Double asymptotic proof: $\mathbf{x}_n \to_d\mathbf{z}\implies\frac{1}{p}\mathbf{x}_n^\top\mathbf{x}_n\to 1$ as $(p,n) \to (\infty, \infty)$

50 Views Asked by At

Let $\mathbf{x}$ be a $p$-dimensional vector, and $\mathbf{z}$ is a $p$-dimensional standard normal vector whose elements are i.i.d. Now we assume for some scaling parameter $n \to \infty$ (and fixed $p$), $$ \mathbf{x} \xrightarrow{\mathcal{L}} \mathbf{z}, $$ where $\xrightarrow{\mathcal{L}}$ is the convergence in law (or convergence in distribution). Then my question is that the following.

Question.

Can we say that for any path $(p,n) \to (\infty, \infty)$, $$ \frac{1}{p} \mathbf{x}^\top \mathbf{x} \xrightarrow{\mathbb{P}} 1?$$

For one path: $n\to \infty$ first, then $p \to \infty$ is quite obvious, but I'm not sure how to handle for any arbitrary path.

I look forward your any help.

Thanks,

1

There are 1 best solutions below

4
On BEST ANSWER

It would not work for all paths. Here is a counterexample, where we first let $p\to\infty$ and then let $n\to\infty$:

Given iid $B_i\sim \text{Bern}(1/n), i=1,...,p$, each independent of $\bf z$, define the $i$th element of $\bf x$ as

$$x_i=z_i+nB_i.$$

Note that as $n\to \infty$,
$$nB_i \xrightarrow{P}0\implies nB_i \xrightarrow{D}0\underbrace{\implies}_{\text{Slutsky's theorem}} x_i\xrightarrow{D}z_i \underbrace{\implies}_{x_i \text{ iid}} {\bf x} \xrightarrow{D} {\bf z}.$$

WLLN implies as $p\to \infty$, $${1 \over p} {\bf x}^\top {\bf x} \xrightarrow{P} E[x_i^2].$$

However, note that $$E[x_i^2]=E[z_i^2]+n^2E[B_i]+2n\underbrace{E[B_iz_i]}_{=0 \text{ by indep}}=1+n\overset{n\to \infty}{\not\to} 1.$$