Bounded derived categories of very simple manifolds (Q763653)

The bounded derived category \(D^{b}(\text{Coh}(X))\) of the abelian category \(\text{Coh}(X)\) of coherent sheaves on a smooth, proper algebraic variety is a \(\mathbb{C}\)-linear Ext-finite triangulated category which is saturated, as shown by \textit{A. Bondal} and \textit{M. van den Bergh} in [Mosc. Math. J. 3, No. 1, 1--36 (2003; Zbl 1135.18302)]. Saturatedness means that any cohomological functor \(H:D^{b}(\text{Coh}(X))\to (\text{Vect-}\mathbb{C})\) of finite type is representable, i. e. there is \(a\in \text{Ob}(D^{b}(\text{Coh}(X))\) such that \(H(-)\simeq \text{Hom}_{D^{b}(\text{Coh}(X))}(-,a)\) as functors. If \(X\) is a smooth, compact complex surface having no complete curve, then Bondal and van den Bergh prove that \(D^{b}(\text{Coh}(X))\) is not saturated. In the present paper, the author generalizes this result to the case of a very simple manifold \(X\), i. e. a compact, connected complex manifold of \(\dim_{\mathbb{C}}(X)\geq 2\) and having no complete irreducible subvariety other than \(X\) and points. If \(X\) is a very simple manifold, the author shows that \(D^{b}(\text{Coh}(X))\) is not saturated by presenting an explicit non-representable cohomological functor of finite type. The proof goes as follows: first, the author defines a torsion pair \((\mathcal{T},\mathcal{F})\) of \(\text{Coh}(X)\) by letting \(\mathcal{T}\) be the full subcategory of \(\text{Coh}(X)\) of torsion sheaves, and \(\mathcal{F}\) the full subcategory of \(\text{Coh}(X)\) of coherent torsion free sheaves. This torsion pair is used to define a \(t\)-structure on \(D^{b}(\text{Coh}(X))\), whose heart is an abelian category \(^{p}\mathcal{C}\). The main technical results is that \(^{p}\mathcal{C}\) is of finite length, i. e. it is a noetherian and artinian abelian category (this is a special case of a more general result of \textit{S. Meinhardt} contained in [Int. J. Math. 23, No. 5, 1250035, 17 p. (2012; Zbl 1240.14008)]). We have then a set \(\{c_{\lambda}\}_{\lambda\in\Lambda}\) of complete representatives of isomorphism classes of simple objects of \(^{p}\mathcal{C}\). If \(^{p}\widehat{\mathcal{C}}\) is the category of left exact functors from \(^{p}\mathcal{C}^{op}\) to \((\text{Vect-}\mathbb{C})\), we have a natural fully faithful, exact, covariant functor \(\iota: ^{p}\mathcal{C}\to ^{p}\widehat{\mathcal{C}}\) sending \(x\) to \(\text{Hom}_{^{p}\mathcal{C}}(\cdot,x)\), inducing a functor \(\widetilde{\iota}:D^{b}(^{p}\mathcal{C})=D^{b}(\text{Coh}(X))\to D^{b}(^{p}\widehat{\mathcal{C}})\). As \(^{p}\widehat{\mathcal{C}}\) is locally noetherian, for every object of \(^{p}\widehat{\mathcal{C}}\) one has an injective hull inside \(^{p}\widehat{\mathcal{C}}\). Letting \(\widehat{c}_{\lambda}\) be an injective hull of \(\iota(c_{\lambda})\), one defines \(e:=\bigoplus_{\lambda\in\Lambda}\widehat{c}_{\lambda}\), which is an injective object of \(^{p}\widehat{\mathcal{C}}\). The author shows that the functor \(E(-):=\text{Hom}_{D^{b}(\text{Coh}(X))}(\widetilde{\iota}(-),e)\) is a non-representable cohomological functor of finite type. The idea of the proof is that if \(E\) was representable, then there would be an object \(a\in D^{b}(\text{Coh}(X))\) such that \(E(-)\simeq \text{Hom}_{D^{b}(\text{Coh}(X))}(-,a)\): using the fact that \(e\) is injective, the author shows that \(a\) can be chosen to be an object of \(^{p}\mathcal{C}\). There are then \(f\in\mathcal{F}\) and \(t\in\mathcal{T}\) and an exact sequence in \(^{p}\mathcal{C}\) of the form \(0\to f\to a\to t[-1]\to 0\). If \(x\) is not in the support of \(t\) nor a singular point of \(f\), then \(\mathbb{C}_{x}[-1]\) is a simple object of \(^{p}\mathcal{C}\). By definition of \(e\) one should have \(E(\mathbb{C}_{x}[-1])=\mathbb{C}\), but the authors shows that \(E(\mathbb{C}_{x}[-1])=0\) by standard cohomological computations, getting a contradiction.