Flops and equivalences of derived categories for threefolds with only terminal Gorenstein singularities. (Q1777969)

One of the main strategies in the birational classification theory of algebraic varieties is the so-called minimal model program (MMP). The underlying idea of the MMP is to find, in each birational equivalence class of varieties, some distinguished representatives (minimal models) which can be effectively used to study the birational properties of algebraic varieties in general. In dimension 2, the feasibility of the MMP is a classical fact, while the situation in higher dimensions appears to be much more complicated. In dimension 3, the MMP was completed by S. Mori in 1988, where it turned out that already in this case some additional birational operations, namely flops and flips, were needed to carry out the MMP. However, the MMP in dimension 3 is still rather specific, and cannot be generalized to higher dimensions without more ado. Therefore it seems to be highly desirable to find a more conceptual approach to the MMP in dimension 3 mainly with a view toward a better understanding of flops, which hopefully allows a generalization to higher dimensions. In this direction, a first break-through was recently achieved by \textit{A. I. Bondal} and \textit{D. O. Orlov} [Compos. Math. 125, No. 3, 327--344 (2001; Zbl 0994.18007)] and by \textit{T. Bridgeland} [Invent. Math. 147, No. 3, 613--632 (2002; Zbl 1085.14017)], who showed that the MMP in dimension 3 could be interpreted in the context of derived categories of coherent sheaves. According to them, the minimal model(s) of a given variety \(X\) might be regarded as certain minimal triangulated subcategories inside the derived category \(D^b(X)\). It is then very natural to interpret flops as replacing equivalent triangulated subcategories by each other, thereby gaining a better understanding of their general properties. In his recent paper cited above, T. Bridgeland studied smooth threefold flops within this framework. Using the smoothness assumption in an essential way, he constructed a fine moduli space of so-called perverse point sheaves for certain ``flopping contractions'' that exactly gave the usual flops. Moreover, he proved that the Fourier-Mukai transform given by the birational correspondence associated with a flop (in this interpretation) is actually an equivalence between the respective bounded derived categories. This long story of fascinating developments is the background of the paper under review. Namely, the author takes up the challenging question of whether Bridgeland's results are also valid for singular threefolds. His main result gives an affirmative answer to that question in the special case of quasi-projective threefolds with at most terminal Gorenstein singularities. In order to modify Bridgeland's approach in a suitable manner, the author's idea is that instead of studying the given singular threefold directly, one should pass over to a smoothing non-singular fourfold of it, make the categorical approach work there, and see how much information about the given singular threefold can be gained from this procedure. This delicate, also technically rather intricate program is carried out in great detail, as it is expected from an excellent Ph.D. thesis like the present one, and the author's main result (Theorem 1.1) reads as follows: Let \(X\) be a quasi-projective threefold with only terminal Gorenstein singularities, and let \(f: X\to Y\) a flopping contraction. Denote by \(W:= W(X/Y)\) the Bondal-Orlov-Bridgeland distinguished component of the moduli space of perverse point sheaves \(M(X/Y)\) and consider the associated canonical morphism \(g: W\to Y\). Then the following holds: (1) \(W\) has only terminal Gorenstein singularities. (2) The Fourier-Mukai transform \(\Psi: D^b(W)\to D^b(X)\) induced by the universal perverse point sheaves is an equivalence of derived categories. (3) The canonical morphism \(g: W\to Y\) is a flop of \(f: X\to Y\). T. Bridgeland's work cited above, which is both the main source and the main reference for the present paper, is briefly sketched in two appendices. This, together with the demonstrated high degree of clarity and rigor, makes the article under review very pleasant to read. All together, the author's fine work provides further evidence to support the categorical approach to the minimal model program in birational geometry.