§1, p. 759, after (1.58) — for EX_i = 0, B_n² < ∞ and x ≥ 2B_n, P(S_n ≥ x) ≥ ½ΣP(X_j ≥ 2x)
ProvedNagaevLD.LowerBound.lower_boundLet be independent real random variables on a probability space with
and let and . Then for every with ,
This is the lower bound that closes §1 of Nagaev's survey. It holds without assuming that the summands are identically distributed or that their tails are regularly varying, and it complements the paper's upper bounds of Fuk–Nagaev type, which have the form : in the range the probability of a large deviation of the sum is, up to the factor and the change from to , at least the probability that one summand alone is large.
Formalization Note The page's hypothesis is stated as "each is square integrable" (Lean's MemLp (X i) 2 P), since Mathlib's variance of a non-square-integrable variable is . The level is positive, as everywhere in the paper (p. 747). The summands are indexed by Fin n.
import Mathlib import Definitions.Def_NagaevLD_LowerBound_Setting open MeasureTheory ProbabilityTheory
namespace NagaevLD.LowerBound
/-- §1, p. 759, after (1.58): for `EX_i = 0`, `B_n² < ∞` and `x ≥ 2B_n`,
`P(S_n ≥ x) ≥ ½ Σ_j P(X_j ≥ 2x)`. -/
theorem lower_bound {Ω : Type*} [MeasurableSpace Ω] (P : Measure Ω) [IsProbabilityMeasure P]
(n : ℕ) (X : Fin n → Ω → ℝ) (hXm : ∀ i, Measurable (X i))
(hindep : iIndepFun (fun i => X i) P)
(hL2 : ∀ i, MemLp (X i) 2 P) (hmean : ∀ i, ∫ ω, X i ω ∂P = 0)
(x : ℝ) (hx : 0 < x) (hxB : 2 * Real.sqrt (NagaevLD.FukNagaev.Bn2 P X) ≤ x) :
(1 / 2) * ∑ j, P.real {ω | 2 * x ≤ X j ω} ≤ P.real {ω | x ≤ NagaevLD.FukNagaev.S n X ω} := by sorry
end NagaevLD.LowerBound
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.