For each $n=1,2,\ldots $, let $\ \xi_{n1},\ldots, \xi_{nn}$ be random and independent variables such that $\mathbb{P}(\xi_{ni}=1)=p_n \ \ $ and $\ \ \mathbb{P}(\xi_{ni}=0)=1-p_n$. Let consider the process: $$ Y_n(t)=\sum_{k=1}^{[nt]}\xi_{nk} \ \ \ \ \ t\in [0,1],$$ where $[x]$ represents the floor of $x$.

Supposing that $np_n\to \lambda\ $ as $n$ goes to $\infty$, with $\lambda>0$, prove that: $$ Y_n \to Y \ \ \ (\text{in distribution), in} \ D[0,1], $$ where $Y$ is a Poisson process with parameter $\lambda$, and $D[0,1]$ is the Skorokhod's (Skorohod's) space.

So what I got is: $$ \xi_{ni}\in Be(p_n),$$ for all $n$ and for all $i=1,\ldots ,n$.
And with this and the independence of $\xi_{n1},\ldots ,\xi_{nn}$, we deduce that: $$ Y_n(t)=\sum_{k=1}^{[nt]}\xi_{nk} \in Bin([nt],p_n)$$ I also know that if $X_n\in Bin(n,p_n) \ \ $ and $\ \ np_n \to \lambda$, then $\ \ X_n\to Poisson(\lambda) \ \ $(in distirbution.)

I dont't know how to continue with this. I am also not sure if I can use that: $$ [nt]p_n \to \lambda t \ \ \text{,and then,} \ \ Y_n(t)\to Poisson(\lambda t)$$ I am really lost from here since even if what I said was true, that doesn't implies it happens in $D[0,1]$.

Any help will be appreciated.


$\def\la{\lambda}$ $\def\To{\Rightarrow}$ $\def\bR{\mathbb{R}}$ $\def\al{\alpha}$ $\def\be{\beta}$ $\def\ep{\varepsilon}$

Yes, you can use what you wrote to conclude that for each fixed $t$, the sequence of random variables $Y_n(t)$ converges in distribution to a Poisson with parameter $\la t$. Or in other words, $Y_n(t) \To Y(t)$ in $\bR$.

This, however, is not enough to show that $Y_n\To Y$ in $D[0,1]$. For this, you will want to show two things:

  1. For every finite set $\{t_1,\ldots,t_k\}\subset[0,1]$, $$ (Y_n(t_1), \ldots, Y_n(t_k)) \To (Y(t_1), \ldots, Y(t_k)) $$ in $\bR^k$, and
  2. The sequence $\{Y_n\}$ is relatively compact in $D[0,1]$.

In any proof of convergence of this type, (2) tends to be the most difficult. Things are sometimes a little easier in $C[0,1]$, but in $D[0,1]$, here is a sufficient condition for relative compactness:

There exists $\al,\be,C>0$ such that for all $t$ and $\ep$ satisfying $0\le t-\ep<t+\ep\le 1$, we have $$ E[|Y_n(t+\ep) - Y_n(t)|^\al|Y_n(t) - Y_n(t-\ep)|^\al] \le C\ep^{1+\be} $$ for all $n$.

This sufficient condition is an immediate consequence of Theorem 3.8.8 in Ethier and Kurtz.


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.