Instantons and curves on class VII surfaces (Q624920)

A class VII surface is a compact non-Kähler surface with first Betti number \(b_1=1\) and Kodaira dimension \(-\infty\). Class VII surfaces with second Betti number \(b_2=0\) are known to be either a Hopf surface or an Inoue surface. On the other hand, for class VII surfaces with \(b_2>0\) it is not known whether they contain a curve. Results which prove the existence of curves on such surfaces seem to be extremely difficult to obtain. Thanks to work of \textit{M. Kato} [Proc.\ Japan Acad.\ 53, 15--16 (1977; Zbl 0379.32023)] class VII surfaces which contain a global spherical shell (an open surface biholomorphic to the standard neighbourhood of \(S^3\) in \({\mathbb C}^2\)) are well understood. In particular it is known that such surfaces contain a cycle of rational curves. A conjecture of Nakamura states that any minimal class VII surface with \(b_2>0\) contains a global spherical shell. This was shown to be true for \(b_2=1\) in a previous article by the author [Invent.\ Math.\ 162, No.\ 3, 493--521 (2005; Zbl 1093.32006)]. On the other hand, \textit{I. Nakamura} [Tohoku Math.\ J., II.\ Ser.\ 42, No.\ 4, 475--516 (1990; Zbl 0732.14019)] has shown that a minimal class VII surface which contains a cycle of curves is a specialisation of blown-up primary Hopf surfaces. The main result of the paper under review states that each minimal class VII surface with \(b_2=2\) contains a cycle of curves. The main tool used in this article is the moduli space of polystable (with respect to a suitable Gauduchon metric) rank-2 vector bundles with \(c_2=0\) and determinant isomorphic to the canonical line bundle. Assuming that a minimal class VII surface with \(b_2=2\) does not contain a cycle of curves, he proves with gauge theoretical techniques that this moduli space is a compact topological manifold and that the subspace corresponding to stable bundles is a complex manifold having a smooth compact connected component which contains only a finite but non-empty subset representing filtrable bundles (bundles which possess subsheaves of rank 1). He then shows that the existence of such a component leads to a contradiction. The main obstacle to applying the same ideas to the case \(b_3\geq 3\) is that the geometry of the moduli space of a polystable bundle will become very complicated.