The Picard group of a \(K3\) surface and its reduction modulo \(p\) (Q435267)

A common strategy to bound the geometric Picard number of a (smooth projective) surface \(X\) defined over a number field \(K\) is to consider reductions of \(X\) modulo good primes. The Picard number of the reduction can be bounded in terms of the zeroes of the zeta function of the reduction of \(X\), and the zeta function can often be computed. A disadvantage of this approach is that the upper bound \(s\) obtained in this way has the same parity as \(\dim H^2(X,\mathbb{C})\). \textit{R. van Luijk} [Algebra Number Theory 1, No. 1, 1--17 (2007; Zbl 1123.14022)] proposed a method to prove that the Picard number is at most \(s-1\) by using a second prime of good reduction. He applied this method in one example. In this example \(X\) is a \(K3\) surfaces and \(s\) equals 2. In this way he could provide an explicit \(K3\) surface with geometric Picard number 1. The main result of this paper is the following. Let \(R\) be a discrete valuation ring with quotient field \(K\) of characteristic zero, residue characteristic \(p>0\) and the ramification degree is at most \(p-2\). Let \(X\to \mathrm{Spec} R\) be a proper smooth morphism of schemes. Let \(k\) be the residue field of \(R\). Then the cokernel of the specialization homomorphism \(\mathrm{Pic}(X_{\overline{K}})\to\mathrm{Pic}(X_{\overline{k}})\) is torsion-free. The authors use this result to prove that a particular \(K3\) surface \(X\) has geometric Picard number one. This is done by first identifying a prime \(p\) such that reduction modulo \(p\) has geometric Picard number 2. Now if the geometric Picard number of \(X\) were 2 then every divisor on the reduction of \(X\) lifts to \(X_{\overline{K}}\). By using a Gröbner basis computation the authors identify an effective divisor on the reduction of \(X\) that does not lift to \(X_{\overline{K}}\).