Seshadri constants in two half-periods (Q2580937)

In the last ten years Seshadri constants of several kind of projective varieties (starting from projective space itself) have been studied intensively; this paper deals with the case of abelian varieties. Let us recall a few notions: If \(X\) is a smooth projective variety and \(L\) a numerically effective line bundle on it, let \(x_1,\dots ,x_r\) be a set of distinct points on \(X\), and \(\pi\) the blow-up of \(X\) at them, with exceptional divisors \(E_1,\dots ,E_r\). Then \(\varepsilon (L,x_1,\dots ,x_r)= \sup \{\varepsilon \mid \pi^*L-\varepsilon\sum_{i=1}^sE_i \text{ is nef}\}\) is called the Seshadri constant of \(L\) at \(x_1,\dots ,x_r\); moreover \(\varepsilon (L,x_1,\dots ,x_r)= \inf\{{LC\over \mu_1C+\cdots +\mu_rC}\}\) where \(C\) runs among all reduced and irreducible curve on \(X\) and \(\mu_iC\) is the multiplicity of \(C\) at \(x_i\). We always have \(0\leq \varepsilon (L,x_1,\dots ,x_r)\leq \sqrt{{L^2\over r}}\); curves \(C\subset X\) for which \(\inf\{{LC\over \mu_1C+\dots +\mu_rC}\}< \sqrt{{L^2\over r}}\) are called submaximal curves. When \(X\) is an abelian surface, it is known a bound for \(\varepsilon (L,x)\), and its exact value is known when \(X\) has Picard number \(=1\); in this work the case \(r=2\) is treated. Let us recall that if \(L\) is a line bundle on an abelian surface \(X={\mathbb C}^2/\Lambda\), we can define the type of \(L\) as the vector \((d_1,d_2)\in {\mathbb N}^2\), where, if \(H: {\mathbb C}^2 \rightarrow {\mathbb C}\) is the hermitian form associated to \(L\), is such that (for an appropriate basis of \(\Lambda\)) \(\operatorname{Im}H = \left(\begin{smallmatrix} 0&D\\ -D&0\end{smallmatrix}\right)\), with \(D= \text{diag}(d_1,d_2)\). The main result in the paper is the following: If \(X\) is an abelian surface with Picard number =1 and \(L\) is an ample line bundle of type \((1,d)\), where \(\sqrt{d}\) is not integer, then - if \(d\) is even and the primitive solution \((k_0,d_0)\) of the diophantine (Pell's) equation \(l^2-dk^2=1\) has \(k_0\) odd, then \(\varepsilon (L,x_1,x_2)=d{k_0\over l_0} \); - in all other cases \(\epsilon (L,x_1,x_2)=2d{k_0\over l_0}\), where \((k_0,d_0)\) is the primitive solution of the diophantine (Pell's) equation \(l^2-4dk^2=1\).