Convergence in distribution of random sequences

76 Views Asked by At

Let's suppose $X,Y \in \mathbb{R}^{\mathbb{N}}$ are random sequences.

I'd like to show that:

$$ \begin{equation} (X_1,...,X_n) \overset{d}{=} (Y_1,...,Y_n) \implies X \overset{d}{=} Y \end{equation}\tag{1}$$

where $X_i,Y_i \in \mathbb{R}$ are random variables and $\overset{d}{=}$ specifies equality in distribution.

In particular, the approach I'd like to use involves convergence of elements in $\mathbb{R}^n$ to elements in $\mathbb{R}^\mathbb{N}$:

$$ \begin{equation} \lim_{n \to \infty}(X_1,...,X_n) \overset{d}{=} X\end{equation} \tag{2}$$

At first glance this might seem silly as $\mathbb{R}^n$ and $\mathbb{R}^\mathbb{N}$ are different spaces but I think this statement can be made precise by arguing that $(2)$ is equivalent to:

$$ \begin{equation} \sum_{n=1}^{\infty} \frac{P(X_n \overset{d}{=} X_{(n)})}{2^n} = 1\end{equation} \tag{3}$$

which may be demonstrated using the definition of convergent series.

To those who are more familiar with probability theory, might there be any important gaps in my line of reasoning?

1

There are 1 best solutions below

1
On BEST ANSWER

It doesn't strictly make sense because, as you say, convergence in distribution only makes sense for a sequence of random objects taking values in the same state space.

However, if you define $X^{(n)}$ as the random sequence $X^{(n)} = (X_1, X_2, \dots, X_n, 0, 0, \dots)$, then you ought to be able to show that $X^{(n)} \to X$ in distribution.

But I suspect this approach is overcomplicated. I would go for a $\pi$-$\lambda$ argument. Show that the collection $\mathcal{L}$ of Borel sets $B \subset \mathbb{R}^{\mathbb{N}}$ for which $P(X \in B) = P(Y \in B)$ is a $\lambda$-system. Now consider the collection $\mathcal{P}$ of sets $B \subset \mathbb{R}^{\mathbb{N}}$ of the form $B = A \times \mathbb{R}^{\mathbb{N}}$ where $A$ is a Borel subset of $\mathbb{R}^n$ for some $n$. Show that this is a $\pi$-system that generates the Borel $\sigma$-algebra of $\mathbb{R}^{\mathbb{N}}$. And your hypothesis (1) is precisely the statement that $\mathcal{P} \subset \mathcal{L}$, so the $\pi$-$\lambda$ theorem implies that $\mathcal{L}$ contains all Borel sets, i.e. $X \overset{d}{=} Y$.