Given $P$ independent and identically distributed random variables, $X_1, X_2, ..., X_P \sim \text{Gamma}(M,2c)$ how can we prove that:

$$U = X_1 + X_2 + ... + X_P$$


$$V = \frac{X_1}{X_1 + X_2 + ... + X_P}$$

are independent, when $U \sim \text{Gamma}(MP,2c)$ and $V \sim \text{Beta}(M,M(P-1))$.


1 Answer 1


In Devroye's Non-uniform random variate generation, Chapter 5 starts with two theorems on Uniform spacings (pp.207-208):

Theorem 2.1 - Given an iid Uniform sample $U_1,\ldots,U_n$, the uniform spacing statistics$$S_1=U_{(1)}\,,\quad S_i=U_{(i)}-U_{(i-1)}\,,\quad i=2,\ldots n\,,\quad S_{n+1}=1-U_{(n)}$$associated with the order statistics are such that $(S_1,\ldots,S_n)$ is uniformly distributed over the simplex$$A_n=\left\{(x_1,\ldots,x_n);x_i\ge 0\,,\,\sum_{i=1}^n x_i\le 1\right\}$$


Theorem 2.2 - $(S_1,\ldots,S_{n+1})$ is distributed as $$\dfrac{E_1}{\sum_{i=1}^{n+1} E_i},\ldots,\dfrac{E_{n+1}}{\sum_{i=1}^{n+1} E_i}$$where $E_1,\ldots,E_{n+1}$ is an iid sequence of exponential random variables. Furthermore, if $G_{n+1}$ is independent of $(S_1,\ldots,S_{n+1})$ and is Gamma$(n+1,1)$ distributed, then $$S_1G_{n+1},\ldots,S_{n+1}G_{n+1}$$is distributed as $E_1,\ldots,E_{n+1}$


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.