\(K3\) surfaces, rational curves, and rational points (Q984835)

In 1981, Bogomolov conjectured that on any \(K3\) surface \(X\) over a finite field or a number field \(k\), every \(\bar k\)-rational point lies on a rational curve \(C\subset X\) which is defined over \(\bar k\). For the finite fields case, evidence has provided in favour of the conjecture when \textit{F. Bogomolov} and \textit{Y. Tschinkel} proved it for Kummer surfaces [Am. J. Math. 127, No. 4, 825--835 (2005; Zbl 1082.14025)]. The present paper develops new techniques in the number field case which may be seen as evidence against Bogomolov's conjecture. The main result concerns generally jacobian elliptic surfaces \(X\) over \(\mathbb{P}^1\) with at least five nodal singular fibers and two-torsion curve irreducible of positive geometric genus: If \(X\) is defined over a number field \(k\) and \(P\) is an algebraic point on \(X\) that is only contained in finitely many rational curves on \(X\), then there is an algebraic point \(Q\) off all rational curves on \(X\). (Some \(2\)-power multiple of \(Q\) returns \(P\).) In particular, this result applies to elliptic \(K3\) surfaces. Note, however, that these always contain infinitely many rational curves. The authors also elaborate on a result concerning rational curves over a fixed field. Along similar lines, the authors work out an example of a \(K3\) surface \(X\) over \(\mathbb{Q}\) of Picard number 4 where all smooth rational curves are defined over \(\mathbb Q\) (as they lie in the orbit of a few curves under the automorphism group). As an application of the above methods, they find an algebraic point \(Q\) on \(X\) that does not lie on any smooth rational curve.