A support theorem for the Hitchin fibration: the case of \(\operatorname{GL}_n\) and \(K_C\) (Q2054203)

The authors compute the supports of the perverse cohomology sheaves of the Hitchin fibration for \(\operatorname{GL}_n\) over the locus of reduced spectral curves. In contrast to the case of meromorphic Higgs fields they find additional supports at the loci of reducible spectral curves. Their contribution to the global cohomology is governed by a finite twist of Hitchin fibrations for Levi subgroups. The corresponding summands give non-trivial contributions to the cohomology of the moduli spaces for every \(n\geq2\). A key ingredient is a restriction result for intersection cohomology sheaves that allows them to compare the fibration to the one defined over versal deformations of spectral curves. This paper is organized as follows: Section 1 is an introduction to the subject and summarizes the main results. In Section 2 the authors set up notation and conventions. In Section 3 they recall the main result from [\textit{L. Migliorini} and \textit{V. Shende}, Algebr. Geom. 5, No. 1, 114--130 (2018; Zbl 1406.14005)], that constrains the potential supports of their perverse cohomology sheaves in terms of higher discriminants. The symplectic structure of Hitchin fibrations allows them to describe these in terms of the action of Jacobians of spectral curves. Section 4 combines the Ngô support theorem and results on compactified Jacobians and concerns the fact that the supports have to be partition strata. Section 5 is devoted to comparison with versal families. Here the authors prove the restriction result for IC-sheaves and show that it applies to the Hitchin fibration by computing the Kodaira-Spencer map for the universal family of spectral curves. Section 6 concerns the fact that the partition strata are supports. Here the authors use the Cattani-Kaplan-Schmid complex for the versal family to translate the problem of determining the generic fibers into a combinatorial problem that they can then solve. They describe the monodromy of these summands, in order to prove that the new summands contribute to the global cohomology of the Hitchin fibration for any \(n\geq2\).