Sum formula for the limiting proportion via Selmer ranks in Frobenian twist families
Establish that, for an elliptic curve E/Q with full rational 2‑torsion, a finite set Σ containing 2, ∞, all places of bad reduction, and all places ramified in a given Galois field L/Q, and a class b∈∏_{v∈Σ}Q_v^×/Q_v^{×2} such that for all d∈F_b(L) the class [C]∈H^1(Q,E[2]) lies in Sel_2(E_d/Q) and the 2^∞‑Selmer rank of E_d/Q is odd, the limiting proportion of d∈F_b(L) with C_d(Q)≠∅ equals ∑_{r≥0} α(2r+m_b)/(2^{n_b+2r+m_b}−1), where α(k)=∏_{j≥1}(1+2^{−j})^{−1}∏_{j=1}^{k} 2/(2^j−1), n_b=dim_F2 S_b is the dimension of the systematic subspace forced into Sel_2(E_d/Q), and m_b∈{0,1} is the parity shift determined in Theorem 1.??.
References
As d varies over F_b(L), it seems natural to model the non-trivial element in δd(E_d(Q)/E[2]) as a uniformly random non-trivial element of Sel_2(E_d/ Q)/δ_d(E_d[2]). This heuristic together with Theorem 1.?? leads to the following concrete conjecture. Conjecture Let b∈∏{v∈Σ}Q_v{×}/Q_v{×2} be such that for all d∈F_b(L) one has [C]∈Sel_2(E_d/ Q) and rk_2(E_d/ Q) odd, and let n_b∈{1,…,4}, m_b∈{0,1}, and α(r) be as in Theorem 1.??. Then we have lim_{X→∞} #{d∈F_b: |d|<X, C_d(Q)≠∅}/#{d∈F_b: |d|<X}= ∑_{r≥0} α(2r+m_b)/(2{n_b+2r+m_b}−1).