Let $X\sim \text{Exp}(\lambda)$ be an exponentially distributed random variable. That is it has the probability density function $f(x)=\lambda e^{-\lambda x}1_{[0,\infty)}(x)$ and cumulative distribution function $$F_X(x)=\int_ {0}^x\lambda e^{-\lambda x}=[-e^{-\lambda x}]_{0}^x=1-e^{-\lambda x}$$

Let $Y:= \frac{1}{X}$. We have for $F_Y(x)$: $$F_Y(x)=\mathbb{P}({Y \leq x})=\mathbb{P}\left(\frac{1}{X}\leq x\right )=\mathbb{P}\left(X\geq\frac{1}{x}\right )=1-\mathbb{P}\left(X < \frac{1}{x} \right)=^!1-F_X\left(\frac{1}{x}\right)$$

The problem at $!$ is that definition of cumulative density function requires a non strict inequality so I don't know why this holds. Aside from that, how does one get the distribution of $Y$ from the cumulative density function/probability density function? How do we deal with $Y$ at $X=0$ (does $X$ attain $0$ as a value even)

  • 5
    $\begingroup$ For continuous r.v. $\mathbb P\{X<x\}=\mathbb P\{X\leq x\}$. $\endgroup$ – Surb Apr 30 '18 at 15:40
  • $\begingroup$ Does the support being positive mean the image of $X$ is positive? $\endgroup$ – why Apr 30 '18 at 15:48
  • 1
    $\begingroup$ The central issue that answers both of your questions is that for a continuous random variable $X$ and any fixed number $a$, $\mathbb P(X = a) = 0$. This is why $\mathbb P(X \leq c) = \mathbb P(X < c)$ -- after all, $\{X \leq c\} = \{X < c\} \cup \{X = c\}$ and the union is disjoint -- and why you don't have to worry about whether or not $X$ might be $0$. $\endgroup$ – Aaron Montgomery Apr 30 '18 at 15:50

Observe that $$\left(-\infty, \dfrac{1}{x}\right) \cup \left\{\dfrac{1}{x} \right\}=\left(-\infty, \dfrac{1}{x} \right]$$ and that the sets $\left(-\infty, \dfrac{1}{x}\right)$, $\left\{\dfrac{1}{x} \right\}$ are disjoint.

Therefore, on the left-hand side, $$\mathbb{P}\left[\left(-\infty, \dfrac{1}{x}\right) \cup \left\{\dfrac{1}{x} \right\}\right]=\mathbb{P}\left[\left(-\infty, \dfrac{1}{x}\right) \right] + \mathbb{P}\left(\left\{\dfrac{1}{x} \right\}\right) = \mathbb{P}\left(X < \dfrac{1}{x}\right)+0$$ because $$\mathbb{P}\left(\left\{\dfrac{1}{x} \right\}\right) = 0$$ due to $X$ being a continuous random variable, and on the right-hand side, we simply have $$\mathbb{P}\left[\left(-\infty, \dfrac{1}{x} \right] \right] = \mathbb{P}\left(X \leq \dfrac{1}{x}\right)$$ Hence, equating the two sides, $$\mathbb{P}\left(X \leq \dfrac{1}{x}\right) = \mathbb{P}\left(X < \dfrac{1}{x}\right)$$

| cite | improve this answer | |
  • $\begingroup$ Thank you for addressing the continuity issue. What can we say about the distribution of $Y$? $\endgroup$ – why Apr 30 '18 at 15:52
  • $\begingroup$ @whycantihavealongername Well, from your work, you have $F_Y(x) = 1 - F_X(1/x) = 1-(1-e^{-\lambda/x}) = e^{-\lambda / x}$. You can use this to find the PDF by taking the derivative if you need it. I don't know if there's a standard name for this distribution. $\endgroup$ – Clarinetist Apr 30 '18 at 15:55
  • $\begingroup$ @whycantihavealongername I suppose you could call it an "inverse exponential distribution:" en.wikipedia.org/wiki/… $\endgroup$ – Clarinetist Apr 30 '18 at 15:56

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.