{"id":"14a8afb2-c4b7-484a-8476-ca50b3c757e4","arxiv_id":"1908.02279","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For an irreducible nodal curve of genus at least two, the cohomology ring of the moduli space of rank-two semistable sheaves with fixed odd determinant obeys the Mumford relations restricted to monodromy-invariant classes, and its Hodge-Poincare polynomial is computed explicitly.","lead":"This paper proves a version of Mumford's conjecture describing the full cohomology ring of the moduli space of rank-two semistable sheaves on an irreducible nodal curve, and gives an explicit formula for its Hodge-Poincare polynomial. The result extends Kirwan's 1992 theorem for smooth curves to singular curves by degenerating a smooth curve to a node and tracking the mixed Hodge structures.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central Mumford decomposition (Theorem 7.2) rests on the unverified limit Mumford–Newstead isomorphism of mixed Hodge structures in Theorem 3.2, which is delegated to the authors' companion paper [3].","rationale":"The reader's verdict is CONDITIONAL with the same weakest assumption: Theorem 3.2 is a load-bearing black box delegated to the authors' companion paper [3]. My reading of the manuscript confirms that this is the most fragile step in the central argument. All of the objects in the statement of Theorem 7.2 — the classes ψ∞_i, the primitive subspaces P∞_i, and the identification H^i(UX0(2,L0),Q) ≅ W_i H^i(G(2,L)∞,Q) — are defined or identified through ~Φ0 and its mixed-Hodge compatibility. If Theorem 3.2 is correct in full, the subsequent use of King–Newstead's basis of the smooth moduli space, the Gysin computation in §4, and the diagram chase in §7 are coherent and mutually consistent. I found no clear internal contradiction; the issue is the unverified external reliance, not a demonstrable error. Therefore the appropriate recommendation is to keep the reader's CONDITIONAL verdict unchanged, with the condition being an independent verification of [3, Proposition 4.1] as spelled out in the concrete test.","tokens_in":25265,"tokens_out":6725,"duration_ms":69494,"concrete_test":"Extract the proof of [3, Proposition 4.1] and verify, in the notation of the present paper, that it establishes all three parts of Theorem 3.2: (a) the extension ~Φ of Φ_Δ* to the canonical extensions; (b) the isomorphism ~Φ0: H^1(~X∞,Q) → H^3(G(2,L)∞,Q); and (c) the compatibility ~Φ0(W_i H^1(~X∞,Q)) = W_{i+2} H^3(G(2,L)∞,Q) and F^p H^1(~X∞,C) ↦ F^{p+1} H^3(G(2,L)∞,C). An independent computational check would be to compute the monodromy logarithm N on H^3(G(2,L)∞,Q) via the Steenbrink spectral sequence of Proposition 2.3 and compare its monodromy weight filtration with the shifted weight filtration of H^1(~X∞,Q). If the weight filtrations do not match, equations (5.2), (6.1), and (7.5) lose their basis, and Theorem 7.2 is not established. If [3, Proposition 4.1] proves only the Hodge-filtration statement, the present proof has a gap at Theorem 3.2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 5 defines every generator ψ∞_i as ~Φ0(e_i), and Theorem 3.2 supplies the only link between the curve-side limit mixed Hodge structure H^1(~X∞,Q) and the moduli-side limit mixed Hodge structure H^3(G(2,L)∞,Q). The claimed weight/Hodge compatibility is then used at three crucial points: to identify Gr^W_3 H^3(G(2,L)∞,Q) with H^3(M_{\\tilde X0}(2,\\tilde L0),Q) in §5.3; to obtain the weight-filtered monomial basis behind equation (5.2); and to identify H^i(UX0(2,L0),Q) with W_i H^i(G(2,L)∞,Q) in equation (7.5), which in turn feeds directly into the Hodge–Poincaré formula of Theorem 7.2. The proof of Theorem 3.2 in this paper is a single sentence: 'See [3, Proposition 4.1] for a proof of the statement.' If that companion proposition proves only the Hodge-filtration part or proves the isomorphism without the weight shift W_i ↔ W_{i+2}, then the generators ψ∞_i, the primitive subspaces P∞_i, and the weight identifications used throughout §§5–7 lose their stated Hodge-theoretic content, and neither the generalized Mumford decomposition nor the displayed Hodge–Poincaré polynomial follows from the argument presented here. I did not find an internal contradiction in the rest of the paper, but a central claim whose proof is physically located in another paper cannot be treated as a verified step in a machine review.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a generalization of Mumford's conjecture for the moduli space U_{X_0}(2,L_0) of rank-2 semistable torsion-free sheaves with fixed odd determinant on an irreducible nodal curve X_0 of genus g≥2. The strategy is to embed X_0 in a one-parameter degeneration, compare the Simpson moduli space with the Gieseker moduli space whose central fiber is a simple normal crossings divisor, and use limit mixed Hodge structures. The central result, Theorem 7.2, gives an isomorphism H^*(U_{X_0}(2,L_0),Q) ≅ ⊕_i P^∞_i ⊗ Q[α∞,β∞,ψ∞]/I^∞_{g-i} and an explicit Hodge–Poincaré polynomial. The proof proceeds through a Gieseker-space version (Theorem 5.1), a Hodge–Poincaré formula for the Gieseker space (Theorem 6.2), and a comparison of Gieseker and Simpson moduli spaces (Proposition 7.1). The main structural steps are the limit Mumford–Newstead isomorphism, the computation of kernels of Gysin morphisms, and the identification of the cohomology of the central Simpson fiber with the monodromy-invariant part of the limit mixed Hodge structure.","tokens_in":25601,"tokens_out":15336,"duration_ms":138964,"significance":"If the proof is completed, this is a substantial and natural extension of the Mumford–Kirwan–Zagier theory to singular curves: it provides the first description of the full cohomology ring of the moduli space of rank-2 sheaves with fixed determinant on an irreducible nodal curve, together with a closed Hodge–Poincaré polynomial. The paper also demonstrates a useful method: passing through the Gieseker compactification and using limit mixed Hodge structures to isolate monodromy-invariant classes. The authors are explicit about the input from their companion paper [3] and from the literature, and the final computations are presented in a checkable algebraic form. The main weakness is that a load-bearing ingredient, Theorem 3.2, is delegated to [3] by a one-sentence citation, and the exact content of that theorem is not reproduced.","major_comments":[{"comment":"The proof of Theorem 3.2 is a one-sentence citation to [3, Proposition 4.1], yet this statement is load-bearing for the whole paper. The generators ψ∞_i in §5.1 are defined via ~Φ0; the identification Gr^W_3 H^3(G(2,L)∞,Q) ≅ H^3(M_{~X0}(2,~L0),Q) in §5.3 uses the weight shift W_i ↔ W_{i+2}; the monomial basis behind equation (5.2) uses the same compatibility; and equation (7.5) identifies H^i(U_{X0}(2,L0),Q) with W_iH^i(G(2,L)∞,Q). The manuscript must either reproduce the proof of [3, Proposition 4.1] or state its exact content in enough detail to verify that it includes the full weight-filtration shift W_iH^1(~X∞,Q) ↔ W_{i+2}H^3(G(2,L)∞,Q) for all i, and not merely an isomorphism of vector spaces or a Hodge-filtered statement without the weight shift.","section":"§3.3, Theorem 3.2"},{"comment":"The proof of (5.2) asserts exactness of the upper row in the displayed commutative diagram and concludes ⊕_i W_iH^i(G(2,L)∞,Q) ≅ ⊕_i P^∞_i⊗Q[α∞,β∞,ψ∞]/I^∞_{g-i}. This is the limit-version of the Mumford–Newstead–Kirwan presentation, but the surjectivity of ν∞ and the identification of its kernel with ⊕_k P^∞_k⊗I^∞_{g-k} are not derived in the text. The cited [23, Theorem 3.2] and [23, Remark 5.3] give the presentation for each smooth fiber G(2,L)_s; the passage to the limit relies again on the weight compatibility of Theorem 3.2 and on the particular definition of P^∞_i. Since (5.2) is the direct algebraic input to Theorem 7.2, the justification should be made explicit.","section":"§5.4, Eq. (5.2)"},{"comment":"There is a determinant inconsistency between two identifications of U_0(L_0). In §7.2 the authors state U_0(L_0) ≅ M_{~X0}(2,~L0(−D)), while in the proof of Proposition 7.1 the cohomology of U_0(L_0) is replaced by H^i(M_{~X0}(2,~L0),Q) and the cokernel is written as ⊕_j H^{i−2j}(M_{~X0}(2,~L0),Q)(ξ'')^j. If the intended isomorphism M_{~X0}(2,~L0(−D)) ≅ M_{~X0}(2,~L0) is obtained by tensoring with a degree-(−1) line bundle, that isomorphism needs to be stated and justified; as written, the cohomological comparison feeds directly into the Hodge–Poincaré computation of Theorem 7.2 without proof.","section":"§7.2 and §7.4, Proposition 7.1"}],"minor_comments":[{"comment":"In the definitions of P_i and P^∞_i, the symbol ⋃ appears where a direct sum or a cup-product action is intended; as written, the displayed domains and targets are not meaningful sets.","section":"§5.2"},{"comment":"The generator list includes ψ∞_g twice: once in the range 1≤i≤2g−1 and once in the product ψ∞_g ψ∞_{2g}. If the intended list is α∞,β∞,ψ∞_1,…,ψ∞_{2g−1},ψ∞_gψ∞_{2g}, this is harmless but should be clarified.","section":"Remark 7.3"},{"comment":"The notation H^{p,q}Gr^W_{p+q}H^{p+q}(G(2,L)∞,C) is used throughout §6 without an explicit definition; a one-line definition would improve readability.","section":"§6"},{"comment":"The structure of G_{X0}(2,L0) as a union of two components, with G_1 and G_0∩G_1 being a P^3-bundle and a P^1×P^1-bundle over M_{~X0}(2,~L0), is cited to [38] and then used repeatedly in §§4–7. A precise statement of which facts are being imported from [38] would make the verification easier.","section":"§3.1"}],"recommendation":"major_revision","confidential_remarks":"The main risk is the single-sentence delegation of Theorem 3.2 to [3]. If [3, Proposition 4.1] indeed proves the full mixed-Hodge-structure compatibility including the weight shift, then the paper's central argument is likely sound; if not, the definitions in §§5–7 and both main theorems lose their foundation. I would ask the authors to state the precise content of [3, Proposition 4.1] and, if possible, include a proof sketch of the weight compatibility. The determinant inconsistency in Proposition 7.1 should also be resolved explicitly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear —,\n\nPunchline: this is a genuinely new result, not a repackaging of Kirwan. The authors prove the Mumford-type presentation for the cohomology ring of the rank-2 fixed-determinant moduli space over an irreducible nodal curve, and they compute the associated Hodge–Poincaré polynomial. The path is sensible: degenerate the curve, work on the Gieseker moduli space whose central fiber is a simple normal crossings divisor, compute the specialization map's kernel via a Gysin computation (Theorem 4.2), then descend to Sun's moduli space.\n\nWhat is good: the monodromy-invariant decomposition and the explicit kernel computation for the Gysin map are real new work; Theorem 4.2 is substantive. The use of King–Newstead's basis and the recursive Mumford ideals is sound. The Hodge–Poincaré polynomial looks plausible and does not appear in the earlier literature.\n\nThe soft spot is exactly where the reader's report puts it: Theorem 3.2, the limit Mumford–Newstead isomorphism of mixed Hodge structures with weight shift W_i ↦ W_{i+2}, is quoted from the authors' companion paper [3] with a one-sentence proof. Everything downstream—the generators ψ∞_i, the primitive subspaces P_i^∞, the identification H^i(U_X0) ≅ W_i H^i(G_∞), and the Hodge–Poincaré formula—depends on that theorem being true exactly as stated. If [3, Prop 4.1] proves only the Hodge filtration part, or the weight shift is different, the main result as presented does not follow. That does not make the paper wrong; it makes it a conditional proof. The authors are honest that the statement is external, and citing a published companion paper is normal practice, but the load-bearing nature is unusually heavy.\n\nMinor caveat: the remark that the papers in the series are “independent and overlap only in background” is a bit generous, since this paper's main theorem leans on [3] for its keystone. Also, the reader cannot check the Hodge-type computations without the companion text; a referee should be asked to look at [3] specifically.\n\nWho this is for: people working on cohomology of moduli spaces over singular curves, or on degeneration of Hodge structures for moduli. I would cite it if I worked in that area.\n\nRecommendation: yes, send it to a serious referee. The referee's main job is to verify Theorem 3.2 against [3, Prop 4.1]. If that checks out, the paper is a solid advance.","headline":"Genuine nodal-curve generalization of Mumford's conjecture, but the proof's keystone is a black box in the authors' companion paper.","tokens_in":26160,"tokens_out":3243,"would_cite":true,"duration_ms":32425,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["32G20","32S35","14D07","14D22","14D20","14H60","55R40"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves Mumford's cohomology-ring conjecture for rank-2 semistable sheaves with fixed odd determinant on an irreducible nodal curve, and computes the Hodge–Poincaré polynomial of this singular moduli space.","keywords":["Mumford conjecture","cohomology ring","moduli of sheaves","nodal curves","limit mixed Hodge structures","Hodge-Poincaré polynomial","rank 2 bundles","Gieseker moduli space"],"falsifier":"For the smallest case $g=2$, expand the claimed Hodge-Poincar\\'e polynomial and compare every coefficient with an independent computation of the Betti numbers of $U_{X_0}(2,L_0)$ from a normalization model or a point count over finite fields; a single mismatch in any $(p,q)$-degree would refute the theorem. A more direct check is that the specialization argument requires $\\mathrm{Gr}^W_4 H^3(G(2,L)_\\infty,\\mathbb{Q})$ to be one-dimensional, generated by $\\psi^\\infty_{2g}$; an explicit calculation of the monodromy invariant cycles in $H^3$ of the smooth fiber that produced a different dimension would also falsify the claim.","tokens_in":25048,"feed_emoji":"📐","tokens_out":12835,"duration_ms":114216,"temperature":0.7,"pith_summary":"This paper proves a generalization of Mumford's conjecture on the cohomology ring of moduli spaces of rank-2 sheaves, moving from smooth curves to an irreducible nodal curve with one node. For an odd-degree line bundle $L_0$, it identifies the rational cohomology ring of the Simpson moduli space $U_{X_0}(2,L_0)$ with a direct sum of monodromy-invariant primitive pieces tensored with polynomial rings in three even-degree classes modulo recursively defined Mumford ideals. The relations are exactly the classical smooth-curve relations, obtained by degeneration; the novelty is that the monodromy action selects which pieces survive to the singular central fiber. As a byproduct the paper computes the Hodge-Poincar\\'e polynomial of this singular moduli space, a quantity previously unknown.","feed_headline":"Mumford cohomology conjecture proven for nodal curves","feed_subtitle":"Rank-2 sheaf moduli on a genus-g nodal curve now have explicit generators, relations, and a Hodge–Poincaré polynomial.","key_machinery":"The load-bearing mechanism is the limit Mumford-Newstead isomorphism $\\tilde{\\Phi}_0: H^1(\\widetilde{X}_\\infty,\\mathbb{Q}) \\xrightarrow{\\sim} H^3(G(2,L)_\\infty,\\mathbb{Q})$ of Theorem 3.2, an isomorphism of mixed Hodge structures that shifts both Hodge and weight filtrations. It transports the symplectic basis $e_i$ of the limit curve's first cohomology to the degree-3 generators $\\psi^\\infty_i$, so the classical Newstead/Mumford structure on a nearby smooth fiber survives in the limit. Around this, the Steenbrink limit-weight spectral sequence for the simple-normal-crossings central fiber $G_{X_0}(2,L_0)$ computes the kernel of the specialization map, and a Gysin-morphism computation (Theorem 4.2) identifies that kernel with the cohomology of the moduli space of the normalized curve. The proper morphism $\\theta: G_{X_0}(2,L_0)\\to U_{X_0}(2,L_0)$ then transfers the resulting description from the Gieseker model to the Simpson moduli space.","core_discovery":"The paper's central claim is Theorem 7.2: for an irreducible nodal curve $X_0$ of genus $g\\geq 2$ with one node and an odd-degree invertible sheaf $L_0$, there is an isomorphism of graded rings $$H^*(U_{X_0}(2,L_0),\\mathbb{Q}) \\cong \\bigoplus_{i=0}^{g} \\frac{P^\\infty_i \\otimes \\mathbb{Q}[\\alpha_\\infty,\\beta_\\infty,\\psi_\\infty]}{I^\\infty_{g-i}},$$ where $P^\\infty_i$ are the monodromy-invariant primitive subspaces of $\\bigwedge^i H^3(G(2,L)_\\infty,\\mathbb{Q})$ with respect to the pairing $\\psi_\\infty$, and $I^\\infty_k$ are the recursive Mumford ideals generated by $\\zeta_k,\\zeta_{k+1},\\zeta_{k+2}$ defined by $\\zeta_0=1$ and $\\zeta_{k+1}=\\alpha_\\infty\\zeta_k+k^2\\beta_\\infty\\zeta_{k-1}+2k(k-1)\\psi_\\infty\\zeta_{k-2}$. Under this identification the ring is generated by $\\alpha_\\infty,\\beta_\\infty,\\psi^\\infty_1,\\ldots,\\psi^\\infty_{2g-1}$, together with $\\psi^\\infty_g\\psi^\\infty_{2g}$. The theorem also gives the Hodge-Poincar\\'e polynomial of $U_{X_0}(2,L_0)$, computed through the weight filtration: $$\\frac{(1+$xy^{2}$)^{g-1}(1+$x^{2}$y)^{g-1}(1+xy+$x^{3}$$y^{3}$)-x^g y^g(1+x)^{g-1}(1+y)^{g-1}(2+xy)}{(1-xy)(1-$x^{2}$$y^{2}$)}.$$","pith_inferences":["Iterating the same degeneration argument over a smoothing with several nodes should express the cohomology ring of a multi-nodal curve's moduli space as the monodromy-invariant part of the smooth fiber, with each node contributing an extra pair of even-degree generators subject to square-zero relations; the Hodge-Poincar\\'e polynomial would then acquire one factor per node.","The mixed-Hodge-structure isomorphism at the center of the proof suggests a motivic refinement: the motive of $U_{X_0}(2,L_0)$ should be a direct summand of the limit mixed motive of the Gieseker family, with the complementary summand accounted for by the normalization moduli space.","A testable extension is to rank $n\\ge 3$: the same degeneration should go through if one adds the extra relations known to be needed for smooth curves, so the monodromy-invariant part would yield the cohomology ring for rank-$n$ sheaves on a nodal curve.","The explicit description of the kernel of the Gysin morphism in Theorem 4.2 could be used to compute intersection-cohomology invariants of the singular moduli space, such as its intersection Betti numbers."],"forward_implications":["The cohomology ring of $U_{X_0}(2,L_0)$ is generated by $\\alpha_\\infty$, $\\beta_\\infty$, $\\psi^\\infty_i$ for $1\\le i\\le 2g-1$, and the single product $\\psi^\\infty_g\\psi^\\infty_{2g}$; all relations are of Mumford type, generated recursively by $\\zeta_k$.","The Hodge-Poincar\\'e polynomial of the singular moduli space is the explicit rational function in Theorem 7.2, and its weight-graded pieces are expressed through the Hodge numbers of a nearby smooth fiber and of the moduli space of the normalized curve.","The kernel and cokernel of the specialization map between the Gieseker central fiber and the limit cohomology are completely described: the kernel is a polynomial ring over $H^*(M_{\\widetilde{X}_0}(2,\\widetilde{L}_0),\\mathbb{Q})$ in two degree-2 classes $X,Y$ subject to $X^2=Y^2=X-Y=0$.","As mixed Hodge structures, $H^i(U_{X_0}(2,L_0),\\mathbb{Q})$ is identified with the weight-$i$ piece $W_iH^i(G(2,L)_\\infty,\\mathbb{Q})$, so the odd cohomology of the nodal moduli space has at most three nonzero weight-graded pieces.","The generalized conjecture reduces the singular case to the monodromy-invariant part of a smooth degeneration, so computing the monodromy action on the cohomology of nearby smooth fibers determines the cohomology of the nodal moduli space."],"supporting_citations":[{"why":"Proves Mumford's conjecture for smooth curves; supplies the relations that this paper degenerates.","marker":"[24]"},{"why":"Gives the limit Mumford-Newstead isomorphism of mixed Hodge structures used as Theorem 3.2.","marker":"[3]"},{"why":"Establishes the classical Mumford-Newstead isomorphism that the relative version extends.","marker":"[26]"},{"why":"Provides Newstead's generators of the cohomology ring of the smooth moduli spaces.","marker":"[28]"},{"why":"Gives the recursive relations and monomial bases used to describe the generic-fiber relations.","marker":"[23]"},{"why":"Constructs the Simpson moduli space of semistable sheaves with fixed determinant on a nodal curve.","marker":"[37]"},{"why":"Describes the irreducible components of the Gieseker central fiber and their projective bundle structures.","marker":"[38]"},{"why":"Supplies Gieseker's degeneration construction of the relative moduli space with normal-crossings central fiber.","marker":"[16]"},{"why":"Provides Schmid's limit mixed Hodge structures and the monodromy weight filtration used to define the limit objects.","marker":"[32]"},{"why":"Provides the limit weight spectral sequence relating central-fiber and limit mixed Hodge structures.","marker":"[36]"}],"fun_headline_variants":["Mumford's conjecture now extends to nodal curves","Explicit cohomology ring for nodal moduli spaces","Nodal curve case yields Mumford relations","Hodge-Poincare polynomial computed for nodal moduli"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Chern-class map identifying the first cohomology of the limit curve with the third cohomology of the limit Gieseker moduli space is an isomorphism of mixed Hodge structures, preserving both the Hodge and weight filtrations; if this identification failed, the generator classes, the primitive subspaces, and the weight filtration used in the proof would collapse, and the generalized Mumford decomposition would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Mumford's conjecture now extends to nodal curves","Explicit cohomology ring for nodal moduli spaces","Nodal curve case yields Mumford relations","Hodge-Poincare polynomial computed for nodal moduli"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001323,"raw_usage":{"total_tokens":5447,"prompt_tokens":1070,"completion_tokens":4377,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":686,"completion_tokens_details":{"reasoning_tokens":4314}},"tokens_in":686,"tokens_out":4377,"duration_ms":29083,"temperature":1.0,"reasoning_tokens":4314,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:17:07.430903+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the smallest case $g=2$, expand the claimed Hodge-Poincar\\'e polynomial and compare every coefficient with an independent computation of the Betti numbers of $U_{X_0}(2,L_0)$ from a normalization model or a point count over finite fields; a single mismatch in any $(p,q)$-degree would refute the theorem. A more direct check is that the specialization argument requires $\\mathrm{Gr}^W_4 H^3(G(2,L)_\\infty,\\mathbb{Q})$ to be one-dimensional, generated by $\\psi^\\infty_{2g}$; an explicit calculation of the monodromy invariant cycles in $H^3$ of the smooth fiber that produced a different dimension would also falsify the claim.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the limit Mumford-Newstead isomorphism of mixed Hodge structures used as Theorem 3.2."},{"cited_title":"Mumford and P","cited_arxiv_id":null,"evidence_quote":"Establishes the classical Mumford-Newstead isomorphism that the relative version extends."},{"cited_title":"Thaddeus","cited_arxiv_id":null,"evidence_quote":"Describes the irreducible components of the Gieseker central fiber and their projective bundle structures."}],"review_version":1}