the left/right partner distinction changes which lookup table is consulted at each end, a short explicit statement of how the partner fields are recorded at model-initialization time would make this reproducible for users writing their own model files.","section":"Section 3.3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"PickLevel is introduced in one sentence as the way to obtain conventional amplitudes by applying the Insertions replacement rules, but no example of its use or output is given, in contrast to the care taken with the CreateFeynAmp example. A two-line illustration would complete the section.","section":"Section 3.5","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The details of the power-set equivalence test and of the starting-topology construction reside in Ref. [9], a Ph.D. thesis distributed only as a PostScript file from the FeynArts website. Restating the essential definitions in the paper itself would make the algorithm self-contained for readers without web access to the thesis.","section":"Section 3.1, Ref. [9]","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"A small typographical point: 'forseeable' in the first paragraph of Section 1 should read 'foreseeable', if this spelling survives in the typeset version.","section":"Section 1","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"minor_revision","confidential_remarks":"This is a solid program-description paper of the kind the journal publishes routinely, and the technical content does not need reworking. The two substantive items I would want to see before acceptance are a short validation paragraph (a cross-check of topology counts and one amplitude against an independent generator or a published result) and a rewording of the abstract's supersymmetry claim, which currently promises more than the distributed package contains. Both are additions of a paragraph or a sentence and should not require another full review cycle. The paper is otherwise well matched to the journal's scope."},"author_rebuttal":null,"desk_editor":{"model":"selfhosted/deepseek-v4.1-flash","letter":"This is the FeynArts 3 writeup — a program description, not a physics result, and it should be read that way. The tool is a workhorse for one- to three-loop diagram generation, and the paper is a clear, well-organized account of how it works: topologies from starting graphs by recursive leg attachment, field insertion by intersecting allowed-field lists at each propagator end, Feynman rules applied symbolically with the generic amplitude kept separate from the level-substitution rules. The three-level field scheme and the flipping-rule treatment of undirected fermion lines are prior work by Eck, Küblbeck and Denner, and the paper says so plainly rather than dressing them up as new. What is genuinely new in v3 is the graphics layer: the custom PostScript prologue, the LaTeX output with feynarts.sty, the Java topology editor, and the roughly fivefold reduction in PostScript file size. That is a real, if unglamorous, contribution, and the LaTeX fragment in Section 3.4 shows it working rather than just asserting it does. The paper is also candid about its own boundaries: the topology algorithm is not self-contained and depends on externally supplied starting topologies (Section 3.1), and the naive canonical-order comparison can miss equivalent graphs when symmetric vertex groups exist, for which the power-set permutation of tagged indices is offered as the repair. The stress-test note is right that the repair comes without a completeness argument — the symmetry-factor bookkeeping is one sentence, and dividing s by the number of versions found is only correct under a free-action assumption the text never states. But this is a validation gap in a paper whose code is open source and downloadable, not a demonstrated error, and I would not weight it heavily. The concrete soft spot is the abstract: it lists 'support for supersymmetric models' as a main feature of version 3, while Section 2 says the MSSM model file 'will be available soon' and no SUSY example appears anywhere. That is a promise in the abstract and nothing in the body to back it. Second, the 2214-topologies-in-ten-seconds figure has no hardware, memory or version attached, and there is no cross-check of diagram counts or symmetry factors against QGRAF or GRACE, which the paper itself names as comparable. Both are minor and fixable. Who should read it: anyone about to use FeynArts, and anyone writing their own generator who wants a clean description of the three-level insertion idea. It deserves referees — they can run the code, which is the right test here.","headline":"A solid, honest program paper for a tool people actually use; the one real mismatch is the abstract's SUSY claim against a body that ships no SUSY example.","tokens_in":11480,"tokens_out":1056,"would_cite":true,"duration_ms":24109,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":null,"created_at":"2026-10-29T18:00:01.580462+00:00","model_set":{"reader":"selfhosted/deepseek-v4.1-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}