Suppose X and Y, independent, are distributed as Uniform[0,1].

(Disclaimer: this relates to a homework problem, but is not itself a homework problem. The problem itself asks to find the conditional density of X given XY=t, i.e. $f_{X|XY=t}(x)$).

Now, I am proceeding by using the definition: $$f_{X|XY=t}(x)=\frac{f_{X,XY}(x,t)}{f_{XY}(t)}$$

I have already calculated the density $f_{XY}(t)$, but have gotten a little lost on the numerator. I am trying to calculate the numerator by first calculating the $F_{X,XY}(x,t)$, and then differentiating with respect to x and t. Now, I am wondering if, using the independence of X and Y, I can split this joint density up as: $$\begin{align} F_{X,XY}(x,t)=P(X \leq x, XY \leq t) \\ = P(X \leq x, Y \leq \frac{t}{X}) \\ = \int_{t}^{1}P(Y \leq \frac{t}{X}|X=s)P(X=s) ds \\ =\int_{t}^{1}P(Y \leq \frac{t}{s})P(X=s) ds \\ =\int_{t}^{1}\frac{t}{s}ds=-t log(t) \end{align}$$

But differentiating this with respect to x and t yields 0, which I know is again wrong.

The areas of concern that I have noted while working through the problem are as follows:

  1. Confusion over the use of the conditional division formula when using CDFs as opposed to pdfs (can we just substitute the pdfs for CDFs?)
  2. Confusion if the conditioning on X=s I use to split up the densities is valid

If someone can point out the error, I would be much obliged.


1 Answer 1

  1. Using $\mathbb{P}(X = x)$ to denote the PDF of $X$ is a very bad convention. It overrides the meaning of $\mathbb{P}(X = x)$ as probability of the event $\{X = x\}$, and there is no clear advantage for doing so.

  2. I am not sure how you came up with integral limits $t$ and $1$.

A systematic approach is as follows. First, using the fact that $X$ has continuous distribution with the PDF $f_X$, we can write

\begin{align*} \mathbb{P}(X \leq x, XY \leq t) &= \int_{-\infty}^{\infty} \mathbb{P}(X \leq x, XY \leq t \,|\, X = s) f_X(s) \, \mathrm{d}s, \\ &= \int_{0}^{1} \mathbb{P}(X \leq x, XY \leq t \,|\, X = s) \, \mathrm{d}s \end{align*}

where the second step follows from the $X \sim \operatorname{Uniform}([0, 1])$.

Next, let us simply the conditional probability. Before doing so, notice that the range of $(X, XY)$ lies in the triangle $\mathcal{T} = \{ (x, t) : 0 \leq t \leq x \leq 1\}$. So, for the purpose of computing the joint PDF of $(X, XY)$, it suffices to assume that $0 < t < x < 1$. Then for $0 < s < 1$,

\begin{align*} \mathbb{P}(X \leq x, XY \leq t \,|\, X = s) &= \mathbb{P}(s \leq x, sY \leq t \,|\, X = s) \\ &= \mathbb{P}(s \leq x, sY \leq t) \\ &= \mathbb{P}(s \leq x, Y \leq t/s) \\ &= \begin{cases} 0, & \text{if } s > x; \\ t/s, & \text{if } s \leq x \text{ and }t \leq s; \\ 1, & \text{if } s \leq x \text{ and }t > s. \\ \end{cases} \end{align*}

Plugging this back, we are led to integrate a piecewisely-defined function, thus we get

\begin{align*} \mathbb{P}(X \leq x, XY \leq t) = \int_{t}^{x} \frac{t}{s} \, \mathrm{d}s + \int_{0}^{t} \mathrm{d}s = t \log (x / t) + t. \end{align*}

Differentiating w.r.t. $x$ and $t$, we get

\begin{align*} f_{X,XY}(x, t) = \frac{\partial^2}{\partial t \partial x}\mathbb{P}(X \leq x, XY \leq t) = \frac{\partial^2}{\partial t \partial x}\left( t \log (x / t) + t \right) = \frac{1}{x}. \end{align*}

We recall that this result is derived on the assumption that $0 < t < x < 1$. In a full-form, $f_{X,XY}$ can be written as

$$ f_{X,XY}(x, y) = \begin{cases} \frac{1}{x}, & \text{if } 0 < t < x < 1; \\ 0, & \text{otherwise}. \end{cases} $$


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.