5
$\begingroup$

Could you tell me how to find the distribution of $Z = X+Y$ if $X$ is a random variable with uniform distribution on $[0,1]$ and $Y$ has uniform distribution on $\{-1,0,1\}$?

$X$ and $Y$ are independent.

I know how to find distributions of sums of random variables if both are discrete or both are continuous. But here we have a mix.

I guess I should consider $P(Z<-1) = P(\{X<-1, Y<0\} \cup \{X<0, Y<-1\})=0$

$P(Z \in [-1,0)) = P(X \in [-1,0), Y=0 \ \text{or} \ X \in [0,1), Y=-1 \ \text{or} \ X \in [-2, -1), Y=1) = \frac{1}{3}$

$P(Z \in [0,1)) = P(X \in [0,1), Y=0 \ \text{or} \ X \in [1,2), Y=-1 \ \text{or} \ X \in [-1, 0), Y=1) = \frac{2}{3}$

$P(Z \in [1,2)) = P(X \in [1,2), Y=0 \ \text{or} \ X \in [2,3), Y=-1 \ \text{or} \ X \in [0, 1), Y=1) = 1$

And so $P(Z \ge 1) = 1$

Is that correct?

$\endgroup$
  • 2
    $\begingroup$ Are you assuming (or have you been told) that $X$ and $Y$ are independent but you have opted to not share this information with us? $\endgroup$ – Dilip Sarwate Feb 28 '15 at 14:08
  • $\begingroup$ I'm so sorry. Yes, they are independed. I haven't noticed that I forgot to write this. $\endgroup$ – Hagrid Feb 28 '15 at 14:36
3
$\begingroup$

Notice, that events $Z \in [-1,0)$, $Z \in [0,1)$ and $Z \in [1,2)$ have empty intersection. But the according to your solution: $$P(Z \in [-1,0)) + P(Z \in [0,1)) + P(Z \in [1,2)) > 1$$ which clearly could not be correct.

So lets count, the desired probabilities once again, using the fact that X,Y are independent: \begin{align} P(Z \in [-1,0)) &= P(X \in [0,1))P(Y=-1) = 1\times\frac{1}{3}=\frac{1}{3}, \\ P(Z \in [0,1)) &= P(X \in [0,1))P(Y=0) + P(X = 1)P(Y=-1) = 1\times\frac{1}{3} + 0\times\frac{1}{3} = \frac{1}{3}, \\ P(Z \in [1,2)) &= P(X \in [0,1))P(Y=1) + P(X = 1)P(Y=0) = 1\times\frac{1}{3} + 0.\times\frac{1}{3} = \frac{1}{3}. \end{align} Of course $P(Z < - 1)= 0$, but we have already counted $$P(Z \geq 1) = P(Z \in [1,2)) + P(Z \geq 2) = \frac{1}{3} + 0.$$

$\endgroup$
  • $\begingroup$ I see where I made a mistake, in the second summands we only take one value of $X$, because the probability that $X$ has values outside $[0,1]$ is zero. Thank you! $\endgroup$ – Hagrid Feb 28 '15 at 15:10
  • $\begingroup$ I have one more question, though. Is this distribution continuous? $\endgroup$ – Hagrid Feb 28 '15 at 15:52
  • 1
    $\begingroup$ In you count the distribution function : $$P(Z \leq z) = \frac{1}{3}P(X \leq z+1) + \frac{1}{3}P(X \leq z) + \frac{1}{3}P(X \leq z-1)$$ for all $z \in [0,1]$, you should see that Z has uniform distribution on $[-1,2]$. $\endgroup$ – iiivooo Feb 28 '15 at 16:03
4
$\begingroup$

You can extend the convolution method for summing continuous independent variables if you identify the "density" of a discrete variable as a sum of Dirac deltas. Here you find that the density of $X$ is $1_{[0,1]}$ while the "density" of $Y$ is $\frac{1}{3} \sum_{i=-1}^1 \delta_i$. So you now need to compute the convolution:

$$g(x) = \frac{1}{3} \sum_{i=-1}^1 \int_{-\infty}^\infty \delta_i(y) 1_{[0,1]}(x-y) dy \\ = \frac{1}{3} \sum_{i=-1}^1 \int_{x-1}^x \delta_i(y) dy$$

By the definition of the Dirac delta, the $i$th summand is $1$ if $i \in (x-1,x)$ and zero otherwise. So exactly one summand will be $1$ if and only if $x \in (-1,2)$, so the density here is $\frac{1}{3} 1_{(-1,2)}$.

In this case this is pretty much the same as the other answer; this method is nicer in more complicated examples.

$\endgroup$

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.