For some set of $n$ i.i.d. variables $\{X \}$ which are Rayleigh-distributed such that

$$ P(X|\sigma) = \frac{X}{\sigma^2}\exp{\left[-\frac{X^2}{2\sigma^2}\right]} $$

I'm interested in anything we can write down analytically about

$$ Y = \ln{\left(\sum_{i=1}^{n} X_{i}^{2}\right)}. $$ Primarily I'm trying to get expressions for $\mathrm{E}[Y]$ and $\mathrm{Var}[Y]$ (clearly they can be calculated numerically via sampling but analytic results would be better). Obviously a closed form for $P(Y|n,\sigma)$ would be great, but I have no idea if one exists.

Im guessing for the mean we could use LOTUS such that $$ \mathrm{E}[Y] = \int_{0}^{\infty}\dots \int_{0}^{\infty} \ln{\left(\sum_{i=1}^{n} X_{i}^{2}\right)} \prod_{i=1}^{n} P(X_i|\sigma) \; \mathrm{d}X_1 \dots \mathrm{d}X_n $$ but I have no idea how to evaluate that, or if there is a simpler way.

  • 5
    $\begingroup$ Since (by inspection) $X^2$ has a Gamma distribution, $Y$ has an exp-gamma distribution (sometimes misleadingly called a "log-gamma distribution"). $\endgroup$
    – whuber
    Feb 23, 2018 at 23:42

1 Answer 1


Whuber points out that if $X \sim \mathrm{Rayleigh}(\sigma)$ then $X^2 \sim \mathrm{Gamma}(1, 2\sigma^2)$, and because Gamma-distributed variables have the property that $$ \sum_{i=1}^{n} g_i \sim \mathrm{Gamma}\left(\sum_{i=1}^{n} k_i, \theta\right) $$ for $g_i \sim \mathrm{Gamma}\left(k_i, \theta\right)$ it is the case that $$ \sum_{i=1}^{n} X^{2}_i \sim \mathrm{Gamma}\left(n, 2\sigma^2\right). $$ We may therefore write $$ P(Y|n,\sigma) = \frac{1}{2^n \sigma^{2n} \Gamma(n)} \exp{\left[n Y - \frac{1}{2 \sigma^{2}}e^{Y}\right]}. $$ Consequently the mean and variance are $$ \mathrm{E}[Y] = \ln{(2\sigma^2)} + \psi_0 (n), $$ $$ \mathrm{Var}[Y] = \psi_1 (n), $$ where $\psi_k$ is the polygamma function.


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.