Abundance theorem for surfaces over imperfect fields (Q2182428)

The author proves the following log abundance theorem for surfaces in positive characteristic. Theorem 1.1. Let \(k\) be a field of positive characteristic. Let \((X,\Delta)\) be a log canonical surface over \(k\), where \(\Delta\) is an effective \(\mathbb{R}\)-divisor. Let \(f:X\rightarrow S\) be a proper \(k\)-morphism to a scheme \(S\) which is separated and of finite type over \(k\). If \(K_X+\Delta\) is nef, then \(K_X+\Delta\) is \(f\)-semi-ample. Over algebraically closed fields of positive characteristic, Theorem 1.1 is due to \textit{E. Bombieri} and \textit{D. Mumford} [in: Complex Anal. Algebr. Geom., Collect. Pap. dedic. K. Kodaira, 23--42 (1977; Zbl 0348.14021)] and \textit{T. Fujita} [J. Fac. Sci., Univ. Tokyo, Sect. I A 30, 685--696 (1984; Zbl 0543.14004)]. The main point of this paper is therefore to treat the case in which \(k\) is imperfect. This generalization is relevant even if one is only interested in varieties defined over algebraically closed fields of positive characteristic. In fact, when studying families of smooth varieties, it is often more natural to look at the generic fiber rather than a general fiber, due to the fact that the latter is not necessarily smooth. Roughly speaking, the proof of Theorem 1.1 goes as follows. Suppose for simplicity that \(\Delta = 0\), \(k\) is separably closed and \(S=\operatorname{Spec}(k)\). Consider the diagram \[ f:Y\rightarrow(X\times_k \bar{k})_{\operatorname{red}} \rightarrow X\times_k \bar{k} \rightarrow X \] where \(Y\) is the normalization of \((X\times_k \bar{k})_{\operatorname{red}}\). An important consequence of this construction is the existence of an effective integral Weil divisor \(E\) such that \(K_Y + E = f^* K_X\). The author then applies the Minimal Model Program for surfaces in positive characteristic to \(Y\) and concludes with a case by case analysis. The first half of the paper is devoted to proving Theorem 1.1 for klt pairs, whereas the second half treats the log canonical case.