Questions: Almost Sure Convergence

Question 1

A sequence Xₙ converges in probability to X — that is, P(|Xₙ−X|>ε) → 0 for every ε > 0. Does this guarantee almost sure convergence?

• A. Yes — convergence in probability is equivalent to almost sure convergence for all sequences
• B. No — convergence in probability only says the probability of being far from X vanishes at each step, but individual sample paths can still oscillate indefinitely without settling
• C. Yes, provided the random variables are bounded
• D. No, but only when the Xₙ are not independent

Answer: (B)

Convergence in probability is a statement about marginal snapshots: at step n, the probability of being far from X is small. But this does not prevent a particular sample path ω from repeatedly wandering away and coming back — the path may never settle. Almost sure convergence requires that for almost every ω, the path eventually stays within ε of X permanently. Almost sure convergence implies convergence in probability, but not vice versa.

Question 2

To prove that Xₙ converges almost surely to X using the Borel-Cantelli lemma, the standard strategy is to show...

• A. That P(|Xₙ−X| > ε) → 0 as n → ∞ for every ε > 0
• B. That the Xₙ are independent and identically distributed with finite mean
• C. That Σₙ P(|Xₙ−X| > ε) < ∞ for every ε > 0, ensuring only finitely many terms deviate beyond ε almost surely
• D. That the sequence is monotone and bounded

Answer: (C)

The first Borel-Cantelli lemma: if Σₙ P(Aₙ) < ∞, then P(Aₙ occurs infinitely often) = 0. Applying this to Aₙ = {|Xₙ−X| > ε}: if the probabilities sum to something finite, then almost surely only finitely many Xₙ deviate from X by more than ε. Since ε was arbitrary, almost surely every path eventually stays within ε of X for all large n — this is almost sure convergence. Option A only gives convergence in probability.

Question 3

Almost sure convergence requires that for almost every individual outcome ω in the sample space, the numerical sequence Xₙ(ω) converges to X(ω) in the ordinary real-analysis sense.

• T. True
• F. False

Answer: True

This is the definition. Recall that each random variable is a function from the sample space Ω to the reals; Xₙ(ω) is just the number that the n-th random variable assigns to outcome ω. Almost sure convergence means that for the set of ω where Xₙ(ω) → X(ω) fails, that set has probability zero. The 'almost' is the one concession — a measure-zero exceptional set is permitted.

Question 4

The Strong Law of Large Numbers and the Weak Law of Large Numbers make the same mathematical statement — that the sample mean converges to the population mean — but the proofs happen to use different techniques.

• T. True
• F. False

Answer: False

They make genuinely different claims. The Weak Law states that the sample mean X̄ₙ converges to μ in probability: for every ε > 0, P(|X̄ₙ − μ| > ε) → 0. The Strong Law states that X̄ₙ → μ almost surely: with probability 1, the actual sample path of averages converges to μ. Almost sure convergence is strictly stronger — it implies convergence in probability, but not conversely. The Strong Law gives a path-level guarantee; the Weak Law only gives snapshot guarantees.

Question 5

Explain the difference between almost sure convergence and convergence in probability using the concept of sample paths.

Model answer: A sample path is the entire sequence of values (X₁(ω), X₂(ω), X₃(ω), ...) for a fixed outcome ω. Almost sure convergence says that for almost every ω, this path eventually settles arbitrarily close to X(ω) and stays there — it is a path-level guarantee. Convergence in probability only says that at each individual time step n, the probability of being far from X is small. A path can simultaneously satisfy 'small probability of large deviation at each n' while oscillating between values far from X, because the oscillations could be rare but persistent. Almost sure convergence rules this out for almost all paths.

The classic demonstration is the 'typewriter sequence' or similar examples where Xₙ → 0 in probability but every individual sample path visits every value in [0,1] infinitely often — a striking failure of almost sure convergence despite perfect convergence in probability.