2
$\begingroup$

I have the following homework assignment: the life expectancy $X$ of a lamp has exponential distribution with rate $\lambda$. The rate depends on the production proccess, such that its population can be described by the uniform distribution $\lambda \sim U[0,1]$. (I'm sorry if the translation is bad, I hope it is understandable).

  1. Describe the joint density of the lamp life $X$ and rate $\lambda$.

This one is just $f_{X \mid\lambda}(x\mid\lambda)f_{\lambda}(\lambda) = \lambda e^{-\lambda x}$ for $x \geq 0$ and $\lambda \in [0,1]$ or $0$ elsewhere.

  1. Find $X$ distribution.

First I tried to find the marginal density of $X$ by integrating the joint density over $\lambda$: $$ f_{X}(x) = \int_{0}^{1} \lambda e^{-\lambda x} \,d\lambda = -\frac{e^{-x}}{x} - \frac{e^{-x}}{x^{2}} + \frac{1}{x^{2}} \quad \text{if $x \geq 0$ and $0$ otherwise}$$ I am already in trouble, somewhere along the way I lost the ability to compute the marginal density for $X = 0$, which was fine for the joint density, I don't know what happened (the integral is correct) nor do I know why was this information lost. It gets even uglier when I try to find the distribution of $X$: $$F(X \leq x) = \int_{0}^{x} -\frac{e^{-x}}{x} - \frac{e^{-x}}{x^{2}} + \frac{1}{x^{2}} \,dx = (-\frac{1}{x} + \frac{e^{-x}}{x})\Big|_{0}^{x} $$ Once again the integral is correct but now I can't even evaluate it because of the lower limit being $0$. So my question is what am I doing wrong? And why did I lose the information on $X = 0$? I know I can just calculate $F(X \leq x \mid \lambda)$ and get some nice results after integrating over $\lambda$ but I really don't know why this straight forward way doesn't really work here...

Thank you for your attention.

$\endgroup$
1
$\begingroup$

The marginal density \begin{align}f(x)&=\int_0^1 \lambda e^{-\lambda x} \text{d}\lambda\\ &= \int_0^1 -\frac{\partial}{\partial x} e^{-\lambda x} \text{d}\lambda\\ &= -\frac{\partial}{\partial x}\int_0^1 e^{-\lambda x} \text{d}\lambda\\ &= -\frac{\partial}{\partial x} \frac{1-e^{-x}}{x}\tag{1}\\ &= \frac{1-e^{-x}}{x^2} -\frac{e^{-x}}{x}\end{align} is well-defined over $(0,\infty)$. The fact that it goes to infinity at $x=0$ does not matter since a density is uniquely defined almost everywhere. Setting $f(0)=0$ or $f(0)=12346$ is equally valid (and irrelevant). The important fact is that $f$ is integrable over $(0,\infty)$, with mass equal to one.

The cdf is found in (1): since $$\lim_{x\to 0} \frac{1-e^{-x}}{x} = \lim_{x\to 0} \frac{\frac{\partial}{\partial x} 1-e^{-x}}{\frac{\partial}{\partial x} x} = \lim_{x\to 0} \frac{e^{-x}}{1} = 1 $$ by L'Hospital's rule, $$F(x)= 1-\frac{1-e^{-x}}{x}$$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ So whenever I run into domain problems within the density function, I can just ignore that value or set it to something convenient? Is there any case where I might run into a problem? Also thank you for taking that limit, I should've thought of it. $\endgroup$ – Rodrigo Meireles Sep 23 at 13:13
  • $\begingroup$ If the joint density is defined almost everywhere, so is the marginal. Hence you should not run into problems. $\endgroup$ – Xi'an Sep 23 at 14:35

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.