Algebraic torus actions on contact manifolds (Q2169816)

A contact complex manifold is a complex manifold \(X\) of odd dimension such that there exists an exact sequence of vector bundles \[ 0 \to F \to TX \to L \to 0 \] where \(L\) is a line bundle, \(TX\) is the tangent bundle to \(X\), \(F\) is a vector subbundle of \(TX\), and the restriction of the derivative \(d\theta|_F : \Lambda^2F\to L\) of the induced twisted form \(\theta \in H^0(\Omega^1X\otimes L)\) is nowhere degenerate. The main result of the article states that if \(X\) is a contact manifold of dimension \(2n+1\), which furthermore is Fano and has a reductive automorphism group \(G\) of rank at least \(n-2\), then \(X\) is a homogeneous space. This result provides a proof of the LeBrun-Salamon conjecture in low dimensions [\textit{C. LeBrun} and \textit{S. Salamon}, Invent. Math. 118, No. 1, 109--132 (1994; Zbl 0815.53078)]. The latter states that a compact simply connected quaternion-Kähler manifold of real dimension \(4n\) with positive scalar curvature is one of the Wolf spaces. This conjecture can be restated in terms of contact complex manifolds via the twistor space construction. The main result of the article stated above implies that the LeBrun-Salamon conjecture is true for \(n=3\) and \(n=4\) (note that the notations in the paper are consistent with the twistor space construction: starting from a quaternion-Kähler manifold of dimension \(4n\), its twistor space is a contact complex manifold of dimension \(2n+1\)). Contact complex manifolds exhibit strong rigidity properties. For example, a projective contact manifold \(X\), with the line bundle \(L\) as above, is either the projectivization \(\mathbb{P}(T^*Y)\) of the cotangent bundle of a projective manifold with \(L\simeq \mathcal{O}_{\mathbb{P}(T^*Y)}(1)\), or \(\mathbb{P}^{2n+1}\) with \(L=\mathcal{O}_{\mathbb{P}^{2n+1}}(2)\), or is Fano with Picard group generated by \(L\). Most of the work in the article takes place in the latter case. In that case, one can show that \(H^0(X,L)\) is isomorphic to \(H^0(X,TX)\) as a representation of (the neutral component of) the group of automorphisms of \(X\) [\textit{A. Beauville}, Comment. Math. Helv. 73, No. 4, 566--583 (1998; Zbl 0946.53046)]. The method of proof in the article relies on a remarkable use of moment polytope(s) and Bialynicki-Birula decomposition. Since the torus actions considered here have high complexity (that is, a generic orbit has high codimension), it is not obvious that meaningful information can be extracted from this data. The authors show that it is possible to do so by using the strong rigidity properties alluded to above. Without going into the details, let us mention one key ingredient introduced by the authors: the notion of \textit{compass}. Let \((X,L)\) be a polarized complex manifold, equipped with an algebraic action of an algebraic torus \(H\) with finite kernel. Let \(y\in X\) be a fixed point under the action of \(H\). Then there is a natural linear action of \(H\) on the cotangent space to \(X\) at \(y\), and we can consider the set of weights of this action counted with multiplicities. It turns out that this set of weights is the same for \(y'\) if \(y'\) and \(y\) are in the same connected component of \(X^H\) (the set of \(H\)-fixed points), and the number of weights equal to zero is the dimension of this connected component. The \textit{compass} is the data of the set of non-zero weights among these, for each connected component of \(X^H\). There is only a finite number of such components, hence it is a combinatorial data. Another combinatorial data, the (moment) polytope of fixed points \(\Delta(X,H,L)\), is the convex hull of all weights for the action of \(H\) on fibers of \(L\) above fixed points. Again, this weight depends only on the connected component in \(X^H\). In the first sections of the paper, the authors prove several general results on this combinatorial data, which are then applied in the specific context of contact complex manifolds. As a basic example, using the notion of compass, they show that the dimension of \(X\) is greater than or equal to the maximal number of edges from any vertex of \(\Delta(X,H,L)\). If \((X,L)\) is a Fano contact manifold with \(\text{Pic}(X)=\mathbb{Z} L\) and automorphism group the exceptional group \(E_6\) for example, then from the above remarks we know that \(\Delta(X,H,L)\) may be identified with the convex hull of all roots of \(E_6\). Counting the edges in this polytope, one deduces that the dimension of \(X\) is at least \(20\). The main result of course relies on more subtle arguments, with some case by case analysis using the classification of simple algebraic groups.