{"id":"37c66fcf-b20e-4c67-ab2d-6fc9da419bbe","arxiv_id":"2507.22703","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a primitive polarization h on a K3 surface S, the Tate-Shafarevich group X(S,h) parametrizes exactly the Pic^0(C_η)-torsors admitting a good hyperkähler model.","lead":"This paper proves that the Tate-Shafarevich group of a polarized K3 surface, defined cohomologically in earlier work, is in bijection with the set of Jacobian torsors over the generic curve in the polarization that admit a good hyperkähler compactification. The result is the higher-dimensional analogue of the classical description of Tate-Shafarevich groups of elliptic K3 surfaces and completes a program begun in the authors' previous paper.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Surjectivity hinges on Kim's preprint [19] and sketched diagram (5.7); a missing detailed proof in this note leaves a genuine load-bearing gap.","rationale":"The reader identified the reliance on Kim [19] and the sketched diagram (5.7) as the weakest assumptions. My reading of the full text supports this: §5.2 and §5.4 import [19, Thm. 1.2, 1.3, 1.6, Prop. 6.32] wholesale, and the isomorphism X(S,h) ≅ H^1(|h|, Pic^{r0s}) is asserted with only a reference to [17, §5.2]. These are genuinely load-bearing because they carry the surjectivity half of Theorem 1.1. I did not find a more fundamental flaw: the injectivity proof in §4 is self-contained modulo standard moduli/Hodge facts and the paper explicitly flags its own limitations in §6.2 (non-primitive h). The concern is not that the authors are wrong, but that the proof as written is not verifiable at its critical junction without access to the details of Kim’s preprint and the unspecified diagram (5.7). Hence the correct verdict is CONDITIONAL: the theorem is plausible and the structure is sound, but the paper should be required to expand the proof of (5.7) and to state precisely which statements from [19] are used in which form. I also note the paper itself points to Remark 5.4 as an alternative route, but explicitly labels it a sketch, so it cannot substitute for the missing details. All fields are therefore consistent with the reader’s verdict; no change is needed.","tokens_in":15988,"tokens_out":2164,"duration_ms":21076,"concrete_test":"Independently verify the isomorphism in diagram (5.7) by spelling out the étale-sheaf sequence 0 → Pic^{r0s} → Pic → Z → 0 on |h| and comparing it with the defining exact sequence 0 → Z/mZ → X(S,h) → Br(S) → 0 of [17, §4]; if the induced edge maps do not glue to the claimed identification, the surjectivity step fails. As a second check, reread Kim [19, Prop. 6.32] and confirm that its hypotheses (including the good-fibre condition and non-separated Néron model) are satisfied verbatim for the moduli space M = M(v_0) of stable sheaves, in particular that the étale restriction H^1(|h|, P) → H^1(η, P_η) is injective under h primitive but without assuming all curves in |h| are integral.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Theorem 1.1’s central bijection has two directionally different halves: injectivity (§4) is argued in the note, but surjectivity (§5.4, Prop. 5.9) depends on three imported ingredients that are not proved here. First, the existence and surjectivity of the Néron model P of Aut^0(X_0/B_0) and the identification P ≅ M^n (after (5.5)), cited to [19, Thm. 1.2, 1.3, Prop. 6.32]; if any of Kim’s statements fails in the present generality (non-separated Néron models, good-fibre condition, reduction over the discriminant locus), the map H^1(|h|, M^n) → H^1(η, Pic^0(C_η)) need not lift to an X(S,h)-class. Second, the asserted isomorphism in diagram (5.7), X(S,h) ≅ H^1(|h|, Pic^{r0s}), is only referenced to [17, §5.2] and sketched; it is exactly the group-theoretic bridge that identifies cohomological X(S,h) with the torsor class rX^n] before applying Cor. 5.8. Third, Cor. 5.8 itself uses that the Néron model M^n is a compactification of Pic^0(C/|h|^sm) with Pic^{r0s} → M^n surjective, which is plausible but rests on Kim’s non-separated Néron-model framework. These are not internal inconsistencies, but the load-bearing steps are imported from a highly technical recent preprint and from [17, §5.2] without enough detail for an independent check. The paper’s own Remark 5.4 indicates an alternative route through the analytic twistor family, but it is explicitly only a sketch. Thus the conditional verdict is justified: a detailed proof of (5.7) and a verification of each application of [19] are needed before the bijection can be certified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves Theorem 1.1: for a complex K3 surface (S,h) with h ample and primitive, there is a natural bijection between the Tate–Shafarevich group X(S,h), defined cohomologically via the special Brauer group, and the set of isomorphism classes of Pic^0(C_η)-torsors admitting a good hyperkähler model, i.e. a Lagrangian fibration X → |h| whose generic fibre is the given torsor and whose every closed fibre contains a smooth point. The injectivity direction is proved in Section 4 by comparing moduli spaces of twisted sheaves and using Hodge isometries and lattice arguments. The surjectivity direction is proved in Section 5 using Kim's recent theory of Néron models of Lagrangian fibrations and a commutative diagram (5.7) identifying X(S,h) with H^1(|h|, Pic^{r0s}). The paper also includes Lemma 5.3, which constructs good hyperkähler models for all classes in X(S,h), and Corollary 5.8, which supplies the key surjectivity of a cohomology map.","tokens_in":16398,"tokens_out":9630,"duration_ms":109314,"significance":"If the proof is correct, the paper gives a clean geometric interpretation of the Tate–Shafarevich group of a polarised K3 surface: it is not merely a cohomological invariant but precisely the group of Pic^0(C_η)-torsors admitting a geometrically well-behaved compactification. This is a natural analogue of the classical description for elliptic K3 surfaces and is likely to be useful for further work on Lagrangian fibrations, twisted sheaves, and Brauer groups. The injectivity part is largely self-contained and relies on standard moduli-space and Hodge-theoretic tools. The surjectivity part is thought-provoking but imports several load-bearing results from the recent preprint [19] and from the authors' earlier work [17]; those imports are identified explicitly, so the argument is transparent about its dependencies.","major_comments":[{"comment":"The isomorphism X(S,h) ≅ H^1(|h|, Pic^{r0s}) is the key bridge between the cohomological definition of X(S,h) and the geometric torsor class rX^n s used in the surjectivity proof. The text only says that 'imitating the discussion in [17, §5.2]' completes the diagram. Since [17, §5.2] concerns elliptic K3 fibrations, the vertical maps and the identification of the kernels Z/mZ in the two exact sequences need to be written out for the present linear system |h|, or a precise statement in [17] that covers this case must be quoted. Without a proof of (5.7), Proposition 5.9 does not produce a class α ∈ X(S,h) from a good hyperkähler model, so this is a load-bearing gap.","section":"§5.4, diagram (5.7)"},{"comment":"In the proof of Lemma 5.3, after extending the generic lift to a trait ψ : Spec(R) → M_α, the text asserts that the closed point ψ(0) is 'necessarily a smooth point' of the fibre M_{α,t}. This is not evident: a section of a proper morphism can specialise to a singular point of the special fibre. Since condition (ii) of Definition 5.1 is exactly this smooth-point condition and is also the hypothesis needed to apply Kim's Néron-model results, this step needs a proof or an explicit reference.","section":"§5.1, Lemma 5.3"},{"comment":"The proof of the surjectivity of (5.3) is compressed. From Remark 5.7 with Z = M^n one obtains a morphism φ : M^n → Pic^{r0s}; from the defining property of the Néron model M^n, applied to Z = Pic^{r0s}, one obtains the morphism (5.2) Pic^{r0s} → M^n. The text should state explicitly that these are the two morphisms being composed and that their composition is the identity by uniqueness of the Néron extension. As written, 'to the inclusion φ0 : Z0 = Pic^0(C/|h|^sm) ↪ Pic^{r0s}(C/|h|) and its compactification Z = M^n' conflates the two different extension properties and makes the proof hard to check.","section":"§5.3, Corollary 5.8"},{"comment":"Proposition 5.9 imports several results from the preprint [19]: the existence and surjectivity of the Néron model P, the identification P ≅ M^n, and Prop. 6.32 used for injectivity of H^1(|h|,P) → H^1(η,P_η). The manuscript does not verify for each application that the hypotheses of the quoted theorems are satisfied by the specific moduli spaces M_α → |h| beyond the 'good fibre' condition. Since the surjectivity half of Theorem 1.1 collapses if any of these imported statements does not apply, the authors should state explicitly which theorem from [19] is used at each step and confirm that its hypotheses are met, especially over the discriminant locus.","section":"§5.2 and §5.4, use of [19]"}],"minor_comments":[{"comment":"There is a typo in 'for the the surjectivity in (ii)', which should read 'for the surjectivity in (ii)'.","section":"§1"},{"comment":"In Definition 5.1(ii), the equivalence between 'contains a smooth point' and 'has at least one generically reduced irreducible component' is stated informally via [5, Rem. 1.3]; adding the precise scheme-theoretic formulation would avoid ambiguity, especially because the paper later discusses non-reduced fibres.","section":"§5.1, Definition 5.1"},{"comment":"The notation Pic^{r0s}(C/|h|) is introduced by analogy with [13], but the superscript 'r0s' is not explained; a one-sentence gloss (total degree zero, as opposed to component-wise degree zero) would improve readability.","section":"§5.3"},{"comment":"In the paragraph after Corollary 5.8, the statement 'as an aside, we observe that the surjection (5.8) is in fact always an isomorphism X(S,h) ≅ H^1(|h|,P)' is interesting but appears to use Corollary 4.4 and [19, Prop. 6.32]; it would be helpful to separate this observation from the proof of Proposition 5.9, since it is not needed for Theorem 1.1.","section":"§5.4"}],"recommendation":"major_revision","confidential_remarks":"The paper is concise and the main theorem is attractive, but the proof of Proposition 5.9 leans heavily on the authors' own [17] and on Kim's unpublished preprint [19]. Before acceptance, I would want the editor to confirm that [19] has been vetted and that the isomorphism (5.7) is either proved here or stated as a theorem in [17] with the same hypotheses as used in this note. The self-citation pattern is not unusual for a companion note, but the load-bearing status of these citations should be made explicit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on arXiv:2507.22703. The paper completes the program from [17]: it proves that the Tate–Shafarevich group of a polarized K3 surface (S,h) with h primitive is in bijection with Pic^0(C_η)-torsors that admit a 'good' hyperkähler model. That is the natural analogue of the elliptic-surface result, and it's the first geometric interpretation of this group. The result is new and, as far as I can assess, plausible; the proof is honest about what is imported.\n\nThe injectivity direction was essentially available in [17], and the genuinely new step is surjectivity, which uses Kim's recent Néron-model results [19] to show that every torsor with a good model is an α-twisted Picard scheme. I appreciate the careful discussion of where primitivity of the polarization enters, and the authors do not hide the limitations: Lemma 5.3 explicitly notes that they don't know whether Pic^0_α(C_{|h|}) is contained in the moduli space M_α, and they then work around it via simple twisted sheaves and Néron models.\n\nThe soft spots are real but local. Diagram (5.7) claims an isomorphism X(S,h) ≅ H^1(|h|, Pic^{r0s}) that is load-bearing: it is the bridge between the cohomological definition and the torsor class. In this note it is only sketched with a reference to [17, §5.2]. A referee should ask for a proof or at least a precise statement from [17] that covers exactly this case. The surjectivity also depends heavily on Kim's preprint [19]—non-separated Néron models, the 'good fibre' hypothesis, and Prop. 6.32. That is a normal research dependency, but it means the paper can't be fully certified without checking Kim's results in the current generality. The alternative route in Remark 5.4 is only a sketch, so it doesn't stand as a backup.\n\nThe citation pattern is fine: self-citation to [17] is appropriate, and Kim's work is exactly what's needed. No circularity: the two sides of the bijection are defined independently.\n\nWho is this for? Anyone working on hyperkähler fibrations, moduli of sheaves, or Brauer groups of K3 surfaces. It's a solid research note that deserves a serious referee; I'd send it out, but I'd ask the authors to expand the proof of (5.7) and to verify each application of [19] before accepting.","headline":"New geometric characterization of the Tate–Shafarevich group of a polarized K3 surface, plausible and useful, but the proof leans on two sketched or imported steps that need referee scrutiny.","tokens_in":16976,"tokens_out":3732,"would_cite":true,"duration_ms":40453,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14J28","14J42","14D10","14F22"],"pacs":[],"model":"deepseek-v4-flash","headline":"A bijection identifies the Tate–Šafarevič group of a polarised K3 surface with the 'good' hyperkähler compactifiable torsors of its generic curve's Jacobian.","keywords":["Tate–Šafarevič group","polarised K3 surfaces","hyperkähler compactifications","Lagrangian fibrations","Néron models","twisted Picard schemes","Brauer group","Weil–Châtelet group"],"falsifier":"For a concrete polarised K3 surface, e.g. one with Picard number one, compute the Néron model $P$ of a good Lagrangian fibration over $|h|$ and compare the image of $H^1(|h|,P)$ under restriction to the generic point with the image of $X(S,h)$ in $H^1(\\eta,\\operatorname{Pic}^0(C_\\eta))$; a torsor class in the difference would disprove surjectivity. Finding two distinct classes $\\alpha_1,\\alpha_2\\in X(S,h)$ with isomorphic $\\operatorname{Pic}^0_{\\alpha_i}(C_\\eta)$ would disprove injectivity.","tokens_in":15755,"feed_emoji":"🌀","tokens_out":10620,"duration_ms":106417,"temperature":0.7,"pith_summary":"This note proves that for a complex K3 surface with an ample primitive polarisation $h$, the Tate–Šafarevič group $X(S,h)$, introduced cohomologically via the special Brauer group, parametrises exactly the $\\operatorname{Pic}^0(C_\\eta)$-torsors that admit a 'good' hyperkähler model: a Lagrangian fibration over $|h|$ whose generic fibre is the torsor and every closed fibre has a smooth point. This is the analogue for polarised K3 surfaces of the classical fact that for an elliptic K3 surface, the Tate–Šafarevič group is the subgroup of the Weil–Châtelet group consisting of twists that can be compactified to a K3 surface. The point of the theorem is to show that a cohomologically defined group has a purely geometric content: an element of $X(S,h)$ is exactly a torsor that can be compactified by a good Lagrangian fibration.","feed_headline":"Good hyperkähler torsors equal the K3 Tate–Šafarevič group","feed_subtitle":"For primitive polarisations, the cohomological group is the same as the set of compactifiable torsors.","key_machinery":"The load-bearing object is the special Brauer group $\\operatorname{SBr}(S,h)\\subset H^2(S,\\mathbb{Q}/\\mathbb{Z})$ and the $\\alpha$-twisted Picard schemes $\\operatorname{Pic}^0_\\alpha(C_\\eta)$ that it produces. These torsors are compactified by moduli spaces $M_\\alpha$ of stable $\\alpha$-twisted sheaves with Mukai vector $(0,h,-h^2/2)$, which are Lagrangian fibrations over $|h|$. On the compactification side the key tool is the Néron model of the relative Jacobian: the paper imports from [19] the theorem that a good Lagrangian fibration admits a Néron model $P\\to|h|$, proves $X(S,h)\\cong H^1(|h|,\\operatorname{Pic}^{r0s})$ via diagram (5.7), and shows that this group surjects onto $H^1(|h|,P)$; restricting to the generic fibre recovers the torsor class. Here $\\operatorname{Pic}^{r0s}$ denotes the étale sheaf of sections of the non-separated relative Picard scheme of total degree zero.","core_discovery":"Theorem 1.1 states that for a complex K3 surface $(S,h)$ with $h$ ample and primitive, there is a natural bijection between the Tate–Šafarevič group $X(S,h)$ and the set of isomorphism classes of $\\operatorname{Pic}^0(C_\\eta)$-torsors admitting a good hyperkähler model. Injectivity is proved by showing that two special Brauer classes producing isomorphic torsors must coincide in $X(S,h)$, using hyperkähler moduli spaces of stable twisted sheaves and a comparison of their transcendental Hodge structures. Surjectivity is proved by starting with any good hyperkähler model, taking its Néron model, and using the identification of $X(S,h)$ with $H^1(|h|,\\operatorname{Pic}^{r0s})$ to show that the generic fibre of the model is a torsor coming from a special Brauer class.","pith_inferences":["A test the authors leave implicit: naturality of the bijection should make $X(S,h)$ into a local system over the moduli space of polarised K3 surfaces, so monodromy would act on the set of good hyperkähler models, and this action might distinguish birational models.","The paper's suspicion that every hyperkähler compactification is good can be checked by searching for a $\\operatorname{Pic}^0(C_\\eta)$-torsor with a Lagrangian fibration over $|h|$ whose closed fibre is everywhere non-reduced; such a torsor would fall outside the current theorem.","The description $X(S,h)\\cong H^1(|h|,\\operatorname{Pic}^{r0s})$ opens a purely sheaf-theoretic route to computing the group without constructing moduli spaces; comparing this cohomology with the Néron-model cohomology in examples would test how much of the 'good' condition is needed."],"forward_implications":["Every class in $X(S,h)$ is witnessed by a good hyperkähler model, so the group is not merely cohomological: its elements are exactly the geometrically compactifiable torsors.","Conversely, every $\\operatorname{Pic}^0(C_\\eta)$-torsor with a good hyperkähler model is of the form $\\operatorname{Pic}^0_\\alpha(C_\\eta)$ for a unique $\\alpha\\in X(S,h)$.","The short exact sequence $0\\to\\mathbb{Z}/m\\mathbb{Z}\\to X(S,h)\\to Br(S)\\to 0$ remains available, so the group is an extension of the Brauer group by the divisibility of $h$; the theorem attaches geometric meaning to that extension.","When all curves in $|h|$ are integral, the Néron model of the Jacobian is $\\operatorname{Pic}^0(C/|h|)$, so $X(S,h)$ can be computed as $H^1(|h|,\\operatorname{Pic}^0(C/|h|))$.","For a non-primitive polarisation the theorem does not apply; the paper suggests that injectivity may persist but a geometric description of the image is still open."],"supporting_citations":[{"why":"Defines $X(S,h)$ and the $\\alpha$-twisted Picard schemes and supplies the kernel computation behind injectivity.","marker":"[17]"},{"why":"Provides the Néron-model existence, surjectivity, and good-fibre results on which Proposition 5.9 depends.","marker":"[19]"},{"why":"Establishes the classical isomorphism for elliptic K3 surfaces that Theorem 1.1 generalises.","marker":"[25]"},{"why":"Supplies the Brauer-group and spectral-sequence results used for the classical case and in diagram (5.7).","marker":"[11]"},{"why":"Describes the weight-two Hodge structure of moduli spaces of sheaves, used to compare transcendental lattices in the injectivity proof.","marker":"[22]"},{"why":"Gives the twisted analogue of the Hodge isometry for moduli spaces of twisted sheaves used in the same argument.","marker":"[26]"},{"why":"Provides the Néron-model theory and properties of $\\operatorname{Pic}^0(C/|h|)$ used in Proposition 5.6.","marker":"[4]"},{"why":"Constructs Néron models of Jacobians over higher-dimensional bases, the foundation for the Néron-model theorems used here.","marker":"[14]"},{"why":"Establishes the link between $X(S,h)$ and the Brauer group of the moduli space, referenced for diagram (5.7).","marker":"[21]"}],"fun_headline_variants":["K3 Tate–Šafarevič group matches hyperkähler torsors","Hyperkähler compactifiable torsors equal the cohomology group","Bijection for primitive polarisations: torsors and Tate–Šafarevič","All good hyperkähler torsors arise from Tate–Šafarevič","Polarised K3: torsors biject with compactifiable twists"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the validity of the Néron-model theorems for Lagrangian fibrations satisfying the 'good fibre' condition, imported from reference [19], together with the identification $X(S,h)\\cong H^1(|h|,\\operatorname{Pic}^{r0s})$ from reference [17], which is sketched rather than fully proved in this note.","fun_headline_variants_meta":{"raw":{"variants":["K3 Tate–Šafarevič group matches hyperkähler torsors","Hyperkähler compactifiable torsors equal the cohomology group","Bijection for primitive polarisations: torsors and Tate–Šafarevič","All good hyperkähler torsors arise from Tate–Šafarevič","Polarised K3: torsors biject with compactifiable twists"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000235,"raw_usage":{"total_tokens":1462,"prompt_tokens":871,"completion_tokens":591,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":487,"completion_tokens_details":{"reasoning_tokens":486}},"tokens_in":487,"tokens_out":591,"duration_ms":6598,"temperature":1.0,"reasoning_tokens":486,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T11:23:28.940340+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a concrete polarised K3 surface, e.g. one with Picard number one, compute the Néron model $P$ of a good Lagrangian fibration over $|h|$ and compare the image of $H^1(|h|,P)$ under restriction to the generic point with the image of $X(S,h)$ in $H^1(\\eta,\\operatorname{Pic}^0(C_\\eta))$; a torsor class in the difference would disprove surjectivity. Finding two distinct classes $\\alpha_1,\\alpha_2\\in X(S,h)$ with isomorphic $\\operatorname{Pic}^0_{\\alpha_i}(C_\\eta)$ would disprove injectivity.","supporting_citations":[{"cited_title":"Huybrechts, M","cited_arxiv_id":null,"evidence_quote":"Defines $X(S,h)$ and the $\\alpha$-twisted Picard schemes and supplies the kernel computation behind injectivity."},{"cited_title":"Huybrechts, P","cited_arxiv_id":null,"evidence_quote":"Provides the Néron-model existence, surjectivity, and good-fibre results on which Proposition 5.9 depends."},{"cited_title":"https://stacks.math.columbia.edu","cited_arxiv_id":null,"evidence_quote":"Establishes the classical isomorphism for elliptic K3 surfaces that Theorem 1.1 generalises."},{"cited_title":"GrothendieckLe groupe de Brauer II","cited_arxiv_id":null,"evidence_quote":"Supplies the Brauer-group and spectral-sequence results used for the classical case and in diagram (5.7)."},{"cited_title":"TateOn the conjectures of Birch and Swinnerton-Dyer and a geometric analog.Séminaire Bourbaki, Vol","cited_arxiv_id":null,"evidence_quote":"Gives the twisted analogue of the Hodge isometry for moduli spaces of twisted sheaves used in the same argument."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Néron-model theory and properties of $\\operatorname{Pic}^0(C/|h|)$ used in Proposition 5.6."},{"cited_title":"Holmes Néron models of jacobians over base schemes of dimension greater than 1.Crelle J","cited_arxiv_id":null,"evidence_quote":"Constructs Néron models of Jacobians over higher-dimensional bases, the foundation for the Néron-model theorems used here."},{"cited_title":"volume 71 of Springer Proc","cited_arxiv_id":null,"evidence_quote":"Establishes the link between $X(S,h)$ and the Brauer group of the moduli space, referenced for diagram (5.7)."}],"review_version":1}