No label defined (Q4345360)

By \textit{T. Ekedahl}'s result [in: The Grothendieck Festschrift, Vol. II, Prog. Math. 87, 197-218 (1990; Zbl 0821.14010)] one is able to extend the methods and results of \textit{A. A. Beilinson, J. Bernstein} and \textit{P. Deligne} [``Faisceaux pervers'', Astérisque 100 (1992; Zbl 0536.14011)] to any base field. Thus one may construct a (bounded) derived category \(D^b_c(X,{\mathbb{Z}}_{\ell})\) of constructible \({\mathbb{Z}}_{\ell}\)-sheaves on a variety over any base field. \(D^b_c(X,{\mathbb{Z}}_{\ell})\) has a canonical \(t\)-structure whose heart is the category of constructible \({\mathbb{Z}}_{\ell}\)-sheaves. However, Ekedahl's formalism does not allow to define a well-behaved notion of weights, and one knows the difference [\textit{U. Jannsen}, ``Mixed motives and algebraic \(K\)-theory'', Lect. Notes Math. 1400 (1988; Zbl 0691.14001)] in case the characteristic of the base field is \(p>0\), which was treated by Beilinson, Bernstein and Deligne (loc. cit.), and the characteristic zero case, in particular, number fields. This situation is remedied in the underlying paper. One fixes a 1-dimensional irreducible regular noetherian connected base scheme \(U_0\). Let \(B=\text{ Spec}(K)\) be its generic point, and for base change, introduce the category \(\mathcal U\) dual to the category of all open subschemes of \(U_0\). A \(U_0\)-scheme \(X\) is called horizontal if it is flat and of finite type over \(U_0\). For such \(X\) let \({\mathcal U}X\) denote the direct system of schemes \(X\times_{U_0}U\) for \(U\) in \({\mathcal U}\). One has an equivalence of categories: \[ \{\text{schemes of finite type over }B\}\leftrightarrow\{\text{direct limits of \({\mathcal U}X\) for horizontal \(X\) over }U_0\}, \] in particular, the limit of \({\mathcal U}X\) is representable by \(X_B\). For a horizontal \(U_0\)-scheme \(X\) one defines the derived category \(D^b_c({\mathcal U}X,{\mathbb{Z}}_{\ell})\) as the direct 2-limit of the categories \(D^b_c(X',{\mathbb{Z}}_{\ell})\) in the sense of Ekedahl for \(X'\) in \({\mathcal U}X\). \(D^b_c({\mathcal U}X,{\mathbb{Q}}_{\ell})\) will be the category obtained by tensoring all morphisms in \(D^b_c({\mathcal U}X,{\mathbb{Z}}_{\ell})\) by \({\mathbb{Q}}\). Much of the theory and results from the Beilinson-Bernstein-Deligne book (loc. cit.) carry over to \(D^b_c({\mathcal U}X,{\mathbb{Z}}_{\ell})\), in particular, it has a \(t\)-structure and the objects in its heart are called \textit{perverse \({\mathbb{Z}}_{\ell}\)-sheaves} on \({\mathcal U}X\). Also, the perverse sheaves in \(D^b_c({\mathcal U}X,{\mathbb{Q}}_{\ell})\) form an artinian and noetherian category whose objects are of finite length. For an immersion \(j:V\rightarrow X\) of horizontal \(U_0\)-schemes with \(V\) smooth over \(U_0\) in \(\mathcal U\), and for an irreducible smooth \({\mathbb{Q}}_{\ell}\)-sheaf \(L\) on \(V\), the direct image \(j_{!*}(L[\dim(V)])\) is a simple perverse \({\mathbb{Q}}_{\ell}\)-sheaf on \({\mathcal U}X\). Conversely, any simple perverse \({\mathbb{Q}}_{\ell}\)-sheaf on \({\mathcal U}X\) is of this kind. An interesting question is whether the natural functor \(\eta^*:D_c^b({\mathcal U}X,\mathbb{Z}_\ell)\to D_c^b(X_B,\mathbb{Z}_\ell)\) induced by \(\eta:B\to U_0\) is an equivalence of categories. It follows from \textit{U. Jannsen}'s book (loc. cit.) that this is not the case. One has the result that the heart of the canonical \(t\)-structure on \(D^b_c({\mathcal U}X,{\mathbb{Z}}_{\ell})\) is equivalent to the full subcategory of constructible \({\mathbb{Z}}_{\ell}\)-sheaves on \(X_B\) that extend to some model of \(X_B\). These sheaves are called horizontal. \(\eta^*:D^b_c({\mathcal U}X,{\mathbb{Z}}_{\ell})\to D^b_c(X_B,{\mathbb{Z}}_{\ell})\) and also \(\overline{\eta}^*:D^b_c({\mathcal U}X,{\mathbb{Z}}_{\ell})\to D^b_c(X_{\overline{B}},{\mathbb{Z}}_{\ell})\) are faithful on \({\mathbb{Q}}_{\ell}\)-perverse sheaves. To define weights in the case of a number field \(K\) take \(U_0=\text{ Spec}({\mathcal O}_K[{1/\ell]})\). Thus the residue fields of the closed points of \(U_0\) are finite. This makes the obvious extension of the notion of mixedness for \({\mathbb{Q}}_{\ell}\)-sheaves on a horizontal \(U_0\)-scheme \(X\) as in the book of Beilinson, Bernstein and Deligne (loc. cit.) possible. Let \(D^b_m({\mathcal U}X,{\mathbb{Q}}_{\ell})\) denote the subcategory of \(D^b_c({\mathcal U}X,{\mathbb{Q}}_{\ell})\) consisting of objects \(C\) such that all \(H^i(C)\) become mixed after base change by suitable \(U\in{\mathcal U}\). Then \(D^b_m({\mathcal U}X,{\mathbb{Q}}_{\ell})\) is stable under the six Grothendieck functors. The perverse \(t\)-structure induces a \(t\)-structure on \(D^b_m({\mathcal U}X,{\mathbb{Q}}_{\ell})\), and any subquotient of a mixed perverse \({\mathbb{Q}}_{\ell}\)-sheaf on \({\mathcal U}\) is mixed. The notion of weight for an object \(C\) of \(D^b_m({\mathcal U}X,{\mathbb{Q}}_{\ell})\) can be carried over from the cited book ``Faisceaux pervers''. Many facts thereof apply in this situation, but contrary to the case treated in this book, there are non-trivial morphisms [see the book by \textit{U. Jannsen} (loc. cit.)] in \(D^b_m({\mathcal U}X,{\mathbb{Q}}_{\ell})\) from \(C\) of weight \(\leq w\) to \(L\) with weights \(>w\). The morphisms in \(D^b_c({\mathcal U}X,{\mathbb{Q}}_{\ell})\) are given by direct limits over Jannsen's continuous étale cohomology groups: For a constructible \({\mathbb{Z}}_{\ell}\)-sheaf \(F\) on the horizontal \(U_0\)-scheme \(X\), one has \[ \text{Hom}_{D^b_c({\mathcal U}X,{\mathbb{Z}}_{\ell})}({\mathbb{Z}}_{\ell},F[i])=\lim_{\mathcal U}H^i_{cont}(X_U,F|{}_U). \] For a \(B\)-scheme \(X_B\), any horizontal model \(X\) of \(X_B\) over an open part of \(U_0\) and a horizontal \({\mathbb{Z}}_{\ell}\)-sheaf \(F\) on \(X_B\), one defines the horizontal étale cohomology \[ H^i_{hor}(X_B,F)=\text{Hom}_{D^b_c({\mathcal U}X,{\mathbb{Z}}_{\ell})}({\mathbb{Z}}_{\ell},F[i]). \] This leads to a good cohomology theory with long exact sequences, Poincaré duality and Chern class morphisms. Embedding \(K\) into \({\mathbb{C}}\) gives rise to a comparison between horizontal perverse \({\mathbb{Z}}_{\ell}\)-sheaves on \(X_B\) and Saito's Hodge modules.