Secant spaces and syzygies of special line bundles on curves (Q2342184)

Let \(C\) be a smooth complex projective curve of genus \(g\) and \(L\) a very ample line bundle of degree \(d\) on \(C\) such that \(h^0(C,L)=r+1\). The graded module \(\gamma(C,L)=\bigoplus_mH^0(C, mL)\) is naturally studied by means of its minimal free resolution over the symmetric algebra \(R=S^*(H^0(L))\). In particular, the Koszul cohomology \(K_{p,q}\) associated to the complexes \[ \bigwedge^{p+1} H^0(L)\otimes H^0((q-1)L) \to \bigwedge^{p}H^0(L)\otimes H^0(qL)\to \bigwedge^{p-1}H^0(L)\otimes H^0((q+1)L) \] is an object of central interest for this study. Properties of type \((N_p)\) focus on the head on the resolution, being \((N_0)\) the surjectivity of the restriction map \(R \to \gamma(C,L)\) and \((N_p)\) the vanishing \(K_{i,j}=0\) for \(j \neq 1\) and \(1 \leq i \leq p\) (plus \((N_0)\)). These properties hold if the minimal resolution behaves well up to the \(p\)-th step. A nice result by \textit{M. Green} and \textit{R. Lazarsfeld} [Invent. Math. 83, 73--90 (1986; Zbl 0594.14010)] states that if \(d \geq 2g+p+1\) then \(L\) is \((N_p)\) and shows the geometric consequences of \(C\) not being \((N_P)\) when \(d=2g+p\) (see Theorem 1.2). Properties of the type \((M_q)\) focus on the \textit{tail} of the resolution, asking for the vanishing of \(K_{n,1}\) for all \(n \geq r-q\). When \(L\) is the canonical line bundle, the self duality of the resolution is showing that property \((M_q)\) is equivalent to \((N_{q-1})\). The main purpose of this paper is to study property \((M_3)\) for special \(L\), generalizing previous results when \(L=K_C\) and when \(d\gg 0\) (see the Introduction of the paper under review and references therein). The main result of the paper (see Theorem 5.4) states that if \(g \geq 14\), \(r \geq 5\), \(L\) special then, under some particular conditions on the geometry of \(C \subset \mathbb{P}(H^0(L))\) (stated in terms of \((\Delta_q)\) defined in Definition 2.3), \(L\) satisfies \((M_3)\) unless its Clifford index is less than or equal to \(2\). The proof of the theorem is based on the fact that, on one hand, nonzero elements in the Koszul cohomology imposes \(C\subset \mathbb{P}(H^0(L))\) to be contained in certain schemes defined by quadrics but, on the other, the geometric properties of the embedding \(C\subset \mathbb{P}(H^0(L))\) prevents \(C\) from this fact.