Let $X_n$ be distributed as a poisson random variable with parameter $n$. Then which of the following are true ?

1.$\underset{n\rightarrow \infty}{\lim} \mathbb{P} (X_n > n + \sqrt n)=0 $

2.$\underset{n\rightarrow \infty}{\lim} \mathbb{P} (X_n \le n)=1 $

3.$\underset{n\rightarrow \infty}{\lim}\mathbb{P} (X_n \le n)=\frac {1}{2} $

The answer is 3. I am not able to think how to start on this. Please any suggestions.


3 Answers 3

  • Let $\lambda$ be the lamda for your sum of independent Poisson random variables ($X_n$)
  • Since $\lambda$ approaches to infinity, we should be able to approximate it with $N(\lambda,\lambda)$.
  • The normal approximation centres on $\lambda$ and is symmetric. You get your answer.

First of all, by elimination the first two are solvable (you can look at the variance and mean of the distribution).

But I strongly agree 3 is tricky, but I came up with an explanation: Poisson distribution has a nice property: $x_i - \operatorname{Poisson}(\lambda_i) \Rightarrow \sum_ix_i-\operatorname{Poisson}(\sum_i\lambda_i) $


$X_n = \sum_{i=1}^{n}X_i, X_i-\operatorname{Poisson}(1)\rightarrow_{n\rightarrow\infty}N\left(\mathbb{E}\left[\sum_iX_i\right],\sum_i\lambda_i\right)=N(n,n) \\\Rightarrow \mathbb{P}(X_n\leq n)=0.5$

The last part is true since Gaussian distribution is symmetric.

BTW - I encourage you to use the same way the calculate the actual probability in 2 and 3


The limiting distribution of $X_{n}$ as $n\rightarrow\infty$ is normal. We can show this using the moment generating function.

Consider a standardized Poisson random variable:


With moment generating function:


Taking the limit:

$$\begin{align} \lim_{n\rightarrow\infty}M_{(X_{n}-n)/\sqrt{n}}(t)&=\lim_{n\rightarrow\infty}E\bigg[\text{exp}\bigg(t\cdot\frac{X_{n}-n}{\sqrt{n}}\bigg)\bigg]\\ &=\lim_{n\rightarrow\infty}\text{exp}\big(-t\sqrt{n}\big)E\bigg[\text{exp}\bigg(\frac{tX_{n}}{\sqrt{n}}\bigg)\bigg]\\ &=\lim_{n\rightarrow\infty}\text{exp}\big(-t\sqrt{n}\big)\text{exp}\big(n(e^{t/\sqrt{n}}-1)\big)\\ &=\lim_{n\rightarrow\infty}\text{exp}\big(-t\sqrt{n}+n(tn^{-1/2}+t^{2}n^{-1}/2+t^{3}n^{-3/2}/6+\ldots)\big)\\ &=\lim_{n\rightarrow\infty}\text{exp}\big(t^{2}/2+t^{3}n^{-1/2}/6+\ldots\big)\\ &=\text{exp}\big(t^{2}/2\big)\\ &=M_{Z}(t) \end{align}$$

Which is the moment generating function of the standard normal random variable, $Z\sim N(0,1)$. Thus, the limiting distribution of our Poisson random variable is simply:

$$\sqrt{n}Z+n\sim N(n,n)$$

So for your options above:

  1. $$\begin{align}\mathbb{P}(X_{n}>n+\sqrt{n})&=\mathbb{P}(N(n,n)>n+\sqrt{n})\\ &=\mathbb{P}(N(0,1)>1)\\ &=1-\Phi(1)\\ &\ne 0 \end{align}$$

  2. $$\begin{align}\mathbb{P}(X_{n}<n)&=\mathbb{P}(N(n,n)<n)\\ &=\mathbb{P}(N(0,1)<0)\\ &=\Phi(0)\\ &\ne 1 \end{align}$$

  3. $$\begin{align}\mathbb{P}(X_{n}<n)&=\mathbb{P}(N(n,n)<n)\\ &=\mathbb{P}(N(0,1)<0)\\ &=\Phi(0)\\ &=0.5 \end{align}$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.