This is a double challenge : tackling reinforcement learning with unfamiliar math,while mastering hard science in English. 😱🤯 I love it
Recursos
Resumen
Take an elliptic curve whose rank is known and whose Sha is finite, and say exactly what is still missing before the BSD leading-term formula holds. Then follow how a twist identity, together with a one-sided inequality, closes that gap at every odd prime.
The gap: one number per prime
Gross–Zagier and Kolyvagin give rank = analytic rank and a finite Sha in analytic rank 0 or 1. Finite is still not a value. The paper puts the missing information into a quotient Q_E (the analytic side divided by everything except Sha) and into the discrepancy X_p(E) = v_p(Q_E) − v_p(#Sha) [0:44]. The exact formula is the statement that X_p(E) = 0 at every prime. If the Selmer corank is 0 or 1 at some prime q, earlier results give the rank and finiteness inputs, and a separate input settles p = 2. What remains is every odd p.
The squeeze
The proof never computes X_p directly [1:29]. It proves two weaker facts:
- X_p(E) ≥ 0 for a non-CM curve of analytic rank ≤ 1;
- X_p(E) + X_p(E^D) = 0 for a suitable imaginary quadratic twist, where a(E) + a(E^D) = 1.
Two quantities that can never be negative and that sum to zero must both be zero. The two facts need different strength. An integral comparison is enough for an inequality. An equality needs a unit.
From integral to unit
Over R = Z_p[[t, u_1, …, u_m]] [2:27], t is a cyclotomic or anticyclotomic variable, and the u_i come from tame characters at movable auxiliary primes. Over K = Q(√D), with p = w·w̄, the strict/full Selmer determinant L_w and the CM-measure series B_w give the quotient U_w = B_w / L_w [3:07].
- Integral [3:36]: residual regularity, divisibility at high characters and Weierstrass division give L_w | B_w, and p ∤ B_w. On its own this yields only an inequality, because a hidden irreducible factor could carry extra powers of p.
- Unit [4:30]: theta congruences make every possible factor regular in the tame variables. The two quotient ideals are then shown to be equal. Extra tame variables push the possible jumps to codimension ≥ 2, so no divisor (codimension 1) survives. Specialization keeps the unit a unit.
- Center [5:13]: the unit equates the two central valuations, the Heegner logarithms cancel, and X(E) + X(E^D) = 0 follows.
Reading test for any leading-term argument [5:53]: is this quotient integral, or a unit?