Theorem 2.5.8 — Kolmogorov's three-series theorem
ProvedDurrettProbability.kolmogorov_three_seriesLet be independent real random variables, fix a truncation level , and set
Then converges almost surely if and only if all three of the following hold:
- ;
- converges;
- .
Three conditions computable from the marginal distributions decide an almost-sure question about the paths, and each does a distinct job. The first says and differ only finitely often, so by Borel–Cantelli the two series converge or diverge together. The third, with the convergence criterion for series of centred independent summands, makes converge. The second puts the means back.
Nothing is assumed about the integrability of : the truncated variables are bounded, so their means and variances exist whatever the tails of do. Nor does the conclusion depend on the choice of — since the left-hand side does not, the three conditions hold for one exactly when they hold for every .
Formalization Note " converges" is convergence of the partial sums to a
real limit, following the book's explicit convention, and is not stated as summability: for real
series Mathlib's Summable means absolute convergence, which would make condition (ii) strictly
stronger and the theorem false as an equivalence. Conditions (i) and (iii) are series of
non-negative terms, where the two notions coincide, and are stated as summability.
Almost-sure convergence of asserts, for almost every outcome, the existence of a real limit of the partial sums. No integrability of the is assumed anywhere. Both implications are part of the statement; Durrett proves sufficiency here and defers necessity to Example 3.4.12, where it follows from the Lindeberg–Feller central limit theorem.
import Mathlib import Definitions.Def_DurrettProbability_Series open MeasureTheory ProbabilityTheory Filter
namespace DurrettProbability
theorem kolmogorov_three_series {Ω : Type*} [MeasurableSpace Ω] {μ : Measure Ω}
[IsProbabilityMeasure μ] (X : ℕ → Ω → ℝ) (hmeas : ∀ i, Measurable (X i))
(hindep : iIndepFun X μ) (A : ℝ) (hA : 0 < A) :
SeriesConvergesAE X μ ↔
(Summable (fun n => (μ {ω | A < |X n ω|}).toReal)
∧ (∃ L : ℝ, Tendsto (fun N => ∑ n ∈ Finset.range N, μ[truncate A (X n)]) atTop (nhds L))
∧ Summable (fun n => Var[truncate A (X n); μ])) := by sorry
end DurrettProbabilityConfirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.