Finding $P(X \leq a | Y = k)$ where $X\sim\operatorname{Exp}(\lambda)$ and $Y = [X]$ where $[\cdot]$ rounds up to the nearest integer.

97 Views Asked by At

$Y = [X]$ where $[\cdot]$ rounds up to the nearest integer. It's given that $X \sim \operatorname{Exp}(\lambda)$. The first part asks me to show that $Y \sim \operatorname{Geo}(1-e^{-\lambda})$. The question asks me to find $P(X \leq a | Y = k) \forall x,y \in \Bbb R$ and any $a>0$. I know the formula for this in terms of the joint distribution and the marginal pmf of $Y$. I tried splitting it into cases where $a<k-1$, $k-1 \leq a < k$, $a \geq k$. For which I think $P(k-1 \leq X \leq a)$ is relevant. Am I on the right track here or not?

Thanks!

1

There are 1 best solutions below

5
On BEST ANSWER

From what you claim is required to show. I think $[.]$ refers to the ceiling function.

Then $P(Y=k)=P(\lceil X\rceil = k) = P(k-1< X\leq k) = \int_{k-1}^{k}\lambda e^{-x\lambda}dx = e^{-(k-1)\lambda}-e^{-k\lambda} = e^{-(k-1)\lambda}(1-e^{-\lambda})$.

So the pmf of $Y$ follows pmf of $\text{Geo}(1-e^{-\lambda})$ distribution.

Next $Y=k$ implies $k-1< X \leq k$.

So $P(X\leq a | Y=k)=\int_{k-1}^{a}\lambda e^{-x\lambda}dx = e^{-(k-1)\lambda}-e^{a\lambda}\,\,$ if $k-1<a< k$ and $P(X\leq a | Y=k)=1$ if $a\geq k$ and is $0$ for all other values of $a$.

So to sum up you have

$\begin{align}P(X\leq a | Y=k)= & {} e^{-(k-1)\lambda}-e^{-a\lambda}\,\,,k-1<a<k\\ & {} =1\,\,, a\geq k\\ &{}=0\,\,,a \leq k-1 \end{align}$

To find the conditional expectation you see that you have already evaluated the conditional cdf.

So the conditional pdf is found by just differentiating wrt a.

Or $f_{X|Y}(x)=\lambda e^{-x\lambda}\,\,,k-1<x\leq k$. And $0$ elsewhere.

So you just find the expectation by simply integrating.

$\int_{k-1}^{k}x\lambda e^{-x\lambda}dx$.

Further more you will get an expression in terms of $k$ and $\lambda$. Then if you sum that expression over all $k$.(Make some rearrangements to make it telescope) . You will get $\frac{1}{\lambda}$. Which is what is "expected" as the expectation of conditional expectation is just expectation of X. $E[X]=E[E[X|Y]]$.

Verify this.