Moduli-stacks for bundles on semistable curves (Q1910171): Difference between revisions

Let \(C\) be a smooth proper curve over a base scheme \(B\), and let \(G\) be a reductive algebraic group. The classification theory of rank-\(r\) vector bundles (or, more generally, of principal \(G\)-bundles) over \(C\) has recently gained crucial significance in the relationship between algebraic geometry and conformal quantum field theory. Especially the construction and the geometric investigation of the relevant moduli spaces of bundles (or, more generally, of the algebraic moduli stacks of \(G\)-torsors) are of fundamental importance. In the present paper, the author exhibits two general constructions for such moduli spaces and moduli stacks, respectively. As to that, the underlying strategy is to show that various invariants are independent of the curve \(C\), and then to let \(C\) degenerate to a rational nodal curve, hoping that in this case the relevant invariants can be explicitly computed. This amounts to study the behavior of moduli spaces and moduli stacks under degenerations. The first construction is rather geometric, but applies only to some particular linear groups \(G\). Nonetheless, this approach leads to global moduli stacks for torsion-free \(G\)-sheaves of given rank \(r\) on \(C\) and, under suitable semi-stability conditions, to classifying algebraic spaces (moduli schemes). The second approach uses loop groups and works for arbitrary reductive structure groups. It is based on an alternative treatment of the local structure of \(G\)-torsors on semi-stable curves and, henceforth, brings about the drawback of being non-canonical and non-global. The resulting object of this construction is a stack which, in characteristic zero, satisfies the valuative criterion of properness. Although, in general, this stack does not classify particular torsion-free sheaves over \(C\), it sometimes maps naturally to one of the previously constructed stacks and, for this reason, might be useful for the purpose of comparison. As the author points out, this second method of construction was inspired by his recent general proof of the Verlinde formula [cf. J. Algebr. Geom. 3, No. 2, 347-374 (1994; Zbl 0809.14009)].