3
$\begingroup$

So I'm trying to get the Weibull distribution moment generating function

$$\sum_{n=0}^\infty \frac{t^n \lambda^n}{n!} \Gamma(1+n/k)$$

(which can be found here https://en.wikipedia.org/wiki/Weibull_distribution)

I'm trying to do it with the definition of moment generating functions, but I can't get to that result. If $X$ has $\mathrm{Weibull}(\lambda, k)$ distribution, then its PDF is

$$f(x) = \begin{cases} \frac{k}{\lambda}(\frac{x}{\lambda})^{k-1} e^{(-x/\lambda)^k}, & \text{if $x\ge0$} \\ 0, & \text{if $x<0$} \end{cases} $$

And its moment generating function should be

$$M(t)=E(e^{tX})=\int_0^\infty e^{tx}\frac{k}{\lambda} \left(\frac{x}{\lambda}\right)^{k-1} e^{-(x/\lambda)^k} dx$$

but I can't get this form from Wikipedia. I have tried many different variable changes, etc., but without good result.

I appreciate any help leading to that form.

$\endgroup$
0
4
$\begingroup$

$$\begin{align} M(t) &= E(e^{tX}) = \int_0^\infty e^{tx}\frac{k}{\lambda} \left(\frac{x}{\lambda}\right)^{k-1} e^{-(x/\lambda)^k} dx &\\ &= \int_0^\infty e^{\lambda tu} ku^{k-1} e^{-u^k} du &\qquad(\text{with }u=x/\lambda\text{, for }\lambda>0) \\ &= \int_0^\infty e^{\lambda tx^{1/k}} e^{-x} dx &\qquad(\text{with }x=u^k\text{, for }k>0) \\ &= \int_0^\infty \sum_{n=0}^\infty {(\lambda t)^n\over n!} x^{n/k} e^{-x} dx \\ &= \sum_{n=0}^\infty {(\lambda t)^n\over n!} \int_0^\infty x^{n/k} e^{-x} dx \\ &= \sum_{n=0}^\infty {\lambda^n t^n \over n!} \Gamma\left({n\over k}+1\right) \end{align}$$ from definitions with two substitutions (as indicated), the series expansion (for $e^{\lambda tx^{1/k}}$), and integrating term by term (with the definition of the Gamma function).

$\endgroup$
3
$\begingroup$

Let $X \sim \operatorname{Weibull}(k, \lambda)$ such that $$F_X(x) = 1 - e^{-(x/\lambda)^k}, \quad x > 0,$$ as described. Then we observe that $$X = \lambda (Y/\lambda)^{1/k},$$ or equivalently, $$(Y/\lambda) = (X/\lambda)^k,$$ where $Y \sim \operatorname{Exponential}(\lambda)$ with $$F_Y(y) = 1 - e^{-y/\lambda}, \quad y > 0.$$ So $X$ can be obtained through a suitable transformation of an exponential distribution. Now recall that the MGF of $Y$ is $$M_Y(y) = \operatorname{E}[e^{tY}] = \int_{y=0}^\infty \frac{1}{\lambda} e^{ty} e^{-y/\lambda} \, dy = \frac{1}{1 - \lambda t} \int_{y=0}^\infty (\lambda^{-1} - t)e^{-y(\lambda^{-1} - t)} \, dy = \frac{1}{1-\lambda t}.$$ Thus $$\operatorname{E}[Y^m] = M_Y^{(m)}(0) = \left[m! \lambda^m (1-\lambda t)^{-(m+1)} \right]_{t=0} = m! \lambda^m,$$ which generalizes to non-integer moments as $$\operatorname{E}[Y^m] = \Gamma(m+1) \lambda^m, \quad m > -1.$$ Now we see that $$\operatorname{E}[X^m] = \operatorname{E}[(\lambda (Y/\lambda)^{1/k})^m] = (\lambda^{m(1 - 1/k)}) \operatorname{E}[Y^{m/k}] = \lambda^{m(1-1/k)} \Gamma(1 + m/k) \lambda^{m/k} = \lambda^m \Gamma(1 + m/k).$$ Since $$M_X(t) = \sum_{n=0}^\infty \frac{M_X^{(n)}(0)}{n!} t^n = \sum_{n=0}^\infty \frac{\operatorname{E}[X^n]}{n!} t^n,$$ it follows $$M_X(t) = \sum_{n=0}^\infty \frac{\lambda^n \Gamma(1 + n/k)}{n!} t^n,$$ which concludes the proof.

$\endgroup$
4
  • $\begingroup$ So much thanks to you! Btw where does that equality come after "Since" $\endgroup$ – DudeWithProblem97 Mar 24 '16 at 19:06
  • $\begingroup$ @DudeWithProblem97 The first equality in that equation is the (Taylor) series expansion about $t = 0$ of the function $M_X(t)$. The second equality arises from the fact that $$\operatorname{E}[e^{tX}] = \operatorname{E}\left[\sum_{n=0}^\infty \frac{(tX)^n}{n!} \right] = \sum_{n=0}^\infty \frac{\operatorname{E}[X^n]}{n!} t^n.$$ $\endgroup$ – heropup Mar 24 '16 at 19:50
  • $\begingroup$ Thanks, you made my day. :) $\endgroup$ – DudeWithProblem97 Mar 24 '16 at 20:01
  • $\begingroup$ @DudeWithProblem97 Please accept the answer and/or upvote it if it has addressed your question to your satisfaction. $\endgroup$ – heropup Mar 24 '16 at 20:08

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.