Webs of Lagrangian tori in projective symplectic manifolds (Q1949229)

An even dimensional vector space \(V\) equipped with a non-degenerate \(2\)-form \(\omega \in \bigwedge ^2V^*\) is a symplectic vector space. Given a subspace \(W\subset V\), define \(W^{\perp}\) to be the set of \(v\in V\) which is perpendicular to \(W\) with respect to \(\omega\), i.e. those \(v\) satisfying \(\omega (v,w)=0\) for all \(w\in W\). \(W\) is said to be Lagrangian if \(W=W^{\perp}\). It follows from the definition that if \(W\) is Lagrangian then \(\dim(V)=2\dim(W)\). Given \(M\) a complex manifold. A symplectic form on \(M\) is a \((2,0)\) form \(\omega\) which is nowhere degenerate. The pair \((M,\omega )\) is called a symplectic manifold. A submanifold \(A\) of a symplectic manifold \((M,\omega )\) is called Lagrangian if for each \(x\in A\) then \(T_x(A)\) is a Lagrangian subspace of \(T_x(M)\). A simply-connected compact Kähler manifold is called a compact hyperkähler manifold if \(H^0(M,\Omega _M^2)\) has dimension \(1\). In particular, such a manifold \(M\) is symplectic. A. Beauville asked the following question: Let \(M\) be a compact hyperkähler manifold having a Lagrangian torus \(A\). Is there a dominant meromorphic map \(f:M\cdots\to B\), a Zariski dense open set \(M^0\subset M\) with \(A\subset M^0\) and a Zariski open dense set \(B^0\subset B\) such that \(f|_{M^0}\) is a proper smooth morphism onto \(B^0\) and \(A\) is a fiber of \(f\)? The paper under review proves the following result which, by an observation of \textit{D. Greb, C. Lehn} and \textit{S. Rollenske} [``Lagrangian fibrations on hyperkähler manifolds -- on a question of Beauville'' \url{arXiv:1105.3410}], gives a complete answer to Beauville's question in the affirmative. { Theorem 1.1.} Let \(M\) be a simply-connected projective manifold with a symplectic form and let \(A\subset M\) be a Lagrangian torus. Then \(M\) has a hypersurface disjoint from a deformation of \(A\). The proof of Theorem 1.1 is quite involved. Here we give a summary of the proof. It is well-known that a Lagrangian torus \(A\subset M\) has trivial normal bundle and is unobstructed. The latter condition means that in the Hilbert scheme \(\mathrm{Hilb}(M)\), then \([A]\) is a smooth point (see Definition 5.1). Let \(\mathrm{Hilb}(M)_A\) be the irreducible component containing \([A]\) of \(\mathrm{Hilb}(M)\), \(\xi :\mathrm{Univ}(M)_A\rightarrow \mathrm{Hilb}(M)_A\) the universal family and \(\eta :\mathrm{Univ}(M)_A\rightarrow M\) the evaluation map. Then by Proposition 5.3, there is a Zariski open subset \(\mathrm{Hilb}(M)_A^0\) of \(\mathrm{Hilb}(M)_A\) parametrizing unobstructed submanifolds with trivial normal bundles. Denote by \(\xi _A^0:\mathrm{Univ}(M)_A^0\rightarrow \mathrm{Hilb}(M)_A^0\) and \(\eta _A^0:\mathrm{Univ}(M)_A^0\rightarrow M\) the restrictions of \(\xi \) and \(\eta\). Then \(\xi _A^0\) is a smooth projective morphism and \(\eta _A^0\) is unramified. By choosing appropriate compactifications of \(\mathrm{Univ}(M)_A^0\) and \(\mathrm{Hilb}(M)_A^0\), we obtain a web of \textit{Lagrangian tori} \(\mathcal{W}=\{\mu :\mathcal{U}\rightarrow M,\rho :\mathcal{U}\rightarrow \mathcal{K}\}\) on \(M\) as in Definition 3.1. Here the map \(\mu\) is finite-to-one. Let \(d\) be the degree of \(\mu\), and \(X=\mu ^{-1}(x)\) where \(x\in M\) is a generic point. There associated finite groups \(G_i\) and \(H_i\), where \(H_i\) is a normal subgroup of \(G_i\) (for \(1\leq i\leq d\)), see Proposition 3.7. They are all subgroups of a finite group \(G\), see Definition 3.6. Roughly speaking, \(G\) is constructed using the action of the fundamental group of an appropriate Zariski open dense set of \(M\) on the finite set \(X\). Thus, its action is transitive. If \(\mathcal{H}\) is the subgroup of \(G\) generated by \(H_1,\ldots ,H_d\), then \(\mathcal{H}\) may not act transitively on \(X\). The remaining of the proof then proceeds by considering these two cases separately. Case 1: \(\mathcal{H}\) does not act transitively on \(X\). Then Proposition 3.10 implies that there is a divisor \(D\subset M\) disjoint from a general member of \(\mathcal{W}\). Case 2: \(\mathcal{H}\) acts transitively on \(X\). Then the triple \(<X,G,H_1>\) is special in the sense of Definition 2.3. The first condition in Definition 2.3 follows readily from the assumption that \(\mathcal{H}\) acts transitively on \(X\). The second condition in Definition 2.3 is proved using Proposition 6.8 that \(\mathcal{W}\) is pairwise integrable, and then applying Propositions 6.9, 5.11, 4.15 and 4.14. Now Theorem 2.4 implies that there is no non-trivial special triples, which means that \(|X|=1\). Thus \(d=1\), i.e. the map \(\mu\) is birational. In this case, it is easy to conclude the proof. Then \(\mu ^{-1}(A)\) is a fiber of \(\rho\), thus \(\rho (\mu ^{-1}(A))\) is a point. We choose \(D'\) to be a divisor disjoint from \(\rho (\mu ^{-1}(A))\). Then \(D=\mu (\rho ^{-1}(D'))\) is a divisor of \(M\) disjoint from \(A\). In Case 2, one important fact used is that the self-intersection number of \(A\) is \(0\), hence if \(A'\) is any point in \(\mathrm{Hilb}(M)_A^0\) then \(A\cap A'\) has no isolated point (see Proposition 5.7).