{"id":"91da0e65-3470-44cb-b06e-7618e3d4fca4","arxiv_id":"2608.13403","paper_version":1,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Twisted hyperholomorphic vector bundles built from transverse Lagrangian fibrations yield proofs of the Lefschetz standard conjecture and D-equivalence for OG10-type hyperkähler manifolds.","lead":"Using two transverse Lagrangian fibrations on a hyperkähler manifold, the authors construct twisted holomorphic vector bundles on products of K3^[n] and O'Grady 10 type manifolds. They then prove the Lefschetz standard conjecture and the derived-equivalence conjecture for all O'Grady 10 hyperkähler manifolds.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The final step of Theorem 6.9 appears to fail: the transported B-field (D+gv)/g is not integral, since it pairs to 1/g with an integral class, so the twist cannot be removed.","rationale":"I read the paper as a good-faith, sophisticated attempt to transfer Markman's hyperholomorphic method to OG10. The construction of Q, the use of Arinkin/Bottini/Yu, and the deformation theory are substantial and largely coherent. The reader's weakest point, Proposition 4.5, is indeed terse, but my disagreement is with the reader's emphasis: the most decisive gap I find is not the ±1 sign but the final integrality/Brauer triviality step in Theorem 6.9. The algebra there is checkable by a single lattice pairing, and it appears to contradict the claim. The rest of the paper, including the Lefschetz standard conjecture (Theorem 6.1), may survive, but the D-equivalence theorem as proved does not go through as written. This justifies moving from conditional acceptance to rejection unless the authors supply a different mechanism for killing the twist that does not require divisibility of a primitive class by g.","tokens_in":29963,"tokens_out":35568,"duration_ms":389186,"concrete_test":"Compute the pairing of the transported B-field with the integral class μ1' from Lemma 6.13. Since (D+gv, μ1')=1, the class [(D+gv)/g] in H^2(X,Q/Z) has nonzero pairing 1/g with μ1', so it is not the zero Brauer class for g>1. If the authors maintain that the B-field is trivial, they must exhibit an integral class W with (D+gv)/g - W having zero image under [−]; the displayed equality in Theorem 6.9 gives no such W. A minimal check is to rerun the paragraph after Corollary 2.20 with the substitutions ρ(-ϵγ2)=D+gv and B'_X0=3η_X0-ϵγ2/g and require (ρ(B'_X0), μ1') ∈ Z; it fails for g>1.","verdict_should_be":"REJECT","load_bearing_attack":"In the proof of Theorem 6.9 the authors need the parallel transported B-field on X to represent the trivial Brauer class. Lemma 6.2 gives B'_X0 = 3η_X0 - ϵγ2/g, and the Eichler isometry ρ is chosen so that ρ(-ϵγ2)=D+gv, where D and v come from Lemmas 6.12 and 6.13. Hence the transported fractional term is (D+gv)/g = D/g+v. The paper asserts that because this lies in Pic(X)_Q + H^2(X,Z) it represents the trivial Brauer class. Under the paper's own definition (Proposition 2.19, Corollary 2.20), the Brauer class of a B-field B is [B] ∈ H^2(X,Q/Z), the image modulo the integral lattice. But Lemma 6.13 constructs D+gv with (D+gv, μ1')=1 for an integral class μ1'. Therefore ((D+gv)/g, μ1')=1/g, so [(D+gv)/g] is nonzero in H^2(X,Q/Z) for g>1. Membership in Pic(X)_Q is not enough: fractional algebraic classes define nontrivial Brauer classes in this formalism. Since the proof requires g large (Proposition 5.3), this is not a harmless edge case. The displayed triviality claim is load-bearing: it is exactly the step that converts the twisted equivalence D^b(X,α_X)≃D^b(Z'',α_Z'') into the untwisted equivalences D^b(X)≃D^b(Z'',α_Z'')≃D^b(X').","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a new construction of twisted hyperholomorphic vector bundles on products of hyperkähler manifolds of K3[n]- and OG10-type, following and generalizing work of Kapustka–Kapustka. Given a hyperkähler manifold Y admitting two transverse Lagrangian fibrations that are Tate–Shafarevich twists of compactified abelian fibrations, the authors form the convolution Q of two twisted Poincaré sheaves and claim in Theorem 5.8 that Q is a twisted hyperholomorphic vector bundle on X × Z. Deforming Q along generic twistor paths yields twisted derived equivalences for arbitrary manifolds of the same deformation type. The two advertised applications are the Lefschetz standard conjecture for OG10-type hyperkähler manifolds (Theorem 6.1) and the D-equivalence conjecture for birational projective OG10-type hyperkähler manifolds (Theorem 6.9). The proof of Theorem 6.9 transports a rational B-field to X and asserts that the transported fractional class represents the trivial Brauer class, thereby removing the twist.","tokens_in":30262,"tokens_out":14456,"duration_ms":149249,"significance":"If the main theorems were established, they would be substantial: the D-equivalence conjecture for OG10-type hyperkähler manifolds is a major open problem, and a proof of the Lefschetz standard conjecture for all OG10-type manifolds would significantly extend known results. The paper is clearly organized, engages seriously with recent work of Arinkin, Bottini, Yu, Dutta–Mattei–Shinder, and Markman, and the construction of Q in Theorem 5.8 is an interesting and potentially reusable technique. The cohomological bookkeeping via LLV-equivariant isometries and the explicit B-field computations are also valuable. However, the proof of the headline D-equivalence theorem contains a load-bearing error in its final step: the transported B-field is claimed to be Brauer-trivial when, under the paper's own definition, it is not. As written, Theorem 6.9 is not proved, and the advertised application to D-equivalence is therefore unsupported.","major_comments":[{"comment":"The assertion that (D+gv)/g ∈ Pic(X)_Q + H^2(X,Z) represents the trivial Brauer class is false under the paper's own definition of the Brauer class of a B-field (Proposition 2.19 and Corollary 2.20). By Lemma 6.13, the isotropic vector D+gv has divisibility one, so there is an integral class μ_1' with (D+gv, μ_1') = 1. Therefore ((D+gv)/g, μ_1') = 1/g, which is nonzero in H^2(X,Q/Z) for g > 1. The proof requires g to be large (Proposition 5.3), so this is not a removable edge case. Since 3η_X is integral and contributes zero to the Brauer class, the parallel transported B-field has nontrivial Brauer class. Consequently the chain of untwisted equivalences D^b(X) ≃ D^b(Z'', α_{Z''}) ≃ D^b(X') is not justified, and Theorem 6.9 is not proved.","section":"§6.3, proof of Theorem 6.9, final paragraph"},{"comment":"The step from fiberwise stability to global stability is asserted in one sentence: 'This is enough to deduce that Q is μ_H-stable.' Slope stability of the restrictions Q|_{x×Z} and Q|_{X×z} with respect to suitable polarizations does not by itself imply slope stability of Q with respect to pr_1^*H_X + pr_2^*H_Z; a proof or a precise reference for this twisted analogue is needed. Since Theorem 2.17, Corollary 5.9, and all subsequent applications rely on the global stability of Q, this is a load-bearing gap in the main construction as written.","section":"§5.1, Theorem 5.8"},{"comment":"The proof that the Poincaré-sheaf isometry acts by ±1 on the integral Lagrangian primitive cohomology H^2_lprim(X,Z) is not fully spelled out. For an arbitrary primitive class λ of large negative square, one must show that there is a deformation of the Beauville–Mukai or LSV system in which λ (or, in the sign-comparison step, a primitive linear combination pλ1+qλ2) is the unique integral Hodge class in H^2_lprim, while the relevant Poincaré sheaf exists and the period stays in the allowed locus. The current text asserts this rather than proving it, and the subsequent sign-agreement argument needs a Hodge-theoretic justification. This matters because Lemma 6.2 and Corollary 6.5, and hence the k-cyclic isometry used in Section 6, depend on this integrality and sign.","section":"§4, Proposition 4.5 and Remark 4.8"}],"minor_comments":[{"comment":"There are two apparent typos in the proof: the saturation should be in the lattice Λ rather than in 'M', and the formula for v_2 should presumably read (p+ka)w_1 + a w_2 rather than (p+ka)v_1 + a v_2. As written, the proof is confusing even though the intended construction is recognizable.","section":"Lemma 4.6, proof"},{"comment":"The displayed computation in the proof of the second diagram has an apparent sign inconsistency: the term (-akϵμ_1 + ϵv_2, B'_Z)f is claimed to equal af, but the signs in the preceding line suggest the opposite. Please verify the calculation, since it feeds directly into Corollary 6.5.","section":"Proposition 6.4, displayed computation"},{"comment":"The proof refers to 'Lemma 6.14 below' before the lemma is stated; this is acceptable stylistically but should be reordered or cross-referenced for readability. More importantly, the final paragraph should be rewritten in light of the Brauer-class issue above.","section":"§6.3, proof of Theorem 6.9"}],"recommendation":"reject","confidential_remarks":"The error in the proof of Theorem 6.9 is not a small presentation issue: the transported B-field genuinely represents a nontrivial Brauer class, and no local modification of the displayed argument appears to remove the twist. The authors would need a substantially different mechanism to prove the D-equivalence theorem. The Lefschetz-standard-conjecture application may survive, but the paper's headline claim as stated is unsupported, so I do not see a path to acceptance without major new work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The paper has a genuinely new construction: twisted hyperholomorphic vector bundles on products X×Z built from two transverse Lagrangian fibrations, covering both K3[n] and OG10, and the applications to the Lefschetz standard conjecture for OG10 look plausible. But the D-equivalence theorem (Theorem 6.9) has a real gap in its final step. The stress-test note is on target: the transported B-field (D+gv)/g is not integral, so it does not represent the trivial Brauer class.\n\nWhat is actually new: the convolution Q of twisted Poincaré sheaves, shown to be a twisted hyperholomorphic vector bundle (Theorem 5.8). If that holds—and I could not fully verify it—then Corollary 5.9 plus Markman's lemma gives Theorem 6.1 quickly. That part is coherent. The paper is careful about what it relies on: Arinkin, Bottini, Yu, Dutta–Mattei–Shinder, etc. No circularity.\n\nThe soft spots: Theorem 5.8's passage from fiberwise stability to global product stability is compressed into a paragraph. It may be fine, but it needs expansion. Proposition 4.5 (Poincaré isometry acts as ±1 on Lagrangian primitive cohomology) leans on a deformation-to-Noether–Lefschetz argument with large-negative-square classes; the sign bookkeeping is delicate and not fully written out. The paper also depends on several unreviewed preprints, which is normal in this area but means the referee has extra homework.\n\nThe main issue is in Section 6.3. Lemma 6.13 produces v with (D+gv, μ1')=1 for an integral class μ1'. The authors transport B'_X0=3η_X0−ϵγ2/g, sending −ϵγ2 to D+gv, so the transported fractional part is (D+gv)/g. They claim that because this lies in Pic(X)_Q + H^2(X,Z), it represents the trivial Brauer class. Under their own definition (Proposition 2.19, Corollary 2.20), the Brauer class of a B-field is its image in H^2(X,Q/Z). Since (D+gv)/g pairs to 1/g with μ1', it is not in H^2(X,Z), and its Brauer class is nonzero. Adding an integral class v doesn't help. So the untwisted equivalence D^b(X)≃D^b(X') is not obtained by this argument. This is a load-bearing point, not a minor typo.\n\nWhere does that leave the paper? The twisted machinery and the LSC result are likely salvageable; the D-equivalence proof needs a new idea or a different choice of B-field. I'd send it to a careful referee, but flag Theorem 6.9 for close scrutiny. If the authors fix the integrality gap, this becomes a strong paper. As written, the main theorem has a genuine hole.","headline":"A serious, well-structured paper whose twisted-bundle construction is new and useful, but the proof of Theorem 6.9 has a load-bearing integrality gap: the transported B-field is not integral, so the untwisted D-equivalence does not follow as written.","tokens_in":30822,"tokens_out":4440,"would_cite":false,"duration_ms":39314,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14J42","14F08","14C25"],"pacs":[],"model":"deepseek-v4-flash","headline":"A twisted hyperholomorphic vector bundle built from two Poincaré equivalences proves the Lefschetz standard conjecture and the D-equivalence conjecture for all projective OG10-type hyperkähler manifolds.","keywords":["hyperkähler manifolds","OG10 type","twisted hyperholomorphic sheaves","Lagrangian fibrations","Lefschetz standard conjecture","D-equivalence conjecture","cubic fourfolds","Tate–Shafarevich twists"],"falsifier":"Compute the action of the Poincaré-sheaf Fourier–Mukai isometry directly on an integral class $\\lambda$ of small negative square (say square $-1$ or $-2$) in $H^2_{\\mathrm{lprim}}(X,\\mathbb{Z})$ for an explicit compactified Jacobian or LSV example; Proposition 4.5 asserts the image is $\\pm \\lambda$ for every class, while its proof only covers classes of sufficiently large negative square, so a single small class with a different image would show the claim is false as stated.","tokens_in":29731,"feed_emoji":"🌀","tokens_out":12176,"duration_ms":105351,"temperature":0.7,"pith_summary":"The paper sets out to prove two long-standing conjectures for hyperkähler manifolds of O'Grady 10 (OG10) type: Grothendieck's Lefschetz standard conjecture, and the D-equivalence conjecture that birational projective varieties have equivalent bounded derived categories. The route is a new supply of twisted hyperholomorphic vector bundles on products of hyperkähler manifolds, formed by composing two twisted Poincaré equivalences attached to transverse Lagrangian fibrations. Because such bundles deform over twistor families, a derived equivalence established at one point of moduli space propagates to every manifold in the deformation family — the same phenomenon that already settled these conjectures for K3^[n]-type manifolds. In the OG10 case the propagation yields both conjectures; a subsidiary output is a twisted derived equivalence between Fano varieties of lines of two cubic fourfolds.","feed_headline":"Two standard conjectures proved for OG10 hyperkähler manifolds","feed_subtitle":"A twisted hyperholomorphic vector bundle from two Lagrangian fibrations drives both proofs","key_machinery":"The load-bearing object is the convolution $$Q = \\Phi_{P_{Y,Z}} \\circ \\Phi_{P_{X,Y}} \\in D^b(X \\times Z, \\alpha_X \\boxtimes \\alpha_Z)$$ of the two twisted Poincaré sheaves, a twisted vector bundle of rank $n!k^n$ whose restrictions to the fibers of either projection are stable twisted vector bundles. Its role is to be a hyperholomorphic kernel: stability with respect to polarizations related by the induced Hodge isometry $\\psi$ makes it deformable over generic twistor paths, so the twisted derived equivalence it defines at one point of moduli space propagates to all points. The explicit formulas for the B-fields $B'_X, B'_Z$ and for the $k$-cyclic isometry on Lagrangian primitive cohomology are what allow the OG10 applications.","core_discovery":"The central claim is that the convolution $Q$ of two twisted Poincaré sheaves — one from $X$ to a common fibered manifold $Y$ and one from $Y$ to $Z$, where $X, Y, Z$ are hyperkähler manifolds of K3^[n]- or OG10-type — is itself a twisted hyperholomorphic vector bundle on $X \\times Z$ (Theorem 5.8). Its fiberwise stability with respect to suitable polarizations is inherited from modular stability theorems for twisted Poincaré sheaves, and the induced Hodge isometry $\\psi$ on $H^2$ is explicitly controlled, so $Q$ deforms along diagonal twistor lines to a twisted vector bundle at every point of the relevant moduli space of Hodge isometries. At each deformed point the bundle still induces a twisted derived equivalence, by the deformation principle for hyperholomorphic bundles. Specializing to OG10-type and combining the resulting equivalences with the known mechanism that turns a nonzero-rank Fourier–Mukai kernel into a proof of the Lefschetz standard conjecture, the paper derives Theorem 6.1; a chamber-by-chamber argument adapting the K3^[n]-type strategy gives Theorem 6.9, the D-equivalence conjecture.","pith_inferences":["If the deformation argument behind Proposition 4.5 is sound, the same two-conjecture pipeline should work for any hyperkähler deformation type admitting transverse Lagrangian fibrations whose fibers are compactified abelian schemes; O'Grady 6-type would be the natural next case once analogues of the Poincaré-sheaf and Tate–Shafarevich inputs exist.","The explicit dependence of the B-fields on the integer $k$ suggests a countable family of twisted derived equivalences between OG10 manifolds, with $k$ as a parameter; testing whether different $k$ give genuinely different Brauer classes would isolate the new content of the construction.","The paper leaves open whether its bundle $Q$ is deformation equivalent to the previously known hyperholomorphic bundle on K3^[n] products along diagonal twistor lines; comparing Chern characters and B-fields on the K3^[n] side would settle that unification question.","One could push the same kernel-based reasoning toward an algebraicity statement for rational Hodge isometries of OG10-type manifolds, by analogy with the K3^[n]-type result, since the construction produces algebraic kernels realizing the isometries."],"forward_implications":["Every projective hyperkähler manifold of OG10-type satisfies the Lefschetz standard conjecture (Theorem 6.1).","Any two birational projective hyperkähler manifolds of OG10-type have equivalent bounded derived categories (Theorem 6.9).","For the very good cubic fourfolds arising in the construction, the Fano varieties of lines carry a twisted derived equivalence $D^b(F_1(C_X), \\delta_X) \\simeq D^b(F_1(C_Z), \\delta_Z)$ (Theorem 6.8).","The same convolution produces new twisted hyperholomorphic vector bundles on products of K3^[n]-type hyperkähler manifolds, beyond the previously known hyperholomorphic sheaf construction (Theorem 1.3).","The isometry induced on Lagrangian primitive cohomology is of $k$-cyclic type and yields explicit B-field identities (Lemma 6.2, Corollary 6.5), which are the computational backbone of the derived-equivalence proof."],"supporting_citations":[{"why":"supplies the autoduality/Poincaré-sheaf derived equivalence for compactified Jacobians, the base case of the kernel.","marker":"[Ari13]"},{"why":"supplies the twisted Poincaré sheaf for Tate–Shafarevich twists, the B-field decomposition, and the modular stability results used to prove hyperholomorphicity.","marker":"[Bot25]"},{"why":"extends autoduality/Poincaré-sheaf equivalences to compactified Prym and LSV fibrations, opening the OG10 case.","marker":"[Yu26]"},{"why":"characterizes Tate–Shafarevich twists of LSV fibrations and their periods, letting the authors realize arbitrary OG10 Lagrangian fibrations in that form.","marker":"[DMS25]"},{"why":"provides the original construction strategy and the deformation principle turning a hyperholomorphic bundle into derived equivalences at every point of moduli space.","marker":"[KK25]"},{"why":"provides the twistor-path deformation machinery for hyperholomorphic sheaves and the lemma converting a nonzero-rank kernel into the Lefschetz standard conjecture.","marker":"[Mar24]"},{"why":"gives the LLV-equivariant isometry theory and extended Mukai lattice formalism used to compute and deform the cohomological action of the kernels.","marker":"[Tae23]"},{"why":"supplies the chamber-by-chamber strategy for the D-equivalence conjecture that the OG10 proof adapts.","marker":"[Mau+25]"}],"fun_headline_variants":["Twisted hyperholomorphic sheaves settle two OG10 conjectures","OG10: Lefschetz and D-equivalence from Lagrangian fibrations","Twisted sheaves from fibrations yield OG10 proofs","Two conjectures down: OG10 via twisted hyperholomorphic sheaves","OG10 proved: Lefschetz and D-equivalence"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Poincaré-sheaf isometry acts on the integral Lagrangian primitive cohomology $H^2_{\\mathrm{lprim}}(X,\\mathbb{Z})$ as multiplication by $\\pm 1$; the proof establishes this by deforming a class to a very general point of the period domain where it is the unique integral Hodge class and then extending the sign to all classes through differences of large-negative-square vectors, and that deformation step is where the argument is most delicate.","fun_headline_variants_meta":{"raw":{"variants":["Twisted hyperholomorphic sheaves settle two OG10 conjectures","OG10: Lefschetz and D-equivalence from Lagrangian fibrations","Twisted sheaves from fibrations yield OG10 proofs","Two conjectures down: OG10 via twisted hyperholomorphic sheaves","OG10 proved: Lefschetz and D-equivalence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00081,"raw_usage":{"total_tokens":3509,"prompt_tokens":856,"completion_tokens":2653,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":472,"completion_tokens_details":{"reasoning_tokens":2564}},"tokens_in":472,"tokens_out":2653,"duration_ms":18635,"temperature":1.0,"reasoning_tokens":2564,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:45:20.865704+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the action of the Poincaré-sheaf Fourier–Mukai isometry directly on an integral class $\\lambda$ of small negative square (say square $-1$ or $-2$) in $H^2_{\\mathrm{lprim}}(X,\\mathbb{Z})$ for an explicit compactified Jacobian or LSV example; Proposition 4.5 asserts the image is $\\pm \\lambda$ for every class, while its proof only covers classes of sufficiently large negative square, so a single small class with a different image would show the claim is false as stated.","supporting_citations":[],"review_version":1}