Splitting mixed Hodge structures over affine invariant manifolds (Q280110): Difference between revisions

The Hodge bundle \(\Omega M_g\) over the moduli space \(M_g\) parametrizes the pairs \((X,\omega)\) of a Riemann surface \(X\) of genus \(g\) and a holomorphic one-form \(\omega \in \Gamma(X,\Omega^1_X)\). Let \(g \geq 2\), choose a partition \(k\) of \( 2g-2\) and consider inside \(\Omega M_g\) the stratum \({\mathcal H}(k)\), this is the family of pairs \((X,\omega)\), also called flat surfaces, such that the divisor of \(\omega\) has multiplicities prescribed by the partition. The support of \(\mathrm{div}(\omega)\) is a set of \(|k|\) points, the first homology of \(X\) relative to it has therefore dimension \(N:=2g + |k| -1\). Fix a basis then integration of \(\omega\) along the chosen generators gives locally a period map from \({\mathcal H}(k)\) to \(\mathbb C ^N\). It turns out that \({\mathcal H}(k)\) is a orbifold and that on the complement of orbifolds points the period maps are an atlas of charts, with transition functions in \(\mathrm{GL}(N,\mathbb Z)\). An affine invariant submanifold is an immersed manifold \(\mathcal Y \rightarrow {\mathcal H}(k)\) such that each point of \(\mathcal Y\) has a neighborhood whose image is cut by linear homogeneous polynomials in the period coordinates with real coefficients. The main result of the paper is the solution of a remarkable question, it is the proof that affine invariant manifolds in strata of flat surfaces are algebraic varieties and are in fact defined over \(\overline{\mathbb Q}\). The author explains that ``the demonstration uses a generalization of a theorem of \textit{M. Möller} [J. Am. Math. Soc. 19, No. 2, 327--344 (2006; Zbl 1090.32004)], to the effect that the image of a certain twisted Abel-Jacobi map lands in the torsion of a factor of the Jacobians. This statement can be viewed as a splitting of certain mixed Hodge structures.'' It is too complicated to recount in detail the precise meaning of the preceding words. Still we recall the exact sequence of vector bundles over \({\mathcal H}(k)\) \(0\to W_0 \to H^1_{\mathrm{rel}} \mathop{\rightarrow}\limits^{p} H^1 \to 0\), it comprises the natural variation of mixed Hodge structures, \(W_0\) being the local system of purely relative cohomology classes, which is just the reduced cohomology of the support of the divisor above. Also it is useful to note that \(H^1_{rel}\) is identified with the tangent bundle \(T{\mathcal H}(k)\). Over the affine invariant submanifold \(\mathcal Y\) the variation of pure Hodge structure given by \(H^1\) splits, according to work of the author [``Semisimplicity and rigidity of the Kontsevich-Zorich cocycle'', \url{arXiv:1307.7314}], as \(H^1 = (H^1_{\iota_0}\oplus\cdots \oplus H^1_{\iota_{r-1}}) \oplus V\), where the indexes correspond to the embeddings of a certain total real number field \(k\). Moreover the Jacobians of the Riemann surfaces corresponding to the points of \(\mathcal Y\) have real multiplication by \(k\). Restricting to the first summand and pulling back via \(p\) one obtains the following extension \(E\), \(0\to W_0 \to E \mathop{\rightarrow}\limits^{p} \left(H^1_{\iota_0}\oplus\cdots \oplus H^1_{\iota_{r-1}}\right) \to 0\) which is a variation of mixed Hodge structures: It happens that the tangent bundle \(T{\mathcal Y} \subset E \subset T{\mathcal H}(k)\) with projection \(p (T{\mathcal Y}) = H^1_{\iota_0}\). Let \( W_0 {\mathcal Y} := W_0 \cap T{\mathcal Y}\), Filip considers the following sequence of variation of mixed Hodge structures: \[ 0\to W_0 / W_0 {\mathcal Y} \to p^{-1} (H^1_{\iota_0}) /W_0 {\mathcal Y} \to H^1_{\iota_0} \to 0. \] The splitting theorem is the statement that the preceding sequence is pointwise split, and so they are the corresponding ones involving the other summands \(H^1_{\iota_n}\). According to the experts, see \textit{A. Wright} in [EMS Surv. Math. Sci. 2, No. 1, 63--108 (2015; Zbl 1372.37090)], the proof crucially uses dynamics and no other demonstration is known. The algebraicity property is eventually obtained by considering the algebraic locus \(\mathcal A\) in \(\mathcal H\) where \((X, \omega)\) admits real multiplication with the same behaviour as in \(\mathcal Y\). By using the splitting above the author is able to control the tangent bundles in such a manner that he can show that a natural algebraic subvariety in \(\mathcal A\) must coincide with \(\mathcal Y\).