2
$\begingroup$

I am trying to prove $$X_n \xrightarrow{d} X, Y_n \xrightarrow{d} a \implies Y_n X_n \xrightarrow{d} aX$$ where $a$ is a constant.
What I tried:
Let $g:\mathbb R\to \mathbb R$ an arbitrary uniformly continuous, bounded function. It suffices to show $\mathbb E[g(Y_n X_n)] \to \mathbb E[g(aX)]$. We have $$\left \lvert \int g(Y_nX_n) - g(aX) \,dP \right \rvert \leq \left \lvert \int g(Y_n X_n) - g(aX_n) \, dP \right \rvert +\left \lvert \int g(aX_n) - g(aX) \, dP \right \rvert,$$ where the right summand goes to $0$ by assumption as $a$ is constant. Now I want to use uniform continuity of $g$ to estimate the left summand: Choose $\delta > 0 $ such that $$\left \lvert g(Y_n X_n) - g(aX_n) \right \rvert < \epsilon,$$ whenever $|Y_n X_n - aX_n| < \delta.$ Since convergence in distribution to a constant implies convergence in probability, we have can control $P(|Y_n - a | > \delta)$. But is it possible to get a bound on $|X_n|$? I assume that probability in distribution does not imply some form of boundedness... Any help is appreciated.

$\endgroup$
1
$\begingroup$

First, you can choose $g$ to be Lipschitz bounded, with $C$ its Lipschitz constant. It follows that $\forall \epsilon >0, A > 0$: $$|\mathbb{E}g(X_n Y_n) - \mathbb{E}g(X_n a)| \leq \mathbb{E}|g(X_n Y_n) - g(X_n a)|1_{|X_n Y_n - X_n a| > \epsilon} + \mathbb{E}|g(X_n Y_n) - g(X_n a)|1_{|X_n Y_n - X_n a| \leq \epsilon} \leq C \epsilon + ||g||_\infty \mathbb{P} (|X_n Y_n - X_n a | > \epsilon) \leq C \epsilon + \mathbb{P} (|X_n Y_n - X_n a | > \epsilon, |X_n|> A) + \mathbb{P} (|X_n Y_n - X_n a | > \epsilon, |X_n| \leq A) \leq C \epsilon + \mathbb{P} (|Y_n - a | > \epsilon / A) + \mathbb{P} (|X_n| \geq A)$$

Now, $\lim_{n \rightarrow \infty} \mathbb{P} (|Y_n - a | > \epsilon / A) = 0$ so $\limsup{n \rightarrow \infty} |\mathbb{E}g(X_n Y_n) - \mathbb{E}g(X_n a)| \leq C \epsilon + \mathbb{P} (|X| \geq A)$. I used here that fact that $X_n$ goes in law to $X$ and that $[A, \infty[$ is closed (see the Portemanteau lemma). You can now let $\epsilon$ goes to zero and $A$ goes to $\infty$.

$\endgroup$
  • $\begingroup$ Does it work without $g$ being Lipschitz continuous? $\endgroup$ – Staki42 May 13 '18 at 12:25
  • $\begingroup$ No I think $g$ have to be Lipschitz. But it's not a problem as Lipschitz bounded test functions characterize the convergence in law $\endgroup$ – Adri May 13 '18 at 12:31
  • $\begingroup$ However if $g$ is uniformly continuous it seems that one can use the same truncation method to control $\mathbb{P}(|X_n||Y_n - a | > \delta)$ by splitting between $|X_n|> A$ and $|X_n| \leq A$ then let $A$ goes to infinity. $\endgroup$ – Adri May 13 '18 at 20:39

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.