MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Is there any "easy" way to calculate fractional moments from Laplace transform. To be more specyfic let us consider the following example. Let $X$ be a positive random variable and $L(\theta) := E \exp (-\theta X)$ be its Laplace transform. Of course it is easy to calculate $E X^n$ where $n$ is a natural number but what with e.g. $E X^{1/2}$.

share|cite|improve this question

For your example of X^{1/2} you can evaluate the fractional half derivative of the Laplace transform (see for example the Wikipedia article on fractional calculus) at theta = 0.

share|cite|improve this answer
I thought about this also. I have even tried to use it but it does not very handy to use. I was hoping for something more "natural". Anyway, I have another question. Let us now drop the assumption that X is positive but assume that the Laplace transform exists. How can one calculate $E|X|$? – Piotr Miłoś Nov 14 '09 at 18:07
If you write the Laplace transform of the probability measure of X as a sum of two pieces, the first corresponds to the integral from -inf to 0 and the second from 0 to +inf, i.e., F(s) = F+(s) + F-(s), then the Laplace transform of the probability measure of |X| will be F+(s)+F-(-s), from which you can derive the moments of |X|. – David Bar Moshe Nov 16 '09 at 6:18
Yes, but the problem is that I have the Laplace transform of "whole" X that is F(s). How can I calculate F+ and F- using F only? May be I missing something elementary, but I can not see how to do it in an easy way (of course in principle one can apply the inverse Laplace transform and so on but this is not easy usualy ). – Piotr Miłoś Nov 17 '09 at 14:31

If $X$ is positive the following works. Let $F(\theta)=E(e^{-\theta X})$ be the Laplace transform. Given $s\in \mathbb{R}$ write $s=n-\alpha$ with $n$ an integer and $\alpha >0$. Then $$E(X^{s}) = (-1)^n\frac{1}{\Gamma(\alpha)} \int_0^\infty F^{(n)} (\theta) \theta^{\alpha} d \theta$$ with $\Gamma$ the usual Gamma function, $$\Gamma(\alpha) = \int_0^\infty \theta^{\alpha -1} e^{-\theta} d \theta.$$

Indeed by Fubini, $$\int_0^\infty F^{(n)}(\theta) \theta^{\alpha-1} d \theta = (-1)^n E (\int_0^\infty X^n e^{-\theta X} \theta^{\alpha-1} d \theta )= (-1)^n E(X^{n-\alpha}) \int_0^\infty \theta^{\alpha-1} e^{-\theta} d \theta,$$ and so long as $\alpha >0$ the integral on the right is convergent.

share|cite|improve this answer
How do I choose the parameter 'Alpha' ? In the question 's'=1/2. Is 'n'=1, 'Alpha'=1/2 the only choice? Or can I use as well 'n'=2, 'Alpha'=3/2 with no difference ? – user1611107 Jan 8 '14 at 10:16
Yes. You can use $alpha =3/2$ and $n=2$. This amounts to integrating by parts once in the integral over $\theta$. – Jeff Schenker Jan 8 '14 at 18:22

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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