{"id":"9b9e3932-69a4-411f-ad2d-c8b73eda868a","arxiv_id":"2505.09912","paper_version":5,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Kovács proves that for varieties with pre-(m-1)-Du Bois singularities, the Grothendieck dual of the m-th graded Du Bois complex injects into the dual of its zeroth cohomology sheaf on cohomology, confirming Conjecture G.","lead":"This paper proves a conjecture about the cohomology of differential forms on spaces with mild singularities. The result is the first general injectivity theorem for higher Du Bois singularities and implies several related conjectures.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the Proposition 5.10 step is delicate but valid once applied to U rather than the projective closure X.","rationale":"The reader's weakest_assumption correctly locates the hyperplane-section step, but my reading is that it is not a substantive flaw. The apparent issue is a notational ambiguity: the proof says 'Let X⊇U be a projective closure... by Proposition 5.10', which could be read as applying the proposition to X, where the hypothesis generally fails. The intended application is to U, with U∩H a general hyperplane section of the affine quasi-projective variety U, and Proposition 5.10 applies there. All surrounding diagram chases, the Nakayama argument, and the éminence grise construction are consistent; the overloaded symbol for β corresponds to the composition eΩ→Ω, and the surjectivity from Theorem 8.1(v)a correctly induces the needed surjection on cokernels. I therefore see no reason to change the reader's ACCEPT verdict, though a clarifying sentence in the proof would remove the ambiguity.","tokens_in":40584,"tokens_out":41574,"duration_ms":368146,"concrete_test":"Independently verify the induction step by formally applying Proposition 5.10 to U: choose a basepoint-free linear system on U whose general member is U∩H, and confirm that the vanishing h^i(Ω^p_U ⊗^L O_{U∩H})=0 follows from the hyperderived spectral sequence using Tor-independence of h^s Ω^p_U with O_{U∩H}. If this check fails, the dimension induction in Theorem 9.1 would need a new argument.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I checked the main inductive step of Theorem 9.1. The only step that could derail the proof is the assertion that V=U∩H inherits pre-(m-1)-Du Bois singularities. As written, the sentence invoking Proposition 5.10 mentions a projective closure X, which generally does not satisfy the hypothesis; the proposition must be applied to U, with H∩U a general member of a basepoint-free linear system on U. This is legitimate because U is quasi-projective and H is chosen general. The proof of Proposition 5.10 itself is sound: the derived tensor product claim follows from the hyperderived spectral sequence with E^2_{r,s}=Tor_r(h^s Ω^p_X, O_H), and the required Tor-independence is supplied by Corollary 2.4 and Lemma 2.6. The subsequent Nakayama argument and the éminence grise step are internally consistent; the apparently problematic cokerβ is the cokernel of the natural map eΩ→Ω, and Theorem 8.1(v)a correctly makes the induced map cokerα→cokerβ surjective. No fatal gap found.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves Theorem 9.1 (Conjecture G of [PSV24]): if U is a variety of pure dimension n with pre-(m-1)-Du Bois singularities, then for every q and every p ≤ m the natural morphism h^q(D_U(Ω^p_U)) → h^q(D_U(eΩ^p_U)) is injective; equivalently, RHom_U(Ω^m_U, ω_U^•) → RHom_U(h^0(Ω^m_U), ω_U^•) is injective on cohomology. The proof develops a substantial machinery of (co-)hyperfiltrations and hyperfiltered connections, establishes preservation of pre-m-Du Bois singularities under general hyperplane sections (Proposition 5.10), uses cyclic covers to obtain a global surjectivity statement (Theorem 8.1, the so-called l'éminence grise), and then converts this global surjectivity into local injectivity via Serre vanishing and a Nakayama argument. Section 10 derives applications, including [PSV24, Conjecture H], a local cohomology surjectivity statement (Theorem 10.3), splitting criteria for pre-m-Du Bois singularities (Theorems 10.4 and 10.5), and implications from higher rational to higher Du Bois singularities (Corollaries 10.11–10.13).","tokens_in":40726,"tokens_out":21654,"duration_ms":194432,"significance":"If correct, the main theorem settles Conjecture G of [PSV24] and, through the arguments indicated in Section 10, also Conjecture H and the intersection-Du-Bois variants. The result is significant because it removes the lci and isolated-singularity assumptions that limited earlier injectivity theorems, and it introduces a new technical apparatus—co-hyperfiltrations, the 'éminence grise' object, and the proper-to-local reduction—that is likely to be useful beyond this paper. The proof is detailed and essentially self-contained: the central chain of implications is accounted for, including the induction through general hyperplane sections, the cyclic-cover comparisons, and the Serre-vanishing/localization step. A particular strength is that the paper is explicit about the provenance of the conjecture and about its overlap with [SVV]. The main risk is the sheer intricacy of the diagram chases; I found no internal inconsistency in the arguments as written.","major_comments":[],"minor_comments":[{"comment":"In the first paragraph of the proof, the sentence 'Then V:=U∩H has pre-(m-1)-Du Bois singularities by Proposition 5.10' is potentially misleading, because Proposition 5.10 as stated requires the ambient variety itself to have pre-(m-1)-Du Bois singularities, and the projective closure X is not assumed to have them. The intended and valid reading is that Proposition 5.10 is applied to the quasi-projective variety U with H∩U a general member of the restricted basepoint-free linear system; please rephrase the sentence to state this explicitly.","section":"Section 9, proof of Theorem 9.1"},{"comment":"The notation in the hypercohomology diagrams is inconsistent: the object G^p_{X,H}(L^{-j}) already contains the twist L^{-j}, yet the displayed cohomology groups write 'G^p_{X,H}(L^{-j})⊗L^{-j}' and 'G^p_X(L^{-j})⊗L^{-j}'. The same proof also uses ν^{p,j}_X both for the morphism G^p_X(L^{-j})→Ω^p_X⊗L^{-j} and for the composition eΩ^p_X⊗L^{-j}→Ω^p_X⊗L^{-j}; this makes it hard to see that cokerβ in (9.1.13) is the cokernel of the latter. Please disambiguate the notation.","section":"Section 8, proof of Theorem 8.1"},{"comment":"The assertion that h^i(Ω^p_X ⊗^L O_H) ≃ h^i(Ω^p_X) ⊗^L O_H is justified only by reference to Corollary 2.4 and Lemma 2.6. Since those results give Tor-independence of each cohomology sheaf with O_H, one uses the hyperderived spectral sequence with E^2_{r,s} = Tor_r(h^s(Ω^p_X), O_H); stating this explicitly would improve readability of an already intricate induction.","section":"Section 5, Proposition 5.10"},{"comment":"There are several typographical slips: 'defintion' in Remark 4.6, 'Sung Gi park' in the Acknowledgments should be capitalized, and 'Fianlly' appears at the end of Section 9. None of these affect the mathematics, but they should be corrected in a final revision.","section":"Throughout"}],"recommendation":"minor_revision","confidential_remarks":"The paper overlaps with [SVV] in definitions and in some results, but the author acknowledges this overlap, and the main theorem proves a conjecture from [PSV24] by a genuinely different route. I see no grounds for concern about attribution or novelty. The unusual reference for 'éminence grise' is harmless. The manuscript is well within the scope of the journal; the only requested changes are local clarifications and typographical fixes."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here is my take: the paper proves the injectivity theorem (Conjecture G of [PSV24]) for arbitrary varieties with pre-(m-1)-Du Bois singularities, with no lci or isolated assumption. That is a genuine advance. The new technical ingredient is the \"\\u00e9minence grise\" object in Section 8, which converts global surjectivity from Hodge theory into the local injectivity you want. Along the way it also gives Conjecture H, a local cohomology surjectivity statement, and splitting criteria for pre-m-Du Bois singularities.\n\nWhat is good: the theorem is new, the main steps are accounted for, and the text is careful. There is some overlap with [SVV] on definitions and corollaries, but the approach and the central theorem are different. The author is also honest about the history: he acknowledges an earlier version had a subtle mistake, now corrected. That is a point in favor of the paper's reliability, not against it.\n\nWhere are the soft spots? The proof is long and intricate, relying on spectral sequence degenerations, 4-lemma/5-lemma chases, and a delicate proper-to-local reduction. I did not machine-check any of it, and the reader did not either. The single most load-bearing step is Proposition 5.10, which says pre-(m-1)-Du Bois persists under general hyperplane section. The stress-test note checked this and concluded it is valid, provided one applies it to the affine variety U rather than the projective closure X. I agree. If that proposition failed, the induction in Theorem 9.1 would collapse; I do not see it failing.\n\nMinor comments: the relationship between the various definitions (pre-m-Du Bois, weakly-m-Du Bois, and the corresponding notions in [SVV]) is a little tangled, though not a mathematical problem. There is a light reference to a Wikipedia page for \"\\u00e9minence grise,\" which is unusual but harmless.\n\nOverall: solid, significant within-field work. It deserves a serious referee and, in my view, acceptance after careful reading. I would bring it to the reading group and would cite it if I worked in the area.\n\nRecommendation: send to peer review, with a referee who can carefully check Proposition 5.10 and the diagram in Theorem 8.1. If those hold up, the paper is correct.","headline":"Kov\\u00e1cs proves Conjecture G of Popa\\u2013Shen\\u2013Vo for arbitrary varieties with pre-(m-1)-Du Bois singularities, removing the lci/isolated restrictions; the proof is intricate but appears sound.","tokens_in":41314,"tokens_out":1780,"would_cite":true,"duration_ms":19231,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14B05","14J17"],"pacs":[],"model":"deepseek-v4-flash","headline":"Injective duality holds for all pre-(m-1)-Du Bois varieties.","keywords":["Du Bois singularities","higher Du Bois singularities","Deligne-Du Bois complex","Grothendieck duality","injectivity theorem","hyperfiltrations","local cohomology","rational singularities"],"falsifier":"A direct refutation would be a specific variety $U$ with pre-$(m-1)$-Du Bois singularities and values $q,p\\le m$ for which $h^q(D_U(\\Omega^p_U)) \\to h^q(D_U(e\\Omega^p_U))$ has a nonzero kernel. A less direct check: exhibit a pre-$m$-Du Bois $X$ whose general hyperplane section is not pre-$m$-Du Bois, which would invalidate Proposition 5.10 and the proof's induction step.","tokens_in":40326,"feed_emoji":"","tokens_out":8555,"duration_ms":72706,"temperature":0.7,"pith_summary":"This paper proves a local injectivity theorem for singular complex varieties: if a variety $U$ has pre-$(m-1)$-Du Bois singularities, then for every $q$ and every $p\\le m$ the natural cohomology map $h^q(D_U(\\Omega^p_U)) \\to h^q(D_U(e\\Omega^p_U))$ is injective, where $e\\Omega^p_U=h^0(\\Omega^p_U)$ is the zeroth cohomology sheaf of the $p$-th graded Du Bois complex. This confirms Conjecture G of Popa, Shen, and Vo, which had previously been established only for local complete intersections and for isolated singularities. The statement is purely local, about injectivity of sheaves, even though the proof has to pass through global surjectivity coming from degeneration of the Frolicher spectral sequence. The result matters because this injectivity underlies vanishing theorems, local cohomology surjectivity, and splitting criteria that feed into birational geometry and moduli theory.","feed_headline":"Injectivity for Du Bois duality proven in general","feed_subtitle":"The key duality map injects on cohomology for every pre-(m-1)-Du Bois variety, settling Popa–Shen–Vo's conjecture.","key_machinery":"The argument is carried by hyper(co)filtrations of the filtered Deligne-Du Bois complex, a derived-category formalism that lets filtrations pass through arbitrary functors. A key auxiliary object, which the paper nicknames `l'eminence grise', is a complex $G^p_X(L^{-j})$ built as a mapping cone interpolating between the $h^0$-complex $e\\Omega^p_X$ and the full graded Du Bois complex $\\Omega^p_X$; it converts the global surjectivity supplied by degeneration of the Hodge-to-de Rham spectral sequence into hypercohomology surjections. Serre vanishing then turns these global surjectivities into the desired local injectivity, and an induction on dimension using general hyperplane sections together with a Nakayama argument (an element $f$ annihilates the kernel) closes the proof.","core_discovery":"The central claim is Theorem 9.1, which the paper identifies as [PSV24, Conjecture G]. Let $U$ be a variety of pure dimension $n$ with pre-$(m-1)$-Du Bois singularities. Then for each $q$ and each $p\\le m$, the natural morphism $h^q(D_U(\\Omega^p_U)) \\to h^q(D_U(e\\Omega^p_U))$ is injective, where $D_U(-)=\\mathrm{RHom}_U(-,\\omega_U^{\\bullet})[-n]$ is the shifted Grothendieck duality functor and $e\\Omega^p_U=h^0(\\Omega^p_U)$. Equivalently, the natural map $\\mathrm{RHom}_U(\\Omega^m_U,\\omega_U^{\\bullet}) \\to \\mathrm{RHom}_U(h^0(\\Omega^m_U),\\omega_U^{\\bullet})$ is injective on cohomology. The theorem removes the earlier lci and isolated-singularity restrictions and does not assume properness of $U$.","pith_inferences":["The cone object at the heart of the proof may transfer to other duality setups, such as relative dualizing complexes for families, which could give deformation-invariance statements for higher Du Bois singularities that the paper does not state.","A natural test is whether the injectivity remains true with $h^k(\\Omega^p_U)$ in place of $h^0(\\Omega^p_U)$ for $k>0$; the paper's arguments only target the zeroth cohomology sheaf.","Because the proof relies on characteristic-zero Hodge degeneration, a meaningful extension would be to formulate the same conjecture for the de Rham-Witt or l-adic intersection complexes in positive or mixed characteristic.","The result suggests that `pre-$m$-Du Bois' is the natural generality for injectivity: the extra codimension and $S_2$ conditions appearing in the definition of $m$-Du Bois singularities are not needed for this particular statement."],"forward_implications":["Conjecture H of [PSV24] follows: if $X$ has pre-$(m-1)$-Du Bois singularities and pre-$m$-Du Bois singularities away from a closed subset of dimension $r$, then $h^i(\\Omega^m_X)=0$ for $0