Duality with expanding maps and shrinking maps, and its applications to Gauss maps (Q2439296)

The author of the paper under review extends to the positive (actually, arbitrary) characteristic case a number of results, previously known only in characteristic 0 (or under some additional assumptions), such as the linearity of general fibers of separable Gauss maps (proven by \textit{P. Griffiths} and \textit{J. Harris} [Ann. Sci. Éc. Norm. Supér. (4) 12, 355--452 (1979; Zbl 0426.14019)]) in characteristic 0 and by \textit{S. Kleiman} and \textit{R. Piene} [Contemp. Math. 123, 107--129 (1991; Zbl 0758.14032)] under the reflexivity assumption) or the characterization of the images of separable Gauss maps due, in characteristic 0, to J.M. Landsberg and, independently, to J. Piontkowski (cf. \textit{T. A. Ivey} and \textit{J. M. Landsberg} [Cartan for beginners: Differential geometry via moving frames and exterior differential systems. Providence, RI: American Mathematical Society (AMS) (2003; Zbl 1105.53001)] and [\textit{G. Fischer} and \textit{J. Piontkowski}, Ruled varieties. An introduction to algebraic differential geometry. Braunschweig: Vieweg (2001; Zbl 0976.14025)]). Before stating the results, we need to recall the definition of expanding maps (due to the author of the paper under review) and of shrinking maps (studied by Landsberg and Piontkowski) and the related formalism. Let \({\mathbb P}^N\) be the projective \(N\)-space over an algebraically closed field \(K\) of arbitrary characteristic, let \(\mathcal X\) be a subvariety of the Grassmannian \({\mathbb G}(m,\, {\mathbb P}^N)\) of \(m\)-dimensional linear subvarieties of \({\mathbb P}^N\), and let: \[ \begin{tikzcd} 0 \rar & S \rar["u"] & \mathcal O_{\mathcal X}\otimes_K W \rar["p"] & Q \rar & 0 \end{tikzcd} \] be the restriction to \(\mathcal X\) of the tautological exact sequence on \({\mathbb G}(m,\, {\mathbb P}^N)\), where \(W = \text{H}^0({\mathbb P}^N, {\mathcal O}_{{\mathbb P}^N}(1))\) and \(\text{rk}\, Q = m + 1\). The composite map: \[ \begin{tikzcd} S \rar["u"] & \mathcal O_{\mathcal X}\otimes_K W \rar["d_{\mathcal X}\otimes \, \mathrm{id}"] & \Omega_{\mathcal X}\otimes_K W \simeq \Omega_{\mathcal X}\otimes_{{\mathcal O}_{\mathcal X}} ({\mathcal O}_{\mathcal X}\otimes_K W) \rar["\mathrm{id} \otimes p"] & \Omega_{\mathcal X}\otimes_{{\mathcal O}_{\mathcal X}} Q \end{tikzcd} \] turns out to be a morphism of \({\mathcal O}_{\mathcal X}\)-modules \(\varphi : S \rightarrow \Omega_{\mathcal X}\otimes Q\). Let \({\mathcal X}^+\) be the open subset of the nonsingular locus \({\mathcal X}^{\text{sm}}\) of \(\mathcal X\) over which \(\varphi\) has maximum rank, let \(S^+ := \text{Ker}\, \varphi \, |_{{\mathcal X}^+}\), \(u^+ : S^+ \rightarrow {\mathcal O}_{{\mathcal X}^+}\otimes_K W\) the restriction of \(u\) and \(Q^+ := \text{Coker}\, u^+\). If \(\text{rk}\, Q^+ = m^+ + 1\) then the exact sequence: \[ \begin{tikzcd} 0 \rar & S^+ \rar["u^+"] &\mathcal O_{\mathcal X^+}\otimes_K W \rar & Q^+\rar & 0 \end{tikzcd} \] defines a morphism \(\gamma : {\mathcal X}^+ \rightarrow {\mathbb G}(m^+,\, {\mathbb P}^N)\) called the \textit{expanding map} of \(\mathcal X\). Applying the same kind of construction to the dual exact sequence: \[ \begin{tikzcd} 0 \rar & Q^\vee \rar & \mathcal O_{\mathcal X}\otimes_K W^\vee \rar & S^\vee \rar & 0 \end{tikzcd} \] one gets a morphism of \({\mathcal O}_{\mathcal X}\)-modules \(\psi : Q^\vee \rightarrow \Omega_{\mathcal X}\otimes S^\vee\). Let \({\mathcal X}^-\) be the open subset of \({\mathcal X}^{\text{sm}}\) over which \(\psi\) has maximum rank, \(Q^- := \text{Coker}\, \psi^\vee \, |_{{\mathcal X}^-}\), \(p^-\) the composite morphism \({\mathcal O}_{{\mathcal X}^-}\otimes_K W \overset{p}\longrightarrow Q \rightarrow Q^-\), and \(S^- := \text{Ker}\, p^-\). If \(\text{rk}\, Q^- = m^- + 1\) then the exact sequence: \[ \begin{tikzcd} 0 \rar & S^- \rar & \mathcal O_{{\mathcal X}^-}\otimes_K W \rar["p^-"] & Q^- \rar & 0 \end{tikzcd} \] defines a morphism \(\sigma : {\mathcal X}^- \rightarrow {\mathbb G}(m^-,\, {\mathbb P}^N)\) called the \textit{shrinking map} of \(\mathcal X\). These constructions are compatible with dominant, \textit{separable} (that is, generically smooth) morphisms \(f : {\mathcal X} \rightarrow {\mathcal Y}\) because for such morphisms the map \(f^\ast \Omega_{\mathcal Y} \rightarrow \Omega_{\mathcal X}\) is, generically, a monomorphism. They can be studied locally using the following observation: if one has a \(K\)-vector space decomposition \(W = W^\prime \oplus W^{\prime \prime}\) then the morphism \(\varphi : {\mathcal O}_{\mathcal X}\otimes_K W^\prime \rightarrow \Omega_{\mathcal X}\otimes_K W^{\prime \prime}\) associated to a short exact sequence of the form: \[ \begin{tikzcd} 0 \rar & \mathcal O_{\mathcal X}\otimes_K W^\prime \rar["\binom{-\text{id}}{\alpha}"] & \mathcal O_{\mathcal X}\otimes_K W \rar["{(\alpha, \mathrm{id})}"] & \mathcal O_{\mathcal X}\otimes_K W^{\prime \prime} \rar & 0 \end{tikzcd} \] is defined by the matrix obtained by applying \(d_{\mathcal X}\) to the entries of the matrix defining \(\alpha : {\mathcal O}_{\mathcal X}\otimes_K W^\prime \rightarrow {\mathcal O}_{\mathcal X}\otimes_K W^{\prime \prime}\). One deduces, in the global case, that if \(\varphi^\prime\) is the composite map: \[ \begin{tikzcd} T_{\mathcal X} \otimes S \rar["\text{id} \otimes \varphi"] & T_{\mathcal X} \otimes \Omega_{\mathcal X} \otimes Q \rar & Q \end{tikzcd} \] (recall that \(T_{\mathcal X} = {\mathcal H}om_{{\mathcal O}_{\mathcal X}}(\Omega_{\mathcal X}, {\mathcal O}_{\mathcal X})\)) then \(\psi^\vee = - \varphi^\prime\). One also deduces that if \(m = 0\), i.e., if \({\mathcal X}\subset {\mathbb P}^N\), then the expanding map \(\gamma\) coincides with the \textit{Gauss map} \({\mathcal X}^{\text{sm}} \rightarrow {\mathbb G}(\dim {\mathcal X},\, {\mathbb P}^N)\) associating to \(x \in {\mathcal X}^{\text{sm}}\) the embedded tangent space \({\mathbb T}_x{\mathcal X} \subset {\mathbb P}^N\). Now, the main results of the paper under review are the following ones: \begin{itemize} \item[(1)] Let \(\gamma : X^{\text{sm}} \rightarrow {\mathbb G}(\dim X,\, {\mathbb P}^N)\) be the Gauss map of a projective variety \(X \subset {\mathbb P}^N\) and \({\mathcal Y} \subset {\mathbb G}(\dim X,\, {\mathbb P}^N)\) the closure of the image of \(\gamma\). If \(\gamma : X^{\text{sm}} \rightarrow {\mathcal Y}\) is separable then the closure in \({\mathcal Y} \times {\mathbb P}^N\) of the graph of \(\gamma\) coincides with the closure of \({\mathbb P}((Q_{\mathcal Y}^-)^\vee) \subset {\mathcal Y}^- \times {\mathbb P}^N\). (The author's convention for projective bundles is \({\mathbb P}(E) := \text{Proj}(\mathrm{Sym} E^\vee)\) so that \({\mathbb P}^N = {\mathbb P}(W^\vee)\) and \({\mathbb P}^{N\vee} = {\mathbb P}(W)\)). \item[(2)] Assume that \({\mathcal X} \subset {\mathbb G}(m,\, {\mathbb P}^N)\) and let \(Y \subset {\mathbb P}(W) = {\mathbb P}^{N\vee}\) be the closure of the image of the projection \({\mathbb P}(S^+) \rightarrow {\mathbb P}(W)\). Since \({\mathbb P}^{N\vee} = {\mathbb G}(N-1,\, {\mathbb P}^N)\) the shrinking map \(\sigma_Y\) associates to \(y \in Y^{\text{sm}}\) the point \(\sigma_Y(y)\) of \({\mathbb G}(N - 1 - \dim Y, \, {\mathbb P}^N)\) corresponding to the intersection of the hyperplanes in \({\mathbb P}^N\) corresponding to the points of the embedded tangent space \({\mathbb T}_yY \subset {\mathbb P}^{N\vee}\). If \(Y\) has dimension \(N - m - 1\) and if \({\mathbb P}(S^+) \rightarrow Y\) is separable then \(\mathcal X\) is the closure of \(\sigma_Y(Y^{\text{sm}})\) and \(\sigma_Y : Y^{\text{sm}} \rightarrow {\mathcal X}\) is separable. \textit{Dually}, assume that \({\mathcal Y} \subset {\mathbb G}(m^+,\, {\mathbb P}^N)\) and let \(X \subset {\mathbb P}(W^\vee) = {\mathbb P}^N\) be the closure of the image of the projection \({\mathbb P}((Q_{\mathcal Y}^-)^\vee) \rightarrow {\mathbb P}(W^\vee)\). If \(\dim X = m^+\) and if \({\mathbb P}((Q_{\mathcal Y}^-)^\vee) \rightarrow X\) is separable then \(\mathcal Y\) is the closure of the image of the Gauss map \(\gamma : X^{\text{sm}} \rightarrow {\mathbb G}(\dim X,\, {\mathbb P}^N)\) and \(\gamma : X^{\text{sm}} \rightarrow {\mathcal Y}\) is separable. \item[(3)] In the final section of the paper under review, the author uses the above results to establish a duality on 1-dimensional developable paramater spaces via expanding and shrinking maps. More precisely, consider \({\mathcal X} \subset {\mathbb G}(m,\, {\mathbb P}^N)\), let \(\pi : {\mathbb P}(Q^\vee) \rightarrow {\mathbb P}^N\) be the projection and put \(X = \pi({\mathbb P}(Q^\vee))\). \(\mathcal X\) is \textit{developable} if, for any general point \(x\) of \(\mathcal X\), the Gauss map \(\gamma_X\) is constant on \({\mathbb P}(Q^\vee(x)) \cap X^{\text{sm}}\). \(\gamma^i_{\mathcal X}\) is defined inductively by \(\gamma^1_{\mathcal X} = \gamma_{\mathcal X} =\) the expanding map of \(\mathcal X\), \(\gamma^i_{\mathcal X} := \gamma_{\gamma^{i-1}{\mathcal X}} \circ \gamma^{i-1}_{\mathcal X}\), and \(\sigma^i_{\mathcal X}\) is defined similarly. Moreover, if \(TX\) denotes the closure in \({\mathbb P}^N\) of \(\bigcup_{x\in X^{\text{sm}}}{\mathbb T}_xX\) then \(T^iX\) is defined inductively by \(T^1X = TX\), \(T^iX = T(T^{i-1}X)\). (If \(C \subset {\mathbb P}^N\) is a curve then, in characteristic 0, \(T^iC\) coincides with the \textit{oscullating scroll} of order \(i\) of \(C\)). \end{itemize} The main result of the final section of the paper under review asserts that if one considers 1-dimensional subvarieties \({\mathcal X} \subset {\mathbb G}(m,\, {\mathbb P}^N)\), \({\mathcal X}^\prime \subset {\mathbb G}(m^\prime ,\, {\mathbb P}^N)\) and a nonnegative integer \(\varepsilon\) then the following are equivalent: \begin{itemize} \item[(a)] \({\mathcal X}^\prime\) is developable, \(X^\prime\) is nondegenerate and is not a cone, \(\gamma^\varepsilon_{{\mathcal X}^\prime}\) is separable and \(\gamma^\varepsilon_{{\mathcal X}^\prime}{\mathcal X}^\prime = {\mathcal X}\); \item[(b)] \({\mathcal X}\) is developable, \(X\) is nondegenerate and is not a cone, \(\sigma^\varepsilon_{\mathcal X}\) is separable and \(\sigma^\varepsilon_{\mathcal X}{\mathcal X} = {\mathcal X}^\prime\). \end{itemize} Moreover, in this case, \(m = m^\prime + \varepsilon\) and \(X = T^\varepsilon X^\prime\).