4
$\begingroup$

Let $X_n$ $(n=1,2,\dots)$ be a sequence of discrete random variables, where the distribution of $X_n$ is the discrete uniform over $\{0, 1/n, 2/n,\dots,1 \}$. Let $U$ be a random variable whose distribution is the continuous uniform over $[0,1]$.

Does $X_n \to U$ almost surely?

My approach is to use Borel-Cantelli to check that $\sum_n\Pr[|X_n-U|>\epsilon] < \infty$, but it appears that for $n>2$ and $\epsilon_n < 1/(n+1)$, we have $\Pr[|X_n-U|>\epsilon_n] = 1-2n{\epsilon}_n/(n+1)$, whose sum does not converge. Is there another approach?

If the convergence does not happen almost surely, is it possible for any discrete random variable to converge almost surely to a continuous random variable? (It is clear in the example above that that $X_n \to^\mathrm{D} U$ i.e. convegence in distribution.)

$\endgroup$
2
$\begingroup$

As written, the statement is false. Suppose the sample space is $\Omega = [0,1]$, with the probability measure $P$ being the Lebesgue measure. Define the random variables $$ X_n : \Omega \to \mathbb R, \ \ \ \ X_n(\omega)=\begin{cases} \frac{\lfloor (n+1)\omega\rfloor}{n} & {\rm if \ } n {\rm \ is \ odd} \\ 1 - \frac{\lfloor (n+1)\omega\rfloor}{n} & {\rm if \ } n {\rm \ is \ even} \end{cases}$$

Notice that the distribution of $X_n$ is discrete uniform over $\{0, 1/n, 2/n, \dots, 1 \}$, and this is true regardless of whether $n$ is odd or even.

And yet, $X_n$ fails to converge pointwise a.s.

$\endgroup$
4
  • $\begingroup$ Nice! From your counter-example, can we conclude that there cannot exist any sequence $X_1, X_2,X_3,\dots$ with discrete uniform over $\{0,1/n,2/n,\dots,1\}$ which converges to $U([0,1])$ a.s? More generally, is it true that if $(Z_n)$ is identically distributed to $(X_n)$, and we know that $(X_n)$ does not converge to $X$ a.s., then $(Z_n)$ cannot converge to $X$ a.s? $\endgroup$
    – jII
    Jun 21 '18 at 23:53
  • $\begingroup$ @jesterII Well, if $X_n(\omega) = \frac{\lfloor (n+1)\omega \rfloor }{n}$ for all $n$ (even or odd), and if $U(\omega) = \omega$, then $X_n$ has the correct discrete distribution and $U$ has the correct continuous distribution, and $X_n$ converges a.s. to $U$. $\endgroup$
    – Kenny Wong
    Jun 21 '18 at 23:56
  • $\begingroup$ It seems then to determine whether a random sequence $X_n$ converges almost surely to $X$, it is not in general enough to only know the distribution $P(X_n \in A)$ of each $X_n$ in isolation, somehow we need to know more about the structure of the random variables or the sampling process. $\endgroup$
    – jII
    Jun 22 '18 at 0:07
  • $\begingroup$ @jesterII Yes, I agree! $\endgroup$
    – Kenny Wong
    Jun 22 '18 at 0:22

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.