Weak convergence on separable and complete product space

42 Views Asked by At

I read a paper in which the authors seem to have a simplified definition of convergence in distribution of random variables in a product space. The paper itself is very specific, so I can link it but i will rewrite the problem in a more universal notation here (https://arxiv.org/abs/1904.02585 , the important results here are Lemma 2.8 and Lemma B.2 in the appendix):

Let $S$ be a separable, complete space and T := $S \times S$ . Furthermore, there are two sequences $(X_n^1)_{n \in \mathbb{N}}$ $(X_n^2)_{n \in \mathbb{N}}$ of random variables taking values in $S$ as well as two random variables $X^1$ and $X^2$ in $S$. Now Lemma 2.8 shows that for any continuous and bounded functions $f_1,f_2 : S \to \mathbb{R}$ that \begin{equation} \label{one} \mathbb{E} \left[f_1(X_n^1)f_2(X_n^2) \right] \to \mathbb{E} \left[f_1(X_1) \right] \mathbb{E} \left[f_2(X_2) \right] \end{equation} So far so good. Now in the proof of Lemma B.2 it seems to me that they implicitly say, that this would already mean that the law of $(X_n^1,X_n^2)$ converges weakly to the measure product of the laws of $X^1$ and $X^2$.

But this is not the definition of weak convergence since there should be more continuous and bounded continuous $f: S \times S \to \mathbb{R}$ than just the products of two functions $f_1,f_2 : S \to \mathbb{R}$.

I have seen (for example here: Approximating continuous functions on a product space) that you can use Stone-Weierstraß Theorem to approximate continuous functions in such a way if the space $S$ is compact. In my scenario this is not the case.

It might be that there is something special in the structure of the space that they are working with in the paper that i have not understood yet. In this case i cannot expect anyone here to help me, since it's very specific as i said.
Though maybe there is a general answer and hence somebody has an idea why it is implicit in the paper that the equation above implies distributional convergence on the product space.

Thank you in advance.

1

There are 1 best solutions below

0
On

This seems very strange, because the condition doesn't depend on joint distribution of $X_1$ and $X_2$ at all, while weak convergence does.

Let all $X_n^1$ and $X_n^2$ be uniform on $\{0, 1\}$ and independent on each other. Let $X_1 = X_2$ also be uniform on $\{0, 1\}$

Then $\mathbb{E} \left[f_1(X_n^1)f_2(X_n^2) \right] = \frac{(f_1(0) + f_1(1))(f_2(0) + f_2(1))}{4} = \mathbb{E} \left[f_1(X_1) \right] \mathbb{E} \left[f_2(X_2) \right]$.

But if we take $f$ s.t. $f(0, 0) = f(1, 1) = 1$, $f(0, 1) = f(1, 0) = 0$, then $\mathbb{E} f(X_n^1, X_n^2) = \frac{1}{2} \not\to 1 = \mathbb{E} f(X_1, X_2)$.