Rees algebras of additive group actions (Q2126081)

The authors introduce a new, geometric method to study the actions of the additive group \(\mathbb{G}_{a}\) on relative affine schemes, using their Rees algebras to construct relative completions. They illustrate it by a comprehensive discussion of classical and new examples. The \(\mathbb{G}_{a}\)-actions give an important measure of flexibility of an affine variety, but they are often very difficult to handle, for example, their rings of invariants might not be finitely generated. Over a field of characteristic zero, every \(\mathbb{G}_{a}\)-action on \(X=\operatorname{Spec} A\) is given by \(A\ni f\mapsto \exp(t \partial)f \in A[t]\) for some locally nilpotent derivation \(\partial\). Then \(\ker \partial\) is the ring of invariants, and \(\ker \partial^{2}\) consists of \emph{local slices}, see [\textit{G. Freudenburg}, Algebraic theory of locally nilpotent derivations. Berlin: Springer (2006; Zbl 1121.13002)]. The \emph{Rees algebra} of such an action is \(R(A,\partial)=\bigoplus_{n\geq 0}\ker \partial^{n}\). Results of Section 1 (see Lemma 6) show that the scheme \(\operatorname{Proj} R(A,\partial)\) is a natural, \(\mathbb{G}_{a}\)-invariant completion of \(X\), with boundary \(\operatorname{Proj} \bigoplus_{n\geq 0} \ker \partial^{n+1}/\ker \partial^{n}\). The latter can be viewed as a well-behaved degeneration of \(X\), cf.\ [\textit{Sh. Kaliman} and \textit{L. Makar-Limanov}, in: Affine algebraic geometry. Dedicated to Masayoshi Miyanishi on the occasion of his retirement from Osaka University. Osaka: Osaka University Press. 231--255 (2007; Zbl 1127.14053)]. Thus whenever \(R(A,\partial)\) is finitely generated, it provides a distinguished compactification of \(X\). An algorithm to compute it is presented in Section 3.4. Nonetheless, as illustrated by Example 26, the singularities at infinity can be rather bad (possibly non \(\mathbb{Q}\)-factorial). When \(X\to X/\mathbb{G}_{a}\) is étale-locally trivial, \(\operatorname{Proj} R(A,\partial)\) is a \(\mathbb{P}^1\)-fibration completing \(X\), explicitly described in Proposition 12. Conversely, Theorem 25 gives a geometric criterion for a \(\mathbb{P}^1\)-fibration \(\bar{X}\) with a section \(B\) guaranteeing that \(\bar{X}\setminus B\) is a \(\mathbb{G}_{a}\)-variety with finitely generated Rees algebra. This approach allows to construct many nontrivial examples. In fact, the above theory is developed in relative setting, with the base \(\operatorname{Spec} k\) replaced by an arbitrary scheme, or even algebraic space \(S\). In practice, one can take \(S=X/\mathbb{G}_{a}\) (if it exists). For example, a Danielewski surface \(X=\{x^{n}z=p(x,y)\}\subseteq \mathbb{C}^3\) is a torsor over an affine line with \(d\)-ple origin. Its completion provided by the Rees algebra is a \(\mathbb{P}^1\)-fibration whose unique degenerate fiber consists of \(d\) lines meeting at an \(A_{k}\)-type singularity, see Section 4.1. More generally, such a relative approach provides a natural setting to perform standard constructions such as affine modifications, or extensions of \(\mathbb{G}_{a}\)-torsors over punctured surfaces, studied e.g.\ in [\textit{A. Dubouloz} et al., Ann. Sc. Norm. Super. Pisa, Cl. Sci. (5) 21, 133--167 (2020; Zbl 1477.14094); \textit{I. Hedén}, Transform. Groups 21, No. 2, 427--449 (2016; Zbl 1391.14116)]. Examples discussed in detail in Section 4.5 give a new perspective towards the cited results. The general theory in Section 1 is developed regardless of the base field. From Section 2, the authors restrict theory attention to characteristic zero. However, most constructions can be adapted to positive characteristic, too. The only technical difficulty is that instead of a single LND \(\partial\) and its iterates, one needs to consider a sequence of higher derivations \(\partial^{(n)}\), subject to usual composition rule explained in Lemma 1.