Compactifying Lagrangian fibrations
Giulia Saccà
TL;DR
The paper develops a general framework to compactify quasi-projective Lagrangian fibrations by holomorphic symplectic varieties, with a criterion guaranteeing a (possibly singular) Q-factorial terminal symplectic compactification extending the fibration over a normal base B. Central tools include extendable cycle-theoretic holomorphic 2-forms and a minimal-model-program approach to obtain nef canonical classes over B, plus a relative Albanese construction that produces a smooth abelian fibration A→B^{nm} acting on X. The author then shows that any almost torsor over A or over a group isogenous to A admits a symplectic compactification, and that if fibers are integral, these compactifications can be smooth and locally isomorphic to the original fibration; this yields new contexts for Shafarevich–Tate twists and links to OG6/OG10-type hyper-Kähler manifolds. The results unify several known examples (e.g., intermediate Jacobian fibrations) under a single compactification criterion and provide a robust framework for extending geometric Lagrangian fibrations while controlling deformation types and Hodge structures.
Abstract
We suggest a general framework for compactifing quasi-projective Lagrangian fibrations of geometric origin by holomorphic symplectic varieties. This framework includes a compactification criterion, which we then apply to various fibrations of geometric origin, and a discussion on holomorphic forms that are defined via correspondences in geometric examples. As application, we show that given a Lagrangian fibration $X \to B$ admitting local sections over an open subset $V$ with codimension $\ge 2$ complement, there exists a (possibly singular) holomorphic symplectic compactification of the Albanese fibration $A \to V$ (which we show exists as a smooth commutative algebraic group with connected fibers acting on $X_{V}$), as well as of any other torsor over $A$, or over any smooth commutative group scheme over $B$ with connected fibers that is isogenous to $A$.
