Compactifications defined by arrangements. I: The ball quotient case. (Q1421141)

In the mid-1980s the author published an announcement [in: Algebraic geometry, Proc. Conf., Vancouver/B.C. 1984, CMS Conf. Proc. 6, 341--364 (1986; Zbl 0624.14008)] of his work on compactifications of locally symmetric varieties. He gave some details, but the full construction was written down only in a University of Nijmegen preprint. This, because of its length and technicality, was never published, but it was not forgotten: two papers of the author's student \textit{H. Sterk} on moduli of Enriques surfaces arose out of it [Math. Z. 207, 1--36 (1991; Zbl 0736.14017) and 220, 427--444 (1995; Zbl 0841.14031)], and several people, the reviewer among them, were directly or indirectly influenced by Looijenga's ideas. The present paper is the first of two, of which the second [Duke Math. J. 119, No. 3, 527--588 (2003; Zbl 1079.14045)] is essentially the Nijmegen preprint, brought up to date. The author's new insight, to which, he says, G. Heckman also contributed, is that the ideas of the Nijmegen preprint are more widely applicable and that they can be seen in their purest form in the case of a ball quotient, rather than in type IV locally symmetric varieties where they first arose. Moreover, the ideas have acquired a new relevance in view of the progress in understanding automorphic forms and their divisors (due to Freitag, Bruinier, Allcock and others, and ultimately to Borcherds), and also because of some progress in desingularising hypersurface arrangements. Let \({\mathbb B}\) be a complex ball and let \(\Gamma\) be an arithmetic group acting on \({\mathbb B}\) and on the automorphic line bundle \({\mathbb L}\). The quotient \(X=\Gamma\backslash{\mathbb B}\) carries an (orbifold) line bundle \({\mathcal L}\) and has a natural (Baily-Borel) compactification which adds finitely many cusps. (Non-isolated cusps are one complicating feature of the type IV case.) Take an arrangement of hypersurfaces in \(X\) satisfying suitable conditions. Then the construction presented here gives a natural, and naturally stratified, compactification \(\widehat{X^\circ}\) of the complement \(X^\circ\) of the divisors. This \(\widehat{X^\circ}\) is also given as a blow-up of the Baily-Borel compactification of \(X\) followed by a blow-down. It frequently happens that \(\widehat{X^\circ}\setminus {X^\circ}\) is of codimension \(\geq 2\). There are many cases where \(X^\circ= Y//G\), where \(Y\) is a quasi-projective variety, \(G\) is a reductive group and \(Y//G\) is a GIT quotient, and \({\mathcal L}\) corresponds to the polarisation of \(Y\), i.e. the linearisation of \(G\). Of course one wants the isomorphism to be the period map. In these cases the period map extends to an isomorphism \(Y^{ss}//G\to\widehat{X^\circ}\), and in practice the stratification of \(\widehat{X^\circ}\) has a geometric interpretation on the GIT side. Section 1 of the paper defines linearisable arrangements of hypersurfaces: very roughly, this means that one may choose isomorphisms between the normal bundle of each hypersurface \(H\) and \({\mathcal L}^*| _H\) that agree on intersections. Many examples are given. Section 2 describes the procedure of blowing up so as to make the arrangement \({\mathcal H}=\{H_1,\dots,H_k\}\) into a normal crossings divisor and keeps track of the bundle \({\mathcal L}({\mathcal H})=\sum{\mathcal L}^{\otimes k}(H_1+\ldots+H_k)\). Section 3 gives conditions for the pullback of \({\mathcal L}({\mathcal H})\) to be semiample and explains how this works for some of the examples given in Section~1. Section 4 does the contraction associated to \({\mathcal L}({\mathcal H})\) in more detail for the ball case: this is the main construction promised in the introduction to the paper. Section 5 makes the extra assumption that the arrangement is arithmetically defined relative to \(\Gamma\), and shows that in this case \({\mathcal L}({\mathcal H})\) is eventually generated: that is, there are lots of associated automorphic forms. Section 6 asks (roughly) whether, if \(\sum n_H{\mathbb B}_H\) is Cartier, it is the divisor of a meromorphic automorphic form. Section 7 ``makes the descent from the general to the specific in Bourbakian style'' and deals with some examples: moduli of quartic curves, rational elliptic surfaces, cubic surfaces, cubic 3-folds, etcetera.