Let r-th moment $E(X^r) = r!\{(-1)^r+(1/2)^r\}/2\;\;$ where $r \in \Bbb N\cup\{0\} $

Then moment generating function $M(t)$ is

$M(t) =\sum_{r=0}^\infty \dfrac{E(X^r)}{r!}t^r= \dfrac{1}{2}\left[\sum_{r=0}^\infty(-1)^rt^r+\sum_{r=0}^\infty\ ({1\over 2})^rt^r\right] = \dfrac{1}{2}\left[\dfrac{1}{1+t} + \dfrac{1}{1-{1\over 2}t}\right] = \int_0^\infty e^{tx}\cdot{1\over 2}e^x+ [?]$

I can't figure out which pdf would the last term will be.

I need to evaluate the proper pdf corresponds to above $E(X^r)$

Any hint?

  • $\begingroup$ Hint: the mgf of a Gamma distribution looks like one of your terms. $\endgroup$
    – whuber
    Jul 2, 2017 at 13:35
  • $\begingroup$ @whuber any other approach not using the cocept of Gamma distribution? I am at little bit before the specific distributions $\endgroup$
    – Beverlie
    Jul 2, 2017 at 13:36
  • $\begingroup$ Since the answer necessarily involves Gamma distributions, it will be hard to avoid using them! $\endgroup$
    – whuber
    Jul 2, 2017 at 13:37
  • $\begingroup$ thx will check it $\endgroup$
    – Beverlie
    Jul 2, 2017 at 13:37
  • 1
    $\begingroup$ or the special case of an exponential distribution if you're more familiar with that $\endgroup$
    – Taylor
    Jul 2, 2017 at 19:59

1 Answer 1


Consider independent but not necessarily identically distributed random variables $X_1, X_2, \ldots, X_n$, with densities $$f_{X_1}(x), f_{X_2}(x), \ldots, f_{X_n}(x),$$ and weights $a_1, a_2, \ldots, a_n \in [0,1]$ such that $\sum_{i=1}^n a_i = 1$. Then the mixture distribution with density $$f_Y(x) = \sum_{i=1}^n a_i f_{X_i}(x)$$ is clearly a density. The MGF of $Y$ is expressible as the weighted sum of the MGFs of the $X_i$s; i.e., $$M_Y(t) = \int_{y\in\Omega} e^{ty} f_Y(y) \, dy = \sum_{i=1}^n a_i \int_{y\in\Omega} e^{ty} f_{X_i}(y) \, dy = \sum_{i=1}^n a_i \operatorname{E}[e^{tX_i}] = \sum_{i=1}^n a_i M_{X_i}(t)$$ for the continuous case.

Since you know that $M_{X_1}(t) = (1+t)^{-1}$ corresponds to the MGF of the density $$f_{X_1}(x) = \begin{cases} e^x, & x < 0 \\ 0, & x \ge 0, \end{cases}$$ (that is to say, $-X_1 \sim \operatorname{Exponential}(\lambda = 1)$), and $M_{X_2}(t) = (1-t/2)^{-1}$ corresponds to the MGF of the density $$f_{X_2}(x) = \begin{cases} 2e^{-2x}, & x > 0 \\ 0, & x \le 0, \end{cases}$$ and the weights are simply $a_1 = a_2 = 1/2$, we find that the random variable characterized by the given MGF has the density $$f_X(x) = \begin{cases}e^{-2x}, & x > 0 \\ e^x/2, & x < 0. \end{cases}$$ The discontinuity at $x = 0$ simply means the CDF is not smooth at this point, but the CDF is unique and well-defined everywhere.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.