Real-analysis facts for large y in [I71, Theorem 4] #
With L = log y, ℓ = log log 3y, Λ = log log y, ξ = ξ(y) = L / ℓ^{11/5} and
w = y^{1/ξ} (so log w = ℓ^{11/5} and pntE w = exp (-κ ℓ^{11/10})):
eventually_delta_mul_le:δ(s₀, y) (1 + C ξ² pntE w) ≤ 1 + Λ² / (8 L²)([I71, (4.6)]);eventually_mul_pntE_le:K ξ² L³ pntE w ≤ 1([I71, (4.2), (4.3)]);eventually_crude: the crude bound2 (e X/(ξ-2))^{ξ-2}for the depth sums atw(withX = log log w + C') timesL⁴is below the Lemma 9 lower bound forG(ξ/2)and belowδ(ξ/2, y)times the Lemma 9 lower bound forG(ξ)([I71, (4.4), (4.5), (4.9), (4.10)]).
L, ξ, w in terms of ℓ = log log 3y #
L⁵ pntE(w) → 0 #
theorem
IwaniecLinearSieve.TheoremFour.tendsto_lin_sub_rpow :
Filter.Tendsto (fun (ℓ : ℝ) => 5 * ℓ - pntκ * ℓ ^ (11 / 10)) Filter.atTop Filter.atBot
The crude bound against the Lemma 9 lower bounds #
theorem
IwaniecLinearSieve.TheoremFour.crude_half_exponent_le
{c ℓ L ξ X : ℝ}
(hℓ_large : 2 ≤ ℓ)
(hxi_large : 8 * ℓ ≤ ξ)
(hlogL_upper : Real.log L ≤ ℓ)
(hlogX_nonneg : 0 ≤ Real.log X)
(hlogX_upper : Real.log X ≤ Real.log ℓ / 4)
(hlog_shift_lower : Real.log (ξ / 2) ≤ Real.log (ξ - 2))
(hlog_half_lower : ℓ / 2 ≤ Real.log (ξ / 2))
(hloglog_half_upper : Real.log (Real.log (ξ / 2)) ≤ Real.log ℓ)
(herror_budget : 2 + c / 2 + 3 / 4 * Real.log ℓ ≤ ℓ / 8)
:
The exponent comparison for [I71, (4.4), (4.9)], after bounding the logarithms.
theorem
IwaniecLinearSieve.TheoremFour.crude_full_exponent_le
{c ℓ L ξ X x : ℝ}
(hℓ_large : 2 ≤ ℓ)
(hxi_large : 8 * ℓ ≤ ξ)
(hlogL_upper : Real.log L ≤ ℓ)
(hlogX_nonneg : 0 ≤ Real.log X)
(hlogX_upper : Real.log X ≤ Real.log ℓ / 4)
(hlogxi_upper : Real.log ξ ≤ ℓ)
(hloglogxi_upper : Real.log (Real.log ξ) ≤ Real.log ℓ)
(hlogx_lower : 3 / 5 * Real.log ℓ - 7 * Real.log 2 ≤ Real.log x)
(herror_budget : 2 + c + 18 * Real.log 2 ≤ Real.log ℓ / 4)
:
The exponent comparison for [I71, (4.5), (4.10)], using the lower bound on log x.
theorem
IwaniecLinearSieve.TheoremFour.crude_alg
{c C' ℓ L ξ : ℝ}
(hC' : 1 ≤ C')
(hℓ : 2 ≤ ℓ)
(hL : L + Real.log 3 = Real.exp ℓ)
(hξ : ξ * ℓ ^ (11 / 5) = L)
(h2 : 2 * Real.log 3 ≤ Real.exp ℓ)
(h3 : 4 * ℓ ^ (11 / 5) ≤ Real.exp (ℓ / 2))
(h4 : 11 / 5 * Real.log ℓ + C' ≤ ℓ ^ (1 / 4))
(h5 : 2 + c + 18 * Real.log 2 ≤ Real.log ℓ / 4)
(h6 : 2 + c / 2 + 3 / 4 * Real.log ℓ ≤ ℓ / 8)
:
0 < 11 / 5 * Real.log ℓ + C' ∧ 2 * Real.exp 1 * (11 / 5 * Real.log ℓ + C') ≤ ξ - 2 ∧ 2 * (Real.exp 1 * (11 / 5 * Real.log ℓ + C') / (ξ - 2)) ^ (ξ - 2) * L ^ 4 ≤ Real.exp (-(ξ / 2 * Real.log (ξ / 2)) - ξ / 2 * Real.log (Real.log (ξ / 2)) - c * (ξ / 2)) ∧ 2 * (Real.exp 1 * (11 / 5 * Real.log ℓ + C') / (ξ - 2)) ^ (ξ - 2) * L ^ 4 ≤ (1 + (ξ / 2) ^ 2 * Real.log (ξ / 2) ^ 5 / L ^ 2) ^ (5 * (ξ / 2)) * Real.exp (-(ξ * Real.log ξ) - ξ * Real.log (Real.log ξ) - c * ξ)
The comparisons [I71, (4.4), (4.5), (4.9), (4.10)] in the variable ℓ = log log 3y, with
L = e^ℓ - log 3 and ξ = L / ℓ^{11/5}, under explicit "ℓ large" hypotheses.
The "ℓ large" hypotheses of crude_alg hold eventually.
theorem
IwaniecLinearSieve.TheoremFour.eventually_crude
(c C' : ℝ)
(hC' : 1 ≤ C')
:
∀ᶠ (y : ℝ) in Filter.atTop, 0 < Real.log (Real.log (y ^ (1 / xi y))) + C' ∧ 2 * Real.exp 1 * (Real.log (Real.log (y ^ (1 / xi y))) + C') ≤ xi y - 2 ∧ 2 * (Real.exp 1 * (Real.log (Real.log (y ^ (1 / xi y))) + C') / (xi y - 2)) ^ (xi y - 2) * Real.log y ^ 4 ≤ Real.exp (-(xi y / 2 * Real.log (xi y / 2)) - xi y / 2 * Real.log (Real.log (xi y / 2)) - c * (xi y / 2)) ∧ 2 * (Real.exp 1 * (Real.log (Real.log (y ^ (1 / xi y))) + C') / (xi y - 2)) ^ (xi y - 2) * Real.log y ^ 4 ≤ delta (xi y / 2) y * Real.exp (-(xi y * Real.log (xi y)) - xi y * Real.log (Real.log (xi y)) - c * xi y)