Theorem 2.5.6 — summable variances give almost sure convergence
ProvedDurrettProbability.series_converges_of_summable_varianceLet be independent and square-integrable with . If
then with probability one the series converges — that is, the partial sums have a real limit for almost every outcome.
This is the workhorse of the section, and its content is that under independence and centring, a condition on second moments alone is enough: no assumption about the shape of the distributions, no absolute convergence, no monotonicity. Absolute convergence would require and is strictly stronger; Durrett's Example 2.5.7 makes the gap concrete with , where the series converges for but converges absolutely only for .
The proof runs through Kolmogorov's maximal inequality applied to the increments beyond time : it bounds by , which tends to zero, so the partial sums are almost surely Cauchy.
Formalization Note Convergence of the series is convergence of the partial sums to a real
limit for almost every outcome, following the book's convention, rather than Mathlib's Summable,
which for real series means absolute convergence and would be a strictly stronger conclusion. The
hypothesis Summable on the variances is the right reading there, since they are non-negative.
import Mathlib import Definitions.Def_DurrettProbability_Series open MeasureTheory ProbabilityTheory Filter
namespace DurrettProbability
theorem series_converges_of_summable_variance {Ω : Type*} [MeasurableSpace Ω] {μ : Measure Ω}
[IsProbabilityMeasure μ] (X : ℕ → Ω → ℝ) (hmeas : ∀ i, Measurable (X i))
(hindep : iIndepFun X μ) (hL2 : ∀ i, MemLp (X i) 2 μ) (hmean : ∀ i, μ[X i] = 0)
(hvar : Summable (fun n => Var[X n; μ])) :
SeriesConvergesAE X μ := by sorry
end DurrettProbabilityConfirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.