Let $(X_n)$ be a sequence of independent and positive random variables. Let $S_n = X_1 + \dots + X_n$. Show that $S_n$ converges almost surely if and only if it converges in probability.

The first implication is well-known. But the other one is none trivial as this exercise is an instance of a statement that is generally false.

The way I think I should organize my proof is as follows:

  1. Define $A_k = \{\lvert S_k - S \rvert \geq \epsilon \}$.
  2. Majorize $\mathbb{P}(\lvert S_k - S \rvert \geq \epsilon)$ using Markov inequality.
  3. Show that the upper-bounding general term series is convergent.
  4. Conclude that $\sum \mathbb{P}(A_k) < \infty$
  5. Conclude by Borel-Cantelli lemma, $\mathbb{P}(\limsup A_k) \rightarrow_k 0$.

However, I am stuck at:

$$\mathbb{P}(\lvert S_k - S \rvert \geq \epsilon) \leq \frac{\mathbb{E}[\lvert S_k - S \rvert]}{\epsilon} \cdot$$

So far I haven't used the fact that $S_n = X_1 + ... + X_n$ nor that $(X_n)$ is a sequence of positive variables.


Independence is not required at all. $(S_n)$ is an increasing sequence of real numbers, so $\lim S_n$ exists a.s. (The limit may be $\infty$). But convergence in probability implies convergence a.s. for a subsequence from which it follows that $\lim S_n$ is finite a.s..


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.