{"id":"1aec8c15-7203-432c-8c20-09e2da9adce1","arxiv_id":"2506.21381","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper proves degree-one Grauert-Riemenschneider vanishing for Cohen-Macaulay klt-type schemes and, in dimension three, full GR vanishing and rational singularities.","lead":"Given a Cohen-Macaulay scheme of klt type and a resolution, the paper proves that the first higher direct image of the canonical sheaf vanishes. This gives Grauert-Riemenschneider vanishing and rationality for three-dimensional Cohen-Macaulay klt singularities in all characteristics.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Claim in Theorem 3.2 can select an exceptional divisor with n_j=0, breaking the coefficient descent; the proof of Theorem 3.2, hence of Theorem 1.2(2), is incomplete.","rationale":"The reader's weakest assumption was the treatment of non-normal exceptional divisors in the Claim. That concern is substantially mitigated by Remark 3.3 and by the fact that each F_j is a Cartier divisor in a regular scheme, hence Cohen–Macaulay and Gorenstein, so Serre duality and adjunction hold without normality. In contrast, the index-selection gap I identify is not addressed in the paper. The Claim's proof selects j as an unconstrained maximizer over J; because a_i can be arbitrarily negative, a divisor with n_i=0 can win, and then the asserted coefficient reduction n_j→n_j−1 is impossible. This is not a mere presentational issue: it blocks the induction that derives R^{d-1}π_*O_Y=0 from the initial vanishing R^{d-1}π_*O_Y(−nE)=0. Since Lemma 3.4, Theorem 1.2(2), and Corollary 1.4 all depend on Theorem 3.2, the central claim is not established as written. The concrete configuration (smooth 5-fold, point plus surface center) shows the failure occurs in a legitimate resolution of a smooth (hence klt-type) variety. The theorem may be repairable by a more careful choice of j or a different descent, but the present proof is incomplete; the honest verdict is unverified rather than accepted or rejected on the merits.","tokens_in":8705,"tokens_out":62664,"duration_ms":674446,"concrete_test":"Exhibit a geometrically realized stage where the Claim's maximizer has n_j=0. Take X a smooth 5-fold, Δ=0, π the blow-up along a disjoint union of a point and a smooth surface. Its exceptional components F_1 (over the point) and F_2 (over the surface) satisfy a_1=−4, a_2=−2, r_1=r_2=n. Trace the iterative reduction of the Claim from (n_1,n_2)=(n,n). At the stage (0,1), compute (n_i−a_i)/r_i: 4/n and 3/n, so the proof selects j=1 with n_1=0. This shows the proof's selection rule does not guarantee n_j≥1, so the descent cannot proceed. A short algebraic calculation or script reproduces the ratios and confirms the selected index is the zero-coefficient one.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"In the Claim of Theorem 3.2, the proof chooses j ∈ J as a maximizer of (n_i−a_i)/r_i, where J = {i | n_i−a_i > 0}. The Claim requires n_j ≥ 1, but nothing in the maximization enforces this. Since the klt discrepancy formula allows a_i to be very negative (positive discrepancy), even n_i = 0 can make n_i−a_i > 0 and the ratio (0−a_i)/r_i can dominate. Concretely, for a blow-up of a point in a smooth d-fold, the discrepancy is d−1, so a_i = −(d−1). During the iterative reduction, some coefficients reach 0 while others remain positive; at that stage the zero-coefficient divisor can be the maximizer. Example: blow up in a smooth 5-fold the disjoint union of a point and a smooth surface, with Δ=0. The exceptional components F_1 (point) and F_2 (surface) have a_1=−4, a_2=−2, r_1=r_2=n. At the stage (n_1,n_2)=(0,1), the ratios are (0−(−4))/n = 4/n and (1−(−2))/n = 3/n, so the selected j=1 has n_1=0. The Claim's conclusion would be a vanishing for O_Y(F_1−F_2), not a reduction toward O_Y. Thus the descent argument for R^{d−1}π_*O_Y=0 stalls, leaving Theorem 3.2 and the main vanishing unproved.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies Grauert-Riemenschneider (GR) vanishing for Cohen-Macaulay schemes of klt type. The main theorem asserts that if X is a Noetherian excellent normal scheme of klt type and Cohen-Macaulay, and π:Y→X is a resolution, then R^1π_*ω_Y=0. The authors combine this with a stated general vanishing R^{d-1}π_*ω_Y=0 to deduce that three-dimensional CM klt-type schemes satisfy full GR vanishing and have rational singularities. A further theorem asserts Q_p-rationality for klt-type schemes over perfect fields of characteristic p>0, using a companion paper by the first author.","tokens_in":8976,"tokens_out":33691,"duration_ms":380659,"significance":"If correct, the paper would answer a natural question raised by the known counterexamples to GR vanishing in positive characteristic: all known counterexamples are non-Cohen-Macaulay, and the paper would show that the CM assumption is enough to restore vanishing in degree one, with a clean three-dimensional corollary. The strategy via codimension bounds on R^iπ_*O_Y and Grothendieck duality is attractive and, if made sound, would be a useful contribution. However, the current proof has two load-bearing gaps: the descent Claim in Theorem 3.2 can select an exceptional divisor whose coefficient is already zero, and Proposition 3.1 is false as stated. These issues affect the main theorem and the advertised corollary respectively.","major_comments":[{"comment":"The descent Claim is not proved. In the Claim, J is defined as {i | n_i−a_i>0}, and the chosen j is a maximizer of (n_i−a_i)/r_i. Nothing forces n_j≥1, because a_i can be negative: with the paper's convention K_Y+Σa_iF_i∼_Qπ^*(K_X+Δ), for a blow-up of a smooth variety along a codimension-c center one has a_i=1−c, which is negative for c≥2. Thus a divisor with n_i=0 can satisfy n_i−a_i=−a_i>0 and can attain the maximum. Concretely, for the blow-up of a smooth 5-fold along the disjoint union of a point and a surface, with Δ=0, the two exceptional components have a_1=−4 and a_2=−2, and r_1=r_2=n. At coefficients (n_1,n_2)=(0,1), the ratios are 4/n and 3/n, so the proof selects j=1 with n_1=0. The conclusion would assert vanishing for O_Y(F_1−F_2), not a reduction toward O_Y. The iterative descent therefore stalls, and Theorem 3.2, and hence Theorem 1.2(2), is not established by the given argument.","section":"Theorem 3.2, Claim (page 5)"},{"comment":"Proposition 3.1 is false as stated. It claims that R^{d-1}π_*ω_Y=0 for every resolution of every d-dimensional variety. For d=2 this would imply every normal surface singularity is rational, since for a resolution of a surface, R^1π_*ω_Y is the local-duality counterpart of R^1π_*O_Y. A cone over an elliptic curve is a normal surface singularity with nonzero R^1π_*O_Y and hence nonzero R^1π_*ω_Y. The proof's reduction to dim 2 and citation of [Kol13, Theorem 10.4] appears to import a statement that does not hold without additional hypotheses. Moreover, the proof invokes 'relative Serre vanishing' to assert R^{d-1}π_*ω_X(H)=0 for a single hyperplane twist; relative Serre vanishing gives vanishing for sufficiently high powers of a relatively ample line bundle, not for the first twist. Since Corollary 1.4 and Remark 1.3(a) rely on Proposition 3.1, this needs to be corrected: either the proposition must be restricted to an appropriate class (e.g. klt type, with a valid proof), or Corollary 1.4 must be proved by a different argument that supplies R^{d-1}π_*ω_Y=0 for the CM klt threefold case.","section":"Proposition 3.1 (page 3)"},{"comment":"The Claim's bigness and duality steps are applied to exceptional divisors F_j that are only integral, not necessarily normal. The displayed Serre-duality isomorphism for F_j and the restriction argument for bigness need a justification in this non-normal setting. The authors state in Remark 3.3 that they work with Q-line bundles, but the proof as written uses a Serre-duality statement involving K_{F_j} for a possibly non-normal divisor. Since F_j is a Cartier divisor on a regular scheme Y, it is Gorenstein and Cohen-Macaulay, so a version of Serre duality may be available, but this is not explained. This issue is less severe than the coefficient-descent gap, but it is load-bearing for the Claim and should be addressed in a revision.","section":"Theorem 3.2, Claim, non-normal F_j (Remark 3.3)"}],"minor_comments":[{"comment":"There is a notational slip: the proof writes R^{d-1}π_*ω_X(H), but the sheaf should live on Y, so it should be R^{d-1}π_*(ω_Y⊗π^*O_X(H)) or similar.","section":"Proposition 3.1 proof (page 3)"},{"comment":"The condition a_i∈Q_{<1} is nonstandard and confusing. For a klt pair written as K_Y+Σa_iF_i+π_*^{-1}Δ∼_Qπ^*(K_X+Δ), the discrepancies normally satisfy a_i>−1. Please clarify the sign convention and the relation to the klt assumption.","section":"Theorem 3.2 proof (page 5)"},{"comment":"There are several small typos, e.g. 'we can reduced' should be 'we can reduce', and the displayed exact sequence after Proposition 3.1 uses ω_X(H) where the intended sheaf is on Y; these should be corrected.","section":"Throughout"},{"comment":"The proof of Theorem 1.5 depends on [Bau25, Theorem 5.1.4], an unpublished companion paper by the first author. This dependency should be explicitly flagged as a preprint dependency, and the referee should be able to check that theorem if the publication decision relies on it.","section":"Theorem 1.5 and Remark 1.6"}],"recommendation":"major_revision","confidential_remarks":"The central idea is promising, and the paper is clearly written in most places, but the two gaps I identified are not merely cosmetic: the descent Claim in Theorem 3.2 appears to fail as written, and Proposition 3.1 is stated too broadly. The former directly undermines Theorem 1.2(2); the latter undermines Corollary 1.4 and Remark 1.3(a). I would be willing to consider a revised version that fixes the descent argument and either repairs Proposition 3.1 under the needed hypotheses or replaces it with a valid proof of the dimension-three vanishing used in Corollary 1.4. The dependence on the companion paper [Bau25] is a concern for the Q_p-rationality theorem, but it is not the main obstruction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper argues that for Cohen-Macaulay klt-type schemes, the first higher direct image of the canonical sheaf under a resolution vanishes, and that in dimension three this gives full GR vanishing and rational singularities. That is a natural and valuable statement, and the surrounding structure is mostly well done. Proposition 3.1 and Lemma 3.4 are standard and clean, the applications to strongly F-regular and +-regular threefolds follow directly, and the text is readable. The citation pattern is honest; the Q_p-rationality theorem depends on the companion paper [Bau25], and the authors say so.\n\nThe soft spot is the Claim inside Theorem 3.2. The proof selects j in J maximizing (n_i - a_i)/r_i, where J = {i | n_i - a_i > 0}. Nothing forces n_j >= 1. Concretely, on a smooth 5-fold, blow up the disjoint union of a point and a smooth surface. The exceptional divisors F_1, F_2 have discrepancies -4 and -2, so a_1 = -4, a_2 = -2. At the stage (n_1, n_2) = (0, 1), both indices lie in J; the ratios are 4/n and 3/n, so the maximizer is j = 1 with n_1 = 0. The Claim's conclusion requires a positive coefficient to reduce, and the proof does not provide one. The induction from R^{d-1} pi_* O_Y(-nE) = 0 to R^{d-1} pi_* O_Y = 0 therefore stalls. This is not a minor technicality; it is a load-bearing gap in the proof of Theorem 3.2, and hence of Theorem 1.2(2) and Corollary 1.4 as written.\n\nThe non-normal exceptional divisor issue flagged in Remark 3.3 is real but secondary; the coefficient-selection problem is more serious. The result may well be true and fixable by a more careful choice of j, but as written the central argument is incomplete. The paper is for researchers in birational geometry and positive-characteristic singularities. It deserves a serious referee, and I would send it to review with the expectation of substantial revision to repair the Claim. I would not yet cite the main theorem in my own work.","headline":"A natural result with a real gap in the key descent claim; the main theorem is unproved as written but the paper deserves referee attention.","tokens_in":9570,"tokens_out":10511,"would_cite":false,"duration_ms":110681,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14F17","14B05","13A35"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that Cohen-Macaulay schemes of klt type satisfy the degree-one Grauert-Riemenschneider vanishing $R^1\\pi_*\\omega_Y=0$, and that three-dimensional such schemes have rational singularities.","keywords":["Grauert-Riemenschneider vanishing","klt singularities","Cohen-Macaulay schemes","rational singularities","positive characteristic","Witt vectors","Q_p-rational singularities","higher direct images"],"falsifier":"Construct a Cohen-Macaulay klt-type threefold over a field of characteristic 2, 3, or 5 with a resolution $\\pi:Y\\to X$ for which $R^1\\pi_*\\omega_Y\\neq0$; that would directly contradict Corollary 1.4. Concretely, take one of the known non-Cohen-Macaulay klt counterexamples to Grauert-Riemenschneider vanishing and test whether a small Cohen-Macaulay modification remains of klt type and still has a nonzero first direct image.","tokens_in":8443,"feed_emoji":"🧮","tokens_out":9804,"duration_ms":96360,"temperature":0.7,"pith_summary":"The paper asks whether Grauert-Riemenschneider vanishing, which is known to fail in positive characteristic, survives for singularities that are both of klt type and Cohen-Macaulay. It proves the first missing case: for any resolution $\\pi:Y\\to X$ of such a scheme, $R^1\\pi_*\\omega_Y=0$. Since the top-degree vanishing $R^{d-1}\\pi_*\\omega_Y=0$ was already known, a three-dimensional Cohen-Macaulay klt-type scheme now satisfies full Grauert-Riemenschneider vanishing and has rational singularities. The same argument yields $\\mathbb{Q}_p$-rational singularities for Cohen-Macaulay klt-type schemes of any dimension over a perfect field of characteristic $p>0$.","feed_headline":"First higher canonical image vanishes for CM klt singularities","feed_subtitle":"Three-dimensional Cohen-Macaulay klt-type spaces then satisfy full Grauert-Riemenschneider vanishing.","key_machinery":"The engine is a codimension bound for higher direct images of the structure sheaf (Theorem 3.2): for a klt-type variety $X$ and resolution $\\pi:Y\\to X$, every $i>0$ satisfies $\\operatorname{codim} \\operatorname{Supp} R^i\\pi_*\\mathcal{O}_Y > i+1$. The proof proceeds by induction on dimension, subtracting exceptional divisors $F_j$ one at a time through short exact sequences; the choice of $F_j$ is governed by klt type, which supplies coefficients $a_i<1$ so that $-(K_Y+\\sum n_iF_i)$ restricts to a big line bundle on $F_j$. Lemma 3.4 then converts this codimension estimate into the desired vanishing: Grothendieck duality and the Cohen-Macaulay property place $R^1\\pi_*\\omega_Y$ inside $H^{-(d-1)}(\\omega_X^\\bullet)$, which vanishes.","core_discovery":"The central result is Theorem 1.2(2): if $X$ is a Noetherian excellent normal scheme of klt type, meaning there is an effective $\\mathbb{Q}$-divisor $\\Delta$ making $(X,\\Delta)$ klt, and $X$ is also Cohen-Macaulay, then every resolution $\\pi:Y\\to X$ satisfies $R^1\\pi_*\\omega_Y=0$. Together with the known vanishing $R^{d-1}\\pi_*\\omega_Y=0$, this gives Corollary 1.4 in dimension three: $R\\pi_*\\omega_Y=\\omega_X$, so $X$ satisfies Grauert-Riemenschneider vanishing and has rational singularities. In positive characteristic, the paper deduces Theorem 1.5: a Cohen-Macaulay klt-type scheme of finite type over a perfect field is $\\mathbb{Q}_p$-rational, and with projectivity plus isolated singularities it is Witt-rational. The authors note that the Cohen-Macaulay hypothesis cannot simply be dropped, because the known positive-characteristic counterexamples to Grauert-Riemenschneider vanishing are all non-Cohen-Macaulay.","pith_inferences":["I infer that Cohen-Macaulayness, rather than characteristic zero, is the structural condition protecting degree-one Grauert-Riemenschneider vanishing; a natural next test is whether the same induction pushes the vanishing to all higher degrees $R^i\\pi_*\\omega_Y=0$ for Cohen-Macaulay klt-type schemes of any dimension.","Because the proof uses only a resolution and not a log resolution, I infer it may adapt to settings where log resolutions are unavailable, such as mixed characteristic or low-dimensional positive characteristic.","The $\\mathbb{Q}_p$-rationality result is stated under Cohen-Macaulayness, but the authors note it only needs $\\mathbb{Q}_p$-Cohen-Macaulayness, a Frobenius-theoretic depth condition preserved under universal homeomorphisms and finite quotients; I infer this makes the result applicable to quotient singularities where ordinary Cohen-Macaulayness fails."],"forward_implications":["Three-dimensional Cohen-Macaulay klt-type schemes satisfy full Grauert-Riemenschneider vanishing, so $R\\pi_*\\omega_Y=\\omega_X$ for every resolution and they have rational singularities.","Strongly F-regular and quasi-F-regular threefolds in positive characteristic satisfy Grauert-Riemenschneider vanishing and have rational singularities, since such varieties are Cohen-Macaulay and of klt type.","Globally +-regular three-dimensional pairs with $\\mathbb{Q}$-Cartier $K_X+\\Delta$ also satisfy Grauert-Riemenschneider vanishing and have rational singularities.","In arbitrary dimension over a perfect field of characteristic $p>0$, a Cohen-Macaulay klt-type scheme is $\\mathbb{Q}_p$-rational; adding projectivity and isolated singularities upgrades this to Witt-rationality.","The degree-one vanishing $R^1\\pi_*\\omega_Y=0$ holds in every dimension for Cohen-Macaulay klt-type schemes, with $\\pi_*\\omega_Y=\\omega_X$ as part of the proof."],"supporting_citations":[{"why":"Defines the klt condition for pairs, hence the class of klt-type varieties the theorem concerns.","marker":"[BMP+23, Definition 2.28]"},{"why":"Supplies the general hyperplane section used in Proposition 3.1 to reduce the top-degree vanishing to lower dimension.","marker":"[BMP+23, Theorem 2.17]"},{"why":"Provides the dimension-two base case of the inductive proof of Proposition 3.1.","marker":"[Kol13, Theorem 10.4]"},{"why":"Gives the degree support bound for the dual of a higher direct image used in Lemma 3.4.","marker":"[Sta25, Tag 0A7U]"},{"why":"Supplies the companion vanishing $R^1\\pi_*\\mathcal{O}_Y=0$ cited in Remark 1.3(b).","marker":"[IY24, Theorem 1.3]"},{"why":"Carries the Witt-vector duality input from the companion paper on which the $\\mathbb{Q}_p$-rationality theorem depends.","marker":"[Bau25, Theorem 5.1.4]"},{"why":"One of the known non-Cohen-Macaulay counterexamples to Grauert-Riemenschneider vanishing, cited to show the Cohen-Macaulay hypothesis is necessary.","marker":"[Ber21]"},{"why":"Known non-Cohen-Macaulay klt counterexample in characteristic two, cited with Ber21 to justify the Cohen-Macaulay assumption.","marker":"[CT19]"},{"why":"Known non-Cohen-Macaulay counterexample for log del Pezzo surfaces, cited to show the Cohen-Macaulay assumption cannot be removed.","marker":"[ABL22]"}],"fun_headline_variants":["R^1 of canonical sheaf vanishes for CM klt schemes","3D CM klt-type schemes get rational singularities","Grauert-Riemenschneider vanishing for 3D CM klt","CM klt type schemes are Q_p-rational over perfect fields","First higher direct image of canonical sheaf vanishes for CM klt type"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the non-normal exceptional divisors in the induction still satisfy the duality and bigness statements used to conclude vanishing, a step the authors flag but do not spell out, and for the $\\mathbb{Q}_p$-rationality half the companion paper's theorem must hold.","fun_headline_variants_meta":{"raw":{"variants":["R^1 of canonical sheaf vanishes for CM klt schemes","3D CM klt-type schemes get rational singularities","Grauert-Riemenschneider vanishing for 3D CM klt","CM klt type schemes are Q_p-rational over perfect fields","First higher direct image of canonical sheaf vanishes for CM klt type"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001789,"raw_usage":{"total_tokens":7020,"prompt_tokens":886,"completion_tokens":6134,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":502,"completion_tokens_details":{"reasoning_tokens":6041}},"tokens_in":502,"tokens_out":6134,"duration_ms":45727,"temperature":1.0,"reasoning_tokens":6041,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T22:27:50.589688+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a Cohen-Macaulay klt-type threefold over a field of characteristic 2, 3, or 5 with a resolution $\\pi:Y\\to X$ for which $R^1\\pi_*\\omega_Y\\neq0$; that would directly contradict Corollary 1.4. Concretely, take one of the known non-Cohen-Macaulay klt counterexamples to Grauert-Riemenschneider vanishing and test whether a small Cohen-Macaulay modification remains of klt type and still has a nonzero first direct image.","supporting_citations":[],"review_version":1}