Relative critical exponents, non-vanishing and metrics with minimal singularities (Q663298)

The article under review gives a direct proof of (a version of) the following theorem: if \((X,\Delta)\) is a klt pair with \(\Delta\) a big \(\mathbb{R}\)-divisor and \(K_X+\Delta\) pseudoeffective, then \(K_X+\Delta\) is \(\mathbb{R}\)-linearly equivalent to an effective divisor (Theorem D in [\textit{C. Birkar} et al., J. Am. Math. Soc. 23, No. 2, 405--468 (2010; Zbl 1210.14019)]). Working in the analytic setting, the author introduces a notion of relative threshold (relative critical exponent) for two closed positive currents with respect to a (singular) measure and is then able to prove a very general non vanishing statement. Theorem. Let \(X\) be a projective manifold and \(\alpha\in \mathrm{NS}_{\mathbb{R}}(X)\) a class in the real Neron-Severi space. Let us consider the following hypotheses: {\parindent=6mm \begin{itemize}\item[(i)] the adjoint class \(c_1(K_X)+\alpha\) contains a closed positive current \(\theta_{K_X+\alpha}\) (i.e. the class is pseudoeffective).\item[(ii)] the class \(\theta\) contains a Kähler current \(\theta_\alpha\) (\(i.e.\) the class is big) such that \(\exp(\varphi_{K_X+\alpha}-\varphi_\alpha)\in L^1\), where \(\varphi_{K_X+\alpha}\) and \(\varphi_\alpha\) denote the local weight of the corresponding currents (this is an analog of the klt condition). \end{itemize}} Under the assumptions (i) and (ii), the adjoint class \(c_1(K_X)+\alpha\) can be represented by an effective \(\mathbb{R}\)-divisor. In [Birkar, Cascini, Hacon and McKernan, loc. cit.], non-vanishing is obtained during the inductive process and is a consequence of the existence of log-minimal models of the pair \((X,\alpha)\). The proof given by M. Păun is by induction on the dimension of \(X\) and goes as follows. (1) A dichotomy according to the numerical dimension of \(c_1(K_X)+\alpha\): if \(\nu=0\) we are done using [\textit{S. Boucksom}, Ann. Sci. Éc. Norm. Supér. (4) 37, No. 1, 45--76 (2004; Zbl 1054.32010)]. In the case \(\nu\geq1\), the holomorphic Morse inequalities (and classical approximation arguments) provide a (1,1) current \(T\in m(c_1(K_X)+\alpha)+\alpha\) (for \(m\) large enough and fixed once for all) with logarithmic poles and singular enough at a very general point of \(X\). (2) Use blow-ups to obtain a SNC situation and then introduce the relative threshold of \(\theta_\alpha\) and of the logarithmic part of \(\theta_{K_X+\alpha}\). This step produces an analog of lc center: a smooth hypersurface \(S\) plus a closed positive current \(\Theta\in c_1(K_X+S)+\alpha\) (satisfying other positivity properties). (3) Restriction to \(S\) and induction; unfortunately the local potentials of \(\Theta\) could be identically \(-\infty\) along \(S\). The author remarks that the regularization process of [\textit{J.-P. Demailly}, J. Algebr. Geom. 1, No. 3, 361--409 (1992; Zbl 0777.32016)] can be used to overcome this difficulty. (4) Once the different objects involved have been restricted to \(S\), the induction hypothesis provides us with an \(\mathbb{R}\)-effective divisor in \(c_1(K_S)+\alpha_S\). A (by now) classical argument of Diophantine approximation combined with a trick of Shokurov is used to turn this effective divisor into a genuine section of a line bundle. (5) The last step is to extend this section to the whole of \(X\). The major problem is the small loss of positivity introduced by both approximation arguments: there is an effective limitation in the number of iterations allowed in the algorithmic method developed by the author in [\textit{M. Păun}, J. Differ. Geom. 76, No. 3, 485--493 (2007; Zbl 1122.32014)]. A detailed study of the situation shows that the number of iterations is precisely controlled by the loss of positivity which occurred in the previous step: the One-Tower algorithm can then be performed sufficiently many times to use at last an Ohsawa-Takegoshi type theorem to extend the section from \(S\) to \(X\). The proof of the main theorem is thus quite long but is decomposed in a series of well delimited steps, each of them being a combination of classical arguments and of clever tricks/observations of the author. Moreover the key points are clearly pointed out along the proof. The end of the paper is concerned with topics related with the main theorem: metrics with minimal singularities, finite generation of adjoint rings and non-vanishing in the line bundle case.