Let $X,Y$ be two independent random variables with exponential distribution with parameter $\lambda > 0$. Let $S = X + Y$ and $T = \frac{X}{S}$. I want to find the joint density function of $(S,T)$ . I want then to calculate the marginals and say whether or not $S$ and $T$ are independent. I start by finding the density function of $S$ using the convolution:

$$ f_S(s) = \int_{-\infty}^{+\infty}f_X(s-t)f_Y(t)dt $$ $$ = \int_{0}^{s} \lambda^2 e^{-\lambda s} dt = \lambda^2 e^{-\lambda s}$$

Then I tried to calculate the density function of $T $ but I am stuck here:

$$ F_T(t) = \mathbb{P}(T \leq t) = \mathbb{P}\left(\frac{X}{X+Y} \leq t\right).$$ Is this the right method of solving this? Should I find the joint density first? (The problem is that I do not know how to to that)

  • 1
    $\begingroup$ Hint: $\frac{X}{X+Y} \leq t \Leftrightarrow X \leq tX+tY \Leftrightarrow (1-t)X \leq tY$ (note that we can assume $t \in [0,1]$) $\endgroup$ – Stockfish Nov 5 '18 at 9:46
  • $\begingroup$ I see. So it is right to find the two density function first? Why can we assume t in that interval? $\endgroup$ – qcc101 Nov 5 '18 at 9:47
  • $\begingroup$ Which values can be attained by T? $\endgroup$ – Stockfish Nov 5 '18 at 9:48
  • $\begingroup$ @qcc101 because $0\leq X\leq X+Y$ a.s. $\endgroup$ – drhab Nov 5 '18 at 9:49
  • 1
    $\begingroup$ There is a mistake/typo in your calculation of $f_S(s)$. It must end with $\cdots=\lambda^2se^{-\lambda s}$. Btw, IMHO it is handsome to concentrate on $\lambda=1$ at first hand. Afterwards the solution can be translated easily to the general case. $\endgroup$ – drhab Nov 5 '18 at 9:58

Assuming you mean $\lambda$ is the rate parameter here (i.e. Exponential with mean $1/\lambda$).

First of all, recheck your density of $S$. The correct density as mentioned in comments is $$f_S(s)=\lambda^2se^{-\lambda s}\mathbf1_{s>0}$$

It is easy to verify that the density of $T$ is $$f_T(t)=\mathbf1_{0<t<1}$$

You can find the joint distribution function of $(S,T)$ as follows:

For $s>0$ and $0<t<1$,

\begin{align} P(S\le s,T\le t)&=P\left(X+Y\le s,\frac{X}{X+Y}\le t\right) \\&=\iint_D f_{X,Y}(x,y)\,dx\,dy\quad\quad,\text{ where }D=\{(x,y):x+y\le s,x/(x+y)\le t\} \\&=\lambda^2\iint_D e^{-\lambda(x+y)}\mathbf1_{x,y>0}\,dx\,dy \end{align}

Change variables $(x,y)\to(u,v)$ such that $$u=x+y\quad,\quad v=\frac{x}{x+y}$$

This implies $$x=uv\quad,\quad y=u(1-v)$$

Clearly, $$x,y>0\implies u>0\,,\,0<v<1$$

And $$dx\,dy=u\,du\,dv$$

So again for $s>0\,,\,0<t<1$,

\begin{align} P(S\le s,T\le t)&=\lambda^2\iint_R ue^{-\lambda u}\mathbf1_{u>0,0<v<1}\,du\,dv\qquad,\text{ where }R=\{(u,v):u\le s,v\le t\} \\&=\left(\int_0^s \lambda^2 ue^{-\lambda u}\,du\right) \left(\int_0^t \,dv\right) \\\\&=P(S\le s)\,P(T\le t) \end{align}

This proves the independence of $S$ and $T$, with $S$ a Gamma variable and $T$ a $U(0,1)$ variable.

And from the joint distribution function, it is readily seen (without differentiating) that the joint density of $(S,T)$ is $$f_{S,T}(s,t)=\lambda^2 se^{-\lambda s}\mathbf1_{s>0,0<t<1}$$

Note that the change of variables isn't really necessary if you are comfortable with the first form of the double integral.


The joined density is defined as

$$f_{S,T}(s,t) \\= \int _0^{\infty }\int _0^{\infty }\lambda ^2 e^{-\lambda (x+y)} \delta (s-x-y) \delta \left(t-\frac{x}{x+y}\right)dydx\tag{1a}$$

Here $\delta(x)$ is the Dirac delta function.

Changing variables $x\to u v, y\to u(1-v)$ gives $u\in (0,\infty), v\in (0,1)$, the modulus of the Jacobian $u$, so that the integral becomes

$$f_{S,T}(s,t) \\=\int _0^{\infty }\int _0^1\lambda ^2 u e^{-\lambda u} \delta (s-u) \delta (t-v)dvdu\\= \lambda ^2 s e^{-s \lambda } \left\{ \begin{array} {ll} 1 & 0\lt t \lt 1 \\ 0 & \, \textrm{otherwise} \end{array} \right.\tag{1b}$$

On the other hand we have for the separate density functions

$$f_{S}(s)\\= \int _0^{\infty }\int _0^{\infty }\lambda ^2 e^{-\lambda (x+y)} \delta (s-x-y)dydx = \lambda ^2 s \; e^{-\lambda s}\tag{2a}$$

$$f_{T}(t) \\= \int _0^{\infty }\int _0^{\infty }\lambda ^2 e^{-\lambda (x+y)} \delta \left(t-\frac{x}{x+y}\right)dydx \\ = \left\{ \begin{array} {ll} 1 & 0\lt t \lt 1 \\ 0 & \, \textrm{otherwise} \end{array} \right.\tag{2b}$$

And, finally, from $(1)$ and $(2)$ we have

$$f_{S,T}(s,t) = f_{S}(s) f_{T}(t)\tag{3}$$

which shows the independence of $s$ and $t$.


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.