Bivariance, Grothendieck duality and Hochschild homology. II: The fundamental class of a flat scheme-map (Q2445858)

Let \(S\) be a fixed Noetherian scheme and \(\mathcal S\) the category of separated, essentially finite presentation \(S\)-schemes. If \(X \rightarrow S\) is an object of \(\mathcal S\) and \(\delta_X : X \rightarrow X\times_SX\) is the diagonal embedding then the \textit{pre-Hochschild functor} is \[ {\mathcal H}_X := \text{L}\delta_X^\ast \text{R}\delta_{X\ast} : \text{D}(X) \longrightarrow \text{D}(X) \] where \(\text{D}(X)\) is the derived category of the category of \({\mathcal O}_X\)-modules. If \(f : X \rightarrow Y\) is a morphism in \(\mathcal S\) then one defines a morphism \(\gamma_f : \text{L}f^\ast{\mathcal H}_Y \rightarrow {\mathcal H}_X\text{L}f^\ast\) by: \[ \text{L}f^\ast\text{L}\delta_Y^\ast \text{R}\delta_{Y\ast} = \text{L}\delta_X^\ast\text{L}(f\times_Sf)^\ast\text{R}\delta_{Y\ast} \stackrel{{\text{L}\delta_X^\ast \theta}}{\longrightarrow}\text{L}\delta_X^\ast\text{R}\delta_{X\ast}\text{L}f^\ast \] where \(\theta : \text{L}(f\times_Sf)^\ast\text{R}\delta_{Y\ast} \rightarrow \text{R}\delta_{X\ast}\text{L}f^\ast\) is the \textit{base-change morphism} associated to the diagram: \[ \begin{tikzcd} X \ar[r,"f"]\ar[d, "\delta_X" '] & Y\ar[d, "\delta_Y"] \\ X\times_SX \ar[r, "f\times_Sf" ']& Y\times_SY \end{tikzcd} \] The authors of the paper under review show that if \(f\) is essentially étale then \(\gamma_f({\mathcal F}^\bullet)\) is an isomorphism for every \({\mathcal F}^\bullet \in \, \text{D}_{\text{qc}}(Y)\) (this means that \({\mathcal F}^\bullet\) has quasi-coherent cohomology sheaves). \textit{S. Nayak} [Adv. Math. 222, No. 2, 527--546 (2009; Zbl 1175.14003)] proved a factorization result allowing one to extend the compactification theorem of Nagata to morphisms of essentially finite type. As a consequence, one can unify the local and global Grothendieck duality theories and show that there exists a contravariant \(\text{D}_{\text{qc}}^+\)-valued pseudofunctor \((-)^!_+\) over \(\mathcal S\) (called the \textit{twisted inverse image}) such that \(f_+^! : \text{D}_{\text{qc}}^+(Y) \rightarrow \text{D}_{\text{qc}}^+(X)\) is a right-adjoint to \(\text{R}f_\ast\) is \(f : X \rightarrow Y\) is proper (recall that \(\text{L}f^\ast\) is a left-adjoint to \(\text{R}f_\ast\)) and such that \(f_+^! = f^\ast\) if \(f : X \rightarrow Y\) is essentially étale. According to Theorem 5.9 in Nayak's paper, if \(f : X \rightarrow Y\) is a \textit{perfect} morphism in \(\mathcal S\) (this means that \({\mathcal O}_{X,x}\) is a finite Tor-dimension \({\mathcal O}_{Y,f(x)}\)-module, \(\forall \, x \in X\)) then one can extend \(f_+^!\) to a functor \(f^! : \text{D}_{\text{qc}}(Y) \rightarrow \text{D}_{\text{qc}}(X)\) defined by the formula \(f^! := f_+^!{\mathcal O}_Y\otimes^{\text{L}}_{{\mathcal O}_X}\text{L}f^\ast\). (This extension is necessary for the purposes of the paper under review because Hochschild homology involves complexes that are bounded above, not bellow). Let \[ \begin{tikzcd} X^\prime \ar[r, "v"] \ar[d, "g" '] & X\ar[d, "f"] \\ Y^\prime \ar[r,"u" '] & Y \end{tikzcd} \tag{\(*\)} \] be a \textit{Cartesian square} in \(\mathcal S\). It is a basic fact of Grothendieck duality theory that if \(f\) is proper and \(u\) is flat then the base-change morphism for the twisted inverse image \(v^\ast f_+^! \rightarrow g_+^!u^\ast\) which is the adjoint to the composite map: \[ \text{R}g_\ast v^\ast f_+^! \overset{{\theta}^{-1}}\longrightarrow u^\ast\text{R}f_\ast f_+^! \longrightarrow u^\ast \] (with the first map deduced from the base-change isomorphism \(\theta : u^\ast\text{R}f_\ast \overset\sim\rightarrow \text{R}g_\ast v^\ast\) and with the second map deduced from the counit \(\text{R}f_\ast f_+^! \rightarrow \text{id}\)) is an isomorphism (when applied to objects of \(\text{D}_{\text{qc}}^+(Y)\)). In their previous paper [Asian J. Math. 15, No. 3, 451--498 (2011; Zbl 1251.14010)], the authors of the paper under review showed that this isomorphism can be extended to an isomorphism \(\text{B} : v^\ast f^! \overset\sim\rightarrow g^!u^\ast\) (if \(f\) is perfect). In the paper under review the authors define, for any flat morphism \(f : X \rightarrow Y\) in \(\mathcal S\), a functor \(\text{c}_f : {\mathcal H}_Xf^\ast \rightarrow f^!{\mathcal H}_Y\) (called the \textit{fundamental class} of \(f\)) and verify its \textit{transitivity} with respect to compositions \(X \overset{f}\rightarrow Y \overset{g}\rightarrow Z\) of flat morphisms and its compatibility with \textit{flat base change}. In order to understand the nature of the difficulties encountered by the authors of the paper under review in proving their results we shall reproduce their definition of the fundamental class. Let \(f : X \rightarrow S\) be a flat morphism in \(\mathcal S\). Consider the diagram \[ \begin{tikzcd} X \ar[r, "\delta_f"] & X\times_YX \ar[d,"p_2" '] \ar[r,"p_1"]& X\ar[d,"f"]\\ {}& X \ar[r,"f" ']& Y \end{tikzcd} \] Let \(\text{R}\delta_{f\ast} \rightarrow p_2^!\) be the morphism deduced from the counit \(\text{R}\delta_{f\ast}\delta_{f+}^!{\mathcal O}_{X\times_YX} \rightarrow {\mathcal O}_{X\times_YX}\) and from the fact that \(p_2 \circ \delta_f = \text{id}_X\) and consider the composite morphism \[ \begin{tikzcd} \mathrm R\delta_{f\ast}f^\ast \rar & p_2^!f^\ast \rar["\mathrm{B}^{-1}"] & p_1^\ast f^! \rlap{\,.} \end{tikzcd} \] Consider, next, the Cartesian square: \[ \begin{tikzcd} X\times_YX \rar["p_1"]\dar["i" '] & X\dar["\Gamma"]\\ X\times_SX \rar["\text{id}_X\times_Sf" '] & X\times_SY \end{tikzcd} \] where \(\Gamma\) is the graph of \(f\). Apply \(\text{R}i_\ast\) to the above defined composite morphism and consider the composite morphism: \[ \begin{tikzcd} \text{R}\delta_{X\ast}f^\ast = \text{R}i_\ast\text{R}\delta_{f\ast}f^\ast \rar \mathrm{R}i_\ast p_1^\ast f^! \rar["\theta^{-1}f^!"] & (\text{id}_X\times_Sf)^\ast \text{R}\Gamma_\ast f^! \end{tikzcd} \] where \(\theta : (\text{id}_X\times_Sf)^\ast \text{R}\Gamma_\ast \rightarrow \text{R}i_\ast p_1^\ast\) is the base-change isomorphism. Applying \(\text{L}\delta_X^\ast = \text{L}\delta_f^\ast \text{L}i^\ast\) to this composite morphism and taking into account that \(p_1 \circ \delta_f = \text{id}_X\) one gets a composite morphism: \[ \text{a}_f : \text{L}\delta_X^\ast \text{R}\delta_{X\ast} f^\ast \rightarrow \text{L}\delta_f^\ast \text{L}i^\ast (\text{id}_X\times_Sf)^\ast \text{R}\Gamma_\ast f^! \overset\sim\longrightarrow \text{L}\delta_f^\ast p_1^\ast \text{L}\Gamma^\ast \text{R}\Gamma_\ast f^! \overset\sim\longrightarrow \text{L}\Gamma^\ast \text{R}\Gamma_\ast f^! \, . \] On the other hand, consider the Cartesian square: \[ \begin{tikzcd} X \rar["f"]\dar["\Gamma" '] & Y\dar["\delta_Y"]\\ X\times_SY \rar["f\times_S\text{id}_Y" '] & Y\times_SY \end{tikzcd} \] Applying \(\text{L}\Gamma^\ast\) to the base-change isomorphism \(\theta : (f\times_S\text{id}_Y)^\ast \text{R}\delta_{Y\ast} \overset\sim\rightarrow \text{R}\Gamma_\ast f^\ast\) one gets an isomorphism \(\phi : f^\ast \text{L}\delta_Y^\ast \text{R}\delta_{Y\ast} \overset\sim\rightarrow \text{L}\Gamma^\ast \text{R}\Gamma_\ast f^\ast\). If \(p : X\times_SY \rightarrow X\) is the projection on the first factor then the \textit{projection formula} provides, for \({\mathcal A}^\bullet,\, {\mathcal B}^\bullet \in \text{D}_{\text{qc}}(X)\), an isomorphism: \[ \beta : \text{L}p^\ast {\mathcal A}^\bullet \otimes_{{\mathcal O}_{X\times_SY}}^{\text{L}} \text{R}\Gamma_\ast {\mathcal B}^\bullet \overset\sim\longrightarrow \text{R}\Gamma_\ast(\text{L}\Gamma^\ast \text{L}p^\ast {\mathcal A}^\bullet \otimes_{{\mathcal O}_X}^{\text{L}}{\mathcal B}^\bullet) \overset\sim\rightarrow \text{R}\Gamma_\ast({\mathcal A}^\bullet \otimes_{{\mathcal O}_X}^{\text{L}} {\mathcal B}^\bullet) \] from which one deduces an isomorphism: \[ \begin{gathered} \begin{tikzcd} \text{L}\Gamma^\ast \text{R}\Gamma_\ast({\mathcal A}^\bullet \otimes_{{\mathcal O}_X}^{\text{L}} {\mathcal B}^\bullet) \rar["\text{L}\Gamma^\ast \beta^{-1}"] & \text{L}\Gamma^\ast (\text{L}p^\ast {\mathcal A}^\bullet \otimes_{{\mathcal O}_{X\times_SY}}^{\text{L}}\text{R}\Gamma_\ast {\mathcal B}^\bullet) \overset\sim\rightarrow \end{tikzcd}\\ \text{L}\Gamma^\ast \text{L}p^\ast {\mathcal A}^\bullet \otimes_{{\mathcal O}_X}^{\text{L}}\text{L}\Gamma^\ast \text{R}\Gamma_\ast {\mathcal B}^\bullet \overset\sim\rightarrow {\mathcal A}^\bullet \otimes_{{\mathcal O}_X}^{\text{L}} \text{L}\Gamma^\ast \text{R}\Gamma_\ast {\mathcal B}^\bullet \, . \end{gathered} \] Let \(\text{b}_f\) be the composite isomorphism: \[ \begin{gathered} \text{b}_f : \text{L}\Gamma^\ast \text{R}\Gamma_\ast f^! = \text{L}\Gamma^\ast \text{R}\Gamma_\ast (f_+^!{\mathcal O}_Y \otimes_{{\mathcal O}_X}^{\text{L}} f^\ast ) \overset\sim\rightarrow f_+^!{\mathcal O}_Y \otimes_{{\mathcal O}_X}^{\text{L}} \text{L}\Gamma^\ast \text{R}\Gamma_\ast f^\ast \rightarrow\\ \begin{tikzcd} {}\rar["\text{id} \otimes \phi^{-1}"] & f_+^!{\mathcal O}_Y \otimes_{{\mathcal O}_X}^{\text{L}} f^\ast \text{L}\delta_Y^\ast \text{R}\delta_{Y\ast} = f^!\text{L}\delta_Y^\ast \text{R}\delta_{Y\ast} . \end{tikzcd} \end{gathered} \] The \textit{fundamental class} \(\text{c}_f : {\mathcal H}_Xf^\ast \rightarrow f^! {\mathcal H}_Y\) is, by definition, \(\text{b}_f \circ \text{a}_f\). The authors of the paper under review prove, after nontrivial verifications of diagram commutativities, that: \begin{itemize} \item[(1)] If \(f\) is essentially étale, so that \(f^! = f^\ast\), then \(\text{c}_f\) is the inverse of the already established isomorphism \(\gamma_f : f^\ast {\mathcal H}_Y \overset\sim\rightarrow {\mathcal H}_Xf^\ast\). \item[(2)] If \(X \overset{f}\rightarrow Y \overset{g}\rightarrow Z\) are flat morphisms in \(\mathcal S\) then \(\text{c}_{gf} = f^!\text{c}_g \circ \text{c}_fg^\ast\). This result shows that fundamental classes are \textit{orientations} for the flat maps in the \textit{bivariant Hochschild theory} constructed by the authors in their previous paper mentioned above. \item[(3)] For any Cartesian square \((*)\) in \(\mathcal S\) with \(f,\, g,\, u,\, v\) flat morphisms, the following diagram commutes: \[ \begin{tikzcd} v^\ast {\mathcal H}_Xf^\ast \rar["v^\ast \mathrm{c}_f"]\dar["\gamma_vf^\ast" '] & v^\ast f^! \mathcal H_Y\dar["\wr" ', "\mathrm B\mathcal H_Y"]\\ \mathcal H_{X^\prime}v^\ast f^\ast \dar[equal] & g^!u^\ast \mathcal H_Y \dar["g^!\gamma_u"]\\ \mathcal H_{X^\prime}g^\ast u^\ast \rar["\mathrm c_gu^\ast" '] & g^!{\mathcal H}_{Y^\prime}u^\ast \end{tikzcd} \] \item[(4)] For any separated, essentially finite type, flat \(S\)-scheme \(\xi : X \rightarrow S\) one gets, by composing the fundamental class map \(\text{c}_\xi({\mathcal O}_S) : {\mathcal H}_X({\mathcal O}_X) \rightarrow \xi^!{\mathcal O}_S\) with the natural product map \({\mathcal H}_X({\mathcal O}_X) \otimes_{{\mathcal O}_X}^{\text{L}} {\mathcal H}_X({\mathcal O}_X) \rightarrow {\mathcal H}_X({\mathcal O}_X)\), a \textit{duality map}: \[ {\mathcal H}_X({\mathcal O}_X) \longrightarrow \text{R}{\mathcal H}om_{{\mathcal O}_X}({\mathcal H}_X({\mathcal O}_X),\, \xi^!{\mathcal O}_S)\, . \] If \(\xi\) is \textit{essentially smooth} then this duality map is an isomorphism. \end{itemize}