Stratification of Hurwitz spaces by closed modular subvarieties (Q1011936)

Hurwitz spaces parametrise Galois covers of (in this case) \({\mathbb P}^1\) with given discrete invariants. The invariants are the Galois group \(G\) as an abstract group, the ramification divisor \({\mathbf t}\) and the inertia canonical invariant \({\mathbf C}\). Over an algebraically closed field \(k\), this is a formal sum of conjugacy classes in \(G\): the sum is over points \(t\in{\mathbf t}\), and the elements of the conjugacy class are the inverse images of a chosen root of unity under the natural map from the inertia group at a place above \(t\) to \(k^\times\). Over a connected \({\mathbb Z}[{{1}\over{|G|}}]\)-scheme \(S\) this still makes sense if we interpret it as meaning the inertia canonical invariant at a geometric point. There is a well-developed (by the first author, Wewers and others) theory of moduli (pre)stacks and schemes for this situation, which is summarised here: the essential point, though, is that there are two notions of isomorphism for Galois covers of \({\mathbb P}^1\), depending on whether one requires the obvious diagram to commute ``on the nose'' or up to automorphisms of \({\mathbb P}^1\). In the latter case the natural classifying object is merely a prestack (i.e.\ the descent data may not be effective), denoted \({\mathcal H}_{\text{G}/\text{PGL}_2}({\mathbf C})\); in the former it is a stack \({\mathcal H}_{\text{G}}({\mathbf C})\). Both, however, have coarse moduli schemes \(H_{\text{G}/\text{PGL}_2}({\mathbf C})\) and \(H_{\text{G}}({\mathbf C})\), defined over some number field. It is these schemes that are given a stratification here. The index of the stratification is the base invariant of a \(G\)-cover \(f: X\to {\mathbb P}^1_{\bar k}\) over a field \(k\) of characteristic zero, which is the conjugacy class \({\mathcal E}_f\) in \({\text{PGL}}_2(\bar k)\) of the stabiliser of the \(G\)-isomorphism class of \(f\) under \({\text{PGL}}_2(\bar k)\). As long as \(\deg{\mathbf t}\geq 3\) this group is contained in the symmetri group of \({\mathbf t}\) and there are only finitely many such conjugacy classes, well-known in group theory. For such a conjugacy class \({\mathcal E}\) we look at the set of \(k\)-rational points in \(H_{\text{G}}({\mathbf C})\) such that \({\mathcal E}_f\supseteq{\mathcal E}\). The main theorem is that these form closed subschemes: then a similar result holds for \(H_{\text{G}/\text{PGL}_2}({\mathbf C})\) also. Two proofs are given. One uses GIT and works directly with \(H_{\text{G}/\text{PGL}_2}({\mathbf C})\), which is shown to be a universal geometric quotient of \(H_{\text{G}}({\mathbf C})\) under \(\text{PGL}_2\). The other uses the properties of these Hurwitz schemes established by [\textit{J. Bertin} and \textit{M. Romagny}, Champs de Hurwitz. Preprint, \url{arXiv:math/0701680}] and is better adapted for calculation. The authors illustrate this last observation by computing some further information about irreducible components of the strata and about \({\mathbb Q}\)-rational points.