The size of Selmer groups for the congruent number problem (Q1210508)

The congruent number problem asks whether a given positive integer \(D\) is the difference of three rational squares in arithmetic progression. This is equivalent to the fact that the elliptic curve \(E_ D: y^ 2= x^ 3- D^ 2x\) defined over \(\mathbb{Q}\) has positive algebraic rank \(r(D)= \text{rank} (E_ D)\), where \(D\) is supposed to be square free. At present there is no known algorithm to decide whether a given \(D\) is congruent. The functional equation of the Hasse-Weil \(L\)-function \(L_ D(s)\) of \(E_ D\) changes sign by \(\varepsilon_ D= \pm1\) when \(s\) is replaced by \(s-2\). From this it follows that \((-1)^{R(D)}= \varepsilon_ D\), where \(R(D)= \text{ord}_{s=1} L_ D(s)\) is the analytic rank of \(E_ D\). Then, assuming the Birch and Swinnerton-Dyer conjecture, \(r(D)>0\) if and only if \(\varepsilon_ D =-1\). From previous calculations by \textit{B. J. Birch} and \textit{N. M. Stephens} [Topology 5, 295-299 (1966; Zbl 0146.424)] one concludes that \(\varepsilon_ D\) equals 1 if \(D\equiv 1,2,3\pmod 8\), 0 if \(D\equiv 5,6,7 \pmod 8\). So this last case gives the congruent numbers. Results of \textit{J. Coates} and \textit{A. Wiles} [Invent. Math. 39, 223-251 (1977; Zbl 0359.14009)], \textit{B. Gross} and \textit{D. Zagier} [Invent. Math. 84, 225- 320 (1986; Zbl 0608.14019)] and \textit{K. Rubin} [Invent. Math. 89, 527-560 (1987; Zbl 0628.14018)] show that \(r(D)= R(D)\) if \(r(D)=0\) or 1. However the situation is still open when \(r(D)\geq 2\). The author attacks this question via ``full 2-descents''. The number of these descents equals the order of the Selmer group \(S^{(2)}\), which is a power of 2 and a multiple of 4 taking into account the rational points of order 2 on \(E_ D\). The author denotes \(\#S^{(2)}= 2^{2+s(D)}\) and calls \(s(D)\) the ``Selmer rank'' of \(E_ D\). The Selmer conjecture states that \(s(D)\) and \(r(D)\) have the same parity. The Birch and Swinnerton-Dyer conjecture makes it ``reasonable'' to expect that \(s(D)\) and \(R(D)\) also have the same parity. Actually it is proved by B. J. Birch and N. M. Stephens (loc. cit.) and \textit{J. Lagrange} [in Sémin. Delange-Pisot-Poitou (Théorie des nombres), No. 16, 1974-75 (1975; Zbl 0328.10013)]\ that the number of first descents has the same parity as \(R(D)\). The author's main result is the following average estimate of \(s(D)\). Let \(h\) be an odd integer and \(S(X,H)= \{D\equiv h\pmod 8\); \(1\leq D\leq X\), \(D\) square free\} then \[ \sum_{D\in S(X,h)} 2^{s(D)}= 3\# S(X,h)+ O(X(\log X)^{-1/4} (\log\log X)^ 8). \] In particular, the average sum of \(2^{r(D)}\) over the above set is less than or equal to the above bound. From his main result one obtains \[ \sum_{D\in S(X,h)} s(D)\leq {\textstyle {4\over 3}} \# S(X,h)+ O(X(\log X)^{-1/4} (\log\log X)^ 8), \] if \(h\) is odd and \(s(D)\) and \(r(D)\) have the same parity, and the same bound for the average of \(r(D)\) over the same set. Unconditional results in both cases imply the same bound replacing \(4/3\) by \(3/2\). An important question related to these averages is whether there exist elliptic curves with arbitrarily high rank and what is the frequency of these high ranks. The work of \textit{F. Gouvêa} and \textit{B. Mazur} [J. Am. Math. Soc. 4, 1-23 (1991; Zbl 0725.11027)] applied to the curves \(E_ D\) shows that \[ \#\{D\in S(X,1);\;R(D)\geq 2\}\gg X^{1/2- \varepsilon}, \quad \text{for each } \varepsilon>0. \] In the case of the Selmer rank, however, the situation is tractable for the Selmer rank and the author proves that for each \(\theta<1\) there exists \(c_ \theta>0\) such that \[ \#\{D\in S(X,1);\;s(D)>c_ \theta \sqrt{\log D}\}\gg_ \theta X^ \theta. \] Finally, whereas it is uncertain that for a fixed rank there exists a positive proportion of elliptic curves, for the Selmer rank this seems likely and the author shows that for each \(n\geq 0\) one has \[ \#\{D\in S(X,h);\;s(D)=n\}\gg_ n {\textstyle{X\over {\log X}}}. \]