Variation of Gieseker moduli spaces via quiver GIT (Q309030)

For a smooth projective complex surface \(X\) and ample class \(L\) on \(X\), the moduli space \({\mathcal M}_L^\mu\) of slope-semistable torsion-free sheaves of a given topological type with large \(c_2\) is irreducible and generically smooth. Moreover, there exists a chamber decomposition of the ample cone with a locally finite collection of rational linear walls such that \({\mathcal M}_L^\mu\) does not change within chambers, while wall-crossings are described by birational flips. There is a similar picture for Gieseker semistable sheaves. When \(\dim X\geq3\), this picture no longer applies. The walls are no longer linear and may contain no rational points at all. To address this problem, the authors of this paper adopt a new strategy. The idea is to work with a stability condition which involves a finite number of ample line bundles and depends on some real parameters. Rather than varying a single ample line bundle, the line bundles can be fixed and the parameters varied. Under a suitable boundedness condition, a good chamber decomposition exists with the wall-crossings described by Thaddeus flips. This works in particular when \(X\) is a smooth projective threefold. The approach depends on a new concept of stability, which the authors call multi-Gieseker stability. Let \(X\) be a smooth projective variety and \(\underline{L}=(L_1,\ldots,L_{j_0})\) a finite collection of ample line bundles on \(X\). Suppose further that \(\sigma=(\sigma_1,\ldots,\sigma_{j_0})\) is a non-zero vector of non-negative real numbers. A torsion-free coherent sheaf on \(X\) is said to multi-Gieseker (semi)stable (or just (semi)stable) with respect to \((\underline{L},\sigma)\) if, for all proper subsheaves \(F\) of \(E\), the inequality \[ \frac{\sum_j\sigma_j\chi(F\otimes L_j^m)}{\text{rank}(F)}(\leq)<\frac{\sum_j\sigma_j\chi(E\otimes L_j^m)}{\text{rank}(E)} \] holds for all sufficiently large \(m\). Of course, this gives the usual Gieseker stability condition when \(j_0=1\). (In fact, the authors give a version of this definition which is valid for any pure coherent sheaf (Definition 2.5).) The new definition satisfies many of the standard properties of Gieseker stability -- existence of a Harder-Narasimhan filtration, Jordan-Hölder filtrations of semistable bundles (hence, S-equivalence), openness of the (semi)stability condition. Multi-Gieseker stability can also be related to a suitable concept of slope stability (Lemma 3.3). The next step is the introduction of chamber structures. For this, one needs two definitions. The topological type of a sheaf \(E\) is the homological Todd class \(\tau_X(E)\) regarded as an element of \(B(X)\otimes_{\mathbb{Z}}\mathbb{Q}\), where \(B(X)\) is the group of cycles on \(X\) modulo algebraic equivalence (Definition 1.4). If all \(L_j\) are very ample and \(p\) is an integer, one says that a coherent sheaf \(E\) is \((p, \underline{L})\)-regular if \(H^i(E\otimes L_j^{p-i})=0\) for \(1\leq j\leq j_0\) (Definition 1.7). Now fix \(\underline{L}\) and consider variations in \(\sigma\). Let \(X\) be an integral projective scheme and \(\Sigma\subset \mathbb{R}^{j_0}_{\geq0}\setminus\{0\}\) a convex subset. Then, for all integers \(p\), the set \(\Sigma\) admits a rational linear chamber structure cut out by finitely many walls with the usual properties for \((p,\underline{L})\)-regular torsion-free sheaves of topological type \(\tau\) (see Section 4 and Proposition 4.2 for a detailed description). Part II of the paper is concerned with the construction of moduli spaces. The first step is to establish a functorial approach using the ideas of \textit{L. Álvarez-Cónsul} and \textit{A. King} [Invent. Math. 168, No. 3, 613--666 (2007; Zbl 1137.14026)]. The quiver \(Q\) that will be used and its representations are described, followed by the introduction of a stability condition analogous to multi-Gieseker stability, which is then linked to GIT stability as in [\textit{A. D. King}, Q. J. Math., Oxf. II. Ser. 45, No. 180, 515--530 (1994; Zbl 0837.16005)]. For \(X\) a projective scheme of finite type over an algebraically closed field of characteristic zero, the functor \(E\mapsto \text{Hom}(T,E)\), where \(T:=\bigoplus_{j=1}^{j_0}L_j^{-n}\oplus L_j^{-m}\) with \(m\gg n\) is shown to be fully faithful on the full subcategory of \((n,\underline{L})\)-regular sheaves of topological type \(\tau\) (Theorem 5.7). The extension of this result to families of sheaves follows, together with the identification of the image of the functor (Proposition 5.9). Next, one needs to establish a boundedness result. For this, \(X\) can be a smooth projective variety over an algebraically closed field of arbitrary characteristic. After fixing \(\underline{L}\) and \(\tau\), we say that a set \(\Sigma\subset \mathbb{R}^{j_0}_{\geq0}\setminus\{0\}\) is bounded if the set of all torsion-free sheaves of topological type \(\tau\) which are semistable with respect to \((\underline{L},\sigma)\) for some \(\sigma\in\Sigma\) is bounded. The positive cone \(\text{Pos}(X)_{\mathbb{R}}\) is defined as the set of classes \(\gamma\in N_1(X)\otimes_{\mathbb{Z}}\mathbb{R}\) of the form \(\gamma=D^{\dim X-1}\) with \(D\) in the real ample cone. The key boundedness result (Corollary 6.11) then states that, if \(\Sigma\subset \mathbb{R}^{j_0}_{\geq0}\) is a closed convex polyhedral cone with the origin removed and \(\sum_j\sigma_jc_1(L_j)^{n-1}\in\text{Pos}(X)_{\mathbb{R}}\) for all \(\sigma\in\Sigma\), then \(\Sigma\) is a bounded set of stability parameters with respect to \(\tau\) and \(\underline{L}\). In three cases (sheaves of rank \(\leq2\), \(\dim X\leq3\), \(X\) has Picard rank at most \(2\)), one can take \(\Sigma=\mathbb{R}^{j_0}_{\geq0}\setminus\{0\}\) (Corollary 6.12). The authors next prove a version of a theorem due to Le Potier and Simpson showing that semistability of a sheaf can be detected by the spaces of sections of its subsheaves (after twisting by \(L_j^n\) for sufficiently large \(n\)) (Theorem 7.2). This and further work then allow the crucial comparison of stability of a sheaf \(E\) and stability of the module \(\text{Hom}(T,E)\). In fact, for all integers \(m\gg n\gg p\gg 0\), a sheaf \(E\) of topological type \(\tau\) is semistable if and only if it is pure, \((p,\underline{L})\)-regular and \(\text{Hom}(T,E)\) is semistable (Theorem 8.1/8.16). Moreover, if \(\sigma\) is positive, \(\text{Hom}(T,\text{gr}E)\simeq\text{gr}\text{Hom}(T,E)\) (Theorem 8.20). The techniques of [Álvarez-Cónsul and King, loc. cit.] now prove that there exists a quasi-projective scheme \(\mathcal{M}_{\underline{L}}^{\sigma-ss}(\tau)\) which corepresents the moduli functor of flat familiies of \((\underline{L},\sigma)\)-semistable sheaves of topological type \(\tau\) and whose closed points correspond to S-equivalence classes of semistable sheaves. Moreover, there is an open subscheme \(\mathcal{M}_{\underline{L}}^{\sigma-s}(\tau)\subset\mathcal{M}_{\underline{L}}^{\sigma-ss}(\tau)\) which corepresents the moduli functor of flat families of stable sheaves and whose closed points correspond to isomorphism classes of stable sheaves. All this is contained in Theorem 9.4. In fact (Theorem 9.6), \(\mathcal{M}_{\underline{L}}^{\sigma-ss}(\tau)\) is proper and therefore projective as a consequence of Langton's theorem; the statement and proof of this result use the assumption that \(\sigma\) is a rational stability parameter, but the authors remark that, as a consequence of the chamber structure, this assumption is not necessary. Part III of the paper consists of three applications. The first involves the construction of a master space for the variation of the moduli space with \(\sigma\) (Theorem 10.1). As a consequence (Corollary 10.2), for any bounded positive stability parameters \(\sigma\) and \(\sigma'\), the moduli spaces \(\mathcal{M}_{\underline{L}}^{\sigma-ss}(\tau)\) and \(\mathcal{M}_{\underline{L}}^{\sigma'-ss}(\tau)\) are related by a finite sequence of Thaddeus flips. The second application is to Gieseker stability with respect to real ample classes and asserts the existence of a projective moduli space for torsion-free sheaves of topological type \(\tau\) on a smooth projective threefold which are \(\omega\)-semistable for some real ample class \(\omega\) (Theorem 11.6). Finally, let \(X\) be a smooth projective threefold over an algebraically closed field of characteristic zero, let \(\tau\) be a topological type and let \(L_1\), \(L_2\) be real ample classes. Then the moduli spaces \(\mathcal{M}_{L_1}\) and \(\mathcal{M}_{L_2}\) of sheaves of topological type \(\tau\) which are Gieseker semistable with respect to \(L_1\) and \(L_2\), respectively, are related by a finite number of Thaddeus flips (Theorem 12.1). This last was a principal motivation for the paper. Apart from giving background and describing the content of the paper, the introduction contains a useful comparison with other works.