[Paper Review] Generalization of the Borel-Cantelli Lemma

Alexei Stepanov|ArXiv.org|Apr 29, 2006

Nonlinear Differential Equations Analysis6 references18 citations

TL;DR

This paper generalizes the Borel-Cantelli Lemma by introducing a new sufficient condition for the probability of infinitely many events occurring to be zero. It shows that if the sum of probabilities of the events where $ A_{n+m} $ occurs after $ m $ consecutive non-occurrences of $ A_n, \dots, A_{n+m-1} $ is finite, then $ P\{A_n\ \text{i.o.}\} = 0 $, extending previous results like Barndorff-Nielsen's lemma and weakening earlier independence or summability assumptions.

ABSTRACT

In the present note a generalization of Borel-Cantelli Lemma is proposed.

Motivation & Objective

• To extend the classical Borel-Cantelli Lemma beyond independence and summability conditions.
• To weaken the sufficient condition for $ P\{A_n\ \text{i.o.}\} = 0 $ by introducing a delayed occurrence structure.
• To generalize Barndorff-Nielsen's Lemma by incorporating longer sequences of non-occurrences before the next occurrence.
• To provide a broader framework for strong limit theorems in probability theory under minimal assumptions.

Proposed method

• Introduce a sequence of events $ A_n $ with $ P(A_n) \to 0 $, ensuring asymptotic rarity.
• Define a new event structure: $ \overline{A}_n \overline{A}_{n+1} \cdots \overline{A}_{n+m-1} A_{n+m} $, representing $ A_{n+m} $ occurring after $ m $ consecutive failures.
• Establish a sufficient condition: if the sum of probabilities of these delayed occurrence events is finite, then $ P\{A_n\ \text{i.o.}\} = 0 $.
• Generalize Barndorff-Nielsen's Lemma by allowing $ m \geq 2 $, making the condition strictly weaker than the original.
• Use measure-theoretic arguments to show that the tail behavior of the sequence is controlled by the summability of these structured events.
• Demonstrate that the condition in (2.1) implies the limsup event has probability zero, even without independence.

Experimental results

Research questions

• RQ1 Can the Borel-Cantelli Lemma be extended to cases where the sum of individual event probabilities diverges, but structured dependencies are present?
• RQ2 How can the condition for $ P\{A_n\ \text{i.o.}\} = 0 $ be weakened beyond Barndorff-Nielsen’s result?
• RQ3 What is the role of delayed occurrences—specifically, $ A_{n+m} $ following $ m $ non-occurrences—in controlling the probability of infinitely many occurrences?
• RQ4 Is there a general condition involving $ m $-step dependencies that ensures $ P\{A_n\ \text{i.o.}\} = 0 $ under $ P(A_n) \to 0 $?
• RQ5 Can the asymptotic behavior of the limsup event be characterized via sums of structured joint probabilities?

Key findings

• The paper establishes that if $ \sum_{n=1}^\infty P(\overline{A}_n \overline{A}_{n+1} \cdots \overline{A}_{n+m-1} A_{n+m}) < \infty $ for some $ m \geq 0 $, then $ P\{A_n\ \text{i.o.}\} = 0 $.
• This condition is strictly weaker than Barndorff-Nielsen’s condition when $ m \geq 2 $, as it accounts for longer sequences of non-occurrences before the next event.
• The result holds without requiring independence among the events $ A_n $, extending the scope of the Borel-Cantelli framework.
• The condition generalizes both the classical Borel-Cantelli Lemma and Barndorff-Nielsen’s result by incorporating memory of length $ m $ in the dependency structure.
• The paper shows that $ P\{A_n\ \text{i.o.}\} = \alpha \in [0,1] $ if and only if $ \lim_{n \to \infty} \sum_{k=0}^\infty P(\overline{A}_n \cdots \overline{A}_{n+k-1} A_{n+k}) = \alpha $, linking the probability of the limsup to a limit of structured sums.
• The result provides a new tool for proving strong limit theorems in probability by relaxing the need for independence or summability of marginal probabilities.