Flops of \(G\)-Hilb and equivalences of derived categories by variation of GIT quotient (Q1881729)

Let \(G\subset\text{SL}(3,\mathbb{C})\) be a finite group. A \(G\)-cluster is a \(G\)-invariant subscheme \(Z\subset \mathbb{C}^3\) of dimension zero with global sections \(H^0(\mathcal{O}_Z)\) isomorphic as a \(\mathbb{C}[G]\)-module to the regular representation \(R\) of \(G\). \textit{I. Nakamura} [J. Algebr. Geom. 10, No.4, 757--779 (2001; Zbl 1104.14003)] introduced the moduli space \(G\text{-Hilb}\) of \(G\)-clusters on \(\mathbb{C}^3\) as a natural candidate for a projective crepant resolution of \(\mathbb{C}^3/G\) and proved it for \(G\) abelian. \textit{T. Bridgeland, A. King} and \textit{M. Reid} [J. Am. Math. Soc. 14, No. 3, 535--554 (2001; Zbl 0966.14028)] subsequently proved the conjecture for all \(G\) by establishing an equivalence of derived categories: \(D(G\text{-Hilb})\sim D^G(\mathbb{C}^3)\). This paper generalises the notion of \(G\)-cluster: a \(G\)-constellation is a \(G\)-equivariant coherent sheaf \(F\) on \(\mathbb{C}^3\) with global sections \(H^0(F)\) isomorphic as a \(\mathbb{C}[G]\)-module to the regular representation \(R\) of \(G\). Set: \[ \Theta:=\left\{\theta\in\text{Hom}(R(G),\mathbb{Q})\mid \theta(R)=0\right\}. \] For \(\theta\in \Theta\), a \(G\)-constellation is said to be stable (resp. semistable) if every proper \(G\)-equivariant coherent subsheaf \(0\subset E\subset F\) satisfies \(\theta(E)>0=\theta(F)\) (resp. \(\geq\)). Generalizing ideas of \textit{A. V. Sardo-Infirri} [Resolutions of orbifold singularities and the transportation problem on the McKay quiver, preprint, \url{arXiv:alg-geom/9610005}] and \textit{A. D. King} [Q. J. Math., Oxf. II. Ser. 45, 515--530 (1994; Zbl 0837.16005)], the authors study the moduli spaces \(\mathcal{M}_{\theta}\) (resp. \(\overline{\mathcal{M}}_{\theta})\) of \(\theta\)-stable (resp. semistable) constellations. Note that \(G\text{-Hilb}\cong \mathcal{M}_{\theta}\) for parameter \(\theta\) in the cone \(\Theta_+:= \{\theta\in\Theta\mid \theta(\rho)>0\) if \(\rho\neq \rho_0\}\), where \(\rho_0\) denotes the trivial representation of \(G\). A parameter \(\theta\in \Theta\) is generic if every \(\theta\)-semistable \(G\)-constellation is \(\theta\)-stable. The method of Bridgeland-King-Reid generalises to show that: If \(\theta\) is generic, there is an equivalence of categories \(D(\mathcal{M}_{\theta})\sim D^G(\mathbb{C}^3)\) and \(\mathcal{M}_{\theta}\rightarrow \mathbb{C}^3/G\) is a projective crepant resolution of singularities. It is then natural to ask whether every projective crepant resolution may be realised as a moduli space \(\mathcal{M}_{\theta}\) for some parameter \(\theta\). The main result of this paper answers this question affirmatively in the abelian case: For a finite abelian subgroup \(G\subset \text{SL}(3,\mathbb{C})\), suppose that \(Y\rightarrow \mathbb{C}^3/G\) is a projective crepant resolution. Then \(Y\cong \mathcal{M}_{\theta}\) for some parameter \(\theta\). For generic \(\theta\), put \(C:=\{\eta\in \Theta\mid\) every \(\theta\)-stable \(G\)-constellation is \(\eta\)-stable\}. This is a convex polyhedral cone (or chamber) in \(\Theta\). The subset \(\Theta^{\text{gen}}\subset \Theta\) of generic parameter is open, dense and is the disjoint union of finitely many open convex polyhedral cones in \(\Theta\). For generic \(\theta\), the moduli space \(\mathcal{M}_{\theta}\) depends only upon the open chamber \(C\subset \Theta\) containing \(\theta\), so we write \(\mathcal{M}_C\) in place of \(\mathcal{M}_{\theta}\) for any \(\theta\in C\). Then the proof's idea is as follows: Since every projective crepant resolution is obtained by a finite sequence of flops from \(G\text{-Hilb}\), it is enough to show that, if \(Y\cong \mathcal{M}_C\) for some chamber \(C\), then for any flop \(Y'\) of \(Y\) there is a chamber \(C'\) (not necessarily adjacent to \(C\)) such that \(\mathcal{M}_{C'}\cong Y'\). Then the first step is to understand the walls of chambers in \(\Theta\) (\S 3) and then how the moduli \(\mathcal{M}_C\) changes as \(\theta\) passes through a wall from \(C\) to another chamber \(C'\). The method uses the description of chambers in terms of Fourier-Mukai transforms.