1
$\begingroup$

The probability density function of a random variable $X$ is a uniform distribution $U(a,b)$. Likewise, the probability density function of a random variable $Y$ is a uniform distribution $U(0, 2\pi)$. $X$ and $Y$ are independent of each other.

What is the probability density function of $Z = X \: \sin{Y}$ ?

$\endgroup$

3 Answers 3

5
$\begingroup$

Recall that a general procedure to compute the distribution of $Z=X\sin Y$ when $(X,Y)$ are independent with densities $f_X$ and $f_Y$ is to compute $E[u(Z)]$ for every bounded measurable function $u$. If it happens that $$ E[u(Z)]=\int_\mathbb Ru(z)g(z)\mathrm dz, $$ then $g$ is the density of $Z$. Here, $$ E[u(Z)]=\iint u(x\sin y)f_X(x)f_Y(y)\mathrm dx\mathrm dy, $$ hence the job is to transform this double integral into the simple integral above. The way to do that is rather clear: change of variables.

Consider $(z,t)=(x\sin y,x\cos y)$ and assume for example that $X$ is uniform on $(0,1)$ and $Y$ uniform on $(0,2\pi)$, then $\mathrm dz\mathrm dt=x\mathrm dx\mathrm dy$ and $x^2=z^2+t^2$ hence $$ E[u(Z)]=\iint_{z^2+t^2\leqslant1} u(z)\frac{\mathrm dz\mathrm dt}{2\pi\sqrt{t^2+z^2}}, $$ and one sees that $$ g(z)=\mathbf 1_{|z|\leqslant1}\int_{-\sqrt{1-z^2}}^{\sqrt{1-z^2}}\frac{\mathrm dt}{2\pi\sqrt{t^2+z^2}}=\frac1\pi\mathbf 1_{|z|\leqslant1}\int_0^{\sqrt{1-z^2}/z}\frac{\mathrm dt}{\sqrt{t^2+1}}, $$ that is, $$ g(z)=\frac1\pi\cosh^{-1}\left(\frac1z\right)\mathbf 1_{|z|\leqslant1}. $$

$\endgroup$
7
  • $\begingroup$ Thank you for your great answer, but I prefer the one of @wolfies because it is more complete. $\endgroup$ Nov 29, 2013 at 13:36
  • 1
    $\begingroup$ @Medicalphysicist No problem (and thanks for the accompanying comment). Are you able to derive the formulas stated in the accepted post? $\endgroup$
    – Did
    Nov 29, 2013 at 13:38
  • $\begingroup$ You are right, this is a big plus of your answer. Maybe I'm more prone to accept the @wolfies answer because as a physicist I've difficulties to understand your answer. It's a pity that I can't choose both. $\endgroup$ Nov 29, 2013 at 13:41
  • $\begingroup$ @Medicalphysicist Difficulties to understand an answer are a parameter you had not mentioned yet. Sorry to hear that. You might want to expand on said "difficulties" (that is, if the learning process of overcoming them interests you--by contrast with, say, simply getting an answer). $\endgroup$
    – Did
    Nov 29, 2013 at 13:46
  • $\begingroup$ @wolfies: You might want to remove your comment since it does not apply at all. To Did: +1 for the "let us calculate ${\rm E}[u(Z)]$ for bounded measurable $u$" approach. $\endgroup$ Nov 29, 2013 at 15:01
0
$\begingroup$

Case 1: $(0 < a < b)$ To proceed, let $X$ have pdf $f(x)$:

where I am initially assuming parameter $a$ is positive. I will consider the alternative cases later. Second, let $W = sin(Y)$ with pdf $\phi(w)$:

The latter is relatively straightforwards to derive.

Next, we seek the pdf of the product of two random variables, namely $Z = X * W$. This can be cumbersome to do by hand, but can be derived easily using the TransformProduct function in the mathStatica suite. In particular, the pdf of $Z$, say $g(z)$ is:

All done. (I should add I am one of the authors of the software.)

Here is a plot of the theoretical pdf derived $g(z)$ (in red) when $a = 2$ and $b = 5$, and superimposed on top a Monte Carlo check (in blue).


The other 2 cases (with parameter $a< 0$) are derived in identical fashion ... just change the assumptions on parameters $a$ and $b$ in the first input. Doing so yields:

Case 2: $(a < 0, b > 0)$

The solution pdf is $g(z)$:

Here is a plot of the solution pdf $g(z)$ when $a = -3, b = 2$:

Case 3: $(a < b < 0)$

The solution pdf is $g(z)$:

Here is a plot of the solution pdf $g(z)$ when $a = -4, b = -2$:

$\endgroup$
0
$\begingroup$

Remember that c. d. f. of $Z$ is $$P\{Z<z\}=P\{X\sin Y<z\}=\iint\limits_{x\sin y<z} p_{XY}(x,y)dxdy=\int\limits_{-\infty}^{+\infty}dx\int\limits_{-\infty}^{z/x}p_{XY}(x,y)dy=\\=\int\limits_a^{b}dx\int\limits_{0}^{z/x}p_{XY}(x,y)dy$$

The probability density function will then be $$\frac{d}{dz}\left(\int\limits_a^{b}dx\int\limits_{0}^{z/x}p_{XY}(x,y)dy\right)$$

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .