{"id":"af81745b-c0f0-4af3-ac67-b71d9877fd5d","arxiv_id":"2502.07233","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The minimal exponent of a reduced hypersurface is characterized by vanishing and isomorphism conditions on higher direct images of twisted logarithmic differential forms on a log resolution.","lead":"This paper gives a new way to compute a refined singularity invariant, the minimal exponent of a hypersurface, directly from a log resolution. The criterion is the birational analogue of the classical formula for the log canonical threshold and yields constancy of the minimal exponent in nice families.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Formula (19), imported from Saito's analytic monomial case and converted to arbitrary algebraic g via a 'standard' citation, is the least secured keystone of Theorem 4.1; a direct algebraic verification is needed before the main theorem is fully supported.","rationale":"I agree with the reader that the analytic-to-algebraic conversion of Saito's formula (19) is the most load-bearing unverified step. The paper's Theorem 4.1 is the technical core; without (19) the filtered resolution of ψ_{g,α} is unsupported, and the main birational description of the minimal exponent collapses. The authors acknowledge the issue in a footnote and point to a standard argument plus a citation, but they do not supply the actual isomorphism or its effect on the filtration and twist. This is a gap in justification rather than a demonstrated counterexample. The proposed concrete test—either an explicit derivation through [CDM24, Rem. 5.10] or a computational check on a non-monomial example—would settle whether the concern lands. Because the fix is likely straightforward and the theorem is plausible, I would not reject the paper; I would recommend conditional acceptance, asking the authors to provide the missing verification or a precise reference covering the algebraic, non-monomial case.","tokens_in":24740,"tokens_out":31363,"duration_ms":275583,"concrete_test":"Write out the isomorphism in [CDM24, Rem. 5.10] relating the V-filtrations of B_g and B_h when g = u·h, and verify that it carries ω_Y(-D_α(h)+E)δ·F_{p+n}(D_Y[θ]) to ω_Y(-D_α(g)+E)δ·F_{p+n}(D_Y[θ]); any shift or twist mismatch would falsify (19). As a computational cross-check, compute the algebraic V-filtration and Hodge filtration for a non-monomial g (e.g., g = x^2+y^3 in A^2) with a D-module package (Macaulay2 or Singular) for α = 1/2 and p = 0,1, and compare with the right-hand side of (19).","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central equivalence (Theorem 1.2 / Corollary 1.3) rests on Theorem 4.1, whose proof uses formula (19) for the Hodge filtration on the V-filtration of B^r_g in the SNC case. This formula is quoted from [Sai90, Prop. 3.5], a result in the analytic setting for monomial g. Footnote 3 asserts that the algebraic version for g = u·h with u invertible follows by a 'standard' passage, citing [CDM24, Rem. 5.10], but the isomorphism relating the V-filtrations of B_g and B_h is not written down, and the behavior under this isomorphism of the filtration shift and of the twist O(-D_α+E) is not verified. Since all subsequent results (Corollaries 5.1-5.3 and Theorem 1.2(ii)) depend on Theorem 4.1, a failure of (19) in the algebraic non-monomial setting would invalidate the main theorem. This is not a demonstrated error, but it is the least secured load-bearing input in the proof and should be made explicit before final acceptance.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proves a birational criterion for Saito's minimal exponent of a reduced hypersurface Z in a smooth complex algebraic variety X. The main result, Theorem 1.2, states that for γ = p + α with α ∈ (0,1) ∩ Q and p ∈ Z_{\\ge 0}, if ~α(Z) ≥ p then the natural morphism (3) between higher direct images of twisted log forms is an isomorphism for q ≠ p and injective for q = p; and if ~α(Z) > p, then ~α(Z) > p + α holds if and only if that morphism is an isomorphism for q = p. Corollary 1.3 reformulates this, via duality, as the vanishing of R^q π_* Ω^p_Y(log E)(⌊αD⌋) for all q ≥ 1. The integer case is handled separately in Theorem 1.1 using results on k-rational singularities. The key technical input is Theorem 4.1, which gives an explicit filtered resolution of the V-filtration pieces of the nearby cycles of g = f ∘ π in the simple normal crossing case. The paper concludes with an application to the constancy of the minimal exponent in proper families admitting a simultaneous log resolution, answering a question of Radu Laza.","tokens_in":24962,"tokens_out":7761,"duration_ms":67416,"significance":"If the main theorem is fully established, it provides a checkable birational description of a subtle Hodge-theoretic invariant, complementing the existing Hodge-ideal description and making the minimal exponent accessible through higher direct images of log differentials. The criterion is parameter-free and does not require constructing a complicated complex on the resolution. The proof is largely self-contained modulo Saito's strictness theorem and the V-filtration formalism, and the authors supply a detailed proof of the filtered resolution in the monomial case, which represents substantial technical work. The application to family constancy is a satisfying consequence. The main caveat, discussed below, is that the algebraic version of the key filtration formula (19) is imported from the analytic monomial case via a brief citation, leaving a load-bearing step to be verified explicitly.","major_comments":[{"comment":"Equation (19), which describes F_p V_{-α} B^r_g in the simple normal crossing case, is quoted from [Sai90, Prop. 3.5] in the analytic setting for monomial g. Footnote 3 asserts that the algebraic version for g = u h follows by a standard passage, citing [CDM24, Rem. 5.10], but the isomorphism relating the V-filtrations of B_g and B_h is not written down, and the behavior under this isomorphism of the filtration shift F_•[−1] and of the twist O(−D_α+E) is not verified. Since Theorem 4.1 is the central technical result and all subsequent results (Corollaries 5.1–5.3, Theorem 1.2, Corollary 1.3, and Theorem 6.1) depend on it, this is a load-bearing point. Please provide a direct algebraic proof of (19) in the SNC case, or state and prove the precise isomorphism between the V-filtrations of B_g and B_h and check the filtration and twist compatibilities. Without this, the main theorem is not fully supported.","section":"Section 4, Theorem 4.1 and Eq. (19); see also Footnote 3"}],"minor_comments":[{"comment":"In the displayed definition of β_i, both the source and the target are written as E^{n-1-i,>α}_{Y/A1}; from the subsequent comparison with γ_i it is clear that the target should be E^{n-1-i,α}_{Y/A1}. Please correct this typo.","section":"Section 5, Corollary 5.3"},{"comment":"The identification of C_{D_α} with the Koszul complex of the left multiplications by the operators in (22) is a bit terse, since C_{D_α} is a complex of right D_Y-modules. A sentence explaining the right-module convention in this identification would improve readability.","section":"Section 4, Eq. (22)"},{"comment":"In the right-left conversion, the filtration on the left D-module ~C^{-q}_G is written as F_k ~C^{-q}_G = F_{k-q-1}D_Y ⊗ ... ; it would be helpful to indicate explicitly how this indexing matches the right-module convention F_{p-n}M^r = ω_X ⊗ F_p M stated in Section 2.","section":"Section 4, Remark 4.4"},{"comment":"The phrase 'Since Z is not reduced, then lct(X,Z)<1 and thus ~α(Z)=lct(X,Z)' is correct, but as written it appears immediately after the definition of D and E; adding a forward reference to [Kol97] would help the reader see why the reduced case is the only nontrivial one.","section":"Introduction, page 2"}],"recommendation":"major_revision","confidential_remarks":"The reliance on formula (19) is the only substantive concern. The paper otherwise reads as a competent and useful contribution, and the self-citations ([MP22], [MP20a], [Che24], [CDM24]) are used with stated hypotheses rather than as substitutes for argument. The typos in Corollary 5.3 and the terseness of a few filtration computations should be easy to fix. I would be happy to accept after the algebraic proof of (19) is supplied."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take. This is a good paper and the main result is real: a birational description of the minimal exponent in terms of higher direct images of twisted log forms. Theorem 1.2 and Corollary 1.3 are new, and the statement is clean enough to be useful. The proof is serious work; the filtered resolution of nearby cycles for SNC divisors (Thm 4.1) is the technical core, and the authors are honest that a related resolution appears in [SS24, Prop 15.12.5]. The application to constancy in a family with simultaneous log resolution answers Laza's question and is a nice payoff.\n\nThe soft spot is exactly the one the stress-test flags: formula (19) for the Hodge filtration on the V-filtration is imported from Saito's analytic monomial case, and the passage to arbitrary algebraic g is delegated to a 'standard' argument in footnote 3. The isomorphism relating V-filtrations of B_g and B_h isn't written down, and neither is the behavior of the filtration shift and the twist under that isomorphism. This is the keystone: without (19), Theorem 4.1 doesn't go through. I don't think this is an actual error—the cited [CDM24, Rem 5.10] likely covers the missing step—but it is a gap in exposition in a load-bearing place. The authors should either prove the algebraic version directly or spell out the isomorphism and its filtered behavior before publication. A referee should check this point carefully.\n\nMinor issues: Corollary 5.3 defines β_i with the same sheaf on both sides; that's a typo. The paper also cites several results from the same group ([MP22], [MP20a], [Che24]), but these are prior theorems with their own proofs, not assumptions of what the paper is trying to prove, so I don't see circularity.\n\nWho is this for? Specialists in singularities and mixed Hodge modules. A general algebraic geometer will find the statement useful as a black box, but the proof is heavy Saito theory. I'd bring it to a reading group that works in the area.\n\nMy recommendation: send it to peer review. It's a solid, novel result with a credible proof. I'd ask the authors to tighten footnote 3 before accepting, but I wouldn't desk-reject this.","headline":"Solid, novel birational description of the minimal exponent; the proof is convincing up to one imported V-filtration formula that should be spelled out.","tokens_in":25527,"tokens_out":3399,"would_cite":true,"duration_ms":29368,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14B05","14F10","14J17","32S25"],"pacs":[],"model":"deepseek-v4-flash","headline":"The minimal exponent of a hypersurface singularity is determined by vanishing of higher direct images of twisted log forms on a log resolution.","keywords":["minimal exponent","V-filtration","nearby cycles","log resolution","log canonical threshold","Hodge ideals","Bernstein-Sato polynomial","logarithmic forms"],"falsifier":"Compute the minimal exponent of a reduced hypersurface with local equation $g=u\\, y_1^{a_1}\\cdots y_r^{a_r}$ for a nonconstant invertible unit $u$, and compare the predicted V-filtration formula (19) and the vanishing condition in Corollary 1.3 against the actual Bernstein-Sato roots; a mismatch would falsify the paper's central criterion.","tokens_in":24530,"feed_emoji":"📐","tokens_out":11848,"duration_ms":91624,"temperature":0.7,"pith_summary":"The paper establishes a birational, sheaf-theoretic description of the minimal exponent, which is a finer measure of hypersurface singularity than the log canonical threshold and is defined through the Bernstein-Sato polynomial (the polynomial governing poles of $f^s$ for a local equation $f$). Its main result says that on a log resolution (a birational morphism $\\pi:Y\\to X$ from a smooth $Y$ for which $\\pi^*(Z)$ is a simple normal crossing divisor $D$ with reduced part $E$), the condition that the minimal exponent exceeds a non-integral value $p+\\alpha$ is equivalent to the vanishing of all higher direct images $R^q\\pi_*\\Omega^p_Y(\\log E)(\\lfloor\\alpha D\\rfloor)=0$ for $q\\ge 1$. This converts a subtle invariant of the rational-indexed V-filtration into concrete cohomology vanishing of twisted logarithmic forms on a resolution. The same criterion yields a new proof that, when the minimal exponent exceeds $p$, the additional jump beyond $p+\\alpha$ is controlled by a single comparison map. A consequence is the constancy of the minimal exponent in proper families of hypersurfaces admitting a simultaneous log resolution.","feed_headline":"Log-form cohomology vanishing determines the minimal exponent","feed_subtitle":"A hypersurface singularity's fine invariant becomes a checkable vanishing condition on a log resolution.","key_machinery":"The load-bearing object is a filtered complex $C_{D_\\alpha}$, built on a log resolution from the sheaves $O_Y(-D_\\alpha)\\otimes\\Omega^{n-1-q}_{Y/\\mathbb{A}^1}(\\log E)\\otimes D_Y$, which gives an explicit filtered resolution of $V_{-\\alpha}B^r_g$ (the $\\alpha$-step of the canonical V-filtration on the graph D-module of $g$; its graded quotient is the nearby-cycles module $\\psi_{g,\\alpha}(O_Y)$ after left-right conversion). Its key input is the explicit V-filtration formula $F_pV_{-\\alpha}B^r_g = \\omega_Y(-D_\\alpha+E)\\delta\\cdot F_{p+n}(D_Y[\\theta])$ for a simple normal crossing divisor, together with the Koszul structure of the associated graded complex in monomial coordinates. This resolution is what lets the paper replace the abstract derived pushforward of a Hodge-module de Rham complex by concrete sheaf cohomology of twisted logarithmic forms.","core_discovery":"On the paper's own terms, the central claim is Corollary 1.3: if $Z$ is a reduced hypersurface in a smooth complex variety $X$, $\\pi:Y\\to X$ is a log resolution, $D=\\pi^*(Z)$, $E=D_{\\mathrm{red}}$, and $\\alpha\\in(0,1)\\cap\\mathbb{Q}$ with $\\bar{\\alpha}(Z)>p$, then $\\bar{\\alpha}(Z)>p+\\alpha$ holds exactly when $R^q\\pi_*\\Omega^p_Y(\\log E)(\\lfloor\\alpha D\\rfloor)=0$ for every $q\\ge 1$. This is obtained from the more precise Theorem 1.2, which describes, for all $q$, when the natural map $R^q\\pi_*\\Omega^{n-p}_Y(\\log E)(-E-\\lfloor\\alpha D\\rfloor)\\to R^q\\pi_*\\Omega^{n-p}_Y(\\log E)(-E)$ is an isomorphism or injection. The proof passes through an explicit filtered resolution of the nearby cycles $\\psi_{g,\\alpha}(O_Y)$ (the D-module tracking the $\\alpha$-eigenspace of monodromy on the Milnor fiber cohomology) in the simple normal crossing case, so the minimal-exponent condition becomes a statement about coherent cohomology of log forms.","pith_inferences":["One could turn Corollary 1.3 into an algorithm: starting from a log resolution, compute a finite list of direct-image cohomology groups; the smallest $p$ for which a vanishing fails would determine the minimal exponent.","The explicit nearby-cycle resolution may give access not only to the minimal exponent but to the full Hodge spectrum of the singularity, so similar birational formulas might refine Du Bois and rational-singularity criteria.","The family constancy theorem might extend to families without a simultaneous log resolution, provided the vanishing conditions in Corollary 1.3 vary flatly over the base; this is a testable extension."],"forward_implications":["For $p=0$, the criterion recovers the classical fact that $\\mathrm{lct}(Z)>\\alpha$ if and only if the multiplier ideal $\\mathcal{J}(\\alpha Z)$ equals $O_X$.","For $\\alpha=1-\\varepsilon$, Corollary 1.3 together with the canonical exact triangle relating log forms to the Du Bois complex recovers the characterization of $\\bar{\\alpha}(Z)\\ge p+1$ via the $p$-th Du Bois complex being an isomorphism.","The vanishing theorem in Corollary 5.2 gives a vanishing statement for higher direct images of twisted log forms, usable independently of the minimal exponent.","Theorem 6.1 shows that the minimal exponent is constant in a proper family of hypersurfaces with a simultaneous log resolution, settling the constancy question for such families.","The description avoids the complicated derived pushforward of Hodge-ideal complexes, replacing it by direct-image sheaves of logarithmic forms."],"supporting_citations":[{"why":"Supplies the analytic V-filtration formula for SNC divisors that equation (19) imports into the algebraic setting, plus the strictness theorem used in push-forward.","marker":"[Sai90]"},{"why":"Defines the minimal exponent through the Bernstein-Sato polynomial and records the rational-singularity criterion used repeatedly.","marker":"[Sai93]"},{"why":"Gives the V-filtration characterization of the minimal exponent that underlies Theorem 2.3.","marker":"[Sai16]"},{"why":"Provides the equivalence between $\\bar{\\alpha}(Z)>p$ and $(p-1)$-rational singularities that Theorem 1.1 is built on.","marker":"[MP22]"},{"why":"Establishes the Hodge-ideal description of the minimal exponent and the semicontinuity/restriction results used in the family application.","marker":"[MP20a]"},{"why":"Provides the vanishing theorem and the exact triangle linking log forms to the Du Bois complex, used in Corollary 5.2 and Remark 5.5.","marker":"[Ste85]"},{"why":"Supplies inversion of adjunction for the minimal exponent, a key step in the constancy proof.","marker":"[Che24]"}],"fun_headline_variants":["Minimal exponent via log-form cohomology vanishing","Cohomology of twisted log forms determines minimal exponent","Vanishing of log-form cohomology gives minimal exponent","Log-resolution cohomology characterizes minimal exponent","Birational description of minimal exponent via log forms"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof rests on an explicit formula for the V-filtration of $B^r_g$ when $g$ defines a simple normal crossing divisor; if the analytic-to-algebraic translation of that formula fails for a non-monomial equation with an invertible unit, the main theorem is not supported.","fun_headline_variants_meta":{"raw":{"variants":["Minimal exponent via log-form cohomology vanishing","Cohomology of twisted log forms determines minimal exponent","Vanishing of log-form cohomology gives minimal exponent","Log-resolution cohomology characterizes minimal exponent","Birational description of minimal exponent via log forms"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000722,"raw_usage":{"total_tokens":3174,"prompt_tokens":817,"completion_tokens":2357,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":433,"completion_tokens_details":{"reasoning_tokens":2282}},"tokens_in":433,"tokens_out":2357,"duration_ms":16815,"temperature":1.0,"reasoning_tokens":2282,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T13:24:17.527087+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the minimal exponent of a reduced hypersurface with local equation $g=u\\, y_1^{a_1}\\cdots y_r^{a_r}$ for a nonconstant invertible unit $u$, and compare the predicted V-filtration formula (19) and the vanishing condition in Corollary 1.3 against the actual Bernstein-Sato roots; a mismatch would falsify the paper's central criterion.","supporting_citations":[],"review_version":1}