On a probabilistic local-global principle for torsion on elliptic curves (Q2155594)

Let \(E\) be an elliptic curve defined over \({\mathbb Q}\). If \(p\) is a good reduction prime and does not divide \(\# E({\mathbb Q})_{\mathrm{tor}}\), it is well-known that \(E({\mathbb F}_p)\) can be realized as a subgroup of \(E({\mathbb Q})_{\mathrm{tor}}\). This allows us to discard possible torsion subgroups and/or point orders for a given curve. Unfortunately, this argument does not always yield the full torsion structure. A famous and easy example is \(E: y^2 = x^3 + x\), which has trivial torsion, while \(E({\mathbb F}_p)\) is divisible by \(4\) for all good reduction primes. The paper under review starts from here and studies further the relationship between \(E({\mathbb F}_p)\) and \(E({\mathbb Q})_{\mathrm{tor}}\), from a probabilistic point of view. The main result is that, if there is a density \(1\) set of primes \(p\) such that \(m | \# E({\mathbb F}_p)\), then the probability of \(m | \# E({\mathbb Q})_{\mathrm{tor}}\) is non-zero for all integers \(m\) that can actually appear as \(\# E({\mathbb Q})_{\mathrm{tor}}\). They, in fact, compute this exact probability for \(m \in \{3,4,5,7\}\), which are mostly the essential cases, thinking in terms of point orders (cases \(m=1,2\) are strightforward). The techniques developed in the paper are diverse and include: \begin{itemize} \item A way of counting curves with prescribed torsion structure and bounded height which refines the results by Harron and Snowden (\textit{R. Harron} and \textit{A. Snowden}, J. Reine Angew. Math. 729, 151--170 (2017; Zbl 1378.11067)). \item A refinement of a result by Katz (\textit{N. M. Katz}, Invent. Math. 62, 481--502 (1981; Zbl 0471.14023)) concerning Galois representations. \item Some ad-hoc calculations in the case--by--case probability computations. \end{itemize} The first two points are, in this reviewer's opinion, really interesting contributions with merit of its own.