6
$\begingroup$

Let $p \in(0,1)$. What is the distribution of the sum of $n$ independent Bernoulli random variables with parameter $p$? Let $0 \leq a < b \leq 1$. Use approprtiate limit theorems to determine how the following depends on $a$ and $b$: $$\lim_{n\to\infty} \sum_{r=\lfloor an \rfloor}^{\lfloor bn \rfloor} {n \choose r } p^r (1-p)^{n-r}\ $$

I know that the sum of $n$ Bernoulli random variables with parameter $p$ has binomial distribution $Bin(n,p)$.

I see that the above is equal to $P(an \leq X \leq bn) $ where $X$ is the sum, and I can see how to use the weak law of large numbers in the cases where $p \neq a$ and $p \neq b$ to find the appropriate limits, but I can't see what I have to do when $p=a$ or $p=b$.

I assume I have to apply the central limit theorem, but I really don't see how to do that. Any help you could give me would be really appreciated.

$\endgroup$
4
$\begingroup$

As you observed, we can express $X=X_1+\dots+X_n$ where the $X_i$ are i.i.d Bernoulli random variables with parameter $p$.

The central limit theorem tells you that $\sqrt{n} \left(\frac{1}{n} X - p\right)$ is distributed as $\mathcal{N}(0, p(1-p))$ when $n$ tends to $\infty$ because $p$ and $p(1-p)$ are respectively the mean and variance of $X_i$ for any $i$. This is true if your variable satisfy the standard conditions for the CLT, which it does but try to check it.

I think that you should be able to conclude now.

EDIT : as you mentioned you are looking for (as $n$ tends to $\infty$) \begin{align*} \mathbb{P}(an\leq X\leq bn) &= \mathbb{P}\left(\sqrt{n}(a-p)\leq \sqrt{n}\left(\frac{1}{n} X - p\right)\leq \sqrt{n}(b-p)\right)\\ &= Q(\sqrt{n}(a-p)) - Q(\sqrt{n}(b-p)) \end{align*}

So now if $a=p$ or $b=p$, one of these two is actually exactly $1/2$, You can just apply the different cases you mentioned to this one.

EDIT 2 : For completeness let me specify the cases. We use the property $Q(0)=1/2$, $\lim_{x\to\infty} Q(x) = 0$ and $\lim_{x\to-\infty} Q(x) = 1$

  • $p\notin [a,b]$ : Then the sign of $a-p$ and $b-p$ are the same and cannot be $0$, this means that for $n\to\infty$, $Q(\sqrt{n}(a-p)) = Q(\sqrt{n}(b-p)) \in \lbrace 0,1\rbrace$ and so $\mathbb{P}(an\leq X\leq bn)=0$
  • $p\in \lbrace a,b \rbrace$ : then the two possible values for $Q(\sqrt{n}(a-p))$ and $Q(\sqrt{n}(b-p))$ are respectively $1/2$, $0$ or $1$, $1/2$. In any case $\mathbb{P}(an\leq X\leq bn)=1/2$
  • $p\in ]a,b[$ : Then $Q(\sqrt{n}(a-p))=1$ and $Q(\sqrt{n}(b-p))=0$ and $\mathbb{P}(an\leq X\leq bn)=1$.
$\endgroup$
3
  • $\begingroup$ I see how $\sqrt{n} \left(\frac{1}{n} X - p\right)$ tends towards $\mathcal{N}(0, p(1-p))$, but I'm struggling to work out how I'm supposed to use that. If it's not too much trouble, would you be able to walk me through it in a bit more depth? $\endgroup$
    – bobphimist
    Sep 11 '18 at 18:21
  • $\begingroup$ no problem at all, see my edit, and if you need anything more, feel free to ask. $\endgroup$
    – P. Quinton
    Sep 11 '18 at 18:38
  • $\begingroup$ Thank you so much! $\endgroup$
    – bobphimist
    Sep 16 '18 at 17:21

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.