{"id":"a8a311de-27b9-4781-81d8-ef076cedc072","arxiv_id":"1908.05888","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Guignard establishes a higher-dimensional factorization of global ε-factors into local contributions via iterated vanishing cycles and refined Artin conductors.","lead":"This paper proves that the determinant of the cohomology of an ℓ-adic sheaf on a proper scheme in positive characteristic splits into a product of local pieces attached to closed points. This is a higher-dimensional analogue of the product formula for curves and yields new twist formulas for global ε-factors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"In the §5.2 gluing, the direct sum defining E(X(x),G) over preimages in the positive-dimensional fiber is not shown finite for local G, so the construction may not land in D^b_c(x,Λ).","rationale":"The paper's central theorem is plausible and the broad strategy is coherent: the projective-space case is reduced to fibers by induction, and the global ε-factor identity in §5.2 is an algebraic consequence of the two localization triangles. However, the gluing step in §5.2 is the place where the proof is least secure. The reader flagged the unverified tensor-compatibility axiom there; the finiteness of the defining direct sum for local objects is an even more basic issue, because without it E(X(x), -) is not even a functor into D^b_c(x,Λ). The manuscript gives no lemma showing that the local functors for X' or Z' vanish on the pullback of a local object outside finitely many points of a positive-dimensional fiber. Such a lemma could be proved by compatibility with the closed fiber, and the inductive construction in §5.4 suggests it is true, but it is not written. This is a genuine gap in the written proof, not an objection to the theorem itself, so the appropriate action is to require that the missing verification be supplied before full acceptance. The independent evidence in the paper—the explicit curve formula from [Gu19], the reduction to projective spaces, and the concrete local constructions via Artin conductors—supports the expectation that the gap is fillable.","tokens_in":18903,"tokens_out":30130,"duration_ms":313896,"concrete_test":"Work out the blow-up example: X a smooth projective surface over an algebraically closed field k, x a closed point, X' = Bl_x X, Z = {x}, Z' = P^1_x. Let G = Λ_{X(x)}. Using the §5.1/§5.4 definitions, compute E(X'(x'), Λ) and E(Z'(x'), Λ)[1] at every closed point x' of the exceptional fiber and determine how many terms in the direct sum defining E(X(x),Λ) are nonzero. If infinitely many are nonzero, the gluing construction does not land in D^b_c(x,Λ). If the sum is finite, repeat with G = j_!Λ for the inclusion j of the generic point of X(x), which tests ramification supported along the fiber.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The most load-bearing gap is in the Chow-lemma gluing of §5.2. For a closed point x ∈ Z, the fiber f^{-1}(x) is a proper k(x)-scheme and can have infinitely many closed points, e.g. the exceptional P^1 of a blow-up of a surface at x. The displayed definition sets E(X(x),G) = E(Z(x), G|_Z) ⊕ ⊕_{x' ∈ |f^{-1}(x)|} Ind(...), but no argument is given that this direct sum has only finitely many nonzero terms for an arbitrary G ∈ D^b_c(X(x),Λ). Theorem 1.2(1) only bounds the set of closed points x with E(X(x), F|_{X(x)}) ≠ 0 for a global F on X; it says nothing about the support of the local functor applied to a pullback f^{-1}G of a local object. The natural fix, namely to identify E(X'(x'), -) on pullbacks from X(x) with the functor for the closed fiber f^{-1}(x) and then apply Theorem 1.2 to that fiber, requires a compatibility statement that is not stated or proved in §5.2. Without finite support, E(X(x),G) is not an object of D^b_c(x,Λ), so the construction of the collection asserted in Theorem 1.2 is incomplete. This is distinct from the tensor-compatibility axiom, though it occurs in the same gluing step.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a factorization theorem for the global ε-factor of a bounded constructible ℓ-adic complex on an arbitrary proper scheme over a perfect field of positive characteristic: the global ε-factor and the Euler characteristic are expressed as the determinant and rank of a finite direct sum of induced Galois modules attached to closed points. The local contributions are given by triangulated functors built from iterated vanishing cycles and from refined Artin conductors, which linearize Artin conductors and local ε-factors. The proof uses the curve case [Gu19] as a base, Chow's lemma to reduce to projective schemes, a reduction to projective spaces, and induction on dimension. The paper also derives applications: an ε-factor analogue of Deligne's theorem on Euler characteristics (Theorem 1.4), an invariance statement for Artin conductors and geometric local ε-factors over a henselian trait (Theorem 1.5), and twist formulas (Theorems 1.6 and 1.7).","tokens_in":19220,"tokens_out":14723,"duration_ms":160926,"significance":"If correct, the main theorem is a substantial advance: it extends Laumon's product formula from curves to arbitrary proper schemes in the geometric setting, and it gives new global consequences such as the ε-factor part of Theorem 1.4 and the base-change invariance in Theorem 1.5. The construction of explicit linearized local invariants via refined Artin conductors is original and is likely to have further applications. The paper is clearly organized and the broad strategy is standard and credible. However, the proof of Theorem 1.2 contains a gap in the Chow-lemma gluing step, where the constructed local functors are not shown to land in the bounded derived category of constructible sheaves; this issue is load-bearing for the main theorem.","major_comments":[{"comment":"In the formula defining E(X(x),G) for x in Z, the direct sum is taken over the closed points x' of the fiber f^{-1}(x). This fiber is a proper k(x)-scheme and can have infinitely many closed points, e.g. the exceptional P^1 of a blow-up of a surface at x. The paper does not prove that only finitely many summands are nonzero for an arbitrary G in D^b_c(X(x),Λ). Theorem 1.2(1) gives finiteness only for the restriction of a global object on X', not for the pullback of a local object, so it cannot be applied directly. Since the theorem requires E(X(x),G) to lie in D^b_c(x,Λ), this is a load-bearing gap rather than a mere omission. A proof is needed, for instance by spreading G out to an étale neighborhood of x and applying the induction hypothesis to the proper fiber f^{-1}(x), together with a compatibility statement for the local functors under étale base change; neither the statement nor the proof of such a compatibility is currently provided.","section":"§5.2"},{"comment":"The verification of the tensor-compatibility axiom (1.2)(2) for the glued functors E(X(x),-) is omitted. The displayed direct-sum definition does not make this compatibility formal: one must check that the restriction to the closed subscheme Z, the functors on X' and Z', and the induction from G_{x'} to G_x interact correctly with the operation F⊗sp^{-1}G, including the projection formula for induction. This verification is necessary because the twist formula in Theorem 1.6 is stated as an immediate consequence of (1.2)(2).","section":"§5.2"}],"minor_comments":[{"comment":"The notation f^{-1}F|_{X'(x')} should be clarified: f^{-1}F is a complex on X'×_X X(x), and one uses the canonical morphism X'(x') -> X'×_X X(x) induced by the universal property of henselization.","section":"§5.2"},{"comment":"The sentence 'This and the induction hypothesis ensure that only finitely many terms contribute in the above sum' is justified for a global object F on P(V), but the analogous finiteness property for the local functors applied to an arbitrary local input is precisely the point that is missing in §5.2.","section":"§5.4"},{"comment":"In the paragraph after Theorem 1.5, 'This a consequence of Theorem 6.1' should read 'This is a consequence of Theorem 6.1'.","section":"Introduction"},{"comment":"In the product over x∈|X| appearing in the twist formula, it would be helpful to state explicitly that the product is finite by Theorem 1.2(1).","section":"Theorem 1.6"}],"recommendation":"major_revision","confidential_remarks":"The finiteness gap in §5.2 is the main obstacle. It is concrete and load-bearing, but it appears fixable by adding a lemma on finite support of the local functors for objects pulled back along a proper morphism; I would not recommend rejection on this basis. The manuscript is within the scope of the journal and the overall strategy is credible."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Main take: this paper proves the first product formula for geometric ε-factors on arbitrary proper schemes over perfect fields of positive characteristic. That is a real theorem, and the construction via iterated vanishing cycles and refined Artin conductors is the right way to get it. The paper is essentially sound; it should go to a serious referee after some tightening.\n\nWhat is new: Theorem 1.2 gives, for every proper k-scheme X, local functors E(X(x),−) such that the global ε-factor factors as a determinant of Galois inductions of these local terms. Deligne's Euler-characteristic statement was already known; the ε-factor part is new. The twist formulas in Theorems 1.6–1.7 and the henselian-trait version in Section 6 are useful consequences. The proof is a standard reduction: Chow's lemma, induction on dimension, and pencils, with the P^1 base case coming from the author's curve result [Gu19]. That prior preprint is a genuine prerequisite, so the reader should know this paper stands on it.\n\nWhere it is soft: the gluing step in §5.2 is compressed. For x in the exceptional locus Z, the definition of E(X(x),G) is a direct sum over all closed points of the fiber f^{-1}(x). That fiber can be a P^1 and have infinitely many closed points, and the paper does not explicitly show the sum is finite. It is not hard to fix: descend G to an étale neighborhood of x, extend by zero to X', and apply property (1) of Theorem 1.2 to that global object. But as written, the reader has to fill that in. Also, verification of the tensor-compatibility axiom (1.2)(2) for the glued functors is not written out; it is formal from the induction hypothesis, but should be stated. The reduction in §5.4 is terse yet clear.\n\nThe main structural risk is the dependence on [Gu19], which is not yet published. If that curve-level result has gaps, the base case collapses. That is a normal preprint reliance, not a red flag, but it is worth knowing.\n\nBottom line: this paper should not be desk-rejected. It deserves a careful referee, and I expect the main theorem to survive with minor revisions. The local finiteness issue and the missing tensor-compatibility verification should be raised as requests for clarification, not as fundamental objections.","headline":"The first higher-dimensional product formula for ε-factors on proper schemes is real and mostly sound, but §5.2 needs a minor finiteness argument.","tokens_in":19717,"tokens_out":6730,"would_cite":true,"duration_ms":66902,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14F20","14G17","11S40"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the global $\\varepsilon$-factor of any $\\ell$-adic sheaf on a proper scheme over a perfect field of positive characteristic factors as a finite product of local contributions at closed points, built from vanishing…","keywords":["ℓ-adic sheaves","ε-factors","vanishing cycles","Artin conductors","product formula","positive characteristic","Galois representations","Chow's lemma"],"falsifier":"Inspect the Section 5.2 assembly for a non-projective proper $X$: compute both sides of Theorem 1.2 for a sheaf supported on the exceptional locus $Z$ and compare with the corresponding computation on the Chow cover $X'$; a failure of the determinant identity or of the tensor-compatibility isomorphism for any such pair would collapse the theorem. A simpler control case is $\\mathbb{P}^1_k$ with $F=\\Lambda$, where the explicit Section 5.1 formula must reproduce the known $\\varepsilon_k(\\mathbb{P}^1_k,\\Lambda)$.","tokens_in":18706,"feed_emoji":"🧮","tokens_out":12413,"duration_ms":101910,"temperature":0.7,"pith_summary":"This paper proves a higher-dimensional product formula: for any proper scheme $X$ over a perfect field of positive characteristic $p\\neq\\ell$, the global $\\varepsilon$-factor of any bounded constructible complex of $\\ell$-adic sheaves is the determinant of a finite direct sum of Galois modules attached to the closed points of $X$. Each local term is obtained by iterating vanishing cycle functors and exact additive functors called refined Artin conductors, whose ranks and determinants recover the usual Artin conductor and local $\\varepsilon$-factor. The same mechanism expresses the Euler characteristic as the rank of the same direct sum. This matters because it reduces a global cohomological invariant to purely local data at closed points, and it yields twist formulas and a stalkwise constancy theorem for $\\varepsilon$-factors.","feed_headline":"Global ε-factors split into local contributions for all proper schemes","feed_subtitle":"Global ε-factors and Euler characteristics of ℓ-adic sheaves are recovered from closed-point data.","key_machinery":"The load-bearing object is the refined Artin conductor $\\mathrm{Art}_\\pi$, an exact additive functor built from the Gabber-Katz extension functor and the Artin-Schreier sheaf $L_\\psi\\{-t\\}$: $\\mathrm{Art}_\\pi(M)=H^1_c(\\mathbb{A}^1_s,r^{-1}_{s,\\pi}M_!\\otimes L_\\psi\\{-t\\})$. When applied to the vanishing cycles of a sheaf on the henselization of a curve at a point, Proposition 4.15 gives $\\mathrm{rk}(\\mathrm{Art}_\\pi R\\Phi_s^{\\mathrm{id}}(F))=a(S(s),F)$ and $\\det(\\mathrm{Art}_\\pi R\\Phi_s^{\\mathrm{id}}(F))=\\varepsilon(S(s),F,d\\pi)$, so the rank and determinant of this linearization reproduce the classical invariants. The higher-dimensional local functors $E(X(x),-)$ are assembled from these curve-level ingredients by iterating vanishing cycles and gluing over a Chow cover, with a shift $[1]$ on the exceptional preimage.","core_discovery":"Theorem 1.2 asserts that for every proper $k$-scheme $X$ there is a collection $(E(X(x),-))_{x\\in|X|}$ of triangulated functors $D^b_c(X(x),\\Lambda)\\to D^b_c(x,\\Lambda)$ satisfying three conditions: local vanishing outside finitely many closed points; compatibility with pullback from the residue field, $E(X(x),F\\otimes\\mathrm{sp}^{-1}G)\\simeq E(X(x),F)\\otimes G$; and the identities $\\varepsilon_k(X,F)=\\det\\big(\\bigoplus_{x\\in|X|}\\mathrm{Ind}_{G_k}^{G_x}E(X(x),F|_{X(x)})\\big)$ and $-\\chi(X,F)=\\mathrm{rk}\\big(\\bigoplus_{x\\in|X|}\\mathrm{Ind}_{G_k}^{G_x}E(X(x),F|_{X(x)})\\big)$. The functors are not unique. The proof reduces to projective space by Chow's lemma, then to $\\mathbb{P}^1$ via a pencil, where the curve product formula supplies explicit local terms. The paper derives from this a stalkwise constancy theorem for $\\varepsilon$-factors and Euler characteristics, a constancy theorem for Artin conductors and local $\\varepsilon$-factors over a henselian trait, and twist formulas for tensor products with local systems.","pith_inferences":["The non-uniqueness of the local functors suggests that the factorization is a property of determinants rather than of the local Galois modules themselves; asking whether a canonical choice exists once a flag or pencil is fixed is a natural next step.","Remark 6.2 points toward a hierarchy of $n$-dimensional local $\\varepsilon$-factors; one could test whether these satisfy a reciprocity law compatible with geometric class field theory, or whether the iteration can be made independent of the auxiliary choices of uniformizers.","The construction works geometrically over any perfect field, so one could test whether the same product formula holds for tame sheaves with general $\\Lambda$-coefficients or for the twisted sheaves mentioned in Section 1.8, where a 2-cocycle on $G_k$ is allowed.","Because the proof passes through Chow's lemma and a pencil, the argument may extend to proper algebraic spaces or stacks if the same reduction steps are available there."],"forward_implications":["Theorem 1.4: if two bounded constructible complexes have isomorphic restrictions to the henselization at every closed point, then their global $\\varepsilon$-factors and Euler characteristics coincide.","Twist formula (Theorem 1.6): for a $\\Lambda$-local system $G$ of constant rank $r$, $\\varepsilon_k(X,F\\otimes G)=\\varepsilon_k(X,F)^r\\prod_{x\\in|X|}\\big(\\det(G_x)\\circ\\mathrm{ver}_{x/k}\\big)^{\\mathrm{rk}(E(X(x),F))}$, and $\\chi(X,F\\otimes G)=r\\chi(X,F)$.","Theorem 1.7 extends the twist formula to objects twisted at each closed point by free $\\Lambda$-modules of rank $r$ with admissible Galois action.","Over a henselian trait, the Artin conductor and geometric local $\\varepsilon$-factor of $Rf_*F$ depend only on the restrictions of $F$ to the henselizations at closed points of the special fiber (Theorem 1.5).","Iterating the construction attaches $(n+1)$-dimensional local $\\varepsilon$-factors to Galois representations of $n$-dimensional local fields, so that the $n$-dimensional local $\\varepsilon$-factor of $R\\Gamma(X_k,F)$ factors into finitely many higher local factors (Remark 6.2)."],"supporting_citations":[{"why":"Supplies the curve-level product formula, the refined Artin conductors, and the definitions of Artin conductor and geometric local $\\varepsilon$-factor that the higher-dimensional construction iterates.","marker":"[Gu19]"},{"why":"Introduces the Gabber-Katz extension and the cohomological construction of the Swan module that the paper's contractions are modelled on.","marker":"[Ka86]"},{"why":"Provides the nearby and vanishing cycle functors and their functoriality properties used in every local contribution.","marker":"[SGA7]"},{"why":"Supplies the universal local acyclicity theorem used to ensure the vanishing-cycle terms are supported on finitely many closed points.","marker":"[SGA 4 1/2]"},{"why":"Proves the original curve product formula over finite fields that this paper extends to arbitrary proper schemes over perfect fields.","marker":"[La87]"},{"why":"Records Deligne's theorem that Euler characteristics agree under stalkwise isomorphism, which Theorem 1.4 complements on the $\\varepsilon$-factor side.","marker":"[Ill81]"}],"fun_headline_variants":["ε-factors factorize locally for all proper schemes","Local ε-factors determine global ε-factors on any proper scheme","Higher-dimensional ε-factor product formula from closed-point data","Product formula: global ε-factors from local contributions on all schemes","Every proper scheme: ε-factors from local data via vanishing cycles"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof that the local functors assembled in Section 5.2 from a projective cover, its exceptional locus, and its preimage satisfy the tensor-compatibility axiom for all pairs of sheaves is not written out in full.","fun_headline_variants_meta":{"raw":{"variants":["ε-factors factorize locally for all proper schemes","Local ε-factors determine global ε-factors on any proper scheme","Higher-dimensional ε-factor product formula from closed-point data","Product formula: global ε-factors from local contributions on all schemes","Every proper scheme: ε-factors from local data via vanishing cycles"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000688,"raw_usage":{"total_tokens":3105,"prompt_tokens":921,"completion_tokens":2184,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":537,"completion_tokens_details":{"reasoning_tokens":2099}},"tokens_in":537,"tokens_out":2184,"duration_ms":15615,"temperature":1.0,"reasoning_tokens":2099,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:01:36.513454+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Inspect the Section 5.2 assembly for a non-projective proper $X$: compute both sides of Theorem 1.2 for a sheaf supported on the exceptional locus $Z$ and compare with the corresponding computation on the Chow cover $X'$; a failure of the determinant identity or of the tensor-compatibility isomorphism for any such pair would collapse the theorem. A simpler control case is $\\mathbb{P}^1_k$ with $F=\\Lambda$, where the explicit Section 5.1 formula must reproduce the known $\\varepsilon_k(\\mathbb{P}^1_k,\\Lambda)$.","supporting_citations":[],"review_version":1}