8
$\begingroup$

Let $\{X_1,X_2,\dots\}$ be a sequence of i.i.d. random variables. Define $M_n=\sup\limits_{k\leq n} |X_k|$. Suppose $X_1$ is $p-$th integrable i.e. $E[|X_1|^p]<\infty$ for some $p\in(0,\infty)$. Show that

$\frac{1}{n^{1/p}}M_n\to 0$

with probability 1.

I have already proven that for an integrable random variable $X$,

$\sum\limits_{n=1}^\infty P(|X|\geq \epsilon n)<\infty$

Hence to prove the statement, let $\epsilon >0$. Define $A_n=\{|X_n|\geq \epsilon n^{1/p}\}$.

Then $\sum\limits_{n=1}^\infty P(A_n)=\sum\limits_{n=1}^\infty P(|X_n|\geq \epsilon n^{1/p})\leq \sum\limits_{n=1}^\infty P(M_n\geq \epsilon n^{1/p})$

I want to show that $\sum\limits_{n=1}^\infty P(A_n)<\infty$ and then use the Borel-Cantelli lemma to show $P(\limsup A_n)=0$.

I don't know how to show $\sum\limits_{n=1}^\infty P(M_n\geq \epsilon n^{1/p})<\infty$. I don't know how to use the condition that $X_1$ is p-th integrable.

$\endgroup$
4
$\begingroup$

Since the $X_j$ are identically distributed, $\mathbb{P}(|X_n| \geq \epsilon n^{1/p}) = \mathbb{P}(|X_1| \geq \epsilon n^{1/p}) = \mathbb{P}(|X_1|^p \geq \epsilon^pn)$. Then as $X_1$ is $p$-th integrable from what you've already shown $\sum_{n=1}^{\infty} \mathbb{P}(|X_n|^p \geq \epsilon^pn) = \sum_{n=1}^{\infty} \mathbb{P}(|X_1|^p \geq \epsilon^pn) < \infty$ so, as you noted, by Borel-Cantelli $\mathbb{P}(\lim \sup A_n) = 0$.

Now we argue that the event $\{\lim_{n \to \infty} \frac{1}{n^{1/p}}M_n < 2 \epsilon \}$ contains the complement of $\lim \sup A_n$ and in particular has probability $1$.

Suppose $\omega \not \in \lim \sup A_n$. Then $|X_n(\omega)| < \epsilon n^{1/p}$ eventually, say for $n \geq N$. So $\frac{1}{n^{1/p}}M_n(\omega) < \frac{1}{n^{1/p}}\max_{k < N} |X_k(\omega)| + \epsilon$ for any $n \geq N$. But since the $\max$ is over finitely many terms the first part of the right hand side tends to $0$ as $n \to \infty$ so there is an $M$ such that for $n \geq M$, $\frac{1}{n^{1/p}}\max_{k < N} |X_k(\omega)| < \epsilon$. Hence for $n \geq \max\{M,N\}$ we have that $\frac{1}{n^{1/p}}M_n(\omega) < 2 \epsilon$ so $\omega \in \{\lim_{n \to \infty} \frac{1}{n^{1/p}}M_n < 2 \epsilon \}$.

To conclude, let $B_k = \{\lim_{n \to \infty} \frac{1}{n^{1/p}}M_n < \frac{2}{k} \}$. By taking $\epsilon = \frac{1}{k}$ in the above $\mathbb{P}(B_k) = 1$ for each $k$ so $\mathbb{P}(\bigcap B_k) = 1$. But since $M_n \geq 0$, $\bigcap B_k = \{\lim_{n \to \infty} \frac{1}{n^{1/p}}M_n = 0 \}$

Edit: I've realised that a problem with the original answer that I give above is that it presupposes the existence of the limit in the third paragraph. An easy fix for this is to just look at the set $\{\lim \sup_{n \to \infty} \frac{1}{n^{1/p}}M_n < 2 \epsilon \}$ instead and then all of the above argument goes through.

$\endgroup$
  • $\begingroup$ Note that $\frac{1}{n^{1/p}} |X_j(\omega)| < ...$ for each $j$ then just take the $\sup$ on the left hand side. $\endgroup$ – Rhys Steele May 6 '17 at 15:25

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.