Positivity vs. slope semistability for bundles with vanishing discriminant (Q2168835)

The authors study nef bundles with \(c_1 \equiv 0\), i.e. numerically flat vector bundles on smooth projective varieties over algebraically closed fields. One of the main results of the paper generalizes Hartshorne's (in char 0) and Barton's (in any characteristics) theorem: a vector bundle of degree zero on a smooth projective curve is nef if and only if it is semistable. The authors establish an equivalence between numerically flat vector bundles and strongly slope semistable torsion free sheaves with numerically trivial Chern classes. This gives new direct proofs of similar results on smooth projective varieties without using non-abelian Hodge theory or reduction to positive characteristics. Strong slope semistability is equivalent to slope semistability over characteristics zero. In positive characteristics strong slope semistability means all pullbacks under Frobenius should be slope semistable in the usual sense. The authors in Theorem C also show that strongly slope semistable reflexive sheaves with discriminant of degree zero are in fact locally free and remain slope semistable under pull-backs and restrictions. Furthermore, using these results the authors compare the the nef cones and the closure of the effective cones in various instances. The authors show that on a smooth projective variety on which every effective divisor is nef, the same holds on the projective bundle \(\mathbb{P}_X(\mathcal{E})\) for a strongly semistable locally free sheaf \(\mathcal{E}\) on \(X\) with trivial discriminant. The paper is very self-contained and also serves as an excellent reference to various properties and results about numerically flat vector bundles. Among other things, in Lemma 3.1 they give a Mehta-Ramanan style restriction theorem for strongly stable torsion-free sheaves \(\mathcal{E}\) with vanishing discriminant (so in particular for numerically flat vector bundles); namely they show that for any ample class \(H\) on \(X\), there exists \(m_0\) depending on \(X\), \(H\), \(\operatorname{rank}(\mathcal{E})\) such that for all \(m\geq m_0\) and any normal divisor \(D\in |mH|\), the restriction \(\mathcal{E}_D\) is again strongly \(\mu_{H_D}\)-stable.