Hitchin map for the moduli space of \(\Lambda \)-modules in positive characteristic (Q6052912)

Let \(\mathbb{K}\) be an algebraically closed field of characteristic \(p > 0\) and \(\pi \colon X \to S\) a morphism of schemes of finite type over \(\mathbb{K}\). A sheaf of rings of differential operators over \(X/S\) is a sheaf \(\Lambda\) of \({\mathcal O}_S\)-algebras over \(X\) admitting a filtration \(\Lambda_0 \subset \Lambda_1 \subset \cdots\) satisfying certain conditions (Definition 2.1). This notion was introduced by Simpson (Zbl 0891.14005) for varieties over \(\mathbb{C}\), as a unifying structure for \({\mathcal D}_X\)-modules and Higgs sheaves over \(X\). Suppose that \(\Lambda_0 \cong {\mathcal O}_X\). Given \(D \in \Lambda_1\), the commutator \([ D, - ]\) defines an \({\mathcal O}_S\)-derivation \(\delta_D \colon {\mathcal O}_X \to {\mathcal O}_X\). This defines a map \(\Lambda_1 \to \mathrm{Der}_{{\mathcal O}_S} ( {\mathcal O}_X, {\mathcal O}_X ) = T_{X/S}\) which factorises via the \textsl{anchor map} \[ \delta \colon H \ \to \ T_{X/S} , \] where \(H\) is the quotient \(\frac{\Lambda_1}{\Lambda_0}\). Generalising Langer's notion (Zbl 1330.14017) of a restricted sheaf of Lie algebroids, the authors define a sheaf \(\Lambda\) of rings of differential operators to be \textsl{restricted} if it admits a \(p\)-structure, which is a map \([p] \colon \Lambda_1 \to \Lambda_1\) extending the \(p\)th power operation on \({\mathcal O}_X\) and satisfying certain conditions (Definition 2.6). They derive various properties of such restricted \(\Lambda\) and give several examples. They also classify all \(p\)-structures on a given \(\Lambda\). Now a \textsl{\(\Lambda\)-module over \(X\)} is a coherent \({\mathcal O}_X\)-module \(E\) equipped with an \({\mathcal O}_X\)-linear structure map \(\nabla \colon \Lambda \otimes_{{\mathcal O}_X} E \to E\) extending the \({\mathcal O}_X\)-structure. Given \(D \in \Lambda_1\), the authors define the \textsl{\(p\)-curvature} by \[ \psi_\nabla (D) \ := \ (\nabla_D )^p - \nabla_{D^{[p]}} . \] This determines a map \(\psi_\nabla \colon \Lambda_1 \to \mathrm{End}_{{\mathcal O}_S} ( E )\). The authors show that this factors via a \(p\)-linear map \(H \to \mathrm{End}_{{\mathcal O}_X} ( E )\), and thus gives an \({\mathcal O}_X\)-linear map \(F^* H \to \mathrm{End}_{{\mathcal O}_X} ( E )\) where \(F\) is the Frobenius map of \(X\). This in turn defines an \(F^* H^\vee\)-valued Higgs field on \(E\), which is also denoted \(\psi_\nabla\). The association \[ ( E, \psi_\nabla ) \ \mapsto \ \left( \mathrm{tr} ( \wedge^k \psi_\nabla ) \right)_{k=1}^r \] defines a \textsl{Hitchin map} \[ h_\Lambda \colon {\mathcal M}_X^\Lambda ( r, P ) \ \to \ {\mathcal A}_r ( X, F^* H^\vee ) \] where \({\mathcal M}_X^\Lambda ( r, P )\) is the moduli space of Gieseker semistable \(\Lambda\)-modules with Hilbert polynomial \(P\) over \(X\) (see Langer [Zbl 1080.14014], [Zbl 1086.14036] and Simpson [Zbl 0891.14005]) and \[ {\mathcal A}_r ( X, F^* H^\vee) \ := \ \bigoplus_{k=1}^r H^0 \left( X, \mathrm{Sym}^k ( F^* H^\vee ) \right) \] is the \textsl{Hitchin base}. The main result of the paper (Theorem 6.6) is that if \(X \to \mathrm{Spec} ( \mathbb{K} )\) is integral and projective and in addition the anchor map \(\delta \colon H \to T_X\) is generically surjective, then the Hitchin map \(h_\Lambda\) factorises \[ {\mathcal M}_X^\Lambda ( r, P ) \ \to \ {\mathcal A}_r ( X, H^\vee ) \ \to \ {\mathcal A}_r ( X, F^* H^\vee ) \] where the second map is pullback of global sections via the Frobenius map. The authors show by example that Theorem 6.6 need not hold without the condition of generic surjectivity, and show that it holds in some interesting situations when \(X\) is a curve. They end with a relative version of Theorem 6.6 for a certain \(\Lambda\)-module over \(X \times \mathbb{A}^1\). This is a highly readable and interesting paper with a generous supply of examples.