Integral \(j\)-invariants and Cartan structures for elliptic curves (Q931793)

Let \(N\geq 3\) be an integer and \(\Gamma\) a subgroup of \(\text{SL}_2({\mathbb Z})\) whose reduction \(\overline{\Gamma} \bmod N\) satisfies \(\det(\overline\Gamma)=(\mathbb Z/N\mathbb Z)^\ast\). Let \(X_\Gamma\) be the corresponding modular curve over \(\mathbb Q\). The first result of this paper is: Theorem. If {Gal}\((\overline{\mathbb Q}/\mathbb Q)\) does not act transitively on the set of cusps of \(X_\Gamma\), then for every point \(P\in X_\Gamma(\mathbb Q)\) with \(j(P)\in\mathbb Z\) we have: \[ \log| j(P)|\leq 21N^2| \overline{\Gamma}|+90N. \] This result is applied to the curves \(X_{\text{split}}(p^n)(\mathbb Q)\), whose corresponding group \(\overline\Gamma\) is the normalizer of a split Cartan subgroup mod \(p^n\). These groups are connected to a question of Serre concerning elliptic Galois representations. The authors prove the following result: Theorem. Assuming the Generalized Riemann Hypothesis (GRH) for zeta functions of number fields, for every large enough prime number \(p\), every point in \(X_{\text{split}}(p^5)(\mathbb Q)\) is either a cusp or a complex multiplication point. It is mentioned in the article that the assumption of GRH could be suppressed, at the expense of replacing \(p^5\) by a higher power of \(p\) in the theorem. More general results are announced to appear in a forthcoming work by the authors (appeared as a preprint in \url{arXiv:0807.4954}).