Cocharacter-closure and the rational Hilbert-Mumford theorem (Q2412524): Difference between revisions

Let $G$ be a reductive group (not necessarily connected) acting on an affine variety $V$ over a field $k$. A subset $X \subset V$ is said to be cocharacter-closed if for any $v \in V$ and any $k$-cocharacter $\lambda$ of $G$, we have $\lim_{a \to 0} \lambda(a)v \in X$ whenever the limit exists in the sense of algebraic geometry. Likewise, one defines the notion of cocharacter-closure of $X$ in $V$. \par The first main Theorem 1.3 of the paper asserts that for all $v \in V$, there is a unique cocharacter-closed $G(k)$-orbit $\mathcal{O}$ in the cocharacter-closure of $G(k) \cdot v$; moreover there exists a $k$-cocharacter $\lambda$ of $G$ that steers $v$ to some point of $\mathcal{O}$ as $a \to 0$. This is a rational counterpart of the Hilbert-Mumford theorem, which is extremely useful in geometric invariant theory when $k = \bar{k}$. \par The second main Theorem 1.5 gathers various properties related to the cocharacter-closedness of $G(k) \cdot v$, assuming the stabilizer $G_v$ to be defined over $k$. In particular, it is shown that Zariski-closedness implies cocharacter-closedness over $k$, and the cocharacter-closedness of $G(k) \cdot v$ behaves well under separable algebraic extensions $k' \mid k$. \par The third main Theorem 1.6 explores the relation between $k$-anisotropy of $G$ and cocharacter-closedness of affine $G$-varieties. \par These variations on the theme of Hilbert-Mumford are immediately useful in arithmetic situations, where $k$ is taken to be a local or global field. They are applicable to Langlands program for the same reason, and can intervene on the dual side of Langlands correspondence through Serre's notion of $G$-complete reducibility; see the works by V.\ Lafforgue on the Langlands parameterization over function fields.