Given two i.i.d uniform random variables $X$ and $Y$, find the joint distribution of $U = X + Y$ and $V = X/(X+Y)$.

135 Views Asked by At

The question asks you to solve this using the jacobian method. From the question, we get that: $X = VU$ and $Y = U(V-1)$ and so the jacobian yields: $|J| = |U|$.

So we get: $f_{UV}(u,v) =|J| \cdot f_{XY}(uv,u(v-1))$. Since X and Y are i.i.d, $f_{XY}(x,y) = 1/4$, if x $\in (0,2)$ and y $\in (0,2)$.

So the final result is: $f_{UV}(u,v) =|u|/4$ if $0<uv<2$ and $0<u(v-1)<2$.

But both regions $0<uv<2$ and $0<u(v-1)<2$ described by this joint distribution have infinite area, so is the final joint distribution valid? Or maybe the jacobian method cannot be applied in this situation?

1

There are 1 best solutions below

2
On BEST ANSWER

First of all there is a mistake. $Y = U(1-V)$ and not $U(V-1)$.

Now, $f_{UV}(u,v) = u \cdot f_{XY}(uv,u(1-v))$. I removed the absolute value as $u$ cannot be negative. You will see why as you go through the answer.

As you said, $0 \lt uv \lt 2$ and $0 \lt u (1-v) \lt 2$

From $0 \lt uv \lt 2$, observe that $u$ and $v$ must have the same sign.

From $0 \lt u(1-v) \lt 2$, $v$ cannot be greater thant $1$ as $u$ and $v$ canno]t have different signs.

Also $u$ and $v$ cannot be negative as for $u, v \lt 0$, $u(1-v) \lt 0$

Now for $0 \lt u \lt 2$ and $0 \lt v \lt 1$, there are no restrictions and any values of $u, v$ work.

Also note intersection of $uv = u (1-v) \implies v = \frac{1}{2}, u = 4$

So for $2 \lt u \lt 4$, we have

$\frac{u-2}{u} \lt v \lt \frac{2}{u}$ and $v$ is always less than $1$

For $u \gt 4$, we get contradiction in value of $v$ from both equations.

So support of the distribution is given by,

$$0 \lt v \lt 1, 0 \lt u \lt 2$$ $$\frac{u-2}{u} \lt v \lt \frac{2}{u}, 2 \lt u \lt 4$$

enter image description here