{"id":"ef5845d4-b9f5-4262-9621-db79c1654492","arxiv_id":"2411.10274","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A torsor under an abelian variety is twisted derived equivalent to another smooth variety exactly when that variety is a moduli space of simple semi-homogeneous sheaves on the torsor, with the universal obstruction as the twist.","lead":"This paper proves that two torsors under abelian varieties are twisted derived equivalent exactly when one is a moduli space of simple semi-homogeneous sheaves on the other, and it proves the converse classification. It extends the Orlov-Polishchuk criterion to torsors, positive characteristic, and Brauer-twisted categories, but two advertised application statements appear to confuse an abelian variety with its dual.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Converse for arbitrary fields hinges on unverified positive-characteristic extension of [dJO22, Thm 1.1/Rmk 1.3]; if that implication fails, Theorem 2.4(3) and Theorem C's only-if direction collapse.","rationale":"The paper's main contribution is the classification of twisted derived equivalences between torsors, and the only-if direction depends on the asserted positive-characteristic validity of a statement whose published proof is characteristic-0 only. The reader's weakest_assumption identifies exactly this dependency, and I agree. The internal inconsistency in Proposition 4.6 (Theorem D) about needing the dual of a Lagrangian subvariety is real and should be fixed, but it concerns the peripheral Theorem D/E, not the central classification; hence it does not change the conditional verdict. A conditional acceptance requiring the author to either supply a proof of the dJO22 implication in positive characteristic or restrict Theorem C accordingly seems right.","tokens_in":27246,"tokens_out":19047,"duration_ms":167858,"concrete_test":"Read [dJO22, Remark 1.3 and equation (6.1)] and determine whether they prove, over an algebraically closed field of arbitrary characteristic, that a simple object F with dim S(F) ≥ dim X is a simple semi-homogeneous sheaf up to shift. If the proof requires characteristic 0, ask the authors whether the implication is known in positive characteristic; alternatively, search for a simple object on an abelian variety in characteristic p with dim S(F) ≥ g that is not a shift of a semi-homogeneous sheaf. A counterexample would invalidate Theorem 2.4(3) and the converse in Corollary 2.8; a proof would remove the concern.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central converse (Corollary 2.8) is powered by Theorem 2.4(3). In the proof, after reducing to an algebraically closed field and showing that the simple object E_y satisfies dim S(E_y) ≥ dim Y = dim X, the argument invokes [dJO22, Theorem 1.1, Remark 1.3, (6.1)] to conclude that E_y is, up to shift, a simple semi-homogeneous sheaf. The introduction explicitly states that [dJO22, Theorem 1.1] is proved only in characteristic 0 and that it is not clear whether that hypothesis is necessary. The workaround for case (3) does not remove this dependency: it only establishes the inequality dim S(E_y) ≥ dim X, after which the same dJO22 implication is applied. Consequently, the 'only if' direction of Corollary 2.8 has no support over arbitrary fields if the positive-characteristic version of that implication fails. This is an external, unreproduced dependency rather than an internal contradiction, but it is directly load-bearing for the headline classification.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves that for a torsor X under an abelian variety, the moduli space SSH^ξ_{X/k} of simple semi-homogeneous sheaves with fixed numerical Chern character carries a universal obstruction μ, and the universal (twisted) sheaf induces a twisted derived equivalence D^b(SSH^ξ_{X/k}, μ^{-1}) → D^b(X) (Theorem A / Theorem 2.1). It then proves a partial converse: under extra hypotheses—characteristic 0, or μ ∈ Br_1(Y), or Y itself a torsor under an abelian variety—any twisted derived equivalence D^b(Y, μ^{-1}) → D^b(X) forces Y to be isomorphic to such a moduli space with μ the universal obstruction (Theorem B / Theorem 2.4). Combining these gives an iff criterion for twisted derived equivalences between torsors under abelian varieties (Theorem C / Corollary 2.8). The paper also establishes a twisted analogue of the Orlov–Polishchuk criterion over algebraically closed fields (Theorem D / Proposition 4.6) and shows that every isometric isomorphism between A× and B×B̂ can be realized by a twisted derived equivalence after passing to suitable torsors (Theorem E / Proposition 4.7).","tokens_in":27302,"tokens_out":2780,"duration_ms":27887,"significance":"If the central theorems are correct, this is a valuable contribution: it generalizes Mukai's and Gulbrandsen's Fourier–Mukai equivalences from abelian varieties to torsors, provides a converse to the moduli-space construction, and extends parts of the classical derived-equivalence criterion for abelian varieties to coefficients in the Brauer group and to arbitrary base fields. The paper is honest about its external dependencies, especially the characteristic-0 hypothesis in [dJO22, Theorem 1.1], and its proofs are largely self-contained modulo the cited results. The construction of SSH^ξ_{X/k} as a torsor under an abelian variety and the Lagrangian-subvariety analysis in Section 3 are independently interesting.","major_comments":[{"comment":"The 'only if' direction of the headline classification for arbitrary fields is load-bearing on an unproved positive-characteristic extension of [dJO22, Theorem 1.1]. In the proof of Theorem 2.4, after establishing dim S(E_y) ≥ dim X, the argument invokes [dJO22, Theorem 1.1, Remark 1.3, (6.1)] to conclude that E_y is, up to shift, a simple semi-homogeneous sheaf. The introduction explicitly states that [dJO22, Theorem 1.1] is proved only in characteristic 0 and that it is not known whether that hypothesis is necessary. The workaround for case (3) does not remove this dependency: it only produces the inequality dim S(E_y) ≥ dim X, after which the same [dJO22] implication is applied. Consequently, if the characteristic-0 hypothesis in [dJO22, Theorem 1.1] turns out to be necessary, Theorem 2.4(3) and Corollary 2.8 lose support over arbitrary fields. The author should either prove the needed positive-characteristic statement, restrict the statement of Theorem 2.4(3) and Corollary 2.8 to characteristic 0 (or to a characteristic in which the [dJO22] result is known), or clearly state this as an open condition in the theorems.","section":"Theorem 2.4(3) and Corollary 2.8"},{"comment":"The verification of condition (1) of Theorem A.40 relies on [HP24, Proposition 6.3] together with an unstated remark from 'Section 6' of [HP24] that is said to imply that all vertical maps in the displayed diagram are isomorphisms. This is a crucial step: it is used to conclude that the map Ext^1(k(y), k(y)) → Ext^1(E_y, E_y) is an isomorphism. The remark is not reproduced, and the hypotheses that make the remark applicable to the present setting (with X = SSH^ξ_{X/k,D} and Y = X_D) are not verified in the text. The author should state the precise remark and check its hypotheses, or replace this step with a direct deformation-theoretic argument. Without this, the full faithfulness part of Theorem 2.1 is not fully established as written.","section":"Theorem 2.1, proof"},{"comment":"Lemma 2.2 asserts that [Or02, Lemma 4.8] applies in arbitrary characteristic because 'its proof does not use the assumptions on the characteristic.' No details or page/equation reference are given. Since this lemma is used to verify condition (2) of Theorem A.40 (the orthogonality condition for non-isomorphic simple semi-homogeneous sheaves), it is load-bearing for Theorem 2.1. The author should supply a proof of the characteristic independence or cite a precise statement in the literature that covers arbitrary fields; the current assertion is not sufficient for a formal proof.","section":"Lemma 2.2"}],"minor_comments":[{"comment":"The sentence 'Let A and B be abelian varieties over field k' is missing the indefinite article; it should read 'over a field k'.","section":"Introduction, Theorem C"},{"comment":"There are several typographical errors: 'digonalizable' in Appendix A, 'vareities' in Theorem E, 'Propoisiton' in the proof of Proposition 2.6, and 'isomorphsim' in the proof of Theorem 2.1. These should be corrected.","section":"Throughout"},{"comment":"The proof of Lemma 1.13 relies on generic flatness and unions over translates, but the statement assumes the orbit map is surjective on k-points; the deduction that the induced map G/stab(x) → X_red is an isomorphism should explicitly justify why the faithfully flat morphism obtained factors through the desired isomorphism, rather than merely asserting it. This is a clarity issue, not a substantive gap.","section":"Section 1, Lemma 1.13"},{"comment":"The phrase 'the map on inertia is the identity map Gm → Gm' is terse; the reason is that the action of the inertia stack on the universal twisted sheaf is by scalar multiplication, but this deserves a sentence of explanation given that the map on inertia is a key step in proving that the induced map of stacks is an isomorphism.","section":"Proposition 2.6, proof"}],"recommendation":"major_revision","confidential_remarks":"The manuscript makes a serious and mostly convincing contribution, but the scope of Theorem B and Corollary 2.8 in arbitrary characteristic is genuinely uncertain because it hinges on an external theorem whose characteristic-0 hypothesis is not known to be removable. This is not an internal inconsistency, but it is a load-bearing dependency that the author explicitly acknowledges. A referee should assess whether the journal is willing to publish a result whose arbitrary-characteristic formulation is contingent on an open problem; if not, the author should restrict the statements accordingly. The other major issues—the unstated [HP24] remark and the characteristic-independence claim for [Or02, Lemma 4.8]—are fixable by adding details or removing the dependency."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThis paper is worth engaging with. Its main contribution is Theorem A: for an A-torsor X, the moduli space of simple semi-homogeneous sheaves with fixed numerical invariants is again a torsor under an abelian variety, and there is a twisted derived equivalence between that moduli space and X. The construction via stabilizers and Lagrangian subvarieties is clean, and it genuinely extends Mukai/Gulbrandsen to arbitrary fields and Brauer twists. The real novelty is the converse (Theorem B / Corollary 2.8), which aims to characterize twisted derived equivalences of torsors as coming from such moduli spaces. That is new.\n\nThe soft spots are real but mostly localized. The converse over arbitrary fields depends on [dJO22, Theorem 1.1], which is proved only in characteristic 0. The author flags this in the introduction, but Theorem 2.4 still states cases (2) and (3) unconditionally, and the proof does not remove the dependency: it shows dim S(E_y) ≥ dim X and then invokes the same dJO22 input. So the 'only if' direction of the classification is unconditional only in characteristic 0; over arbitrary fields it is conditional on a characteristic-free version of that theorem. That should be stated as a caveat in the theorem, not just in the introduction.\n\nThe more serious local problem is Theorem D (Proposition 4.6) and, to a lesser extent, Proposition 4.7. The construction in Proposition 3.2 gives a moduli space that is a torsor under the dual of the Lagrangian stabilizer, not under the Lagrangian itself. As stated, the criterion involving B as a Lagrangian subvariety of A× does not follow; the correct statement should involve the dual (or the quotient). Proposition 4.7's 'for any A-torsor X' also exceeds what Corollary 3.8 provides, which is only one specific A-torsor. These are genuine errors, but they are peripheral: Theorems A–C do not depend on them.\n\nSmaller issues: Lemma 2.2 asserts without proof that [Or02, Lemma 4.8] is characteristic-independent, and the proof of Theorem 2.1 relies on the unreproduced [HP24, Proposition 6.3]. Both are probably fixable, but a referee should check.\n\nThis paper is for people working on Fourier-Mukai partners of abelian varieties and torsors over non-algebraically-closed fields; the appendix on twisted derived categories is a useful reference in its own right.\n\nRecommendation: send to a serious referee. The core theorems are plausible and important; the referee should require a corrected statement/proof of Theorem D/E and an explicit caveat about the positive-characteristic converse. Not a desk reject.","headline":"A serious paper with a real converse theorem, but the headline classification is conditional on a characteristic-0 input and Theorem D is misstated.","tokens_in":28004,"tokens_out":8508,"would_cite":true,"duration_ms":75078,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14F08","14K05","14F22","14D20"],"pacs":[],"model":"deepseek-v4-flash","headline":"For torsors under abelian varieties, twisted derived equivalence partners are exactly moduli spaces of simple semi-homogeneous sheaves.","keywords":["twisted derived categories","semi-homogeneous sheaves","moduli spaces of sheaves","abelian varieties","torsors","integral transforms","Brauer group","derived equivalence criterion"],"falsifier":"Over an algebraically closed field of positive characteristic, take a simple object $E\\in\\mathrm{D}^b(A)$ on an abelian variety of dimension $g$ with $\\dim\\operatorname{Ext}^i(E,E)=\\binom{g}{i}$ for $i=0,1$, and compute the dimension of its stabilizer under the $A\\times\\hat A$-action. If this stabilizer has dimension $<g$, then the classification result used for the converse fails outside characteristic 0, and Theorem 2.4's general statement loses its support.","tokens_in":26827,"feed_emoji":"🔁","tokens_out":18571,"duration_ms":159088,"temperature":0.7,"pith_summary":"This paper classifies, geometrically, which smooth projective varieties can be derived equivalent to a fixed torsor under an abelian variety, allowing the target category to be twisted by a cohomological obstruction. For an $A$-torsor $X$, the moduli space $\\mathrm{SSH}^\\xi_{X/k}$ of simple semi-homogeneous sheaves on $X$ with fixed numerical Chern character $\\xi$ is again a torsor under an abelian variety, and its universal sheaf induces a twisted derived equivalence $\\mathrm{D}^b(\\mathrm{SSH}^\\xi_{X/k},\\mu^{-1})\\to\\mathrm{D}^b(X)$ with $\\mu$ the universal obstruction. The paper proves the converse in a large class of cases, including all untwisted equivalences: if $Y$ is smooth projective and $\\mathrm{D}^b(Y,\\mu^{-1})\\simeq\\mathrm{D}^b(X)$, then under mild hypotheses $Y$ is one of these moduli spaces and $\\mu$ is the universal obstruction. Hence, for two torsors under abelian varieties, twisted derived equivalence is equivalent to being the other's semi-homogeneous-sheaf moduli space. The author also derives a twisted analogue of the standard derived-equivalence criterion for abelian varieties over algebraically closed fields, expressed in terms of Lagrangian subvarieties of $A\\times\\hat A$.","feed_headline":"Semi-homogeneous sheaves classify twisted derived equivalences","feed_subtitle":"The converse holds: every twisted equivalence to a given torsor forces the target to be one of these moduli spaces.","key_machinery":"The load-bearing object is the moduli space $\\mathrm{SSH}^\\xi_{X/k}$ of simple semi-homogeneous sheaves on a torsor $X$, cut out by fixing the numerical Chern character $\\xi$. The stabilizer $S(F)$ of a point sheaf under the natural $A\\times\\hat A$-action is a $g$-dimensional abelian subvariety; the common stabilizer $S(\\xi)$ is Lagrangian in the symplectic product $A\\times\\hat A$, and its quotient $A(\\xi)=(A\\times\\hat A)/S(\\xi)$ acts simply transitively, making $\\mathrm{SSH}^\\xi_{X/k}$ a torsor. The proof of the equivalence runs through a twisted full-faithfulness criterion (Theorem A.40): the universal sheaf's fibers satisfy the required Ext-vanishing and Ext-dimension conditions, Grothendieck duality and Serre functors turn full faithfulness into an equivalence, and the converse uses the classification result [dJO22, Theorem 1.1] together with deformation theory from [HP24] to identify the fibers of a given equivalence kernel as simple semi-homogeneous sheaves.","core_discovery":"The central claim is that the family of moduli spaces $\\mathrm{SSH}^\\xi_{X/k}$ of simple semi-homogeneous sheaves on an $A$-torsor $X$ parametrizes its twisted derived equivalence partners. A sheaf is simple semi-homogeneous when it is simple and its stabilizer under the $A\\times\\hat A$-action on the moduli stack of simple sheaves has dimension $g=\\dim X$; such sheaves have the same self-Ext dimensions as a skyscraper sheaf, and any two with the same numerical Chern character have no extensions between them. The paper shows each nonempty $\\mathrm{SSH}^\\xi_{X/k}$ is itself a torsor under the abelian variety $A(\\xi)=(A\\times\\hat A)/S(\\xi)$, carries a universal obstruction $\\mu$, and the universal twisted sheaf $E$ defines an integral transform $\\Phi_E:\\mathrm{D}^b(\\mathrm{SSH}^\\xi_{X/k},\\mu^{-1})\\to\\mathrm{D}^b(X)$ that is an equivalence. Conversely, under the hypotheses of Theorem 2.4, any twisted derived equivalence $\\Phi_E:\\mathrm{D}^b(Y,\\mu^{-1})\\to\\mathrm{D}^b(X)$ forces $Y$ to be isomorphic to $\\mathrm{SSH}^\\xi_{X/k}$, with $\\mu$ the universal obstruction and $E$ a shift of the universal sheaf. Combined, these two directions give the iff statement of Corollary 2.8: for two torsors $X,Y$ under abelian varieties, $\\mathrm{D}^b(Y,\\mu^{-1})\\simeq\\mathrm{D}^b(X)$ exactly when $Y$ is a moduli space of simple semi-homogeneous sheaves on $X$ and $\\mu$ is universal.","pith_inferences":["If the characteristic-0 hypothesis in the cited classification result is removable, the converse becomes unconditional for every smooth projective $Y$; a positive answer would upgrade Theorem 2.4 to a bijection between twisted equivalence kernels and moduli spaces.","The stabilizer map $\\xi\\mapsto S(\\xi)$ embeds the classification of twisted partners into the symplectic geometry of $A\\times\\hat A$; enumerating Lagrangian abelian subvarieties that occur as stabilizers would yield an explicit count of derived-equivalence partners of a given torsor.","The paper leaves open whether every isometric isomorphism of $A\\times\\hat A$ is realized by a derived equivalence between torsors; a natural test is whether adding Galois-equivariance to the isometry repairs the failure exhibited in its Example 4.4.","Over finite fields, Proposition 4.7 predicts untwisted derived equivalence from isometric isomorphisms; since derived equivalent varieties share zeta functions, this gives a numerical check: compute the zeta functions of $A$ and $B$ for such an isometry and compare."],"forward_implications":["Every smooth projective twisted partner $Y$ of a torsor $X$ that falls under one of the three hypotheses in Theorem 2.4 is itself a torsor under an abelian variety, isomorphic to $\\mathrm{SSH}^\\xi_{X/k}$ for some numerical Chern character $\\xi$.","All untwisted derived equivalences $\\mathrm{D}^b(Y)\\simeq\\mathrm{D}^b(X)$ between a torsor and a smooth projective variety force $Y$ to be a fine moduli space of simple semi-homogeneous sheaves on $X$.","For two torsors $X$ and $Y$ under abelian varieties, $\\mathrm{D}^b(Y,\\mu^{-1})\\simeq\\mathrm{D}^b(X)$ holds exactly when $Y$ is a moduli space $\\mathrm{SSH}^\\xi_{X/k}$ and $\\mu$ is the universal obstruction; no other twisted partners occur.","Over algebraically closed fields, the twisted analogue of the usual abelian-variety criterion holds: $\\mathrm{D}^b(B,\\beta^{-1})\\simeq\\mathrm{D}^b(A)$ iff $B$ is isomorphic to a Lagrangian abelian subvariety of $A\\times\\hat A$.","Any isometric isomorphism between $A\\times\\hat A$ and $B\\times\\hat B$ produces, for every $A$-torsor $X$, a $B$-torsor $Y$ and a twisted equivalence with obstruction in $\\mathrm{Br}_1(Y)$; over finite fields this yields untwisted derived equivalence $\\mathrm{D}^b(B)\\simeq\\mathrm{D}^b(A)$."],"supporting_citations":[{"why":"Supplies the classification, in characteristic 0, of simple objects with the Ext-dimensions of a point sheaf as simple semi-homogeneous sheaves; this powers the converse.","marker":"[dJO22]"},{"why":"Provides the generalized full-faithfulness criterion and the deformation-theoretic Proposition 6.3 used to verify the Ext^1 condition for the universal kernel.","marker":"[HP24]"},{"why":"Establishes the stabilizer, Ext-vanishing, and Chern-character facts for semi-homogeneous bundles that make the moduli spaces torsors.","marker":"[Mu78]"},{"why":"Supplies the Ext-vanishing lemma and the isometric-isomorphism framework for derived equivalences of abelian varieties.","marker":"[Or02]"},{"why":"Supplies the symplectic-biextension and Lagrangian-subvariety tools behind Propositions 3.2 and 4.6.","marker":"[Pol12]"},{"why":"Gives the Picard-scheme dimension comparison used to handle the case $\\mu\\in\\mathrm{Br}_1(Y)$ in the converse.","marker":"[Hon18]"},{"why":"Shows uniqueness of dg-enhancements, letting the proof pass from derived equivalence to autoequivalence group schemes.","marker":"[CNS22]"},{"why":"Describes the identity component of derived autoequivalence groups, used to bound the stabilizer dimension.","marker":"[Ros09]"}],"fun_headline_variants":["Semi-homogeneous sheaves parametrize twisted equivalences","Twisted equivalences from semi-homogeneous sheaf moduli","Semi-homogeneous sheaves classify torsor duals","Torsor equivalences arise from semi-homogeneous sheaves"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The converse direction rests on a characteristic-0 theorem, noted by the author as possibly unnecessary, that any simple object whose self-extensions have the same dimensions as those of a structure sheaf is, up to shift, a simple semi-homogeneous sheaf; the forward direction also leans on a cited deformation-theoretic proposition whose proof is not included.","fun_headline_variants_meta":{"raw":{"variants":["Semi-homogeneous sheaves parametrize twisted equivalences","Twisted equivalences from semi-homogeneous sheaf moduli","Semi-homogeneous sheaves classify torsor duals","Torsor equivalences arise from semi-homogeneous sheaves"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001013,"raw_usage":{"total_tokens":4295,"prompt_tokens":981,"completion_tokens":3314,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":3240}},"tokens_in":597,"tokens_out":3314,"duration_ms":25351,"temperature":1.0,"reasoning_tokens":3240,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:52:39.637186+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Over an algebraically closed field of positive characteristic, take a simple object $E\\in\\mathrm{D}^b(A)$ on an abelian variety of dimension $g$ with $\\dim\\operatorname{Ext}^i(E,E)=\\binom{g}{i}$ for $i=0,1$, and compute the dimension of its stabilizer under the $A\\times\\hat A$-action. If this stabilizer has dimension $<g$, then the classification result used for the converse fails outside characteristic 0, and Theorem 2.4's general statement loses its support.","supporting_citations":[],"review_version":1}