Suppose $z$ is a random variable taking values in $\mathbb{C}^n$ and admitting the complex Gaussian density $p(z;W) \propto \exp{(-\frac{1}{2}z^*Wz)}$, where $W$ is Hermitian. Let $r$ be the vector of moduli and $\theta$ be the vector of phases; e.g., $r_j = |z_j|$ and $\theta_j =$ arg$(z_j)$.

I would like to compute or at least approximate the marginal density $p(\theta; W)$. Andersen et al.'s book on the complex normal distribution provides results for marginals on partitions of $z$, but nothing else, as far as I can tell. Other than that, I have only found a reference to the $r$ marginal, which apparently follows a multivariate central chi-square distribution with correlations induced by $W$ and has no closed form.

The integral $\int_r p(r, \theta ;W) \ dr$ seems difficult to calculate because of the couplings created by $W$ prevent me from factoring the density. I could diagonalize $W$, but I want to ensure this won't interfere with my ultimate goal, approximating the MLE for $W$ over some data set via gradient descent.

Any resources or suggestions for performing this calculation would be greatly appreciated. Thank you in advance.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.