Foliations with positive slopes and birational stability of orbifold cotangent bundles (Q2418564)

Let \(X\) be a complex projective manifold of dimension \(n\), and denote by \(\Omega_X\) its cotangent bundle. If the manifold \(X\) is not covered by rational curves, a classical theorem of \textit{Y. Miyaoka} [Adv. Stud. Pure Math. 10, 449--476 (1987; Zbl 0648.14006)] states that for a curve \(C \subset X\) that is a general complete intersection of sufficiently ample divisors, the restriction \(\Omega_X|_C\) of the cotangent bundle to the curve is a nef vector bundle. By a theorem of \textit{S. Boucksom} et al. [J. Algebr. Geom. 22, No. 2, 201--248 (2013; Zbl 1267.32017)] the hypothesis in the theorem of Miyaoka-Mori is equivalent to the pseudoeffectivity of the canonical bundle \(K_X:=\det \Omega_X\), so one obtains a relation between the positivity of the canonical bundle and generic positivity of the cotangent bundle. A theorem of \textit{F. Bogomolov} and \textit{M. McQuillan} [in: Foliation theory in algebraic geometry. Proceedings of the conference, New York, NY, USA, September 3--7, 2013. Cham: Springer. 21--51 (2016; Zbl 1337.14041)], see also \textit{J.-B. Bost} [Publ. Math., Inst. Hautes Étud. Sci. 93, 161--221 (2001; Zbl 1034.14010)] and \textit{S. Kebekus} et al. [J. Algebr. Geom. 16, No. 1, 65--81 (2007; Zbl 1120.14011)], makes the connection between rational curves and positivity of the cotangent sheaf more precise: let \(\mathcal F \subset T_X\) be a (singular) foliation, that is a saturated subsheaf that is closed under the Lie bracket. Let \(C \subset X\) be a curve that is disjoint from the singular locus of the foliation such that the restriction \(\mathcal F|_C\) is an ample vector bundle. Then the leaves of the foliation \(\mathcal F\) passing through \(C\) are algebraic and the closure of the general leaf is rationally connected. If the curve \(C\) is a complete intersection of sufficiently ample divisors, the theorem of Mehta-Ramanathan allows to reduce Miyaoka's theorem to this result. \newline The first main result of this paper is an analogue of the Bogolomolov-McQuillan theorem for movable classes: let \(N_1(X)_{\mathbb R}\) be the vector space generated by the numerical equivalence classes of curves, and let \(\alpha \in N_1(X)_{\mathbb R}\) be a movable class, i.e. a class such that for every effective divisor \(D\) one has \(D \cdot \alpha \geq 0\). Given a non-zero torsion-free sheaf \(\mathcal E\) one defines the slope with respect to the class \(\alpha\) by \(\mu_\alpha(\mathcal E):= \frac{c_1(\mathcal E) \cdot \alpha}{\mathrm{rk} \mathcal E}\) and the minimal slope \[ \mu_{\alpha, \min} (\mathcal E):= \inf \{ \mu_{\alpha}(\mathcal Q): \mathcal E \rightarrow \mathcal Q \rightarrow 0 \}, \] where \(Q\) is any non-zero torsion-free quotient. Suppose now that \(\mathcal F \subset T_X\) is a foliation such that \(\mu_{\alpha, \min} (\mathcal F)>0\). The authors show that a general leaf of the foliation \(\mathcal F\) is algebraic and its closure is rationally connected. There is a crucial difference between this statement and earlier results: the proof does not proceed by restricting to some well-chosen curve \(C \subset X\), so it is not necessary to control the stability properties of \(\mathcal F\) under restriction. This allows some extra flexibility which can actually be extended even further: \textit{S. Druel} observed in [Invent. Math. 211, No. 1, 245--296 (2018; Zbl 1419.14063)] that the proof shows that if the dual sheaf \(\mathcal F^*\) is not pseudoeffective, then the general leaves of the foliation \(\mathcal F\) are algebraic (but not necessarily rationally connected). This has lead to further progress for foliations with numerically trivial canonical divisor, e.g. [\textit{A. Höring} and \textit{T. Peternell}, Invent. Math. 216, No. 2, 395--419 (2019; Zbl 07061101)]. \newline The paper also contains several important applications and extensions of the main result: let \(X\) be a projective manifold and let \(\mathcal F \subset T_X\) be a foliation with at most canonical singularities (in the sense of \textit{M. McQuillan} [Pure Appl. Math. Q. 4, No. 3, 877--1012 (2008; Zbl 1166.14010)]). Assume that the canonical bundle of the foliation \(K_{\mathcal F} := \det (\mathcal F^*)\) is pseudoeffective. For any \(m>0\), consider any quotient \(\bigotimes^m \mathcal F^* \rightarrow \mathcal Q \rightarrow 0\). Then the determinant \(\det(\mathcal Q)\) is also pseudoeffective. \newline This statement actually holds in the more general setting of (smooth) orbifolds introduced by the first-named author in [Ann. Inst. Fourier 54, No. 3, 499--630 (2004; Zbl 1062.14014)], i.e. pairs \((X, \Delta)\) where \(X\) is a projective manifold and \(\Delta\) a boundary \(\mathbb Q\)-divisor with simple normal crossings support. In analogy to the logarithmic cotangent bundle the authors define the orbifold cotangent bundle \(\Omega^1(X, \Delta)\) for this setup and introduce the notion of adapted covers \(\pi: X_\Delta \rightarrow X\) for the orbifold \((X, \Delta)\). Assume that the canonical divisor \(K_X+\Delta\) is pseudoeffective. For any \(m>0\), consider any quotient \(\otimes^m \pi^* \Omega^1(X, \Delta) \rightarrow \mathcal Q \rightarrow 0\). Then for any movable class \(\alpha\) on \(X\), one has \(\det(\mathcal Q) \cdot \pi^* \alpha \geq 0\). \newline Last but not least, the paper contains a substantially simplified proof of the main result of the authors' earlier paper [Ann. Inst. Fourier 65, No. 2, 835--861 (2015; Zbl 1338.14012)]: in the situation above, let \(L\) be a big line bundle such that for some \(m \in \mathbb N\) there exists a non-zero morphism \(\pi^* L \rightarrow \otimes^m \pi^* \Omega^1(X, \Delta)\). Then the log-canonical divisor \(K_X+\Delta\) is pseudoeffective. By earlier results of Viehweg and Zuo this immediately implies Viehweg's conjecture: let \(V \rightarrow B\) be a family of canonically polarised manifolds over a quasi-projective base \(B\). If the variation of the family is maximal, then for any good compactification \(B \subset X\) with SNC boundary \(\Delta\), the pair \((X, \Delta)\) is of general type. Using the theory of Hodge modules \textit{M. Popa} and \textit{C. Schnell} have recently extended this theorem to families of varieties of general type [Invent. Math. 208, No. 3, 677--713 (2017; Zbl 1375.14043)].