The birational geometry of moduli spaces of sheaves on the projective plane (Q476380): Difference between revisions

In order to state properly the results of the paper under review we need to recall some (actually, quite a lot of) definitions and facts. Let \(X\) denote complex projective plane \({\mathbb P}^2\), \(\text{D}^b(X)\) the bounded derived category of coherent sheaves on \(X\) and \(\text{K}(X)\) the Grothendieck group of \(X\) (which becomes a ring with \(1 = [{\mathcal O}_X]\) and \([F_1]\cdot [F_2] = [F_1\otimes^{\text{L}} F_2]\); recall, also, that the pairing \((u,v) \mapsto \chi(u\cdot v)\) is nondegenerate on \(\text{K}(X)_{\mathbb Q}\)). The Chern character\(\, :\) \[ \text{ch} = (r,\, c_1,\, \text{ch}_2 = \frac{1}{2}c_1^2 - c_2) : \text{K}(X)_{\mathbb Q} \rightarrow \text{H}^\ast(X,{\mathbb Q}) \] is, in our case, an isomorphism of rings. Fix a primitive element \(v = (r,\, c_1,\, \text{ch}_2)\) of the lattice \({\mathbb Z} \oplus \text{NS}(X) \oplus \frac{1}{2}{\mathbb Z}\), with \(r > 0\), and let \(M(v)\) denote the moduli space of Gieseker semistable torsion free sheaves on \(X\) of topological type \(v\) (with respect to the polarization \(H\) defined by the class of a line). If non-empty, \(M(v)\) is smooth and irreducible and the anti-canonical bundle \(-K_{M(v)}\) is nef and big , hence \(M(v)\) is a \textit{Mori dream space} and there is a finite rational polyhedra decomposition into \textit{Mori chambers and walls} of the pseudo-effective cone \({\overline{\text{NE}}}^1(M(v))\) according to the stable base locus of the divisors by \textit{C. Birkar} et al. [J. Am. Math. Soc. 23, No. 2, 405--468 (2010; Zbl 1210.14019)]. Moreover, by the work of \textit{J. M. Drezet} and \textit{J. Le Potier} [Ann. Sci. Éc. Norm. Supér. (4) 18, 193--243 (1985; Zbl 0586.14007)] and \textit{J.-M. Drezet} [Ann. Inst. Fourier 38, No. 3, 105--168 (1988; Zbl 0616.14006)], \(M(v)\) has Picard number 2, in general, and 1 for some particular \(v\)'s. On the other hand, for every \((s,t) \in {\mathbb R}\times {\mathbb R}_+\) one defines a \textit{Bridgeland stability condition} (see \textit{T. Bridgeland} [Duke Math. J. 141, No. 2, 241--291 (2008; Zbl 1138.14022)]) \(({\mathcal A}_s \subset \text{D}^b(X),\, Z_{s,t} : \text{K}(\text{D}^b(X)) \rightarrow {\mathbb C})\) as follows\(\, :\) let \({\mathcal Q}_s\) be the full subcategory of \(\text{Coh}(X)\) consisting of sheaves \(Q\) which are either torsion or with \(\mu_{\text{min}}(Q/Q_{\text{tors}}) > s\) and let \({\mathcal F}_s\) consists of the torsion free coherent sheaves \(F\) with \(\mu_{\text{max}}(F) \leq s\). Here \(\mu := c_1\cdot H/r\) is the \textit{Mumford slope} and \(\mu_{\text{min}}\) and \(\mu_{\text{max}}\) are defined with respect to the Harder-Narasimhan filtration. Then\(\, :\) \[ \begin{gathered} {\mathcal A}_s := \{E \in \text{D}^b(X)\, | \, {\mathcal H}^{-1}(E) \in {\mathcal F}_s, \, {\mathcal H}^0(E) \in {\mathcal Q}_s,\, {\mathcal H}^i(E) = 0\;\text{otherwise}\}\\ Z_{s,t}(E) := - \text{deg}\left(e^{-(s+it)H}\text{ch}(E)\right)_2 \end{gathered} \] and the associated slope function is \[ \mu_{s,t}(E) = \frac{\text{ch}_2 - sc_1 + \frac{1}{2}r(s^2 - t^2)}{t(c_1 - rs)} \] (identifying \(c_1 \in \text{NS}(X)\) with \(c_1 \cdot H \in {\mathbb Z}\)). By \textit{D. Abramovich} and \textit{A. Polishchuk} [J. Reine Angew. Math. 590, 89--130 (2006; Zbl 1093.14026)] the moduli stack \(M_{s,t}(v)\) of \(\mu_{s,t}\)-stable objects of \({\mathcal A}_s\) is proper (recall that \(v\) is primitive). \textit{D. Arcara} et al. [Adv. Math. 235, 580--626 (2013; Zbl 1267.14023)] showed that there exists an integer \(k\) such that, for any \(\mu_{s,t}\)-stable object \(E\) of \({\mathcal A}_s\), either \(E\) or \(E[1]\) is of the form\(\, :\) \[ {\mathbb C}^{n_0}\otimes{\mathcal O}_{{\mathbb P}^2}(k-2) \longrightarrow {\mathbb C}^{n_1}\otimes{\mathcal O}_{{\mathbb P}^2}(k-1) \longrightarrow {\mathbb C}^{n_2}\otimes{\mathcal O}_{{\mathbb P}^2}(k)\, . \] Moreover, in the case where \(E\) is of that form, there exists \(\vec{a} = (a_0,a_1,a_2) \in {\mathbb Z}^3\) with \(a_0n_0 + a_1n_1 + a_2n_2 = 0\) such that a complex of the above form is a \(\mu_{s,t}\)-stable object of \({\mathcal A}_s\) if and only if it is \textit{quiver stable} with respect to \(\vec{a}\) which means that for every proper subcomplex \[ {\mathbb C}^{d_0}\otimes{\mathcal O}_{{\mathbb P}^2}(k-2) \longrightarrow {\mathbb C}^{d_1}\otimes{\mathcal O}_{{\mathbb P}^2}(k-1) \longrightarrow {\mathbb C}^{d_2}\otimes{\mathcal O}_{{\mathbb P}^2}(k)\, . \] one has \(a_0d_0 + a_1d_1 + a_2d_2 < 0\). Then, by work of \textit{A. D. King} [Q. J. Math., Oxf. II. Ser. 45, No. 180, 515--530 (1994; Zbl 0837.16005)], \(M_{s,t}(v)\) is projective. Arcara et al. [Zbl 1267.14023] also show that if \(s < \mu\) and \(t \gg 0\) then \(M_{s,t}(v)\) is isomorphic to \(M(v)\). Now, there exists a decomposition of \({\mathbb R} \times {\mathbb R}_+\) into \textit{Bridgeland chambers and walls} such that, in the same chamber, the moduli spaces \(M_{s,t}(v)\) are isomorphic, while crossing a wall means that \(M_{s,t}(v)\) changes. More precisely, if \(v^\prime\) is another element of \({\mathbb Z} \oplus \text{NS}(X) \oplus \frac{1}{2} {\mathbb Z}\), not proportional to \(v\), then one defines, in \({\mathbb R} \times {\mathbb R}_+\), the \textit{potential wall}\(\, :\) \[ W_{v,v^\prime} := \{(s,t) \, | \, \mu_{s,t}(v) = \mu_{s,t}(v^\prime)\}\, . \] Potential walls are nested semicircles. A potential wall \(W_{v,v^\prime}\) is an \textit{actual wall} if there exist \(\mu_{s,t}\)-semistable objects \(A \subset E\) of \({\mathcal A}_s\) with \(\text{ch}(E) = v\) and \(\text{ch}(A) = v^\prime\). Choose an integral element \(w_{s,t}\) of \(\text{H}^\ast(X,{\mathbb R})_{\text{alg}}\), perpendicular to \(v\) and \(v^\prime\) under the non-degenerate pairing defined by \(\chi\) and oriented such that \(\chi(w_{s,t} \cdot \text{ch}(C)) > 0\), \(\forall \, C \in {\mathcal A}_s\) with \(\mu_{s,t}(C) < \mu_{s,t}(v)\). Choose, also, a complex \(F_{s,t}\) such that \(\text{ch}(F_{s,t}) = w_{s,t}\). First of all, the authors of the paper under review provide a formula for the quiver polarization \(\vec{a}\), referred to above, namely they show that one can take \(a_i = (-1)^i\chi(F_{s,t}(k-2+i))\), \(i = 0,\, 1,\, 2\). Then they show that if \[ {\mathcal E}\, :\, {\mathcal O}_{{\mathbb P}^2}(k-2) \boxtimes U \longrightarrow {\mathcal O}_{{\mathbb P}^2}(k-1) \boxtimes V \longrightarrow {\mathcal O}_{{\mathbb P}^2}(k) \boxtimes W \] is a flat family of \(\mu_{s,t}\)-stable objects of \({\mathcal A}_s\) of topological type \(v\) on \({\mathbb P}^2 \times S\) then\(\, :\) \[ \text{det}\text{R}p_{2\ast}(p_1^\ast F_{s,t}\otimes {\mathcal E}) \simeq (\overset{n_0}{ \bigwedge} U)^{\otimes \, a_0} \otimes (\overset{n_1}{ \bigwedge} V)^{\otimes \, a_1} \otimes (\overset{n_2}{ \bigwedge} W)^{\otimes \, a_2} \, . \] One gets, in this way, an ample line bundle \(\lambda_{s,t}\) on \(M_{s,t}(v)\). On the other hand, if \({\mathcal E}_{\text{univ}}\) is the universal family on \({\mathbb P}^2 \times M(v)\) then one can consider the line bundle \[ \lambda(w_{s,t}) := \text{det}\text{R}p_{2\ast}(p_1^\ast F_{s,t}\otimes {\mathcal E}_{\text{univ}}) \] on \(M(v)\). Let \(M_{s,t}^P(v)\) denote the main component of \(M_{s,t}(v)\) whose generic point corresponds to a sheaf (not to a complex of sheaves). For \(s < \mu\) and \(t \gg 0\) one has \(M_{s,t}^P(v) \simeq M(v)\) and the restriction of \(\lambda_{s,t}\) to \(M_{s,t}^P(v)\) is isomorphic to \(\lambda(w_{s,t})\). Choose an \(s < \mu\) such that the ray \(\{s\} \times {\mathbb R}_+\) intersects all actual walls. The authors of the paper under review assert that if one decreases \(t\) then \(\lambda(w_{s,t})\) moves in \(\text{N}^1(M(v))_{\mathbb R}\) ``toward the side of the nef cone opposite to \(-K_{M(v)}\). This corresponds to running a directed MMP on \(M(v)\) and we get \(M_{s,t}^P(v)\) as birational models of \(M(v)\)''. Then they analyse what happens when an actual wall is crossed. Assume that the ray \(\{s\} \times {\mathbb R}_+\) hits an actual wall at \((s,t_0)\) and choose \(t_0^- < t_0 < t_0^+\) such that \((s,t_0^-)\) and \((s,t_0^+)\) belong to adjacent chambers. The main technical tool used by the authors in their analysis of wall crossings is a lemma asserting that the dimension of the exceptional locus of the contraction \(\pi_0^- : M_{s,t_0^-}(v) \rightarrow M_{s,t_0}(v)\) is strictly less than the dimension of the exceptional locus of the contraction \(\pi_0^+ : M_{s,t_0^+}(v) \rightarrow M_{s,t_0}(v)\). Most of the above results work, also, for moduli spaces of torsion sheaves of pure dimension 1 on \({\mathbb P}^2\). Using the results of \textit{J.-M. Drézet} and \textit{M. Maican} [Geom. Dedicata 152, 17--49 (2011; Zbl 1236.14012)] and of \textit{M. Maican} [Ill. J. Math. 55, No. 4, 1467--1532 (2011; Zbl 1273.14027)], the authors of the paper under review describe explicitly the flips that appear in the above described process for the topological types \(v = (0,\, 4,\, -4)\) and \(v = (0,\, 5,\, -\frac{15}{2})\). One should also mention that the case \(v = (1,\, 0,\, -n)\), corresponding to the Hilbert scheme \({\mathbb P}^{2[n]}\), has been extensively studied in the paper of Arcara et al. [Zbl 1267.14023].