On the splitting of Lazarsfeld-Mukai bundles on \(K3\) surfaces (Q897773)

Lazarsfeld-Mukai bundles are a very important class of vector bundles on \(K3\) surfaces. They have applications, among others, to Brill-Noether theory and the classification of Fano varieties; see [\textit{M. Aprodu}, Commutative algebra. Expository papers dedicated to David Eisenbud on the occasion of his 65th birthday. New York: Springer. 1--23 (2013; Zbl 1266.14034)] for an overview. Concretely, for \(C\subset X\) a smooth curve in a \(K3\) surface, such that \(\mathcal O_X(C)\) is globally generated, and a base point free divisor \(Z\) on \(C\), the associated Lazarsfeld-Mukai bundle \(E_{C,Z}:=F_{C,Z}^\vee\) is given by the dual of the kernel of the evaluation map \[ 0\to F_{C,Z}\to H^0(\mathcal O_C(Z))\otimes \mathcal O_X\rightarrow{\mathsf{ev}} \mathcal O_C(Z)\to 0\,. \] If \(\text{rank} E_{C,Z}=h^0(\mathcal O_C(Z))=2\) and \(E_{C,Z}\) is non-simple, which means that \(\text{hom}(E_{C,Z}, E_{C,Z})\geq2\), the question whether or not the Lazarsfeld-Mukai bundle splits is settled in Section 3 of [\textit{M. Aprodu} and \textit{G. Farkas}, Compos. Math. 147, No. 3, 839--851 (2011; Zbl 1221.14039)]. Namely, there is a unique extension \[ 0\to M\to E_{C,Z}\to N\otimes \mathcal I_{Z'}\to 0, \] called the \textit{Donagi-Morrison extension}, where \(M,N\in \text{Pic}(X)\) such that \(N\) is globally generated, \(h^0(M),h^0(N)\geq 2\), and \(Z'\subset X\) is a zero-dimensional subscheme. The bundle \(E_{C,Z}\) splits into a direct sum of two line bundles if and only if the Donagi-Morrison extension splits, in which case \(Z'=\emptyset\) holds. In the paper under review, the author studies the question whether a simple Lazarsfeld-Mukai bundle of rank 2 can be written as an extension of two line bundles under the additional assumption that \(\mathcal O_X(1):=\mathcal O_X(C)\) is very ample. First, he shows that, under this assumption, \(E_{C,Z}\) is an \textit{arithmetically Cohen-Macaulay (ACM)} bundle, which means that \(H^1(E_{C,Z}(\ell))=0\) for all \(\ell\in \mathbb Z\), or, equivalently, that \(\Gamma_*(E_{C,Z})\) is a maximal Cohen-Macaulay module over \(\Gamma_*(\mathcal O_X(1))\). The main result states that, if \(E_{C,Z}\) can be written as an extension by line bundles \[ 0\to L_1\to E_{C,Z}\to L_2\to 0 \] (necessarily of the form \(L_2=L_1^\vee(1)\) as \(c_1(E_{C,Z})=[C]\)) with \(h^1(L_1)=0\), then both line bundles \(L_1\) and \(L_2\) are already ACM and satisfy some additional numerical properties. The author conjectures that the condition that \(h^1(L_1)=0\) can be removed. In the special case that \(X\) is a quartic surface in \(\mathbb P^3\) the paper provides further results. In this case, if an extension as above exists, the intersection numbers are given by \(L_2\cdot \mathcal O_X(C)=5\) and \(L_2^2=2\).