Point configurations and translations (Q487168): Difference between revisions

From the introduction: ``In the present paper we examine point configurations on the projective line up to translations. In general, let us consider \(n\) distinct points on \(\mathbb{P}_1\). Then the open subset \(U \subseteq \mathbb{P}_{1}^{n}\) consisting of pairwise different coordinates is the space of possible configurations. For an algebraic group \(G\) acting on \(\mathbb{P}_1\) the question arises what the resulting equivalence classes of configurations are, i.e., we ask for a quotient \(U/G\) of the diagonal action and a possible canonical compactification. In the case of the full automorphism group \(G = \mathrm{PSL}(2, K)\) this problem has been thoroughly studied. The space of configuration classes is canonically compactified by the famous Grothendieck-Knudsen moduli space \(\overline{M}_{0,n}\), i.e., we have \[ M_{0,n}=U/\mathrm{PSL}(2,K) \subseteq \overline{M}_{0,n}. \] Originally introduced as moduli space of certain marked curves Kapranov shows in [\textit{M. M. Kapranov}, in: I. M. Gelfand seminar. Part 2: Papers of the Gelfand seminar in functional analysis held at Moscow University, Russia, September 1993. Providence, RI: American Mathematical Society. 29--110 (1993; Zbl 0811.14043)] that \(\overline{M}_{0,n}\) has (among others) the following two equivalent descriptions. Firstly it arises as the GIT-limit of \(\mathbb{P}_{1}^{n}\) with respect to the \(G\)-action, i.e., the limit of the inverse system of Mumford quotients. Secondly, it can be viewed as the blow-up of \(\mathbb{P}_{n-3}\) in \(n-1\) general points and all the linear subspaces of dimension at most \(n-5\) spanned by them. Later this setting has been studied in the case where the full automorphism group was replaced by its maximal torus \(\mathbb{G}_m \subseteq \mathrm{PSL}(2,K)\). Similarly, it turns out that the Losev-Manin moduli space \(\overline{L}_n\) coincides with the the GIT-limit, which in this case is the toric variety associated to the permutahedron. Again, the GIT-limit arises in a sequence of (toric) blow-ups from projective space, see [\textit{I. M. Gelfand} et al., Discriminants, resultants, and multidimensional determinants. Reprint of the 1994 edition. Modern Birkhäuser Classics. Boston, MA: Birkhäuser (2008; Zbl 1138.14001)], \textit{M. M. Kapranov} et al. [Math. Ann. 290, No. 4, 643--655 (1991; Zbl 0762.14023)], \textit{A. Losev} and \textit{Y. Manin} [Mich. Math. J. 48, 443--472 (2000; Zbl 1078.14536)]. In this paper we treat point configurations on \(\mathbb{P}_1\) up to the action of the maximal connected unipotent subgroup \(\mathbb{G}_a \subseteq \mathrm{PSL}(2,K)\). It consists of upper triangular matrices with diagonal elements equal to \(1_K\) and can be thought of as group of translations. Since \(\mathbb{G}_a\) is not reductive, we are faced with the additional problem of first finding a suitable replacement for the GIT-limit, i.e., assigning a canonical quotient to this action. This will be overcome in the following manner. \textit{B. Doran} and \textit{F. Kirwan} [Pure Appl. Math. Q. 3, No. 1, 61--105 (2007; Zbl 1143.14039)] the notion of finitely generated semistable points admitting so-called enveloped quotients. Moreover, in [\textit{I. V. Arzhantsev} et al., J. Algebra 387, 87--98 (2013; Zbl 1300.14043)] Arzhantsev, Hausen and Celik proposed a Gelfand-MacPherson type construction which allows one to apply methods from reductive GIT to obtain these enveloped quotients. Building on this work we obtain again an inverse system and the corresponding GIT-limit. In general the enveloped quotients are not projective, hence one cannot expect the GIT-limit to be so. We then show that (up to normalisation) the limit quotient, i.e., a canonical component of the GIT-limit, is canonically compactified by an iterated blow-up of \(\mathbb{P}_{1}^{n-1}\). To make this a little more precise consider a subset \(A \subseteq \{2,\ldots, n\}\). Denoting by \(T_2, S_2,\ldots, T_n, S_n\) the homogeneous coordinates on \(\mathbb{P}_{1}^{n-1}\) we associate to \(A\) a subscheme \(X_A\) on \(\mathbb{P}_{1}^{n-1}\) given by the ideal \[ \langle T_i^2, T_j S_k-T_kS_j; i,j,k \in A, j < k \rangle. \] The scheme-theoretic inclusions give rise to a partial order of these subschemes. Let \(Bl(\mathbb{P}_1^{n-1})\) denote the blow-up of \(\mathbb{P}_1^{n-1}\) in all these subschemes in non-descending order. {Theorem .} If \(\widetilde{\mathbb{P}_{1}^{n}/\mathbb{G}_a}\) and \(\widetilde{Bl(\mathbb{P}_1^{n-1})}\) denote the normalisations of the limit quotient and the above blow-up of \(\mathbb{P}_1^{n-1}\) respectively, then we have open embeddings \[ U / \mathbb{G}_a \subseteq \widetilde{\mathbb{P}_{1}^{n}/\mathbb{G}_a} \subseteq \widetilde{Bl(\mathbb{P}_1^{n-1})}. \] The paper is organized as follows. In Sect. 2 we recall the results of [Zbl 1300.14043] and introduce the non-reductive GIT-limit and limit quotient. In the following Sect. 3 we apply these constructions to the action of \(\mathbb{G}_a\) on \(\mathbb{P}_{1}^{n}\). We discuss explicitly the GIT-fan which contains the combinatorial data needed to make the limit quotient accessible. The blow-ups of \(\mathbb{P}_1^{n-1}\) will be dealt with in a mostly combinatorial way, i.e., as proper transforms with respect to toric blow-ups. For this we prove a result on combinatorial blow-ups in the spirit of \textit{E.-M. Feichtner} and \textit{D. N. Kozlov} [Sel. Math., New Ser. 10, No. 1, 37--60 (2004; Zbl 1068.06004)]. This will be carried out in Sect. 4. The final Sect. 5 then is dedicated to the proof of the main theorems.''