{"id":"952856cc-3b0c-4e22-9ef5-5d820b10e25f","arxiv_id":"1908.03131","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The Brauer class obstructing the relative Picard functor of a smooth proper curve carries an involution of the second kind and splits at generic points of theta divisors.","lead":"A short algebraic geometry paper studies the Brauer class that blocks the existence of a universal line bundle on the Picard scheme of a curve. It proves that the associated division algebra has a natural involution, and that the class vanishes at the generic points of certain theta divisors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.8's proof chooses a K-rational linear form in the open set of perfect pairings, but σ-equivariance of the resulting pairing requires the form to be fixed by the involution σ; the proof does not justify the existence of such a σ-invariant K-point.","rationale":"The reader's weakest_assumption focused on the reducedness of the generalized theta divisor Theta_E in the splitting theorems. That is a legitimate concern, and the paper itself restricts to the universal-curve setting to make it hold. However, I find a more direct gap in the paper's central involution theorem: the K-rational point in the open set O is only shown to be Galois-invariant, not σ-invariant, and σ-invariance is exactly what is needed to obtain a σ-semilinear isomorphism in Proposition 3.6. This affects Theorem 3.8, one of the two main advertised results, and is not addressed by the paper's discussion of reducedness. Since the theorem could still be true and the gap might be repairable (e.g., by averaging or by a finer descent argument), the appropriate verdict remains CONDITIONAL, matching the reader's judgment. The reader's concern and mine are distinct but complementary, hence partial agreement.","tokens_in":10740,"tokens_out":54015,"duration_ms":572237,"concrete_test":"For a concrete genus 3 curve X over a field k and a degree-0 line bundle L on X_{k^sep}, compute the incidence variety Z of Lemma 3.7 and the open O ⊂ P^∨(H^0(ω^⊗2)). Explicitly check whether O contains a point of the fixed locus P^∨(H^0(X_k,ω^⊗2)) embedded in P^∨(H^0(X_K,ω^⊗2)); if O ∩ Fixed = ∅, the proof of Theorem 3.8 has a gap.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 3.8 obtains a pairing u': H^0(ω⊗L) × H^0(ω⊗L^∨) → K' by composing the natural σ-equivariant bilinear map B with a K-rational linear form ℓ on H^0(ω^⊗2). Such an ℓ is G-invariant, but the σ-equivariance of u' requires ℓ(σ(v)) = σ(ℓ(v)) for all v, which is equivalent to ℓ being fixed by the involution σ. The paper's invocation of Lemma 3.7 and 'K is infinite' only guarantees a K-rational point in the open set O of perfect pairings; it does not ensure that this point lies in the σ-fixed locus. Nonempty G-invariant open subsets of projective space over an infinite field can fail to contain any σ-fixed K-point (e.g., the open set G_m ⊂ P^1 with x ↦ -x has no fixed point in characteristic ≠ 2). Thus the construction in Theorem 3.8 may not produce a σ-linear isomorphism as required by Proposition 3.6, so the existence of the involution of the second kind extending σ is not established by the given argument.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies the Brauer class α ∈ Br(Pic_{X/k}) obstructing the representability of the relative Picard functor for a smooth proper curve X over a field k. The author proves three main results: (i) the division algebra representing the restriction α_0 at the generic point of Pic^0_{X/k} admits an involution of the second kind extending the natural dual involution (Theorem 3.8); (ii) the class α_{g-1} splits at the generic point of the theta divisor Θ ⊂ Pic^{g-1}_{X/k} (Theorem 4.4); and (iii) under the universal-curve hypothesis, the class α_0 splits at the generic point of a generalized theta divisor Θ_E ⊂ Pic^0_{X/k} for a semistable rank-two slope-(g-1) vector bundle E (Theorem 5.5). The proofs use standard techniques: Leray spectral sequences, Brauer-Severi varieties, Galois descent, and projectivizations of direct images.","tokens_in":10941,"tokens_out":31872,"duration_ms":354457,"significance":"If the results are correct, they supply concrete structural information about an explicit Brauer class associated with the Picard scheme, a class that has been studied mainly through its period and index. The involution statement gives a positive answer for α_0 to a natural algebra-structure question, and the splitting results identify height-one loci where the obstruction vanishes, which is of interest for constructing Azumaya orders and for the geometry of theta divisors. The paper is careful to restrict the generalized-theta result to the universal curve, where reducedness of Θ_E is available. The manuscript also contains useful descent lemmas (Lemmas 4.2 and 4.3) that may be of independent use. However, as written, the proofs of the involution theorem contain several gaps that are central to the claim, so the paper needs substantial revision before the results can be considered established.","major_comments":[{"comment":"The proof of Theorem 3.8 chooses a K-rational point in the open set O of perfect pairings, but it does not justify that the corresponding linear form ℓ is fixed by the involution σ. The pairing u obtained as ℓ ∘ B is σ-equivariant only if ℓ(σ(w)) = σ(ℓ(w)) for all w, which is equivalent to ℓ being a σ-fixed point of the induced action on the dual projective space. The argument 'K is infinite' only guarantees a K-rational point, hence a G-invariant point, not a σ-fixed one. This is load-bearing because Proposition 3.6 explicitly requires a σ-equivariant pairing. The gap is repairable by observing that O, being defined over K and stable under σ, descends to a nonempty open subset of the projective space over K^σ, which must contain a K^σ-point since K^σ is infinite; but this argument is absent from the manuscript.","section":"§3.4, Theorem 3.8"},{"comment":"The proof of Lemma 3.4 asserts that σ^*L ⊗ L is the trivial line bundle on X_{K'} and then says 'Take any K-section of σ^*L ⊗ L would yield a G-equivariant isomorphism σ^*L ≅ L^∨.' Triviality over K' does not by itself provide a K-section, i.e., a G-invariant nonzero section. One must argue that the line bundle is trivial over X_K, for instance by the injectivity of Pic(X_K) → Pic(X_{K'}) (which follows from Hilbert 90 because H^0(X_{K'}, O^*) = K'^*), or by explicitly constructing the descent data. As written, the proof skips this step, and Proposition 3.6 relies on the existence of a G-invariant isomorphism φ: L^∨ → σ^*L.","section":"Lemma 3.4"},{"comment":"Even if a σ-linear isomorphism Q: H^0(ω⊗L) → H^0(ω⊗L)^∨ is produced, the conclusion of Theorem 3.8 requires an involution of the second kind, i.e., a σ-semilinear anti-automorphism whose square is the identity. The manuscript proceeds 'by Proposition 3.6 and Lemma 3.5' without explaining how a σ-linear isomorphism of vector spaces yields an order-two anti-automorphism of the central simple algebra. The natural adjoint anti-automorphism attached to Q has square equal to an inner automorphism in general, unless Q satisfies a σ-Hermitian (or skew-Hermitian) condition. The proof does not verify this condition, nor does it invoke a standard result (such as the cohomological criterion in Lemma 3.1 or Albert's symmetrization theorem) to bridge the gap. Thus the existence of an involution extending σ is not established by the given argument.","section":"Theorem 3.8 and Lemma 3.5"}],"minor_comments":[{"comment":"The statement of Lemma 5.2 is garbled: 'the numerical class is always a multiple of 2g−2 / g. c. d(2g−2,d +g−1) Θ' should be written as a multiple of ((2g-2)/gcd(2g-2, d+g-1))·Θ, and the proof would benefit from a few more details relating the Néron-Severi class to the torsor class in H^1(k, Pic^0_{X/k}).","section":"Lemma 5.2"},{"comment":"The sentence 'But T_U is reduced, so it is regular at the generic point' is imprecise: a reduced scheme is regular at its generic points, but this is being used for an irreducible divisor, so the wording should specify that T_U (or its component at the relevant generic point) is integral.","section":"Lemma 4.1"},{"comment":"The application of Lemma 4.2 to F = π'_*(L') does not explicitly verify the hypotheses: the line bundle N on T_U ×_T T_U and the isomorphisms φ and β with the cocycle condition are not identified. The argument is plausible because the descent data come from the étale cover and the Brauer class, but the proof should spell out the data or cite the construction explicitly.","section":"Theorem 4.4"},{"comment":"There are numerous small typographical errors and missing spaces in the text (e.g., 'thetha' in the abstract, 'Kk^sep' for K k^sep, and inconsistent use of 'i' as an index starting from 0 in the matrices of Proposition 5.3). These do not affect the mathematics but should be corrected.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper is unusually terse in places, and the proof of the main involution theorem has real gaps: the σ-invariance of the chosen linear form, the K-section existence in Lemma 3.4, and the passage from a σ-linear isomorphism to an involution of order two. Each of these appears repairable within the scope of the manuscript, but as written the arguments do not establish the theorem. The splitting results are more convincing, though they would also benefit from additional detail in the descent lemmas. I recommend major revision rather than rejection, because the underlying strategy is sound and the fixes are local rather than fundamental."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper's flagship theorem, the second-kind involution on the division algebra at the generic point of Pic^0, is not proved as written. The stress-test note points at the real gap: Theorem 3.8 finds a K-point in the open set of perfect pairings, but σ-equivariance of the resulting pairing requires that point to be fixed by σ. A K-rational point in a σ-invariant open subset need not be σ-fixed; the example of G_m inside P^1 with x ↦ -x shows the issue is concrete. The proof would need an additional argument, for instance showing the open set has a σ-fixed point defined over the fixed field, to work.\n\nThat said, the paper is not sloppy. The splitting theorems are genuinely new, and the approach is coherent. The idea of testing whether the obstruction class dies at the generic point of the theta divisor by descending the projectivization of a direct image is elegant, and the descent lemmas (4.2, 4.3) are carefully written. The paper also deserves credit for raising the Azumaya-order question, which is a good research prompt.\n\nThe soft spots are concentrated in proof details. Lemma 3.4 asserts a Galois-equivariant isomorphism σ*L ≅ L^∨ from the triviality of σ*L ⊗ L with a one-sentence argument; a K-section of the tensor product does not automatically give a G-equivariant isomorphism after base change. This affects the setup of Proposition 3.6, though it may be fixable by choosing the section more carefully. Lemma 4.1's corank argument is compressed: reducedness of the theta divisor implies regularity at the generic point, but the step from corank ≥2 to non-regularity needs more spelling out. The generalized theta case in Section 5 explicitly relies on the universal-curve hypothesis for reducedness of Θ_E, and the paper is honest about that limitation, so I would not count that as a flaw beyond scope.\n\nOverall, the main theorem likely needs nontrivial repair, but the splitting results may survive. This is a worthwhile preprint for anyone working on Brauer groups of Picard schemes, and it deserves a serious referee, though not acceptance as is. I would not cite the involution theorem without checking a corrected version; the splitting results might be citable after verification.\n\nRecommendation: send to a good referee, but expect a report requiring substantive revision.","headline":"The involution theorem is not proved as written—the chosen pairing need not be σ-invariant—but the theta-splitting results are genuinely interesting and worth referee time.","tokens_in":11485,"tokens_out":3344,"would_cite":false,"duration_ms":39024,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14H60","14K30","14F22","16K50"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the Brauer class obstructing the relative Picard functor of a curve admits an involution of the second kind on its division algebra and splits at the generic point of the theta divisor.","keywords":["Brauer class","Picard scheme","division algebra","involution of the second kind","theta divisor","generalized theta divisor","universal curve","tautological line bundle"],"falsifier":"Find a smooth proper curve $X$ and a semistable rank-2 bundle $E$ for which the generalized $\\theta$ divisor $\\Theta_E$ is nonreduced, and compute the generic corank of the connecting homomorphism $\\delta'$ in the exact sequence of Section 5.4. If the corank is at least 2, then $\\pi'_*(E'\\otimes\\mathcal{L}')$ has generic rank greater than 1, so the projectivization in Theorem 5.5 is not birational and the argument for splitting at $k(\\Theta_E)$ breaks down.","tokens_in":10487,"feed_emoji":"📐","tokens_out":16525,"duration_ms":149613,"temperature":0.7,"pith_summary":"The paper studies the Brauer class $\\alpha$ on the Picard scheme of a smooth proper curve that measures the obstruction to the relative Picard functor being representable by an actual line bundle on the product of the curve with its Picard scheme. It establishes that the division algebra representing $\\alpha_0$ at the generic point of $\\mathrm{Pic}^0_{X/k}$ has an involution of the second kind extending the natural involution induced by dualizing line bundles. It also proves that the restriction of the class to the generic point of the $\\theta$ divisor in $\\mathrm{Pic}^{g-1}_{X/k}$ is zero, and that in the universal genus-$g$ case the restriction to generalized $\\theta$ divisors in $\\mathrm{Pic}^0_{X/k}$ is zero. These splitting results say the obstruction disappears at height-one loci of the Picard scheme, so a tautological line bundle exists over the generic points of those divisors. The interest is that this gives concrete structural information about a natural Brauer class and its division algebra, including a property relevant to the open question of whether that division algebra contains an Azumaya order.","feed_headline":"Brauer obstruction gains involutions, splits on theta divisors","feed_subtitle":"The Brauer class blocking a universal line bundle carries an involution and splits at theta divisors.","key_machinery":"The central object is the obstruction class $\\alpha = d_0^{0,1}(1_{\\mathrm{Pic}_{X/k}}) \\in \\mathrm{Br}(\\mathrm{Pic}_{X/k})$ coming from the low-degree terms of the Leray spectral sequence for the projection $X \\times \\mathrm{Pic}_{X/k} \\to \\mathrm{Pic}_{X/k}$. On the generic point, $\\alpha_0$ is represented by the endomorphism algebra of $H^0(X_{K^{\\mathrm{sep}}}, \\omega_{X_{K^{\\mathrm{sep}}}} \\otimes \\mathcal{L})$ for the unique tautological line bundle $\\mathcal{L}$. For the involution, the load-bearing identity is a $\\sigma$-equivariant, Galois-equivariant perfect pairing between $H^0(\\omega \\otimes \\mathcal{L})$ and $H^0(\\omega \\otimes \\mathcal{L}^\\vee)$, obtained from multiplication of sections into $H^0(\\omega^{\\otimes 2})$; Skolem–Noether converts this pairing into the required semilinear automorphism of the algebra. For the splitting theorems, the key mechanism is the projectivized direct image $\\mathbb{P}(\\pi'_*\\mathcal{L}')$ over the $\\theta$ divisor, with descent data that make the tautological line bundle descend to the total space; because $\\pi'_*\\mathcal{L}'$ has generic rank 1, the projectivization is birational and the descended bundle gives the splitting over $k(\\Theta)$ (and over $k(\\Theta_E)$ in the universal case).","core_discovery":"At the generic point of $\\mathrm{Pic}^0_{X/k}$, the Brauer class $\\alpha_0$ is represented by a division algebra $D$ over $K = k(\\mathrm{Pic}^0_{X/k})$. The paper establishes three properties. First, $D$ carries an involution of the second kind extending the involution $\\sigma$ on $\\mathrm{Pic}^0_{X/k}$ induced by $L \\mapsto L^\\vee$. Second, the class $\\alpha_{g-1}$ restricts to zero in $\\mathrm{Br}(k(\\Theta))$ for the $\\theta$ divisor $\\Theta \\subset \\mathrm{Pic}^{g-1}_{X/k}$. Third, under the universal-curve hypothesis, $\\alpha_0$ restricts to zero in $\\mathrm{Br}(k(\\Theta_E))$ for the generalized $\\theta$ divisor attached to a semistable rank-2 slope-$(g-1)$ vector bundle $E$. The splitting results are proved by descending a tautological line bundle over the projectivization of the direct image of the tautological bundle restricted to the divisor, and the descent is birational because the direct-image sheaf has generic rank 1.","pith_inferences":["The paper's method suggests a broader criterion: whenever a divisor in a Picard scheme is reduced at its generic point and the relevant direct image has generic rank 1, the same projectivization-and-descent construction should split the Brauer class over the divisor's function field.","If the division algebra of $\\alpha_0$ contains an Azumaya order compatible with the constructed involution, the paper's open question would be answered affirmatively; if not, the natural involutions could still help build an unramified division algebra without an Azumaya order, in the direction of the reference [AW14].","A testable next step is to compute the generic rank of $\\pi'_*(E'\\otimes\\mathcal{L}')$ on a nonreduced generalized theta divisor, such as those cited from [HP15]; a rank jump would show exactly where the universal-curve hypothesis is needed."],"forward_implications":["Because $\\alpha_{g-1}$ vanishes over $k(\\Theta)$, a tautological line bundle exists over the generic point of the theta divisor, so the obstruction splits at that height-one point of $\\mathrm{Pic}^{g-1}_{X/k}$.","The division algebra of $\\alpha_0$ has an involution of the second kind extending the dual involution $L \\mapsto L^\\vee$, and the same construction gives an analogous involution for $\\alpha_{g-1}$ extending $L \\mapsto \\omega_X \\otimes L^\\vee$.","For the universal genus-$g$ curve, $\\alpha_0$ splits at the generic point of every generalized theta divisor $\\Theta_E$ built from a semistable rank-2 bundle of slope $g-1$.","The proof's template shows that any height-one point of the Picard scheme where the restricted direct-image sheaf has generic rank 1 yields a splitting of the corresponding restriction of $\\alpha$."],"supporting_citations":[{"why":"Defines the Brauer class obstructing the existence of a tautological sheaf on the product of a variety with its moduli space, the central object of the paper.","marker":"[Cal00]"},{"why":"Supplies the representability of the relative Picard functor and the descent-effectiveness results used to define the Picard scheme and to descend projective bundles to the theta divisor.","marker":"[Bos90]"},{"why":"Gives the Brauer–Severi description of the class as the descended projective bundle of the direct image of the canonical bundle twisted by the tautological bundle.","marker":"[Gir71]"},{"why":"Provides the cohomological criterion (vanishing of the appropriate corestriction) for the existence of involutions of the second kind, invoked in Lemma 3.1.","marker":"[KMRT98]"},{"why":"Supplies the Skolem–Noether theorem identifying isomorphisms of endomorphism algebras with isomorphisms of the underlying vector spaces, used for descent.","marker":"[GS17]"},{"why":"Gives the base-change/canonical isomorphism of global sections used in Lemma 3.4 to identify the pullback of the Brauer class with the algebra built from the dual tautological bundle.","marker":"[Har77]"},{"why":"Supplies the vanishing theorem for semistable bundles of Euler characteristic zero and the numerical equivalence of the generalized theta divisor with twice the classical theta divisor, used in Section 5.","marker":"[Ray82]"},{"why":"Supplies the strong Franchetta fact that the Picard group of the universal curve is generated by the canonical class, used to prove semistability and control numerical classes.","marker":"[Sch03]"},{"why":"Provides the classical construction of the theta divisor as the degeneracy locus of a connecting homomorphism, on which the proof of Theorem 4.4 relies.","marker":"[ACGH85]"}],"fun_headline_variants":["Brauer class gains involutions, splits on theta divisors","Involutions on Brauer algebra, theta-divisor splitting","Brauer class: involutions and splitting at theta divisors","Division algebra gets involutions, splits at theta divisors","Brauer obstruction: new involutions, theta splitting"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the direct-image sheaf on the theta divisor has generic rank 1, which the paper derives from reducedness of the divisor; in the generalized-theta case that reducedness is known only for universal curves and can fail for arbitrary curves or higher-rank bundles.","fun_headline_variants_meta":{"raw":{"variants":["Brauer class gains involutions, splits on theta divisors","Involutions on Brauer algebra, theta-divisor splitting","Brauer class: involutions and splitting at theta divisors","Division algebra gets involutions, splits at theta divisors","Brauer obstruction: new involutions, theta splitting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000181,"raw_usage":{"total_tokens":1254,"prompt_tokens":838,"completion_tokens":416,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":454,"completion_tokens_details":{"reasoning_tokens":336}},"tokens_in":454,"tokens_out":416,"duration_ms":4098,"temperature":1.0,"reasoning_tokens":336,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:23:58.683394+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a smooth proper curve $X$ and a semistable rank-2 bundle $E$ for which the generalized $\\theta$ divisor $\\Theta_E$ is nonreduced, and compute the generic corank of the connecting homomorphism $\\delta'$ in the exact sequence of Section 5.4. If the corank is at least 2, then $\\pi'_*(E'\\otimes\\mathcal{L}')$ has generic rank greater than 1, so the projectivization in Theorem 5.5 is not birational and the argument for splitting at $k(\\Theta_E)$ breaks down.","supporting_citations":[],"review_version":1}