Espaces Homogènes Principaux et Points de 2-Division de Courbes Elliptiques (Q3150996)

Consider an elliptic curve \(E\) which has everywhere good reduction over the number field \(L\). If \(a\) is a nonzero endomorphism of \(E\) (so \(a\) is an integer or in the CM case possibly a quadratic irrationality), then there is a Picard invariant map \(\psi\) which attaches to every \(P\in E(L)\) a class in \(Cl({\mathcal A})\), where \(\mathcal A\) is the dual of the Hopf algebra that represents the \(O_L\)-scheme \(E[a]\) of \(a\)-torsion points. Essentially, \(\psi(P)\) describes the Galois module structure of a certain natural order in the field gotten by adjoining a point \(Q\) with \(aQ=P\). Taylor conjectured that \(\psi\) is trivial on torsion points. This was confirmed for most cases by results of Agboola and of Pappas (cf.~also work of Srivastav and Taylor), but the case that \(a\) is a 2-power shows peculiarities. \textit{W. Bley} and \textit{M. Klebel} [Pac. J. Math. 185, 221--235 (1998; Zbl 0979.11032)] found examples without CM that have 2-torsion points over \(L\) on which \(\psi\) is nontrivial. The article under review produces CM examples for this phenomenon; actually a stronger statement of nontriviality is proved. To wit: if we assume \(E[2]\subset L\), \(P\in E[2]\), and let \(L_P\) the field obtained by adjoining \(Q\) with \(2Q=P\), then \(\psi(P)\) will be nontrivial whenever the extension \(L_P/L\) (which is trivial, quadratic, or biquadratic a priori) cannot be gotten by adjoining square roots of units. This is exactly what may happen, as the authors show. They take \(E\) with CM by an imaginary quadratic field \(k\) of class number 1 in which \(2\) is inert: this permits the five discriminants \(-p\), \(p=11,19,43,67,163\). By ramification considerations at level \(k\) they show that \(L_P\) is generated by the square roots of \(-1\) and \(\alpha\) over \(L\), where \(\alpha\) is associated to \(\sqrt{-p}\). Of necessity \(p\) must ramify from \(k\) to \(L\), since the bad reduction of \(E\) is supposed to turn good over \(L\). One actually looks for \(L\) as an overfield of a quadratic extension \(K/k\) in which \(\sqrt{-p}\) ramifies. Thus, the ideal generated by \(\alpha\) will certainly be a square, and the point is to force that the ideal square root \(\sqrt{(\alpha)}\) is nonprincipal. This is done by setting \(L\) equal to \(Kk(E[2])\), \(K/k\) quadratic as said above, and such that \(\pm\sqrt{-p}\) is not a norm in \(K/k\). The authors provide an explicit choice of \(K\) for all five values of \(p\). It seems likely that there are actually infinitely many choices for \(K\).