My setup is the following.

Consider a sequence of random variables $(X_n)_n$ such that for every $\delta>0$ \begin{align*} \lim_{n,m\rightarrow\infty}\mathbb{P}[\sup_{m<k\leq n}|X_m-X_k|\geq \delta]=0 \end{align*}

And I have to prove that there is a random variable $X$ such that the sequence converges to $X$ almost surely.

What I have: The assumption implies the sequence is Cauchy in measure so there is a random variable $X$ such that the sequence converges to $X$ but in measure. Naturally, that is my aspirant to the a.s. limit. Furthermore, there exists a subsequence that converges a.s. to $X$ but I can't manage to prove the whole sequence converges.

Any suggestions? Thanks a lot!


I believe we can do this without appealing to the existence of a random variable $X$ to which the sequence converges in measure. My argument is a little informal, so please let me know if you find something confusing or if you suspect something is incorrect.

Suppose that $(X_{n})_{n\in\mathbb{N}}$ does not have an a.s.-limit. Then $\{(X_{n})_{n\in\mathbb{N}}\text{ is not Cauchy}\}$ has positive measure. In particular, there exists some $j\in\mathbb{N}$ such that $\{\limsup_{n,m\to\infty}|X_{m}-X_{n}|>1/j\}$ has positive measure, for $$ \{(X_{n})_{n\in\mathbb{N}}\text{ is not Cauchy}\}=\bigcup_{j\in\mathbb{N}}\{\limsup_{n,m\to\infty}|X_{m}-X_{n}|>1/j\} $$ and if all the terms in the union have measure zero then countable subadditivity tells us that the lefthand side must have measure zero. $\{\limsup_{n,m\to\infty}|X_{m}-X_{n}|>1/j\}$ having positive measure contradicts your initial hypothesis.

  • $\begingroup$ Thank you for your answer. I believe the idea is nice but I have problems comparing the sets $\{\limsup_{m,n}|X_n-X_m|\geq 1/j\}$ and $\{\sup_{m<k\leq n}|X_k-X_m|\geq1/j\}$ $\endgroup$
    – user152168
    Sep 16 '15 at 3:40
  • $\begingroup$ If $\omega\in\{\limsup_{n,m\to\infty}|X_{n}-X_{m}>1/j\}$, then for arbitrarily large $N,M\in\mathbb{N}$ there exist $n\geq N$ and $m\geq M$ such that $|X_{n}(\omega)-X_{m}|>1/j$. But we might not have $\omega\in\{\sup_{m<k\leq n}|X_{k}-X_{m}|>1/j\}$ if, for example, you take $n$ and $m$ to always differ by a fixed constant. So I think it depends on how you take $n,m\to\infty$. Sorry for the lack of clarity; I'm not sure exactly what's meant in this context by the limit as $n$ and $m$ approach $\infty$ at the same time! $\endgroup$ Sep 16 '15 at 14:10

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.