Corollary 4.16 — universal STS expected-length lower bound
ProvedGlynnSTS.Length.corollary_4_16confidence-intervalexpected-lengthlower-boundp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1simulation-output-analysisstandardized-time-series
Let be nonnegative on , and suppose Assumption (2.1) holds for the output process, with Brownian scale . Choose and so that . If is the length of the STS interval, then
This is the paper's universal lower bound for the expected length of an asymptotic STS confidence interval.
Formalization Note The normal quantile is supplied as with . Extended nonnegative expectations retain the case . The process is measurable and integrable on finite intervals so its integrated average is defined.
Preamble
import Mathlib import Definitions.Def_GlynnSTS_Length_Setting open MeasureTheory ProbabilityTheory Filter Topology open scoped NNReal ENNReal
Formal statement
namespace GlynnSTS.Length
/-- Corollary 4.16, p. 12: every nonnegative STS interval has the normal-quantile lower bound. -/
theorem corollary_4_16 {Ω : Type*} [MeasurableSpace Ω]
(P : Measure Ω) [IsProbabilityMeasure P]
(B : Ω → C(unitInterval, ℝ)) (hB : GlynnSTS.Limit.IsStdBMC P B)
(g : C(unitInterval, ℝ) → ℝ) (hg : GlynnSTS.Limit.ClassM P B g)
(hg0 : ∀ x, 0 ≤ g x)
(Y : ℝ → Ω → ℝ) (Ybar : ℕ → Ω → C(unitInterval, ℝ))
(μ σ : ℝ) (h21 : Assumption21 P Y Ybar μ σ B)
(δ : ℝ) (hδ0 : 0 < δ) (hδ1 : δ < 1)
(α β : ℝ) (hαβ : H P B g β - H P B g α = 1 - δ)
(p : ℝ) (hp : cdf (gaussianReal 0 1) p = 1 - δ / 2) :
ENNReal.ofReal (2 * σ * p) ≤
liminf (fun n : ℕ => ENNReal.ofReal (Real.sqrt n) *
∫⁻ ω, ENNReal.ofReal (Ln g Ybar α β n ω) ∂P) atTop := by sorry
end GlynnSTS.Length
Source
Glynn & Iglehart, Simulation output analysis using standardized time series, Math. Oper. Res. 15 (1990), p. 12, Corollary 4.16
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.