1
$\begingroup$

This is a homework question for a course in asymptotic statistics.

Definitions.

Let $X_1, X_2, \ldots$ be sequence of random variables with $|X_n| < B$ almost surely for all $n$.

Let $X_n \xrightarrow[]{p} X$ denote convergence in probability: $\lim_{n\to \infty} P(|X_n -X| > \epsilon ) = 0.$

Let $X_n \xrightarrow[]{r} X$ denote convergence in the rth mean: $\lim_{n\to \infty} E(|X_n -X|^r) = 0.$

Let $r > 0$.

Problem.

Show that $X_n \xrightarrow[]{p} X$ if and only if $X_n \xrightarrow[]{r}X$.

Work.

I have a proof for the if part, but not for the "only if" part: $$X_n \xrightarrow[]{r} X \Leftarrow X_n \xrightarrow[]{p} X.$$

I tried to show that $E|X_n - X|^r \leq P(|X_n-X| \geq \epsilon)$.

First: I tried expanding $|X_n - X|$ using the binomial theorem. Could only find a bound in terms of a sum over $B$.

Second: Tried writing the expectation in terms of a probability density integral. Again I can only find a bound in terms of $B$.

Third: I tried splitting the $E|X_n - X|^r$ using indicator functions. I ended up getting that $E|X_n - X|^r \geq E|X_n - X|^r$. I don't see how to do this better. I feel like it should involve $B$ somehow.

Status.

I am very stuck.

$\endgroup$
  • 1
    $\begingroup$ When $|X_n-X|<\epsilon$ fails, which it might sometimes, $B^r$ bounds the difference $|X_n-X|^r$. The hit you take (in bounding $E|X_n-X|^r$) from this is bounded by $B^r$ times the chance that $|X_n-X|\ge \epsilon$ . $\endgroup$ – kimchi lover Sep 23 '17 at 21:08
1
$\begingroup$

First off, notice that we have $|X| \leq B$ almost surely. Let $\epsilon > 0$. We find \begin{align*} \mathbb{E}[|X_n - X|^r] &= \int_{|X_n -X| < \epsilon} |X_n -X|^r dP + \int_{|X_n -X| > \epsilon}|X_n -X|^r dP \\ &\leq \epsilon^r + \int_{|X_n -X| > \epsilon} (2B)^r dP = \epsilon^r + (2B)^r \cdot P(|X_n -X| > \epsilon) \end{align*} Taking the lim sup with respect to $n$ of both sides, we obtain $$\limsup_{n\to\infty} \mathbb{E}[|X_n -X|^r] \leq \epsilon^r$$ for any $\epsilon > 0$, which proves the claim.

$\endgroup$

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.