{"schemaVersion":1,"records":[{"schemaVersion":1,"status":"published","id":"ym-wilson-20261028","slug":"wilson-limit-yang-mills","title":"A Wilson lattice limit for four-dimensional Yang–Mills theory","summary":"The Yang–Mills Group judged that OpenAI’s October submission constructs a relativistic quantum field theory from pure Wilson lattice gauge theory for each compact simple gauge group. The infinite-volume step selects a subsequence. Uniqueness and a mass gap remain open.","topic":"Constructive quantum field theory","aliases":["Yang-Mills","Yang–Mills","Wilson lattice gauge theory","Yang Mills","Wightman axioms","Osterwalder–Schrader reconstruction","mass gap","lattice gauge theory","continuum limit","formal verification","Lean","OpenAI"],"firstPublished":"2026-11-20","scientificUpdated":"2026-11-20","lastSourceChecked":"2026-11-20","sections":[{"id":"result","title":"What the group assessed","paragraphs":["OpenAI submitted Lean code on 28 October 2026. The shepherding committee announced its release on 9 November, and the group issued its assessment on 10 November. The group concluded that the construction yields a field theory satisfying the Wightman axioms. To identify it as Yang–Mills, the assessment examines where its correlations come from, which local fields survive the limit, and how those fields behave at short distances.","The result concerns a pure gauge field in four dimensions. In the fixed-volume theorem, the gauge group is represented as a closed subgroup of the unitary group with simple real Lie algebra. Gauge-invariant local fields are built from curvature and its covariant derivatives. The limiting argument smears correlations with smooth, compactly supported joint test functions. Their support avoids coincident insertion points and, at fixed volume, lies strictly inside the comparison cell."],"sources":["assessment","kitp","statement"]},{"id":"limits","title":"Two limits, taken in order","paragraphs":["Start on a periodic Euclidean lattice and hold its physical side length fixed. As the spacing shrinks along the sequence supplied by the construction, the bare coupling runs and the local lattice observables are renormalized. Their smeared correlations converge along that entire chosen sequence. The theorem does not assert convergence for arbitrary choices of spacing and coupling.","Next increase the physical side length. Bounds on correlations and their variation on compact sets allow a simultaneous subsequence to be extracted for all the correlators. The resulting Euclidean correlations can mix phases. The construction decomposes the limit into clustering components and applies Osterwalder–Schrader reconstruction to obtain relativistic theories from them.","Different choices of infinite-volume subsequence have not been shown to agree. The construction also leaves open whether the Hamiltonian has a positive spectral gap above the vacuum. The existence assessment therefore settles neither uniqueness nor the mass-gap part of the Clay problem."],"sources":["assessment"]},{"id":"wilson-law","title":"Checking the finite-lattice starting point","paragraphs":["Two implementations can both call their distribution the Wilson law while assigning a different coefficient to the same plaquette. The group compared the submission to a separate lattice gauge theory formalization called lgt. Both put a group element on each oriented link of a periodic four-dimensional lattice. Following the four links of a plaquette in order, with inverses on the return edges, gives its holonomy.","Let t be the real part of the trace of that holonomy in the chosen representation, and let m be the matrix dimension of the representation. The submission uses the cost 1 − t/m per plaquette. The lgt convention uses m − t. Equating their contributions to the action fixes the conversion below.","The action identity is only the first step. The Lean bridge reindexes the links, preserves their product Haar measure, and matches the Boltzmann weights and partition functions. Its law_preserving theorem consequently identifies the normalized probability laws on each finite lattice. An integrable observable therefore has the same expectation after reindexing and converting the coupling."],"sources":["submission","bridge","assessment"]},{"id":"use","title":"Using the comparison","paragraphs":["To compare a plaquette expectation against the October construction, first check the trace convention in the action. With a two-dimensional representation the submission’s coefficient is twice the lgt coefficient. Comparing at the same numerical coefficient would generally compare different probability measures. After the conversion, check the plaquette orientation and the normalization of the observable itself.","This identity establishes the Wilson origin of the finite-lattice integrals. Passing from those integrals to continuum correlations still requires the renormalization and extraction arguments. For those steps, read Theorems 4.1 and 5.1 of the assessment in that order."],"sources":["bridge","assessment"]},{"id":"attribution","title":"Who did what","paragraphs":["OpenAI reports that its models produced a natural-language existence proof and that a Lean formalization followed. The group based its November assessment on the submitted Lean source. Michael R. Douglas, Arthur Jaffe, Ron Nissim and Fred Rajasekaran prepared the assessment with contributions from additional colleagues, and the group approved it. They credit OpenAI’s Astra with helping translate Lean into informal mathematics and develop the additional tests in Part II.","The group reports a clean build, an axiom audit and Comparator checks. Its supplementary record for the finite-lattice worked example records ordinary Lean compilation against an existing acceptance build. Our examination followed the published definitions and theorem statements. We did not run Lean or Comparator."],"sources":["openai","assessment","record"]}],"equation":{"tex":"\\beta_{\\rm sub}\\!\\left(1-\\frac{t}{m}\\right)=\\beta_{\\rm lgt}(m-t),\\qquad \\beta_{\\rm sub}=m\\,\\beta_{\\rm lgt}.","caption":"The coefficient conversion makes the actions agree for every plaquette holonomy. Here m is the dimension of the matrix representation.","source":"bridge"},"sources":[{"id":"assessment","title":"Validation of the OpenAI Yang–Mills construction","authors":"Yang–Mills Group","date":"2026-11-10","url":"https://github.com/yang-mills-group/YM/blob/b379615f559643d30b49fb2e05e1efb83d755fc8/docs/assessment-1.pdf","locator":"p. 1, initial assessment and scope. §4, eq. (3) and Theorem 4.1. §§5–6, Theorem 5.1 and phase reconstruction. §8, open questions. §§9–12, checks. p. 22, acknowledgements."},{"id":"submission","title":"PublicWilson.lean","authors":"OpenAI","date":"2026-10-28","url":"https://github.com/yang-mills-group/YM/blob/b379615f559643d30b49fb2e05e1efb83d755fc8/submissions/openai-20261028/submission/PublicWilson.lean#L20-L33","locator":"October submission. YangMills.plaquette, wilsonEnergy and wilsonLaw define the ordered product, normalized trace cost and probability law."},{"id":"bridge","title":"LGTWilsonBridge.lean","authors":"Yang–Mills Group","date":"2026-11-10","url":"https://github.com/yang-mills-group/YM/blob/b379615f559643d30b49fb2e05e1efb83d755fc8/submissions/openai-20261028/checks/LGTWilsonBridge.lean#L35-L133","locator":"YMLGTBridge.plaquette_eq, action_eq (lines 67–79), law_preserving (107–120), integrable_iff and expectation_eq (124–133)."},{"id":"record","title":"Supplementary worked-example compilation record","authors":"Yang–Mills Group","date":"2026-11-10","url":"https://github.com/yang-mills-group/YM/blob/b379615f559643d30b49fb2e05e1efb83d755fc8/submissions/openai-20261028/evidence/examples-2026-11-10/additional/README.md","locator":"A3/A5 lattice-and-scale example and the stated limits of ordinary compilation against an existing build."},{"id":"kitp","title":"Statement by the Yang-Mills Working Group","authors":"Kavli Institute for Theoretical Physics","date":"2026-11-09","url":"https://www.kitp.ucsb.edu/statement-by-the-yangmills-working-group","locator":"November 9 update and its link to the third committee statement."},{"id":"statement","title":"Third statement by the Shepherding Committee","authors":"Michael R. Douglas, David J. Gross, Arthur Jaffe, Igor R. Klebanov and Nikita Nekrasov","date":"2026-11-09","url":"https://www.kitp.ucsb.edu/sites/default/files/3rd-Statement-About-Quantum-YangMills-Theory_Nov-9.pdf","locator":"p. 1, announcement of the two repositories and the October submission date."},{"id":"openai","title":"Working with academic communities to build scientific understanding","authors":"OpenAI","date":"2026-10-26","url":"https://openai.com/index/building-scientific-understanding-with-academic-communities/","locator":"Physics section, paragraph beginning “Our models produced”."}],"verification":{"method":"Primary-source comparison","scope":"Read the dated assessment, the submitted Wilson law and the group’s bridge. Compared the plaquette normalization, the direction of the coupling conversion and the order of the two limits. Lean and Comparator were not replayed.","leanReplayed":false},"limits":{"massGap":false,"uniqueInfiniteVolume":false}},{"schemaVersion":1,"status":"published","id":"single-minus-gluons-2602-12176","slug":"single-minus-gluons","title":"Where a single-minus gluon tree amplitude survives","summary":"One negative-helicity gluon and three positive-helicity gluons have a vanishing tree amplitude at generic momenta. In real split signature, a distribution supported on half-collinear momenta can survive. Its coefficient changes as the momenta cross chamber boundaries.","topic":"Scattering amplitudes","aliases":["single-minus","single minus","half-collinear","helicity","gluon","Klein signature","split signature","Berends–Giele recursion","GPT-5.2 Pro","arXiv:2602.12176"],"firstPublished":"2026-11-20","scientificUpdated":"2026-11-20","lastSourceChecked":"2026-11-20","primarySource":"paper","calculationSlug":"single-minus","sections":[{"id":"support","title":"Where the usual vanishing argument breaks","paragraphs":["Write each massless momentum as an undotted spinor times a dotted one. Angle brackets contract the first pair of spinors and square brackets the second. For generic momenta, using the negative-helicity leg as a reference spinor for the positive-helicity polarizations makes the usual tree-level vanishing argument work. On the half-collinear locus every angle bracket vanishes. That reference choice now divides by zero.","In real signature (2,2), square brackets can still differ from zero on this locus. Guevara and colleagues derive a color-ordered amplitude whose delta functions enforce half-collinearity and momentum conservation. The quantity to compare between allowed momenta is the coefficient that remains after those factors and the helicity prefactor are removed."],"sources":["paper"]},{"id":"formula","title":"A four-point chamber coefficient","paragraphs":["Take leg 1 to have negative helicity and the other three to have positive helicity. Work in the decay region R₁, where a frame exists with negative frequency for leg 1 and positive frequencies for the remaining legs. In this region the authors reduce their recursion to a product of signs. For four legs it has two factors.","Write sᵢⱼ for the sign of the square bracket of legs i and j. In s₁,₂₃ the second spinor is the sum of the dotted spinors of legs 2 and 3. Each sign is constant until its bracket crosses zero. The equation therefore assigns a constant coefficient to the interior of each chamber."],"sources":["paper"]},{"id":"contribution","title":"From recursion to a product","paragraphs":["The Berends–Giele recursion builds currents from lower-point currents. Its four-point coefficient, Eq. (30), is already short. At higher multiplicity the recursion yields many more terms. The all-multiplicity product in Eq. (39) gives a compact answer in R₁. The authors credit GPT-5.2 Pro with conjecturing that formula and an internal OpenAI model with producing a proof. They report checking the solution by hand against the recursion."],"sources":["paper"]},{"id":"comparison","title":"Comparing two expressions at the same momenta","paragraphs":["Our exact calculation compared the four-point recursion result to the four-point specialization of the product. We set all undotted spinors to (1,0), chose positive frequencies for legs 2 through 4, and fixed the dotted spinor of leg 1 by momentum conservation. With dotted spinors (−6,−4), (1,−2), (2,0), (3,6), both expressions give +1.","Across twelve retained inputs we found four values each of +1, −1 and 0, with agreement at every point. Six additional inputs had a vanishing pair bracket and were excluded. Reversing the sign of one term in Eq. (30) produced a discrepancy at every retained point. The linked calculation gives the spinors and the code for repeating the comparison."],"sources":["paper"]}],"equation":{"tex":"A_{1234}\\big|_{\\mathcal R_1}=\\tfrac14(s_{23}+s_{12})(s_{34}+s_{1,23})","caption":"Equation (39) at four points in the paper’s fixed spinor frame. The signs use the square-bracket orientation in Eq. (5).","source":"paper"},"sources":[{"id":"paper","title":"Single-minus gluon tree amplitudes are nonzero","authors":"Alfredo Guevara, Alexandru Lupsasca, David Skinner, Andrew Strominger and Kevin Weil","authorList":["Guevara, Alfredo","Lupsasca, Alexandru","Skinner, David","Strominger, Andrew","Weil, Kevin"],"date":"2026-03-09","url":"https://arxiv.org/html/2602.12176v2","locator":"arXiv:2602.12176v2. Introduction, model and author contributions. Eqs. (2)–(5), spinor frame. §I.1, Eqs. (13)–(16) and footnote 6, support and signature. §I.4, Eq. (30), four-point coefficient. §§II.1–3, Eqs. (33) and (39), decay region and product."}],"verification":{"method":"Exact evaluation on half-collinear momenta","scope":"Evaluated Eqs. (30) and (39) for four external gluons on twelve chamber-interior inputs in R₁, using an independently written script. This checks sampled coefficients. The distributional derivation and all-multiplicity proof were outside the calculation."},"limits":["The calculation concerns a real split-signature distribution on special kinematics. Generic Lorentzian scattering remains outside this result.","The equations and numbering follow arXiv v2. Footnote 6 leaves the continuation to complex momenta open."]},{"schemaVersion":1,"status":"published","id":"learning-amplitudes-2408-04720","slug":"learning-amplitudes","title":"Learning to simplify a gluon amplitude","summary":"Cheung, Dersy and Schwartz trained networks to find groups of spinor-helicity terms that simplify together. A four-gluon example reduces 17 terms to one. Its value can be checked against the known Parke–Taylor amplitude using exact arithmetic.","topic":"Scattering amplitudes","aliases":["spinor-helicity","MHV","maximally helicity violating","Parke–Taylor","Parke-Taylor","gluon","transformer","contrastive learning","Schouten","arXiv:2408.04720"],"firstPublished":"2026-11-20","scientificUpdated":"2026-11-20","lastSourceChecked":"2026-11-20","primarySource":"paper","calculationSlug":"four-gluon","sections":[{"id":"method","title":"Choosing which terms to simplify","paragraphs":["A sum of tree diagrams can conceal a short answer. In a spinor-helicity representation, momentum conservation and the Schouten identity let visibly different products of brackets represent the same amplitude. The difficulty for a simplifier is deciding which terms to combine before applying those identities.","Cheung, Dersy and Schwartz generated training pairs by expanding short expressions with known identities. They trained an encoder-decoder transformer to reverse the process on short inputs. For longer inputs, a separate model trained with contrastive learning groups terms likely to simplify together. The sequential algorithm sends those groups through the transformer. The authors tested the method on amplitudes computed with established physics techniques."],"sources":["paper"]},{"id":"formula","title":"A model output with an analytic benchmark","paragraphs":["Consider the four-dimensional massless gluon tree amplitude with helicities (--++), cyclic color order (1,2,3,4), and all momenta outgoing. Appendix G gives a 17-term input for a particular choice of reference spinors. The model returns the expression on the left below. The right side is the known Parke–Taylor formula with negative helicities on legs 1 and 2.","Angle and square brackets are antisymmetric spinor contractions. Keep the cyclic denominator in order, especially the final angle bracket from leg 4 back to leg 1. Reversing that bracket changes the sign. Both forms use the same stripped normalization and apply at on-shell, momentum-conserving points where the displayed denominators are nonzero."],"sources":["paper"]},{"id":"comparison","title":"A check with exact momenta","paragraphs":["We chose undotted and dotted spinors with rational components and solved momentum conservation before evaluating either formula. The spinors may vary independently at complexified kinematics. At the three points in the calculation file, both formulas gave, in order, 1/2, 3/70 and 49/40.","A second check changes the spinors of one leg while leaving its momentum fixed. Multiplying its undotted spinor by 3/2 and its dotted spinor by 2/3 multiplies the amplitude by 9/4 for a negative-helicity gluon and by 4/9 for a positive-helicity gluon. Both formulas gave the expected weight for each leg at all three points. Flipping the overall sign broke equality, and the script rejected an input that violated momentum conservation."],"sources":["paper"]},{"id":"beyond","title":"What the benchmark tells us","paragraphs":["The four-gluon example tests whether a learned simplification recovers the value and helicity weights of a known amplitude. The paper also reports a reduction from 298 terms to two for three massless scalars and two positive-helicity gravitons in a theory with minimal gravitational coupling and cubic scalar self-interaction, Eq. (4.7). That example and the input-length dependence in Fig. 9 are separate parts of the authors’ benchmark.","For another proposed compact amplitude, start by checking that all terms have the same little-group weights. Then test the expression against an independent representation at nonsingular, momentum-conserving points. A mismatch at one admissible point disproves an identity. Agreement at finitely many points still leaves a symbolic identity to establish."],"sources":["paper"]}],"equation":{"tex":"-\\frac{\\langle12\\rangle[34]^2}{\\langle34\\rangle[14][23]}=\\frac{\\langle12\\rangle^4}{\\langle12\\rangle\\langle23\\rangle\\langle34\\rangle\\langle41\\rangle}","caption":"The model’s compact output after Eq. (G.2) and the n = 4 specialization of Eq. (1.1), with coupling and overall phase stripped.","source":"paper"},"sources":[{"id":"paper","title":"Learning the Simplicity of Scattering Amplitudes","authors":"Clifford Cheung, Aurélien Dersy and Matthew D. Schwartz","authorList":["Cheung, Clifford","Dersy, Aurélien","Schwartz, Matthew D."],"date":"2024-11-19","url":"https://arxiv.org/html/2408.04720v2","locator":"arXiv:2408.04720v2. §1, Eq. (1.1), Parke–Taylor benchmark. §§4.2–4.3, grouping and sequential simplification. §4.4, Fig. 9 and Eq. (4.7), physical examples. Appendix G, four-gluon input in Eq. (G.2) and the unnumbered output immediately following."}],"verification":{"method":"Exact rational evaluation and helicity scaling","scope":"Compared the published compact output with Parke–Taylor at three rational points and twelve individual leg rescalings. The site calculation did not run model training or inference, transcribe the full 17-term input, or evaluate the five-point example."},"limits":["The comparison tests one tree-level model output. It supplies neither a general proof of equivalence nor a guarantee that learned simplifications are minimal.","The equation locators and tested expression follow arXiv v2 of 19 November 2024."]}]}