The Tate-Shafarevich group for elliptic curves with complex multiplication (Q731238)

This work explores the connection between the Birch and Swinnerton-Dyer conjecture and Iwasawa theory for elliptic curves. Quoting the introduction of the article, ``it has long been known that the main conjectures of Iwasawa theory provide a precise link between the \(\mathbb Z_p\)-corank of the \(p\)-primary subgroup of the Tate-Shafarevich group of an elliptic curve and the multiplicity of the zero of certain \(p\)-adic \(L\)-functions at the point \(s=1\)\dots''. The authors translate this link into concrete results for the particular case of elliptic curves with complex multiplication. Given an elliptic curve \(E\) over \(\mathbb Q\), let \(g_{E/\mathbb Q}\) be its rank, and denote by \(\text Ш(E/\mathbb Q)\) its Tate-Shafarevich group and by \(t_{E/\mathbb Q,p}\) the \(\mathbb Z_p\)-corank of the \(p\)-primary part of this group, i.e., the number of copies of \(\mathbb Q_p/\mathbb Z_p\) occurring in it. The authors are able to establish: \textbf{Theorem 1.1:} If \(E\) has complex multiplication, for each \(\epsilon>0\) there exists an explicitly computable number \(c(E,\varepsilon)\) such that \[ t_{E/\mathbb Q,p}\leq (1+\varepsilon)p-g_{E/\mathbb Q} \] for all primes \(p\geq c(E,\varepsilon)\) of ordinary good reduction. A similar bound for the corresponding invariants over the field of definition of the complex multiplication of the curve is also given. The interest of these results is mainly theoretical, since the bound is obviously very weak. Indeed, the same authors give an improved version in a continuation of the article [Milan J. Math. 78, No. 2, 395--416 (2010; Zbl 1295.11059)]. Besides the work of the Dokchitser brothers [\textit{T. Dokchitser} and \textit{V. Dokchitser}, Ann. Math. (2) 172, No. 1, 567--596 (2010; Zbl 1223.11079)], very little was known about \(t_{E/\mathbb Q,p}\). Let \(K\) be the field of the complex multiplication of \(E\), \(\mathbb Z_K\) its maximal order and \(w\) the number of roots of unity in \(K\). Assume that the class number \(h(K)\) of \(K\) is one. Then, the period lattice of the Néron differential form of \(E\) can be written as \(\Omega_{\infty}\mathbb Z_K\) for certain \(\Omega_\infty\in\mathbb C^\ast\). The smallest real positive period \(\Omega_\infty^+\) of \(E\) satisfies \[ \Omega_{\infty}^+=\Omega_{\infty}\alpha(E) \] for some non-zero integral \(\alpha(E)\in\mathbb Z_K\). Along the proof of Theorem 1.1, the authors provide a more practical result concerning the value \[ c_p^+(E)=(\Omega_\infty^+)^{-p}L(\overline{\psi}_E^p,p). \] Here \(L(\overline{\psi}_E^p,p)\) denotes the value at \(s=p\) of the primitive Hecke \(L\)-function of \(\overline{\psi}_E^p\), where \(\psi_E\) is the Hecke-Grössencharacter attached to \(E\). It is shown: \textbf{Theorem 2.2:} Assume that \(K\) has class number 1 and that the analytic rank of \(E\) has the same parity of its algebraic rank \(g_{E,\mathbb Q}\). Let \(p\) be a prime number of good reduction of \(E\), totally split in \(K\) and coprime with \(w\alpha(E)\). If we have \[ \text{ord}_p(c_p^+(E))< g_{E/\mathbb Q}+2, \] then the \(p\)-primary part \(\text Ш(E/K)(p)\) of \(\text Ш(E/K)\) is finite. Moreover, if \[ \text{ord}_p(c_p^+(E))= g_{E/\mathbb Q}, \] and \(\sharp E(\mathbb F_p)\) is prime to \(p\) with \((p,6)=1\), then \(\text Ш(E/K)(p)=0\). The theorem gives a practical way to prove the triviality of \(\text Ш(E/K)(p)\) in practical cases, when combined with a known expression of \(c_p^+(E)\) in terms of the trace of certain value of an Eisenstein series at a proper torsion point of \(E\). (Note that the expression given in the article (formula (24) and consequently formula (53)) have a misplaced factor \(w\) which should be corrected, as the authors point out in their later work [Zbl 1295.11059]. The second part of the paper is devoted to explain how to compute in practice \(c_p(E)=\Omega_\infty^{-p}L(\overline{\psi}_E^p,p)\in K\) for the family of curves \(E_D: y^2=x^3-Dx\), with \(D\) a fourth-power free non-zero rational integer. The authors made explicit computations for the two curves with \(D=17\) and \(D=-14\), for all the primes \(p<13500\) congruent to \(1\mod 4\) (the prime 17 excluded for the first curve). For all these primes (except \(p=29, 277\) for the second curve), it turns out that \(\text{ord}_p(c_p^+(E))=2\). The values of \(c_p^+(E) \mod p^3\) for \(p\in [5,1000]\cup[11000,12000]\) are given in the final tables (which carry the misplaced factor mentioned above). It is worth mentioning that these computations are widely extended in Part II [Zbl 1295.11059], which includes also the correct version of the tables.