Rank jumps on elliptic surfaces and the Hilbert property (Q2152619)

Computing the ranks of elliptic curves is a central problem in number theory. In order to find curves of high rank over some given number field \(K\), it was pioneered by Néron to start with a (rational) elliptic surface over \(K\) of large rank (i.e. where the generic fibre is an elliptic curves of large Mordell-Weil rank over some function field (mostly \(K(t)\))). Then one uses specialization to exhibit suitable fibres of high rank. This is supported by a famous theorem of \textit{J. H. Silverman} [J. Reine Angew. Math. 342, 197--211 (1983; Zbl 0505.14035)] asserting that the rank can drop only at finite many places. In fact, often the goes up even further upon specialization, thus the rank jumps mentioned in the title of the paper. Notably, extended to K3 surfaces, this has led Elkies to discover the current record curves of rank \(28\) over \(\mathbb Q\). Naturally this leads to the question about the nature of the set of places where the rank jumps -- either generally or by some given number (at least). Building on previous work for the general problem, the authors obtain the following main result: Let \(X\to\mathbb P^1\) be a geometrically rational elliptic surface of Mordell-Weil rank \(r\) over some number field \(K\). Assume that \(X\) admits a bisection of arithmetic genus zero and that the generic fibre is not the quadratic twist of a constant elliptic curve. Then the set \[ \{v\in\mathbb P^1(K);\; \mathrm{rank}(X_v)\geq r+2\} \] is not thin. In particular, the above set of rank jumps is infinite -- despite a conjecture of Silverman stating that 100\% of the fibres of \(X\) when order by height should tend to have rank \(r\) or \(r+1\). The paper also gives results for quadratic twists of constant elliptic curves at two places, thus covering part of the surfaces excluded in the main result. The proof of the main result uses conic bundle structures -- quite a natural technique going back to the classification of surfaces which also renders the assumption on the bisection quite natural. The interested reader may also want to consider previous work by the authors which partly builds to these results, as well as competing work by \textit{J.-L. Colliot-Thélène} [Acta Arith. 196, No. 1, 93--108 (2020; Zbl 1460.14096)] which proves rank jumps by one under the assumption of `weak weak approximation', and by two in the special setting where the generic fibre has complex multiplication.