Convexity and Zariski decomposition structure (Q343521): Difference between revisions

Zariski decomposition is a fundamental tool for studying linear systems on an algebraic surface, and its generalizations to divisors on higher dimensional varieties have played a relevant role in birational geometry. In the paper under review the authors study Zariski decomposition for big curve classes on a projective variety \(X\). In this setting, letting \(n=\dim X\), a Zariski decomposition for a big curve class \(\alpha\) (i.e.\ lying in the interior of the pseudo-effective cone of curves \(\overline{\text{Eff}}_1(X)\)) is a decomposition \((*)\) \ \(\alpha = B^{n-1}+\gamma\), where \(B\) is a big and nef \(\mathbb R\)-Cartier divisor class, \(\gamma\) is pseudo-effective and \(B \cdot \gamma=0\). In fact this notion generalizes Zariski's original definition, and the authors prove that any big curve class admits a unique Zariski decomposition. The authors derive it from a general Zariski decomposition theory in an abstract framework, by applying it to the natural duality between the nef cone of divisors on \(X\), \(\text{Nef}^1(X)\) and the cone \(\overline{\text{Eff}}_1(X)\). A sequel is announced, focusing on the duality between the pseudo-effective cone of divisors and the movable cone of curves, in which the \(\sigma\)-decomposition for divisor classes can also be interpreted in terms of their general theory. As to the general context, let \(\mathcal C\) be a full dimensional closed proper convex cone in a finite dimensional real vector space, and let \(\text{HConc}_s(\mathcal C)\) be the set of functions \(f:\mathcal C \to \mathbb R\) that are upper semi-continuous, homogeneous of weight \(s > 1\), strictly positive on the interior \(\mathcal C^{\circ}\) of \(\mathcal C\), and \(s\)-concave, in the sense that \(f(v)^{1/s} + f(x)^{1/s} \leq f(v+x)^{1/s}\) for any \(v,x \in \mathcal C\). Let \(\mathcal C^*\) be the dual cone of \(\mathcal C\). The polar transform (suggested by the Legendre--Fenchel transform of classical convex analysis) maps every \(f \in \text{HConc}_s(\mathcal C)\) to the function \(\mathcal Hf \in \text{HConc}_{s/s-1}(\mathcal C^*)\), defined as \(\mathcal Hf(w^*) = \text{inf}_{v \in \mathcal C^{\circ}} \big(\frac{w^* \cdot v}{f(v)^{1/s}}\big)^{s/s-1}\) for every \(w^* \in \mathcal C^*\). In this general setting the authors define what it means for \(f \in \text{HConc}_s(\mathcal C)\) to have a Zariski decomposition structure and prove, among other things, that this property is connected with a differentiability condition for \(\mathcal Hf\). Essentially, this structure can be viewed as a decomposition of the elements of \(\mathcal C^{\circ}\) into ``positive parts'' retaining the value of \(f\) and ``negative parts'', along which the strict log-concavity of \(f\) fails. In the specific case of a projective variety \(X\) of dimension \(n\), letting \(\mathcal C=\text{Nef}^1(X)\), \(\mathcal C^*=\overline{\text{Eff}}_1(X)\) and taking as \(f\) the volume function \(\text{vol}\) (in particular, \(\text{vol}(D) = D^n\) for any nef divisor \(D\)), it turns out that the polar transform \(\mathcal Hf\) is the function \(\widehat{\text{vol}}\) defined by the second author in [Math. Ann. 369, 1527--1555 (2017; Zbl 1379.32015)] as \(\widehat{\text{vol}}(\alpha) = \text{inf}_A \big(\frac{A \cdot \alpha}{\text{vol}(A)^{1/n}}\big)^{n/n-1}\), for every pseudo-effective curve class \(\alpha\), the infimum being taken for \(A\) ranging among the big and nef divisor classes. This point of view allows the authors to develop the theory of the volume function \(\widehat{\text{vol}}\), with particular attention to the log-concavity. Concerning the Zariski decomposition \((*)\) of the big curve class \(\alpha\), which derives from the general theory, it turns out that \(\widehat{\text{vol}}(\alpha)=\widehat{\text{vol}}(B^{n-1}) = \text{vol}(B)\). In particular, \(B\) computes \(\widehat{\text{vol}}(\alpha)\). Moreover, any big and nef divisor computing \(\widehat{\text{vol}}(\alpha)\) is proportional to \(B\). As the authors stress, the volume function for curves shares many properties of the volume function for divisors; in particular, a description of the derivative is provided, which mirrors results of \textit{S. Boucksom} et al. [J. Algebr. Geom. 18, No. 2, 279--308 (2009; Zbl 1162.14003)] and of \textit{R. Lazarsfeld} and \textit{M. Mustaţă} [Ann. Sci. Éc. Norm. Supér. (4) 42, No. 5, 783--835 (2009; Zbl 1182.14004)]. The function \(\widehat{\text{vol}}\) is of class \textit{C}\(^{\;1}\) on the big cone of curves. More precisely, let \(\alpha\) be a big curve class on \(X\) with Zariski decomposition \((*)\); then for any curve class \(\beta\) one has \(\frac{d}{dt}\big|_{t=0} \widehat{\text{vol}}(\alpha + t \beta)= \frac{n}{n-1} B \cdot \beta\). The final sections of the paper are devoted to several interesting geometric interpretations of the theory for varieties with some special structures.