Assume $U_k$ are correlated standard normal random variables.

Let $R_k := a_k U_k^2$, with $a_k > 0$ and $\sum_{k=1}^{\infty} a_k < \infty$.

How can we prove that $S_p:= \frac{1}{p}\sum_{k=1}^{p}R_k$ converge to 0 in probability?

  • $\begingroup$ Is this a homework problem or textbook question? If so, please tag as self-study. $\endgroup$ – Arya McCarthy Apr 11 at 0:00
  • $\begingroup$ @AryaMcCarthy it's not, but if it's a trivial result, a link to some textbook with the solution would also suffice. I'm just not sure how to handle the dependency between the $U_k$ to write this down analytically. $\endgroup$ – runr Apr 11 at 0:04

The first thing to try is always Chebyshev's inequality -- especially with well-behaved distributions where moments are likely to be enough to decide the question.

$$\mathrm{var}[S_p]=\frac{1}{p^2}\sum_{i,j=1}^p \mathrm{cov}[a_iU_i^2,a_jU_j^2]$$

Now, correlations are bounded by 1,so: $$\left|\mathrm{cov}[a_iU_i^2,a_jU_j^2]\right|\leq \sqrt{\mathrm{var}[a_iU_i^2]\mathrm{var}[a_jU_j^2]}=|a_ia_j|C$$ for some finite $C$ that we could work out if we cared.

So, $$\mathrm{var}[S_p]\leq C\sum_{i,j=1}^p|a_ia_j|$$ and this bound is sharp; it is achieved when the $U_i$ are perfectly correlated. We need only ask if $$\lim_{p\to\infty}\frac{1}{p^2}\sum_{i,j=1}^p|a_ia_j|= 0$$

There can be only finitely many $a_i>1$; we can delete an initial segment of the series if necessary and assume there are none, so $a_ia_j<a_i$. Then $$\frac{1}{p^2}\sum_{i,j=1}^p|a_ia_j|< \frac{1}{p^2}\sum_{i=1}^pp|a_i|=\frac{1}{p}\sum_{i=1}^p|a_i|$$

Since $\sum_{i=1}^p|a_i|$ is bounded, $$\frac{1}{p}\sum_{i=1}^p|a_i|$$ converges to zero.

So, $S_p$ converges to zero in mean square and thus in probability.


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.