Convergence of random series, truncation, and the exchangeable sigma-field
DefinitionDurrettProbability_SeriesFour notions from section 2.5 of Durrett, Probability: Theory and Examples.
Partial sums. For a sequence of random variables, .
Almost sure convergence of a series. The series converges almost surely when for almost every there is a real number with . Following Durrett, "the series converges" means the partial sums have a limit, not that the series converges absolutely.
Truncation. For a level , the truncation of is
the variable that agrees with where and vanishes elsewhere. It is bounded, so its mean and variance always exist — which is why Kolmogorov's three-series theorem can dispense with any integrability assumption on itself.
Permutable events. A finite permutation of is a bijection fixing all but finitely many indices. An event in a sequence space is permutable, or exchangeable, when rearranging finitely many coordinates leaves it unchanged: for every finite permutation . Examples are and , since permuting finitely many coordinates does not change for large ; every tail event is permutable, and the first example shows the converse fails. The permutable events form the exchangeable σ-field .
Formalization Note Convergence of a series is the convergence of the sequence of partial sums
to a real limit, not Mathlib's Summable, which for real series means unconditional and hence
absolute convergence. The distinction matters in condition (ii) of the three-series theorem.
Almost sure convergence asserts the existence of a limit for almost every outcome without claiming
the limit is a measurable function. A permutable event is defined directly by its invariance
property rather than by constructing the exchangeable σ-field as an object.
import Mathlib
open MeasureTheory ProbabilityTheory Filter
namespace DurrettProbability
/-- The partial sums `S_n = X_1 + ⋯ + X_n` of a sequence of random variables. -/
def partialSum {Ω : Type*} (X : ℕ → Ω → ℝ) (n : ℕ) (ω : Ω) : ℝ := ∑ i ∈ Finset.range n, X i ω
/-- The series `∑ X_n` **converges almost surely**: for almost every `ω` the partial sums
`S_N(ω)` converge to some real limit. Durrett, *Probability: Theory and Examples*, section 2.5. -/
def SeriesConvergesAE {Ω : Type*} [MeasurableSpace Ω] (X : ℕ → Ω → ℝ) (μ : Measure Ω) : Prop :=
∀ᵐ ω ∂μ, ∃ L : ℝ, Tendsto (fun N => partialSum X N ω) atTop (nhds L)
/-- The truncation `Y = X · 1(|X| ≤ A)` used in Kolmogorov's three-series theorem. -/
noncomputable def truncate {Ω : Type*} (A : ℝ) (X : Ω → ℝ) : Ω → ℝ := fun ω => if |X ω| ≤ A then X ω else 0
/-- A **finite permutation** of `ℕ`: a bijection moving only finitely many indices. -/
def FinitelySupported (σ : Equiv.Perm ℕ) : Prop := ∀ᶠ i in Filter.cofinite, σ i = i
/-- A **permutable** (exchangeable) event: a measurable set of sequences whose occurrence is
unaffected by rearranging finitely many coordinates. These form the exchangeable σ-field `ℰ` of
Durrett, section 2.5. -/
def IsPermutable {S : Type*} [MeasurableSpace S] (A : Set (ℕ → S)) : Prop :=
MeasurableSet A ∧ ∀ σ : Equiv.Perm ℕ, FinitelySupported σ → (fun ω => ω ∘ σ) ⁻¹' A = A
end DurrettProbability