{"id":"4c558349-4ca6-480c-a2ba-8e9f6d67d644","arxiv_id":"2608.03646","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Intersection matrices of geometric-ordered Feynman integral bases are Laurent polynomials or, after a power-of-epsilon factor, integers, which enables systematic elimination of redundant auxiliary functions on the maximal cut.","lead":"This paper shows that intersection matrices of Feynman-integral master integrals from geometric ordering are simpler than expected: Laurent polynomials in epsilon for one basis, integers for another. The result yields an algorithm that removes redundant transcendental auxiliary functions in epsilon-factorised differential equations.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The general claims (Laurent-polynomial and integer intersection matrices) rest on an unproven existence condition: boundary constants must solve the algebraic system in Algorithm 1 step 4 with det R2 ≠ 0; the paper supplies examples, not a proof, and its own eq.","rationale":"The reader's weakest assumption is also the load-bearing point in my reading: the paper's headline claims are conditional on the existence of boundary values for the auxiliary functions that make the algebraic system of Algorithm 1 step 4 solvable with det R2 nonzero. The paper demonstrates this in examples and even shows that a generic choice of one boundary constant breaks constancy, but no general existence proof or degree-of-freedom count is supplied. This does not invalidate the worked examples, and the paper is candid about the conditionality, so the appropriate verdict remains conditional acceptance. No change to the reader's verdict is needed.","tokens_in":23768,"tokens_out":9017,"duration_ms":88944,"concrete_test":"Pick a sector from refs. [19,20] not treated in the paper, e.g. the five-loop equal-mass banana or a two-loop non-planar diagram with internal masses. Write R2 with all auxiliary-function boundary constants as symbols, expand C = R2^{-1} ˜C (R2^{∨T})^{-1} in ε, and assemble the equations (91): C(k)=0 for k<0 and C(0)=N with N a symbolic symmetric invertible constant matrix. Count the number of independent equations against the number of free boundary constants, and compute the Jacobian with respect to those constants. If the system has a solution with det R2 ≠ 0 and det N ≠ 0, the existence premise is supported; if it is overdetermined or forces det R2 = 0, the general claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central structural statement, abstract and §2.1, has two parts. For the ε-factorised basis, 'entries are integers up to an overall power of ε if the boundary values ... are chosen appropriately' requires that there exist integration constants for the auxiliary functions in R2 such that, after expanding C = R2^{-1} ˜C (R2^{∨T})^{-1} = Σ C(k) ε^k, one has C(k)=0 for k<0 and C(0)=N with N a constant symmetric invertible matrix. This is precisely the system in Algorithm 1 step 4, eq. (91). The paper never proves this system is solvable: no counting argument (number of boundary constants vs NF(NF+1)/2 equations), no genericity statement, and no theorem that any filtration-compatible basis from [19,20] admits such boundary values. The verification step 3 (eq. 90) merely throws an exception when the required identity d_B C(k)=0 fails, so the algorithm assumes the property rather than establishes it. The dependence on boundary values is real: in example 6.5, eq. (168)-(169), a non-zero boundary constant N(-2)_34 forces d_B C(-1)_24 = N(-2)_34 R33/R22 ≠ 0, so only special constants work. Since eq. (39) implies constant C requires AC = C A^T, a nontrivial algebraic constraint on A, it is not automatic that the freedom in R2 is sufficient. The examples are consistent with the claim, but the headline assertion as a general mathematical statement is unproven.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies intersection matrices of Feynman integrands on the maximal cut, for two bases produced by the geometric Laporta algorithm of refs. [19,20]: the filtration-compatible basis \\tilde{J} and the \\epsilon-factorised basis K. The main claims are that the intersection matrix of \\tilde{J} has entries that are Laurent polynomials in \\epsilon, and that for K the entries are, after an overall power of \\epsilon is factored out, integers, provided the integration constants (boundary values) of the auxiliary functions in the rotation R2 are chosen appropriately. The paper formulates an algorithm that exploits these properties to eliminate algebraically redundant auxiliary functions, and illustrates the claims on examples ranging from simple rational cases to elliptic curves, Calabi-Yau banana integrals, and higher-genus necklace integrals. The paper also clarifies that intersection numbers must be defined on symmetrised integrands in order to be meaningful at the level of Feynman integrals.","tokens_in":24096,"tokens_out":4889,"duration_ms":47416,"significance":"If the structural claims hold, the paper provides a useful practical tool: it shows that constant intersection numbers give more constraints than self-duality, and it gives a systematic, differential-equation-based algorithm for reducing the number of auxiliary transcendental functions in \\epsilon-factorised systems. The worked examples are detailed and the dependence on boundary constants is demonstrated explicitly, e.g. in the four-loop banana example. The differential-equation derivations in Sections 2-4 are coherent, and the clarification about symmetrised integrands addresses a real subtlety. However, the general claims are not proved; they are supported by examples and by the phrase 'in all examples we checked'. Because the central statements are formulated as general findings, the missing proof of existence of suitable boundary constants is a load-bearing gap that must be addressed.","major_comments":[{"comment":"The existence of a solution to the system C(k)_ij = 0 for k < 0 and C(0)_ij = N_ij with det N ≠ 0 is assumed, not proved. The verification step in eq. (90) only checks that a candidate solution satisfies d_B C(k) = 0 and throws an exception otherwise, so the algorithm presupposes the very property it is designed to exploit. This matters because eq. (39) shows that a constant intersection matrix requires AC - C A^T = 0, a nontrivial algebraic constraint on A; it is not automatic that the integration constants available in R2 are sufficient to satisfy it. The paper gives no counting argument relating the number of free boundary constants to the number of equations, and no genericity statement. Since the abstract's 'if the boundary values ... are chosen appropriately' depends directly on this solvability, the authors should either prove existence for the bases constructed in refs. [19,20], state clearly that this is an additional assumption, or explicitly formulate the claim as a conjecture.","section":"§5, Algorithm 1 step 4, eq. (91)"},{"comment":"The general statements that the intersection matrix for a filtration-compatible basis is a Laurent polynomial in \\epsilon, and that the \\epsilon-factorised basis gives an integer matrix up to a power of \\epsilon, are supported only by examples. The text before eq. (82) says 'in all examples we checked', and no proof is supplied for either the Laurent-polynomial structure or the integrality. Eq. (87) determines \\tilde{C} up to a prefactor from a differential equation, but it does not by itself imply the claimed rationality or Laurent-polynomial form. If these are meant as theorems, a proof should be provided; if they are empirical observations, the abstract and Section 2.1 should be reworded so that the conditional status is explicit.","section":"Abstract and §2.1, eqs. (82)-(83)"},{"comment":"The sentence 'Given that the entries of the intersection matrix are rational, the integer condition follows easily from an appropriate rescaling' is not correct as written. Rationality plus proportionality to a power of \\epsilon does not imply integer entries after factoring out only that power: for example, C = (1/2)\\epsilon^{-1} is rational and proportional to \\epsilon^{-1}, but its entries are not integers and cannot be made integer without introducing a rational prefactor. The examples achieve exact integer entries, but the argument in the introduction needs to be made precise, either by allowing an additional constant prefactor or by proving integrality directly.","section":"§1, paragraph on the integer condition"}],"minor_comments":[{"comment":"For the four-loop banana, the paper states that 10 of the 20 auxiliary functions can be eliminated, but no explicit list of the eliminated functions or the resulting reduced system is given. A table or an explicit list would make the algorithm's output easier to check and would strengthen the example.","section":"§6.5, discussion after eq. (164)"},{"comment":"The step from the differential equation for the symmetrised forms to the statement that CFeynman is the intersection matrix of the symmetrised forms is stated rather than derived. A short justification that the symmetrisation does not alter the differential equation in the presence of the block structure of eq. (68) would be helpful.","section":"§4, eq. (80)"},{"comment":"The sentence 'We expect this to be true beyond the maximal cut' makes clear that the results are only established on the maximal cut. This limitation should also appear in the abstract, since the current abstract states the claims without this restriction.","section":"§7, Conclusions"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely to be of interest to the Feynman-integral community, but the central structural claims are currently presented as established facts while the evidence is a finite set of examples. The main issue is not the internal consistency of the examples but the unproved existence of boundary constants solving Algorithm 1 step 4. This can be fixed by adding a proof for the relevant classes, or by reframing the paper as presenting a conjectural pattern together with a conditional algorithm. The latter would still be useful, but the abstract and introduction must be adjusted accordingly. I do not see a fundamental correctness error in the worked derivations, so I would not recommend rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Mark,\n\nHere's my take on arXiv:2608.03646. The paper's most valuable part is the systematic elimination algorithm for auxiliary transcendental functions on the maximal cut (Section 5). That is a direct, usable tool: you compute the intersection matrix via a differential equation, rotate to the factorised basis, and solve algebraic equations for boundary constants. The observation that constant intersection numbers give NF(NF+1)/2 constraints versus NF(NF-1)/2 for self-duality is correct and explains why this method eliminates more functions. For people doing multi-loop work with Calabi-Yau or higher-genus sectors, this matters.\n\nThe paper also does several things well. The differential-equation derivations in Sections 2–4 are coherent, and the discussion in Section 4 of why you need symmetrised integrands when symmetry relations exist is a real clarification, not a footnote. Examples 5 and 6 are worked out in enough detail to show the mechanism, including the concrete way a generic boundary constant (N(-2)_34 in the four-loop banana) destroys the constancy of C. The self-citations to [19,20] are appropriate; the present paper extends that algorithm.\n\nThe soft spot is in the headline statements. The abstract and Section 2.1 present the Laurent-polynomial and integer-entries results as general facts, but the evidence is 'in all examples we checked' (Section 4) plus an algorithm whose verification step throws an exception when the property fails. In particular, the integer statement is conditional on the existence of boundary values that solve the algebraic system in Algorithm 1 step 4. The paper never proves such boundary values exist: no counting argument, no genericity statement, no theorem for the bases produced by [19,20]. It shows in each example that they do exist and shows a generic choice fails, which is honest, but the general existence question remains open. That gap is real, but it is not fatal: the practical contribution does not depend on the strongest form of the claim, and the conditions are stated explicitly.\n\nIs the central argument sound? The differential-equation plus algebra structure is sound; the claims are empirical regularities presented as findings. A referee should ask the authors to separate observation from theorem and, ideally, prove existence for a non-trivial family or state the conjecture with a precise genericity hypothesis.\n\nWho is this for? Realistically, the Feynman-integral computation community. I would send it to peer review; it deserves careful refereeing, not desk rejection. If I were working on epsilon-factorised systems, I'd cite it for the algorithm.","headline":"A genuinely useful practical algorithm for eliminating auxiliary functions, wrapped in two structural claims that are well-evidenced but not actually proven.","tokens_in":24625,"tokens_out":3284,"would_cite":true,"duration_ms":30464,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Intersection matrices for geometrically ordered Feynman bases collapse to Laurent polynomials in ε, and to integers times ε^{-n} after ε-factorisation.","keywords":["Feynman integrals","intersection numbers","twisted cohomology","epsilon-factorised differential equations","Laporta algorithm","maximal cut","auxiliary functions","geometric order relation"],"falsifier":"Compute one entry of $\\tilde C$ for the four-loop equal-mass banana directly from the defining intersection integral rather than from the differential equation; the claimed structure predicts a Laurent polynomial in $\\varepsilon$ with lowest power $\\varepsilon^{-2}$. Finding a term such as $\\ln x$ or a pole below $\\varepsilon^{-2}$ would refute the claim.","tokens_in":23527,"feed_emoji":"📐","tokens_out":15733,"duration_ms":123749,"temperature":0.7,"pith_summary":"The paper asks what happens to the intersection matrix, the pairing between a Feynman integrand and its dual, for the master integrals selected by a geometric order relation in the Laporta algorithm. It claims two simplifications on the maximal cut: for a filtration-compatible basis the entries are Laurent polynomials in the dimensional-regularisation parameter $\\varepsilon$, and for an $\\varepsilon$-factorised basis, with a suitable choice of boundary values for the auxiliary functions entering the rotation, the entries are integers up to an overall power $\\varepsilon^{-n}$. The practical payoff is that the auxiliary transcendental functions introduced in the second step of the geometric construction are not all independent; the paper gives an algorithm that eliminates the redundant ones and minimises the number of required computations. If correct, this keeps the size of intermediate expressions under control in precision calculations of Feynman integrals.","feed_headline":"Intersection matrices become integers times ε^{-n}","feed_subtitle":"With geometric ordering, intersection matrices simplify to integers, so fewer auxiliary functions are computed.","key_machinery":"The load-bearing object is the intersection matrix $C_{ij}=\\langle\\Psi_i|\\Psi_j^\\vee\\rangle$ of integrand classes, paired with the dual basis obtained by $\\varepsilon\\to-\\varepsilon$, together with the rotation matrix $R_2$ that connects the step-1 basis $J$ to the step-2 basis $K$. The identity that runs through the paper is the differential equation $d_B\\tilde C=\\tilde A\\tilde C+\\tilde C\\tilde A^{\\vee T}$, which fixes $\\tilde C$ up to an $\\varepsilon$-dependent prefactor; applying the rotation through $C=R_2^{-1}\\tilde C(R_2^{\\vee T})^{-1}$ and requiring $d_BC=0$ converts the demand of constant intersection numbers into algebraic equations among the auxiliary functions. Algorithm 1 reads these equations off from the bottom up in powers of $\\varepsilon$, eliminating a subset of the auxiliary functions while keeping the number of required intersection-number computations small. For Feynman integrals proper, the machinery first replaces integrands by their symmetry-averaged versions, so that the intersection numbers are well defined at integral level and independent of the integral representation.","core_discovery":"The paper's central claim is that the intersection matrices of the bases produced by the geometric Laporta algorithm are far simpler than generic rational functions would suggest. Working on the maximal cut, with the dual basis defined by the substitution $\\varepsilon\\to-\\varepsilon$ and with integrands symmetrised under the symmetry relations of the Feynman family, the paper finds that the intersection matrix $\\tilde C$ of the filtration-compatible basis $J$ is a Laurent polynomial in $\\varepsilon$ whose lowest power is at least $\\varepsilon^{-n}$; and that the intersection matrix $C$ of the $\\varepsilon$-factorised basis $K$ is constant in the kinematic variables and, up to an overall factor $\\varepsilon^{-n}$, has integer entries, whenever the boundary values of the auxiliary functions are chosen so that $d_B C=0$. Because constant intersection numbers impose $N_F(N_F+1)/2$ constraints whereas weak self-duality imposes only $N_F(N_F-1)/2$, the constant-intersection condition detects algebraic relations among auxiliary functions that self-duality misses. The resulting Algorithm 1 eliminates the redundant auxiliary functions on the maximal cut and is verified in examples that include an elliptic curve, the three-loop electron self-energy with one zero mass, the four-loop equal-mass banana (a Calabi-Yau three-fold), and higher-genus necklace diagrams.","pith_inferences":["The parity structure of the Laurent-polynomial expansion suggests a diagnostic: if an even-loop family shows a lowest $\\varepsilon$-power on the diagonal that reaches $-l+1$, either a symmetry relation has been missed or the basis is not filtration-compatible.","The algebraic-equation viewpoint opens an optimisation problem: choose the rotation ansatz so that the elimination constraints are triangular, thereby minimising the number of auxiliary functions that ever have to be integrated.","If the integer-entry property of $C$ survives beyond the maximal cut, intersection matrices could serve as a normalisation-independent certificate that an $\\varepsilon$-factorised basis is well chosen, independent of the representation used for the integrals."],"forward_implications":["For any Feynman family for which the two-step geometric construction exists, the redundant auxiliary transcendental functions on the maximal cut can be eliminated automatically, shrinking the $\\varepsilon$-factorised differential system.","Constant intersection numbers become a well-defined criterion for fixing the integration constants of auxiliary functions: choose boundary values so that $d_BC=0$.","The counting difference ($N_F(N_F+1)/2$ equations versus $N_F(N_F-1)/2$ for weak self-duality) means the constant-intersection condition should remove at least as many auxiliary functions as self-duality in every example.","In the four-loop equal-mass banana, the algorithm reduces the twenty auxiliary functions of the rotation to ten, and the surviving functions are expressed through a period of the Calabi-Yau three-fold and related quantities.","The same simplification is expected beyond the maximal cut once relative twisted cohomology is brought in, which would extend the elimination algorithm to complete Feynman integrals."],"supporting_citations":[{"why":"Supplies the two-step geometric Laporta algorithm whose bases $J$ and $K$ and rotation $R_2$ this paper studies.","marker":"[19, 20]"},{"why":"Defines symmetrised integrands by group averaging so that intersection numbers are well-defined at the level of Feynman integrals.","marker":"[21, 22]"},{"why":"Introduces weak self-duality through a constant symmetric intersection matrix, the property the paper compares with constant intersection numbers.","marker":"[23]"},{"why":"Uses constant intersection numbers to obtain algebraic relations among auxiliary functions, the approach this paper systematises into an algorithm.","marker":"[24]"},{"why":"Provides the differential-equation method for intersection numbers that keeps the algorithm's required calculations minimal.","marker":"[34]"},{"why":"Supplies the twisted bilinear relation from which the differential equation for the intersection matrix is derived.","marker":"[35]"},{"why":"Provides earlier examples of auxiliary-function elimination in banana integrals, which the paper extends and generalises.","marker":"[28–30]"},{"why":"Offers symbolic methods for solving the rational differential equation in Algorithm 1's first step.","marker":"[37, 38]"}],"fun_headline_variants":["Feynman intersection matrices simplify to integers via geometric bases","Integer intersection matrices for geometric-ordered Feynman bases","Eliminate auxiliary functions via simpler intersection matrices","Geometric ordering yields integer intersection matrices for Feynman integrals","Master integrals with integer intersection matrices from geometric order"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"All of the simplification rests on being able to solve the equations that fix the auxiliary functions of the rotation so that the unwanted negative powers of $\\varepsilon$ vanish; the paper verifies this in every worked example but does not prove that such a solution exists for every Feynman family.","fun_headline_variants_meta":{"raw":{"variants":["Feynman intersection matrices simplify to integers via geometric bases","Integer intersection matrices for geometric-ordered Feynman bases","Eliminate auxiliary functions via simpler intersection matrices","Geometric ordering yields integer intersection matrices for Feynman integrals","Master integrals with integer intersection matrices from geometric order"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000584,"raw_usage":{"total_tokens":2758,"prompt_tokens":969,"completion_tokens":1789,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":585,"completion_tokens_details":{"reasoning_tokens":1711}},"tokens_in":585,"tokens_out":1789,"duration_ms":11747,"temperature":1.0,"reasoning_tokens":1711,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:48:16.179884+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute one entry of $\\tilde C$ for the four-loop equal-mass banana directly from the defining intersection integral rather than from the differential equation; the claimed structure predicts a Laurent polynomial in $\\varepsilon$ with lowest power $\\varepsilon^{-2}$. Finding a term such as $\\ln x$ or a pole below $\\varepsilon^{-2}$ would refute the claim.","supporting_citations":[{"cited_title":"Matsubara-Heo and N","cited_arxiv_id":null,"evidence_quote":"Provides the differential-equation method for intersection numbers that keeps the algorithm's required calculations minimal."},{"cited_title":"Cho and K","cited_arxiv_id":null,"evidence_quote":"Supplies the twisted bilinear relation from which the differential equation for the intersection matrix is derived."}],"review_version":1}