{"id":"e16b6e4b-995d-46e9-9554-7638c10cb794","arxiv_id":"1908.05333","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Irreducible holomorphic bundles on the twistor space of a simple hyperkähler manifold are generically fibrewise stable when their rank is 2 or 3, or when they are simple on some fibre.","lead":"A new theorem shows that on the twistor space of a hyperkähler manifold, irreducible vector bundles of rank 2 or 3, or bundles that are simple on at least one fibre, are automatically stable on most fibres of the twistor projection. This gives a partial converse to a known result of Kaledin and Verbitsky.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The final contradiction in the general-rank converse (Theorem 4.1, Step 4) is unjustified: the paper's appeal to the transition matrix B ignores the left vertical identification η(-D), so the proof as written is incomplete.","rationale":"I read the paper in good faith. The rank-2 and rank-3 arguments are clear and convincing. The general-rank proof is intricate, and the final step is the least secure part. The reader's stated weakest assumption about a smooth family of Gauduchon metrics is easily bypassed because the natural metrics G_I on Z_I are Kähler, hence Gauduchon, and vary smoothly with I; one can take G'_I = G_I. The actual soft spot is the unjustified final contradiction in Step 4: the paper's appeal to B being a matrix of functions ignores the left vertical arrow η(-D). A simple linear-algebra model shows the claimed contradiction does not follow without proving that η also has scalar-matrix form. Since that missing lemma is likely true, the verdict remains CONDITIONAL rather than REJECT, but the proof needs a substantive revision in this paragraph.","tokens_in":27652,"tokens_out":43001,"duration_ms":434853,"concrete_test":"Write out the final diagram of Theorem 4.1 Step 4 with both vertical arrows expressed in the decomposition E_1⊕...⊕E_d. Under the assumption π_*(E^*⊗E)=O_{CP^1}, prove that every morphism E_j(-D)→E_i is a pullback of a section of O_{CP^1}(-D_j+D+D_i), and that the left vertical isomorphism η(-D) has components in the same spaces. If both vertical arrows are scalar matrices, then γ = A η^{-1} is a scalar matrix, forcing each diagonal block F|_{U_i}→E|_{U_i} to be scalar multiplication, impossible for rank s<r. If this missing lemma cannot be established, the contradiction in Step 4 fails and the general-rank case of Theorem 4.1 is unproved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Reader's flagged concern about a smooth family of Gauduchon metrics in Proposition 3.5 is not actually load-bearing: each G_I is a Kähler metric on Z_I, hence Gauduchon, and G_I depends smoothly on I, so one can simply take G'_I = G_I and the functions C1, C2, C3 are continuous. The genuinely insecure point is the final contradiction in the general-rank case of Theorem 4.1 (Step 4, pp. 24-27). The paper claims that the upper description of γ|_U as a direct sum of inclusions F|_{U_i} → E|_{U_i} (whose projections to every summand are nowhere surjective, since rank s < r) contradicts the lower description A whose projection to the first t summands is an isomorphism, 'since the right-hand vertical arrow B is a matrix of functions'. This is insufficient: the commutative diagram also contains the left vertical isomorphism η(-D) between the two domain trivializations. If η(-D) is an arbitrary isomorphism, no contradiction follows; for example, with r=2, s=1, d=2, t=1, the map γ(x,y)=((x,0),(0,y)) has no surjective coordinate projection, but for a suitable scalar matrix B and a suitable domain identification, B^{-1}γη can have surjective first coordinate. The argument only works if η(-D) is also proved to be a matrix of scalar functions; this does follow from π_*(E^*⊗E)=O_{CP^1}, but the paper neither states nor proves it. Thus the main general-rank converse rests on an unproved identification.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies holomorphic vector bundles E on the twistor space Tw(M) of a compact simple hyperkähler manifold M, viewed as a family {E_I} over the fibres M_I of the twistor projection π:Tw(M)→CP^1. The main results are: (1) Theorem 3.2, which establishes Zariski openness of fibrewise stability and semi-stability in this non-product family by adapting Teleman's argument; (2) Theorem 4.1, which proves the forward implication that generic fibrewise stability implies irreducibility (following Kaledin–Verbitsky) and establishes a partial converse: an irreducible bundle is generically fibrewise stable for rank 2 and 3, and for general rank when at least one fibre restriction is simple. The general-rank converse is proved via a relative Quot/Douady construction, a multisection lemma for projective morphisms to a curve, and a careful analysis of the pushforward of a subsheaf along a branched cover.","tokens_in":27954,"tokens_out":22249,"duration_ms":252500,"significance":"If the general-rank converse is correct, the paper gives a satisfying structural statement: irreducible bundles on twistor spaces of simple hyperkähler manifolds are generically fibrewise stable under a mild simplicity assumption, and the higher-rank obstruction is not merely a failure of the exterior-monomial cone. The Zariski-openness theorem is a genuine extension of Teleman's result to the twistor-family setting, and the rank-2 and rank-3 proofs are transparent and appear sound. The paper is largely self-contained and carefully written. However, the proof of the general-rank converse in Step 4 of Theorem 4.1 contains a load-bearing gap concerning change of trivialization, so the headline general-rank claim is not yet established as written. The rank-2 and rank-3 results, together with the openness theorem, remain solid contributions.","major_comments":[{"comment":"The final contradiction is not justified as written. The upper description of γ|_U as a direct sum of inclusions F|_{U_i}→ϕ^*E|_{U_i} is made in the geometric trivialization by the sheets U_i, while the lower description A is obtained after composing with the left vertical isomorphism η(-D) and the right vertical isomorphism B. The property 'no projection onto a direct summand of E|_U^{⊕d} is surjective at any point' is not invariant under even scalar changes of the d copies of E. For example, with r=2, s=1, d=2, t=1, the map γ(x,y)=((x,0),(0,y)) has no surjective coordinate projection to either E-summand, but for a suitable 2×2 scalar transition matrix B and η=id, the conjugate map B^{-1}γη has surjective projection to the first E-summand. The paper only notes that B is an everywhere nonsingular matrix of holomorphic functions; that fact is true of every vector-bundle isomorphism and does not by itself yield a contradiction. To make Step 4 work one would need to prove that η(-D) and B respect the direct-sum decomposition in a stronger sense, or replace the coordinate-projection argument by a filtration or degree argument. Since this is the only step that proves the general-rank converse, the theorem as stated is not established.","section":"§4, Step 4 (pp. 27–28; diagram after Eq. (4.7))"}],"minor_comments":[{"comment":"The asserted smooth family of Gauduchon metrics G'_I can be taken to be G_I itself, because each G_I is Kähler and hence Gauduchon, and G_I depends smoothly on I. This would remove an unnecessary and lightly justified step.","section":"§3, Proposition 3.5"},{"comment":"The statement that if K_j→E_j is generically an isomorphism for every j then rk φ_*(F) = rk(E_1⊕...⊕E_d) is not immediate. It is true, but it needs a short proof: the kernels K_j are linearly independent subspaces of φ_*(F), so their ranks add.","section":"§4, Step 3 (p. 25)"},{"comment":"There are numerous typographical errors and OCR artifacts (e.g., 'holomoprhic', 'satisifes', and doubled symbols in the TeX source) that should be cleaned up before publication.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The rank-2 and rank-3 converse results and the Zariski-openness theorem appear sound and publishable. The general-rank converse is the headline new claim, and its proof currently rests on an unjustified change-of-trivialization argument in Step 4. I recommend major revision rather than rejection because the gap is local and the surrounding strategy is plausible; the authors should either repair Step 4 or explicitly restrict the general-rank statement to the cases that are proved."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a serious paper, but the main general-rank converse is not proved as written. The rank-2 and rank-3 cases are clean and convincing, and the extension of Teleman's Zariski-openness theorem to the twistor family is a genuine contribution. The general-rank simple-on-a-fibre proof has a load-bearing gap in the final step.\n\nWhat's new: the partial converse for ranks 2 and 3, and the observation that the Kaledin–Verbitsky irreducibility statement can be pushed to a converse in those cases. The proof of the forward direction is a clean adaptation. Theorem 3.2, Zariski openness for the non-product twistor family, is a real extension of Teleman's argument and deserves credit. The citation pattern is fine: [To2] supplies the counterexample to the full converse and is not doing any circular work.\n\nSoft spots: the reader worried about the 'smooth family of Gauduchon metrics' in Proposition 3.5. That worry is misplaced: G_I is Kähler, hence Gauduchon, and it varies smoothly in I, so G'_I = G_I solves the problem. The relative Picard space description is somewhat sketched; a referee should ask for details, but it looks plausible.\n\nThe real problem is in Step 4 of the general-rank converse (pp. 24–27). The paper compares the upper description of γ|_U (a direct sum of inclusions F|_{U_i} → E|_{U_i}, whose projections to every summand are nowhere surjective) with the lower description A (whose projection to the first t summands is an isomorphism), and says they contradict 'since the right-hand vertical arrow B is a matrix of functions'. That ignores the left vertical arrow η(-D). The actual relation is γ = B A η(-D)^{-1} up to the diagram's conventions. Surjectivity of coordinate projections is not stable under arbitrary invertible changes of basis, and the author never proves η(-D) is a matrix of scalar functions in the relevant trivializations. The stress-test note's example with r=2, s=1, d=2, t=1 shows the claimed contradiction is not a formal consequence. This is load-bearing: without it, the simple-on-a-fibre case is unproved.\n\nWho this is for: people working on stability on non-Kähler and twistor spaces. The rank-2/3 results and the openness theorem are worth having even if the general-rank case needs repair. I would send this to a serious referee, not desk-reject, but I'd flag Step 4 explicitly.","headline":"The rank-2 and rank-3 converse is solid and the Teleman-style openness theorem is a genuine extension, but the general-rank simple-on-a-fibre converse has a load-bearing gap in Step 4: the argument ignores the left vertical isomorphism η(-D), so the final contradiction does not follow.","tokens_in":28516,"tokens_out":5754,"would_cite":true,"duration_ms":61781,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14D20","14J60","53C26"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that an irreducible holomorphic bundle on the twistor space of a compact simple hyperkähler manifold is generically fibrewise stable when its rank is 2 or 3, and also for any rank when some fibre restriction is simple.","keywords":["hyperkähler manifold","twistor space","fibrewise stability","stable vector bundles","irreducible bundles","Zariski openness","relative Quot scheme","simple bundle"],"falsifier":"Find a compact simple hyperkähler manifold $M$ and an irreducible rank-3 vector bundle $E$ on $\\mathrm{Tw}(M)$ whose restriction $E_I$ is unstable for every $I \\in \\mathbb{CP}^1$. The theorem's rank-3 converse says such a bundle cannot exist, so a single explicit example of this kind would disprove the paper's main claim. Equivalently, for a candidate rank-3 irreducible bundle, compute the two Zariski-closed loci of destabilizing line subsheaves and destabilizing rank-2 subsheaves; the proof predicts they cannot together cover $\\mathbb{CP}^1$ unless a global subsheaf of $E$ exists.","tokens_in":27421,"feed_emoji":"🌀","tokens_out":20846,"duration_ms":192719,"temperature":0.7,"pith_summary":"A compact simple hyperkähler manifold has a twistor space: a complex manifold fibred over the projective line, whose fibres are the same underlying manifold equipped with its various induced complex structures. A holomorphic vector bundle on this twistor space can therefore be viewed as a family of bundles on those Kähler fibres. This paper proves that fibrewise stability is a Zariski-open condition: for a fixed bundle, the set of fibre parameters $I \\in \\mathbb{CP}^1$ for which the restriction $E_I$ is stable or semistable is Zariski open. The main theorem is a partial converse to the known fact that generic fibrewise stability implies irreducibility: if $E$ is irreducible and has rank 2 or 3, or if it is simple on at least one fibre, then it is generically fibrewise stable. The interest is that irreducible bundles cannot be built by extensions and carry no obvious stability data, so this gives a way to detect their stability from the individual Kähler fibres.","feed_headline":"Rank 2 and 3 bundles: irreducible implies generic fibrewise stability","feed_subtitle":"On hyperkähler twistor spaces, the stability of an irreducible bundle can be read fibre by fibre.","key_machinery":"The load-bearing object is the relative Quot space $\\mathrm{Quot}^1_{lf,\\mathbb{CP}^1}(E)$: the analytic space parametrizing, over each $I$, quotient sheaves $E_I \\to Q_I$ whose kernel is a line bundle. The proof embeds this space into the relative space of effective divisors of the projectivized bundle $\\mathbb{P}(E^*)$, where a compactness theorem for divisors gives properness of the level sets that impose degree bounds on the kernel. A second device is the cone of exterior monomials $C_s(E) \\subset \\Lambda^s E$; a rank-$s$ subsheaf of $E$ corresponds to a line subbundle of $\\Lambda^s E$ whose image lies in this cone, and fibrewise stability is exactly the non-existence of such line bundles of non-negative slope. For the converse, the pushforward of $L^* \\otimes \\Lambda^s E$ to $\\mathbb{CP}^1$ is the space over which the candidates for destabilizing maps live; a multisection, i.e. a finite cover of $\\mathbb{CP}^1$ together with a section of the pulled-back cone locus, constructs a subsheaf on the pullback twistor space, and the decomposition of the pushforward of the structure sheaf of that cover into line bundles turns the final comparison into a matrix of meromorphic functions.","core_discovery":"On a compact simple hyperkähler manifold, the paper establishes a fibrewise converse for irreducibility on the twistor space. A known theorem says that if the restrictions $E_I$ of a holomorphic bundle $E$ are stable for a Zariski-generic $I$, then $E$ itself has no proper subsheaves of lower rank, i.e. is irreducible. The paper shows the reverse holds in low ranks: an irreducible rank-2 or rank-3 bundle is generically fibrewise stable. It also holds for irreducible bundles of any rank provided at least one restriction $E_I$ is simple, meaning $\\mathrm{Hom}(E_I,E_I) = \\mathbb{C}$. The route goes through a Zariski-openness theorem for the family of restrictions: the locus of $I$ for which $E_I$ is stable or semistable is Zariski open in $\\mathbb{CP}^1$, so failure of generic fibrewise stability means instability on every fibre. This lets a single rank-$s$ destabilizing subsheaf direction, encoded as a line bundle $L$ with maps $L_I \\to \\Lambda^s E_I$ landing in the cone of exterior monomials, be propagated over the whole twistor line. In ranks 2 and 3 these maps glue directly into a global subsheaf of $E$ or $\\Lambda^2 E$, contradicting irreducibility; in higher rank the gluing is done over a finite cover, and the simplicity hypothesis makes the resulting endomorphisms reduce to a matrix of meromorphic functions, which gives the contradiction.","pith_inferences":["The proof leaves open whether the family of Gauduchon metrics it assumes to vary smoothly can actually be constructed; if it can, the same Zariski-openness argument would apply to any family of fibre metrics varying continuously over a compact base, not just the twistor projection.","The rank restriction in the converse points to the cone of exterior monomials being a proper subbundle when the subsheaf rank is strictly between 1 and $r-1$; a testable extension is to build irreducible bundles of higher rank from low-rank ones by direct image or pullback constructions and check whether they automatically satisfy the simplicity-on-a-fibre hypothesis.","A neighbouring problem made tractable by the openness theorem is how the canonical destabilizing filtration of $E_I$ varies with $I$; if its type is Zariski lower-semicontinuous, the general-rank converse might hold without the rank or simplicity assumptions.","The rank-2 and rank-3 results suggest that, in moduli problems over non-algebraic twistor spaces, fibrewise stability could serve as an open chart condition defining the moduli of irreducible bundles, much as stability over a fixed Kähler class defines moduli in the algebraic setting."],"forward_implications":["The theorem implies that for rank-2 and rank-3 bundles on the twistor space of a compact simple hyperkähler manifold, being irreducible and being generically fibrewise stable are the same condition.","Fibrewise stability and semistability are Zariski open in the twistor parameter, so stability of one generic fibre spreads to a Zariski-open neighbourhood, while instability on a Zariski-dense set forces instability on every fibre.","For bundles of general rank, simplicity of a single fibre restriction is enough to force generic fibrewise stability, so the converse to the known irreducibility implication holds in a much wider class than low rank.","Because twistor spaces are never projective, this Zariski-openness statement goes beyond the classical algebraic families of bundles and applies to families in which the complex structure of the fibre varies.","The results give a practical way to test irreducibility of low-rank bundles on twistor spaces: check whether the restrictions are stable on the generic fibre rather than searching for all subsheaves over the whole twistor space."],"supporting_citations":[{"why":"It establishes the forward direction the paper builds on: generic fibrewise stability implies irreducibility of the bundle on the twistor space.","marker":"[KV]"},{"why":"It supplies the proof of Zariski openness of stability in holomorphic families that Section 3 adapts to the twistor projection.","marker":"[Te]"},{"why":"It provides the Hermitian-Einstein correspondence and the degree identities used to relate fibre degrees to volumes.","marker":"[LT]"},{"why":"It gives the correspondence between rank-s subsheaves and line subbundles of exterior powers lying in the cone of exterior monomials, plus the semicontinuity theorem.","marker":"[OSS]"},{"why":"It supplies an example of a stable bundle on a twistor space that is nowhere fibrewise stable, showing the unqualified converse is false.","marker":"[To2]"},{"why":"It builds the relative Quot space for proper morphisms, the parameter space for fibrewise quotients with invertible kernel.","marker":"[P]"},{"why":"It gives the compactness theorem for effective divisors of bounded volume used to prove properness of degree level sets.","marker":"[Bi]"},{"why":"It supplies the analytic coherence and base-change results used to identify fibres of pushforwards in the general-rank gluing step.","marker":"[GR]"},{"why":"It provides the pushforward identity for a finite cover that decomposes the pulled-back bundle into direct sums of twisted copies of E.","marker":"[Ba]"}],"fun_headline_variants":["Irreducibility on twistor space: stability on generic fibre","Partial converse: irreducibility yields fibrewise stability in low rank","Twistor stability: low-rank irreducibility implies generic fibrewise stability","Hyperkähler twistor bundles: converse for stability in ranks 2-3","Irreducible rank 2-3 twistor bundles: fibrewise stability generic"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on a smooth choice, as the parameter $I$ varies over the twistor line, of Gauduchon metrics (a standard normalization of Hermitian metrics) on the associated projective bundles; if that smooth family does not exist, the continuity of the constants connecting fibre degrees and volumes fails and the properness argument that powers the openness and converse theorems collapses.","fun_headline_variants_meta":{"raw":{"variants":["Irreducibility on twistor space: stability on generic fibre","Partial converse: irreducibility yields fibrewise stability in low rank","Twistor stability: low-rank irreducibility implies generic fibrewise stability","Hyperkähler twistor bundles: converse for stability in ranks 2-3","Irreducible rank 2-3 twistor bundles: fibrewise stability generic"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001076,"raw_usage":{"total_tokens":4576,"prompt_tokens":1089,"completion_tokens":3487,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":705,"completion_tokens_details":{"reasoning_tokens":3385}},"tokens_in":705,"tokens_out":3487,"duration_ms":24539,"temperature":1.0,"reasoning_tokens":3385,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:18:37.197783+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a compact simple hyperkähler manifold $M$ and an irreducible rank-3 vector bundle $E$ on $\\mathrm{Tw}(M)$ whose restriction $E_I$ is unstable for every $I \\in \\mathbb{CP}^1$. The theorem's rank-3 converse says such a bundle cannot exist, so a single explicit example of this kind would disprove the paper's main claim. Equivalently, for a candidate rank-3 irreducible bundle, compute the two Zariski-closed loci of destabilizing line subsheaves and destabilizing rank-2 subsheaves; the proof predicts they cannot together cover $\\mathbb{CP}^1$ unless a global subsheaf of $E$ exists.","supporting_citations":[],"review_version":1}