Theorem 3.2(i) — upper tail: t^κP(R > t) = C₊ − (1/2π)ℜ∮ e^{−iθ log t} ĝ₁(θ)/(1 − η̂(θ)) dθ + O(t^{−β/2})
OpenGoldieRenewal.Rates.theorem_3_2_iLet be a probability space, , , and let and be independent real random variables with
and (3.4): the conditional law of given is spread out. Let be the tilted law, and suppose that satisfies the decomposition (3.1) and the subsequent conditions of Theorem 3.1 at this . Let (3.5), let be the constant (2.12), and let be any admissible rectangle contour in the strip enclosing all zeros of in . If
then, as ,
This is part (i) of Goldie's Theorem 3.2, the rate-of-approach refinement of the implicit renewal theorem for the upper tail: the second-order term of is an explicit contour integral determined by the zeros of in the strip, up to an error .
Formalization Note Same conventions as GoldieRenewal.Rates.theorem_3_2: (3.7) is printed with and is read with ; integrability on is IntegrableOn; the expansion is stated with the exponent that the proof establishes (ExpansionPos); moments are lower Lebesgue integrals; almost surely.
import Mathlib import Definitions.Def_GoldieRenewal_Rates_Transforms import Definitions.Def_GoldieRenewal_Rates_Model import Definitions.Def_GoldieRenewal_Rates_Expansion open MeasureTheory ProbabilityTheory
namespace GoldieRenewal.Rates
/-- **Theorem 3.2(i)** (Goldie 1991, p. 132; rate of approach in the implicit renewal theorem,
upper tail, `M ≥ 0`). Let `M ≥ 0` be independent of `R` with (2.3) `E|M|^κ = 1`, (3.3)
`E|M|^{κ+β} < ∞` for some `β > 0`, and (3.4) the conditional law of `log|M|` given `M ≠ 0`
spread out. Suppose `η(dx) := e^{κx}P(log M ∈ dx)` satisfies (3.1) and the subsequent
conditions of Theorem 3.1 (`StoneConditions η β`). Then (3.7)
`∫₀^∞ |P(R > t) − P(MR > t)| t^{κ+β−1} dt < ∞` implies (3.8):
for every admissible rectangle contour `𝒞`,
`t^κ P(R > t) = C₊ − (1/2π) ℜ ∮_𝒞 e^{−iθ log t} ĝ₁(θ)/(1 − η̂(θ)) dθ + O(t^{−β/2})`.
Formalization Note: this is part (i) of `GoldieRenewal.Rates.theorem_3_2`, with the same
conventions: (3.7) is read with `P(R > t)`; "`∫₀^∞ |h(t)| t^{κ+β−1} dt < ∞`" is
`IntegrableOn (h · t^{κ+β−1}) (0, ∞)`; (3.8) is `ExpansionPos` (exponent `−iθ log t`). -/
theorem theorem_3_2_i {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(M R : Ω → ℝ) (hM : Measurable M) (hR : Measurable R) (hM_nonneg : ∀ᵐ ω ∂P, 0 ≤ M ω)
(h_indep : IndepFun R M P) (κ β : ℝ) (hκ : 0 < κ) (hβ : 0 < β)
(h_2_3 : ∫⁻ ω, ENNReal.ofReal (|M ω| ^ κ) ∂P = 1)
(h_3_3 : ∫⁻ ω, ENNReal.ofReal (|M ω| ^ (κ + β)) ∂P < ⊤)
(h_3_4 : SpreadOut (logLawGivenNonzero P M))
(h_stone : StoneConditions (tiltedLaw P M κ) β)
(h_3_7 : IntegrableOn (fun t : ℝ =>
(P.real {ω | t < R ω} - P.real {ω | t < M ω * R ω}) * t ^ (κ + β - 1)) (Set.Ioi 0)) :
ExpansionPos P M R κ β := by sorry
end GoldieRenewal.Rates