2
$\begingroup$

I have found an equation for the entropy of a $p$-variate Cauchy distribution here [page 70]:

$H(X,R) = \frac{1}{2}\log(\det(R))+f(p)\,,$

where $X=(X_1,X_2,\dots,X_p)$ is vector of random variables having a $p$-variate Cauchy distribution, and $R$ is the correlation matrix of $X$, and $f$ is some function.

I have problem understanding how one can define a correlation matrix for a multivariate Cauchy distribution. Normally, one would define the a correlation $R_{ij}$ via $Cov(X_i,X_j)$ and the variances of $X_i$ and $X_j$. But for the Cauchy distribution, the variances are undefined. So how would I define correlation for the multivariate Cauchy distribution?

$\endgroup$
  • 4
    $\begingroup$ That's not actually a correlation matrix, it's a scale matrix. It so happens that for the multivariate Normal distribution with all the variances equal to 1, the scale matrix is a correlation matrix, but otherwise not. So... it will exist even for a Cauchy distribution. $\endgroup$ – jbowman Feb 9 '18 at 23:30
0
$\begingroup$

Answered in comments:

That's not actually a correlation matrix, it's a scale matrix. It so happens that for the multivariate Normal distribution with all the variances equal to 1, the scale matrix is a correlation matrix, but otherwise not. So ... it will exist even for a Cauchy distribution. – jbowman

$\endgroup$

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.