On the Jordan-Hölder property for geometric derived categories (Q2445895)

The derived category \(D^b(X)\) of coherent sheaves and its variants are fundamental invariants of algebraic varieties (in the present article \(X\) is always a smooth projective variety over \(\mathbb{C}\)). In particular, finding structures in \(D^b(X)\) is important for many applications. The most frequently considered such structures are semi-orthogonal decompositions, that is, when \(D^b(X)\) is decomposed into generating (admissible) subcategories \(\mathcal{S}_i\) that are semi-orthogonal to each other: \(\mathrm{RHom}^\bullet(\mathcal{S}_i, \mathcal{S}_j)=0\) for \(i<j\). It is a natural question then whether there is a maximal (with respect to the number of elements) such decomposition, and whether such decomposition is unique up to reordering and equivalences of the \(\mathcal{S}_i\). The former is called the Noetherian property, while the latter is called the Jordan-Hölder property in the article. The article shows that in general the Jordan-Hölder property does not hold (assuming the Noetherian property). More precisely, it shows that the standard Godeaux surface (X:=\(\{x_0^5 + x_1^5 + x_2^5 + x^3_5=0\}/\mathbb{Z}_5\)) has two maximal exceptional sequences of different lengths: one with 9 and one with 11 elements. Here an exceptional sequence is a sequence \((E_i \in D^b(X))\) such that \(\mathrm{RHom}^{\bullet}(E_i,E_j)=\mathbb{C}\) if \(i=j\) and \(0\) if \(i<j\). Such sequences yield semi-orthogonal decompositions by taking \(\mathcal{P}_i\) to be the generated subcategories by \(E_i\), and also adding an orthogonal complement to the \(\mathcal{P}_i\). Since both exceptional sequences are maximal, any two maximal semi-orthogonal extensions of the semi-orthogonal decomposition induced by them are different. The sequence with 11 elements was found in an earlier paper of the authors [\textit{C. Böhning} et al., Adv. Math. 243, 203--231 (2013; Zbl 1299.14015)]. The main content of the article is finding a maximal exceptional sequence of length 9 by using the isomorphism \(\mathrm{Pic}(X)/(\mathrm{tors}) \cong \mathrm{Pic}(S)\) (and its corresponding version with \(K_0\)), where \(S\) is a Del-Pezzo surface of degree \(1\).