(2.48), p. 768 — ∫_{u<0} e^{hu} dF_j(u) ≤ exp{g(0) + hx/n − g(x/n)} ∫_{u<0} e^{hu} dF_j(u), h = g′(x/n)
ProvedNagaevLD.GenMoment.eq_2_48exponential-momentslarge-deviationsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a real random variable with distribution function . Let have, on , a positive nondecreasing derivative , let be an integer, , and . Then
The factor is at least , so the negative part of can be multiplied by it; this is what makes , rather than the bare integral, appear in the per-summand bound of Theorem 2.5.
Formalization Note is the integral over the event . Since the integrand lies in there, so the integral is finite and no integrability hypothesis is needed. The hypotheses on are imposed on only. The implicit of the paper is explicit in Lean.
Preamble
import Mathlib import Definitions.Def_NagaevLD_GenMoment_Setting open MeasureTheory ProbabilityTheory
Formal statement
namespace NagaevLD.GenMoment
/-- (2.48), p. 768: for `h = g′(x/n)`,
`E[e^{hX_j}; X_j < 0] ≤ exp{g(0) + hx/n − g(x/n)} · E[e^{hX_j}; X_j < 0]`. -/
theorem eq_2_48 {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(n : ℕ) (hn : 0 < n) (X : Fin n → Ω → ℝ) (hXm : ∀ j, Measurable (X j))
(g g' : ℝ → ℝ) (hg : ∀ u : ℝ, 0 ≤ u → HasDerivAt g (g' u) u)
(hg'pos : ∀ u : ℝ, 0 ≤ u → 0 < g' u) (hg'mono : MonotoneOn g' (Set.Ici 0)) (j : Fin n) (x : ℝ) (hx : 0 < x) (h : ℝ) (hh : h = g' (x / n)) :
∫ ω in {ω | X j ω < 0}, Real.exp (h * X j ω) ∂P
≤ Real.exp (g 0 + h * (x / n) - g (x / n))
* ∫ ω in {ω | X j ω < 0}, Real.exp (h * X j ω) ∂P := by sorry
end NagaevLD.GenMoment
Source
Nagaev, Large deviations of sums of independent random variables, Ann. Probab. 7 (1979), p. 768, proof of Theorem 2.5, (2.48)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.