Reciprocal sieve threshold for every real z at least 1.3 million
ProvedPrimePairSieve.reciprocal_threshold_tail_1300000For every real , the mission's literal squarefree reciprocal sum satisfies
where is the mission's twin-prime constant and is zero for non-squarefree , otherwise the product of at the prime 2 and at odd prime factors. The real variable, natural floor, and threshold are unchanged.
This refines the already Proved tail above two million. It uses the same three accepted analytic inputs: the explicit quadratic expansion, second logarithmic moment bound, and sieve constants certificate. The improvement comes from retaining the sharper elementary constant-term bound , instead of rounding it down to . The existing inequalities give .
For , certify using , and using . The logarithmic correction is bounded by . The exact rational comparison is
The proof checks this comparison and its monotone extension to the entire unbounded interval in Lean. This is a formal numerical refinement of the existing argument, not a new sieve estimate. It shortens the finite certificate obligation; it does not by itself close the full reciprocal threshold or five-primes mission.
Explicit expansion, second-moment bound, constants certificate.
import Mathlib import Definitions.Def_TaoFivePrimes_SingularSeries open scoped BigOperators set_option autoImplicit false
theorem PrimePairSieve.reciprocal_threshold_tail_1300000 :
∀ z : ℝ, 1300000 ≤ z →
(Real.log (4 * z)) ^ 2 / (2 * TaoFivePrimes.twinPrimeConstant) /
(1 - ((z + 1) * ((Real.log (4 * z)) ^ 2 /
(2 * TaoFivePrimes.twinPrimeConstant)) / (16 * z ^ 2)))
≤ ∑ q ∈ Finset.Icc 1 ⌊z⌋₊,
(if Squarefree q then
∏ p ∈ q.primeFactors, if p = 2 then (1 : ℝ) else 2 / ((p : ℝ) - 2)
else 0) / (1 + (q : ℝ) / z) := by sorry