A note on nonvanishing and applications (Q1320643)

Let \(X \subset \mathbb{P}^N\) be a smooth projective variety of dimension \(n\) defined over the field of complex numbers. By \(L\) let us denote the restriction of \({\mathcal O}_{\mathbb{P}^N} (1)\) to \(X\), and by \(K_X\) let us denote the canonical divisor of \(X\). A (classical) version of adjunction theory concerns the studying of an adjoint linear system \(|K_X + rL |\) for a suitable chosen positive integer \(r\). In particular, a typical problem is to decide whether the adjoint divisor \(K_X + rL\) is nef or for which value of \(r\) it becomes ample or even very ample. If \(K_X + rL\) is nef, one may want to find out whether the linear system \(|K_X + rL |\) has base points or for which positive integer \(m\) its multiple \(|m (K_X + rL) |\) becomes base-point free. The most interesting case concerns the situation when the adjoint divisor \(K_X + rL\) is nef but not ample. Then, although not much can be said about the system \(|K_X + rL |\) itself, a Kawamata-Shokurov contraction theorem asserts that some of its multiple, \(|m (K_X + rL) |\), is base-point free for \(m \gg 0\) and defines an adjoint contraction morphism \(\varphi : X \to Z\) onto a normal projective variety \(Z\), with \(\varphi_* {\mathcal O}_X = {\mathcal O}_Z\). Understanding this map seems to be very important for any classification theory of higher-dimensional manifolds. In the present paper, we study the situation when \(L\) is merely ample and the adjoint contraction morphism has fibers of ``small'' dimension. This last hypothesis allows us to apply an inductive method which is typical of this theory, called ``Apollonius method'' by \textit{T. Fujita} in his book: ``Classification theories of polarized varieties'', Lond. Math. Lect. Note Ser. 155 (1990; Zbl 0743.14004). In the present paper we call it a horizontal slicing argument; it can be briefly summarized as follows. Consider a general divisor \(X'\) from the linear system \(|L |\) (a hyperplane section of \(X\) if \(L\) is very ample), and assume that it is a ``good'' variety of dimension \(n - 1\) (i.e., it has the same singularities as \(X)\). By adjunction, \(K_{X'} = (K_X + L) \mid X'\), and by the Kodaira-Kawamata-Viehweg vanishing theorem, if \(r > 1\), the linear system \(|m (K_{X'} + (r - 1) L) |\) is just the restriction of \(|m (K_X + rL) |\), so that the adjoint contraction morphism of \(X'\) can be related to the one of \(X\). Moreover, fibers of the adjoint morphism of \(X'\) will be usually of smaller dimension, and an inductive argument can be applied. The Kawamata-Kollár method together with the horizontal slicing allows us to extend the nonvanishing to ``relative-spannedness''. Namely we prove the following: Theorem. Let \(X\) be a normal variety such that \(K_X\) is \(\mathbb{Q}\)-Cartier and \(X\) has at worst log-terminal singularities; let \(L\) be an ample line bundle over \(X\). Let \(\varphi : X \to Z\) be an adjoint contraction supported by \(K_X + rL\). Let \(F\) be a fiber of \(\varphi\). Assume moreover that either \(\dim F < r + 1\) if \(\dim Z < \dim X\), or \(\dim F \leq r + 1\) if \(\varphi\) is birational. Then the evaluation morphism \(\varphi^* \varphi_*L \to L\) is surjective at every point of \(F\). As a consequence of this, in the above hypothesis on the dimension of the fibers, one can apply the horizontal slicing (or Apollonius) argument to varieties with log-terminal singularities.