Flops and mutations for crepant resolutions of polyhedral singularities (Q525711)

Suppose that \(G\subset\mathrm{SO}(3)\) is a polyhedral group of type \(\mathbb Z/n\mathbb Z\), \(D_{2n}\) or tetrahedral. By the result of Bridgeland-King-Reid [\textit{T. Bridgeland} et al., J. Am. Math. Soc. 14, No. 3, 535--554 (2001; Zbl 0966.14028)] and also Gomi-Nakamura-Shinoda [\textit{Y. Gomi} et al., Asian J. Math. 4, No. 1, 51--70 (2000; Zbl 0981.14002)] it is known that \(G\)-\(\text{Hilb}(\mathbb C^3)\) is always a projective crepant resolution of \(\mathbb{C}^3/G\) and the fiber over \(0\) is 1-dimensional with irreducible rational components in bijection with nontrivial irreducible representations of \(G\). By the result of \textit{Y. Ito} and \textit{H. Nakajima} [Topology 39, No. 6, 1155--1191 (2000; Zbl 0995.14001)] \(G\)-\(\text{Hilb}(\mathbb C^3)\) can be realized as a moduli space of stable representations of the McKay quiver \((Q,W)\) with suitable relations (arising from the potential \(W\)). The space of generic stability conditions of the McKay quiver is the disjoint union of finitely many chambers. For a given chamber \(C\) let \(\mathcal M_C\) be the corresponding moduli space. Bridgeland-King-Reid [Zbl 0966.14028] more generally prove that each \(\mathcal M_C\) is in fact a projective crepant resolution of \(\mathbb{C}^3/G\). \textit{A. Craw} and \textit{A. Ishii} [Duke Math. J. 124, No. 2, 259--307 (2004; Zbl 1082.14009)] conjectured that conversely every projective crepant resolution of \(\mathbb{C}^3/G\) is of the form \(\mathcal M_C\) for some chamber \(C\). This conjecture is proven when \(G\) is abelian. The paper under review proves this conjecture for any \(G\) as above. It moreover proves that there exists a one-to-one correspondence between flops of \(G\)-\(\text{Hilb}(\mathbb C^3)\) and those mutations of \((Q,W)\) which do not mutate the trivial vertex. As a result to each \(\mathcal M_C\) above is associated an iterated quiver \(Q_C\). The paper under review proves that the dual graph of the fiber over \(0\) of the crepant resolution \(\mathcal M_C\) is the same as the graph of \(Q_C\) with the trivial vertex removed, and that the number of loops at a vertex of \(Q_C\) determines the normal bundle of the corresponding rational component of the fiber.