{"id":"ab74afc2-6c17-4753-b7b5-a1b92f45b52c","arxiv_id":"2506.18452","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For several multiloop Feynman integral systems, the monodromy group has a basis where its generators are matrices over Z[z,1/z] with z = exp(iπd), obtained by a numerical recognition method.","lead":"This paper computes the monodromy matrices of multiloop Feynman integral differential equations as explicit functions of the spacetime dimension d. It finds that, for several integral families, these matrices can be written with entries that are integer-coefficient Laurent polynomials in z = exp(iπd), and it checks bilinear relations that connect d and -d.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Single- d PSLQ recognition underpins the Z[z,1/z] claim; no second- d validation is provided.","rationale":"The reader's weakest_assumption correctly identifies the single-point PSLQ recognition as the load-bearing step. I agree that the lack of a second d check and the partly self-referential bilinear consistency leave the central claim in the status of a very well-supported observation rather than a demonstrated theorem. The high precision makes a spurious integer relation extremely unlikely, so I do not see grounds to reject the results; the appropriate posture is a conditional acceptance pending an independent check at a second d value. The concrete test proposed here would directly settle whether the recognized Laurent polynomials are the true analytic functions of d and would strengthen the paper substantially.","tokens_in":14208,"tokens_out":5432,"duration_ms":58905,"concrete_test":"Re-run the Section 5 pipeline at a second transcendental point, e.g., d=7/pi, for at least Example 1 (sunrise) and Example 3 (vertex). Construct the numerical monodromy generators from high-precision connection matrices; substitute z=exp(7i) into the paper's M_a(z) (Eqs. (6.10) and (6.30)); compare each matrix entry to 500 significant digits. Also verify the bilinear relation M^T(1/z) B(z) M(z)=B(z) using B(z) from Eqs. (6.12)/(6.33) at the new d. If any entry differs beyond rounding error, the recognized Laurent polynomials are not the true monodromy functions, and the central Z[z,1/z] claim fails for those examples.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central assertion — that for each treated family there is a basis in which all monodromy generators have entries in Z[z,1/z] — rests on PSLQ recognition of rational functions from high-precision connection matrices at the single transcendental point d=12/pi (Section 5 and Examples 6.1-6.3). No rigorous error bound is given, and no second numerical value of d is used to confirm the recognized Laurent polynomials are the true analytic functions of d. The consistency checks are necessary but not independent: the bilinear matrix B(d) is solved from the very monodromy matrices being tested (e.g., Eq. (6.11)), and the group relation M_infty = (M1 M_1/9 M0)^{-1} is enforced by construction. Step 4's assumption that a non-degenerate eigenvalue exactly proportional to z^k exists to seed the basis is verified only by explicit choice in each example; Appendix A proves only that eigenvalues of the residue matrices are rational powers of z, not that such a non-degenerate eigenvalue exists or that the resulting basis is well-defined for all d. Because the matrices are explicit and notebooks are shipped, the claim is checkable; the concern is that the evidence is heuristic rather than demonstrative.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the monodromy of regular Fuchsian differential systems satisfied by multiloop Feynman integrals in dimensional regularization.  It recalls the standard construction of monodromy generators from generalized series solutions and connection matrices, and it derives two structural properties: (i) there exists a basis in which the monodromy is periodic under d -> d+2, and the eigenvalues of the residue matrices are linear in d with rational slope; (ii) the twisted Riemann bilinear relations imply the bilinear constraint rho^T(gamma,-d) B(d) rho(gamma,d) = B(d) for a d-dependent matrix B(d).  The new content is a heuristic PSLQ-based method for finding monodromy matrices as functions of d: the authors compute high-precision connection matrices at a single transcendental d, here d = 12/pi, choose a basis from eigenvectors, and recognize the matrix entries as rational functions of z = exp(i pi d).  The method is applied to the two-loop equal-mass sunrise, a three-loop forward-box elliptic sector, and a three-loop off-shell massless vertex.  In each example the recognized generators are Laurent polynomials in z with integer coefficients, so the monodromy image is a subgroup of GL(n, Z[z,1/z]); a matrix B(d) satisfying the bilinear relation is also found.","tokens_in":14506,"tokens_out":6015,"duration_ms":66947,"significance":"If correct, the Z[z,1/z] structure is a striking and potentially general property of multiloop monodromy, connecting the present examples to known results for hypergeometric functions and to Shimada's Picard-Lefschetz framework.  The paper's rigorous contributions are the periodicity statement in Section 4.1 and the eigenvalue theorem in Appendix A; these are clean and useful.  The main new observation is explicitly heuristic, but the paper ships Mathematica notebooks and all recognized matrices are explicit, so the central claim is machine-checkable and reproducible.  The paper does not prove the Z[z,1/z] assertion, and the numerical evidence for it is concentrated on a single value of d; this is the main factor limiting the current strength of the paper.","major_comments":[{"comment":"The central assertion that monodromy generators can be chosen in GL(n, Z[z,1/z]) rests on PSLQ recognition of rational functions from high-precision numerical values at the single transcendental point d = 12/pi.  No second value of d is used, and no rigorous error bounds are given for the truncation errors of the series or for the PSLQ identification.  A finite-precision integer relation can in principle be spurious, and the recognized Laurent polynomials are not certified to be the true analytic functions of d.  I request at least one independent check at another generic d, for example d = 5/pi, comparing the directly computed high-precision monodromy generators with the proposed formulas; if the formulas are exact, this check is essentially free and would turn a single-point fit into a nontrivial test.","section":"Section 5, Step 5; Examples 6.1-6.3"},{"comment":"The bilinear-relation checks are internal rather than independent.  In each example the matrix B(d) is solved from the very equations rho^T(1/z) B(d) rho(z) = B(d) that are then said to be satisfied.  This verifies compatibility of the four recognized generators with each other, but it does not tie B(d) to an independently derived intersection form or to the geometry of twisted cycles.  The relation (4.8) is a necessary condition, so this check has some content, but the text should state clearly that it is a self-consistency test rather than an independent confirmation of the PSLQ recognition.","section":"Section 4.2 and Eqs. (6.11), (6.25), (6.32)"},{"comment":"The basis construction assumes the existence of a non-degenerate eigenvalue proportional to z^k for one of the monodromy generators, and this is verified only by explicit choice in each example.  Appendix A proves a weaker statement: eigenvalues of the residue matrices are of the form c1 d + c0 with rational c1.  It does not prove that the corresponding monodromy eigenvalue is non-degenerate and proportional to z^k with integer k, nor that the basis obtained from it remains well-defined for all d.  If the Z[z,1/z] observation is intended as a general statement, this gap should be discussed; if it is an example-specific input to the heuristic, the text should say so explicitly.","section":"Section 5, Step 4; Appendix A"}],"minor_comments":[{"comment":"The conclusion states that the examples 'show' the group is isomorphic to a subgroup of GL(n, Z[z,1/z]); given the heuristic nature of the identification in Section 5, I recommend wording such as 'provide evidence' or 'observe' to avoid overstating the status of the result.","section":"Section 7"},{"comment":"The expansion of lambda(d) at d = infinity is written with a set S of leading powers whose fractional parts are distinct; it would be clearer to state explicitly that these powers are rational numbers, since the argument comparing fractional powers is otherwise implicit.","section":"Appendix A, Eq. (A.2)"},{"comment":"The sentence introducing the six relations says 'they should at least satisfy the six relations', but no proof or reference is given that these are sufficient for the fundamental group of the complement of the reducible curve; please clarify that these are only the relations checked in the example.","section":"Example 6.3, around Eq. (6.31)"},{"comment":"The notation fM and fMa is used inconsistently in the text around Eqs. (6.15)-(6.16); please unify the notation for the transformed system.","section":"Section 6.2"},{"comment":"The Mathematica notebooks are a valuable addition; a short README describing the exact input values, the PSLQ configuration, and the number of series terms used would make the numerical recognition easier to reproduce and audit.","section":"Ancillary files"}],"recommendation":"major_revision","confidential_remarks":"The paper is a well-written report of a heuristic observation with high-precision numerical support.  The main concern, shared by the stress-test note, is that the central Z[z,1/z] claim is supported only at a single value of d; adding a second independent numerical check is feasible and would materially increase confidence.  I do not think rejection is warranted, because the rigorous parts are sound and the observation is clearly labeled as heuristic, but the conclusion should be tempered and the internal bilinear check should be described more carefully."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the paper gives a concrete recipe—high-precision connection matrices at one transcendental d, PSLQ recognition, change of basis—for getting monodromy matrices as explicit functions of d, and applies it to three nontrivial Feynman integral families. Second, the advertised structure GL(n,Z[z,1/z]) is genuinely interesting but is supported by numerical recognition at a single d, not by proof. The first is useful; the second is plausible.\n\nWhat is new: explicit monodromy matrices for the two-loop sunrise, three-loop forward box, and three-loop off-shell vertex as functions of d don't appear in earlier literature as far as the citations show. The observation that all treated systems admit a basis with entries Laurent polynomials in z with integer coefficients is a real pattern. The bilinear relation rho^T(γ,-d)B(d)rho(γ,d)=B(d) is also tested and holds for the found matrices; that gives a consistency check of the whole package.\n\nThe clean parts: Section 4.1's periodicity argument via dimensional recurrence is textbook-simple and sound. Appendix A's proof that residue eigenvalues are rational in d plus a constant is solid, and I don't see a gap there. The examples are reproducible: notebooks are shipped, and the method's mechanics (Libra series, connection matrices) is standard high-precision numerics.\n\nSoft spots, in order of importance. The central claim rests on PSLQ recognition at the single transcendental point d=12/π. There are no rigorous error bounds, and no second d value is used to confirm the recognized Laurent polynomials interpolate the true analytic functions. With hundreds of digits a coincidence is very unlikely, so this is a softness, not a fatal flaw—but it means the paper's title-level claim is \"observed in examples\" until either a proof or a second check appears. The basis construction in Step 4 of Section 5 assumes a non-degenerate eigenvalue proportional to z^k; Appendix A proves eigenvalues are rational powers of z but not that such a clean eigenvalue exists in every family. And the B(d) check is partly self-referential: B is solved from the same monodromy matrices it is then used to verify. That is a consistency test, not an independent constraint.\n\nWho this is for: anyone computing multiloop integrals who needs monodromy data, and people working on the arithmetic structure of Feynman integral differential equations. It is a methods paper with a conjectural-pattern novelty, not a theorem. I would send it to peer review. The right referee will ask for a second d validation or an explicit statement that the Z[z,1/z] property is an observation for these examples.","headline":"Handy numerical-to-analytic method for monodromy matrices, with a plausible but unproven GL(n,Z[z,1/z]) pattern; worth refereeing.","tokens_in":14964,"tokens_out":2274,"would_cite":true,"duration_ms":22781,"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":"This paper introduces a numeric-to-exact method that expresses the monodromy of multiloop Feynman integrals as a function of the dimension $d$, finding in every example matrices with entries in $\\mathbb{Z}[z,1/z]$, $z=\\exp(i\\pi d)$.","keywords":["monodromy group","multiloop integrals","dimensional regularization","Feynman integrals","differential equations","Laurent polynomials","integer relation algorithms","twisted Riemann bilinear relations"],"falsifier":"Run the same connection-matrix calculation at a second transcendental dimension, such as $d=5/\\pi$, and test whether every entry of the monodromy generators matches the paper's Laurent-polynomial formulas to hundreds of digits; a mismatch at this second point would mean the recognized formulas are not the true monodromy.","tokens_in":14051,"feed_emoji":"🔁","tokens_out":14320,"duration_ms":130280,"temperature":0.7,"pith_summary":"This paper tries to establish that the monodromy group governing the analytic continuation of multiloop Feynman integrals in dimensional regularization is an arithmetic object. For each integral family studied, there is a basis of solutions in which every monodromy generator is a matrix with entries that are Laurent polynomials in $z=\\exp(i\\pi d)$ with integer coefficients, i.e., the monodromy group is a subgroup of $GL(n,\\mathbb{Z}[z,1/z])$. If true, the branch structure of these integrals is controlled by a single root of unity $z$, the representation is automatically periodic under $d\\mapsto d+2$, and the monodromy matrices can be written down explicitly as rational functions of the spacetime dimension. The paper further derives, from twisted Riemann bilinear relations, the constraint $\\rho^{\\top}(\\gamma,-d)\\,B(d)\\,\\rho(\\gamma,d)=B(d)$ with a fixed matrix $B(d)$ whose determinant is a power of $z$ times a product of cyclotomic polynomials, and verifies this constraint in all computed examples. The reason to care is that monodromy is a compact invariant of a differential system: knowing it exactly as a function of $d$ strongly constrains the system itself and the class of transcendental functions its solutions can build.","feed_headline":"Multiloop monodromy matrices become integer Laurent polynomials","feed_subtitle":"The full branch structure of Feynman integrals becomes explicit and periodic in the spacetime dimension.","key_machinery":"The load-bearing machinery is the connection-matrix factorization of monodromy in a global normalized Fuchsian form: a rational gauge in which every singular point has non-resonant residue matrix. In this gauge the generator around a singular point $a$ is $C_a^{-1}e^{2\\pi i M_a}C_a$, where $C_a$ is a product of connection matrices built from overlapping generalized power series. The method evaluates these connection matrices numerically at a transcendental $d$ (so $z$ is transcendental, e.g. $d=12/\\pi$), chooses a basis seeded by an eigenvector with eigenvalue proportional to $z^k$, and lifts the numeric entries to rational functions of $z$ with an integer-relation algorithm (PSLQ). The identity $\\rho(\\gamma,d)=\\rho(\\gamma,d+2)$, which follows from compatibility of the differential system with the dimensional recurrence, is what makes $z=\\exp(i\\pi d)$ the correct variable, and the twisted Riemann bilinear relation $\\rho^{\\top}(\\gamma,-d)B(d)\\rho(\\gamma,d)=B(d)$ supplies the independent constraint used to check the recovered matrices.","core_discovery":"The paper's central claim is that, after a suitable change of basis, the monodromy generators of the multiloop differential systems considered are matrices in $GL(n,\\mathbb{Z}[z,1/z])$, $z=\\exp(i\\pi d)$, with determinant $\\pm z^k$. The claim is demonstrated on three explicit families: the two-loop equal-mass sunrise, the three-loop forward box (an elliptic sector), and the three-loop off-shell massless vertex, the last with kinematic variables on $\\mathbb{CP}^2$. For each family the paper presents the generators around all singular points and shows they satisfy the bilinear identity $\\rho^{\\top}(\\gamma,-d)\\,B(d)\\,\\rho(\\gamma,d)=B(d)$ for a single matrix $B(d)$ independent of the loop, with $\\det B(d)$ a constant multiple of a product of cyclotomic polynomials times a power of $z$. An appendix proves that the eigenvalues of the monodromy generators must be proportional to rational powers of $z$ under the assumed periodic structure, which explains why a Laurent-polynomial ansatz in $z$ can succeed.","pith_inferences":["If the $\\mathbb{Z}[z,1/z]$ property is generic rather than an accident of these examples—an extension of the paper's claim—then the analytic-continuation group of Feynman integrals is a subgroup of a fixed finitely generated matrix group, so the spacetime dimension enters only through one root of unity $z$.","A cheap testable extension is to repeat the connection-matrix computation at a second transcendental dimension, such as $d=5/\\pi$ or $d=10/\\pi$, and verify that PSLQ-recognized entries coincide with the reported Laurent polynomials; this would either harden or break the heuristic.","The cyclotomic factors in $\\det B(d)$ suggest a general rule: transitions from irreducible to reducible monodromy occur exactly at rational dimensions where an invariant subspace is killed by the kernel of $B(d)$, a criterion that could be used without computing all connection matrices.","The same integer-recognition strategy could generate explicit monodromy conjectures for Euler-type and $A$-hypergeometric integrals, where known results already exhibit the same $\\mathbb{Z}[z,1/z]$ shape."],"forward_implications":["For the three families treated, the exact monodromy matrices become explicit bookkeeping objects: substituting any complex $d$ gives the analytic continuation around a singular point without re-solving the differential system.","The periodicity $d\\mapsto d+2$ is manifest in the Laurent-polynomial form, so resonance phenomena tied to differences of exponents can be read off directly, and at roots of unity where $B(d)$ degenerates the representation becomes reducible.","The matrix $B(d)$ gives a contour-independent bilinear invariant pairing dimension $d$ with $-d$ for the same monodromy representation, providing a practical check and a link to known quadratic relations among Feynman integrals.","Because the method works with a generic one-dimensional section of a higher-dimensional kinematic space and uses the corresponding generators of the fundamental group, it extends to multiloop systems in several variables; the $\\mathbb{CP}^2$ vertex example demonstrates this."],"supporting_citations":[{"why":"Supplies the monodromy theory, generalized series solutions, and connection matrices on which the method is built.","marker":"[12]"},{"why":"Shows how to reduce multiloop systems to global normalized Fuchsian form, the gauge required by the method.","marker":"[17]"},{"why":"Supplies the fixed-order recurrence that makes high-precision generalized-series coefficients computationally efficient.","marker":"[18]"},{"why":"Provides the software implementation used to construct the generalized series solutions and connection matrices.","marker":"[19]"},{"why":"Supplies the earlier example of a monodromy representation into $GL(n,\\mathbb{Z}[z,1/z])$ that the paper's observation parallels.","marker":"[20]"},{"why":"Supplies the twisted intersection theory and Riemann period relations used to derive Eq. (4.8).","marker":"[21]"},{"why":"Supplies the derivation of twisted Riemann bilinear relations for Feynman integrals used for the bilinear checks.","marker":"[27]"},{"why":"Supplies the integer-relation algorithm used to recognize rational functions of $z$ from numeric data.","marker":"[28]"},{"why":"Provides the three-loop forward-box integral family used as the elliptic-sector example.","marker":"[29]"}],"fun_headline_variants":["Monodromy matrices: integer Laurent polynomials in z=exp(iπd)","Multiloop monodromy: integer Laurent polynomials in exp(iπd)","Branch structure of multiloop integrals becomes explicit in d","Monodromy group of Feynman integrals: integer Laurent polynomials"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central evidence is read off from high-precision numbers at one value of the dimension, $d=12/\\pi$, and the paper supplies no proof that the integer-relation search at that single point gives the true functions of $z$ rather than a numerical coincidence.","fun_headline_variants_meta":{"raw":{"variants":["Monodromy matrices: integer Laurent polynomials in z=exp(iπd)","Multiloop monodromy: integer Laurent polynomials in exp(iπd)","Branch structure of multiloop integrals becomes explicit in d","Monodromy group of Feynman integrals: integer Laurent polynomials"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001399,"raw_usage":{"total_tokens":5621,"prompt_tokens":875,"completion_tokens":4746,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":491,"completion_tokens_details":{"reasoning_tokens":4666}},"tokens_in":491,"tokens_out":4746,"duration_ms":32006,"temperature":1.0,"reasoning_tokens":4666,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:48:13.717900+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same connection-matrix calculation at a second transcendental dimension, such as $d=5/\\pi$, and test whether every entry of the monodromy generators matches the paper's Laurent-polynomial formulas to hundreds of digits; a mismatch at this second point would mean the recognized formulas are not the true monodromy.","supporting_citations":[{"cited_title":"Haraoka, Linear Differential Equations in the Complex Domain: From Classical Theory to Forefront, vol","cited_arxiv_id":null,"evidence_quote":"Supplies the monodromy theory, generalized series solutions, and connection matrices on which the method is built."},{"cited_title":"Shimada, Picard-Lefschetz theory for the universal coverings of complements to affine hypersurfaces, Publications of the Research Institute for Mathematical Sciences 32 (1996) 835","cited_arxiv_id":null,"evidence_quote":"Supplies the earlier example of a monodromy representation into $GL(n,\\mathbb{Z}[z,1/z])$ that the paper's observation parallels."},{"cited_title":"Cho and K","cited_arxiv_id":null,"evidence_quote":"Supplies the twisted intersection theory and Riemann period relations used to derive Eq. (4.8)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the derivation of twisted Riemann bilinear relations for Feynman integrals used for the bilinear checks."},{"cited_title":"Ferguson, D","cited_arxiv_id":null,"evidence_quote":"Supplies the integer-relation algorithm used to recognize rational functions of $z$ from numeric data."}],"review_version":2}