Terminal Fano threefolds and their smoothings (Q658344)

Let \(X\) be a (almost) Fano threefold with \textit{terminal} Gorenstein singularities (\(-K_X\) is always at least nef and big). Then, due to \textit{T. Minagawa} [Osaka J. Math. 38, No. 3, 533--540 (2001; Zbl 1073.14540)] and \textit{Y. Namikawa} [J. Algebr. Geom. 6, No. 2, 307--324 (1997; Zbl 0906.14019)], there exists a \textit{smoothing} of \(X\), i.e., a flat projective morphism \(\pi: \mathcal{X} \to \Delta\), where \(\mathcal{X}\) is a reduced complex space, \(\Delta\ni 0\) is a unit disk, \(\mathcal{X}_0 := \pi^{-1}(0) = X\), and \(\mathcal{X}_t := \pi^{-1}(t)\) is smooth ((almost) Fano) for all \(t\neq 0\). The main result of the paper under review (see Theorem 1.4 in the text) asserts that \(\text{Pic}(\mathcal{X})\simeq\text{Pic}(\mathcal{X}_t)\) for all \(t\). Let us sketch the authors' argument. Firstly, one may notice (Proposition 1.1) that \(\mathcal{X}\) is normal, with at most isolated terminal factorial singularities. This, together with Kawamata--Viehweg vanishing Theorem and Leray spectral sequence gives \(\text{Pic}(X) \simeq H^2(X, \mathbb{Z}) \simeq \text{Pic}(\mathcal{X}) \simeq H^2(\mathcal{X}, \mathbb{Z})\) (note also that the inclusion \(X \hookrightarrow \mathcal{X}\) is a deformation retract). So it suffices to show the natural restriction map \(r: H^2(\mathcal{X}, \mathbb{Z}) \to H^2(\mathcal{X}_t, \mathbb{Z})\), \(t \neq 0\), is an isomorphism. Note that the injectivity of \(r\) is obvious. Further, after a base change one may assume the monodromy of \(\pi\) to be trivial. In particular, if \(L_t\) is a line bundle on \(\mathcal{X}_t\), we may suppose \(L_t\) is defined on \(\mathcal{X}^* := \mathcal{X}\setminus{X}\). One thus has to extend \(L_t\) to \(\mathcal{X}\). Let \(\mathcal{Y}\) be a log resolution of \((\mathcal{X}, X)\) and \(\mathcal{W}\) a semistable reduction of \(\mathcal{Y}\). Then the pullback of \(L_t\) to \(\mathcal{W}^*\) is the restriction of a class \(L_{\mathcal{W}} \in H^2(\mathcal{W}, \mathbb{Q})\) (by the \textit{Local Invariant Cycle Theorem}). Now, as the rational map \(\mathcal{W} \dashrightarrow \mathcal{X}\) is Galois over \(\mathcal{X}_{\text{reg}}\), the Galois trace of \(L_{\mathcal{W}}\) determines a class \(\xi \in H^2(\mathcal{X}_{\text{reg}}, \mathbb{Q})\). The latter extends \(L_t\) to \(\mathcal{X}_{\text{reg}}\). Furthermore, since \(\mathcal{X}_{\text{reg}}\) is Cohen-Macaulay, we have \(H^2(\mathcal{X}_{\text{reg}}, \mathcal{O}_{\mathcal{X}_{\text{reg}}}) = 0\), which together with the exponential exact sequence shows that a multiple of \(\xi\) comes from a line bundle on \(\mathcal{X}_{\text{reg}}\). But \(\mathcal{X}\) is normal and factorial. So a multiple of \(\xi\) induces a line bundle on \(\mathcal{X}\). Now it is easy to see that \(L_t\) also extends to \(\mathcal{X}\). In Section 3, the authors consider examples of Fano threefolds with \textit{canonical} Gorenstein singularities. They construct non-smoothable such Fano (\(\mathbb{P}(3,1,1,1)\) in Example 3.1), one for which the Picard number jumps in the smoothing (\(\mathbb{P}^1 \times \mathbb{P}(1,1,2)\) in Example 3.2), and also a Fano with non-unique smoothing (cone in \(\mathbb{P}^7\) over del Pezzo surface of degree \(6\) in Example 3.3).