the problem in question is: Let $\{X_n\mid n\ge 1\}$ and $Y$ be random variables on probability space $(\Omega, \mathcal F, \mathbb P)$. Suppose that the $X_n$ are independent, $E[X_n] = 0$ and $E[X_n^2] = 1$ for all $n$, and $Y$ is square-integrable. Show that $E[X_n Y] \to 0$ as $n\to \infty$.

My attempt at a solution:

First define $Z_n = E[X_n Y]X_n$. The $Z_n$ are independent, and $E[Z_n^2] = E^2[X_nY]E[X_n^2] = E^2[X_nY]$. By Cauchy Schwarz, this last probability is bounded above by $E[Y^2] = M <\infty$. In particular, the $Z_n$ are uncorrelated and have uniformly bounded second moment, and $E[Z_n] = 0$, so $E[\sum_{i=1}^n Z_n] = 0$. We can apply a SLLN theorem and get $$\frac{\sum_{i=1}^n Z_i}{n} \to 0$$ almost surely. I'm not sure where to go from here or if all this is exactly correct... the prof left a hint to consider $\sum_{i=1}^n E[X_i Y]X_i$

  • $\begingroup$ You need to include some additional conditions/check the existing ones, because $E(X_nY)\to 0$ simply doesn't have to hold. $\endgroup$ – A.S. Nov 3 '15 at 18:47
  • $\begingroup$ @A.S. Ah, sorry I forgot to include that the $X_n$ are independent - obviously that's a big assumption to add. $\endgroup$ – user62748 Nov 3 '15 at 19:02

It suffices to show that if $\left(n_j\right)_{j\geqslant 1}$ is an increasing sequence of real numbers such that the sequence $\left(\mathbb E\left[X_{n_j}Y\right]\right)_{j\geqslant 1}$ converges to some real number $l$, then $l=0$. Indeed, the sequence $\left(\mathbb E\left[X_{n}Y\right]\right)_{n\geqslant 1}$ is bounded, hence it is possible to extract a convergent subsequence. By what we assumed, the potential limit is $0$. Applying this reasoning to each subsequence, we can deduce that the sequence $\left(\mathbb E\left[X_{n}Y\right]\right)_{n\geqslant 1}$ converges to $0$.

Now, consider an increasing sequence of real numbers $\left(n_j\right)_{j\geqslant 1}$ and define $c_j:=\mathbb E\left[X_{n_j}Y\right]$.

Here are some hints to show the wanted claim.

  • The sequence $\left(1/n\sum_{j=1}^nc_j\right)_{n\geqslant 1}$ converges to $0$.

  • If $c_j\to l$, then $\left(1/n\sum_{j=1}^nc_j\right)_{n\geqslant 1}$ converges to $l$.

  • Conclude.

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.