0
$\begingroup$

The differential entropy of the multivariate student-t distribution when the covariance matrix is the identity matrix is given by $$ h = - \log \frac{ \Gamma \left( \frac{\nu+d}{2} \right) } { \Gamma \left( \frac{\nu}{2} \right) (\nu \pi)^{\frac{d}{2}} } + \left( \frac{\nu+d}{2} \right) \left( \psi \left( \frac{\nu+d}{2} \right) - \psi \left( \frac{\nu}{2} \right) \right) $$ (source: Shannon Entropy and Mutual Information for Multivariate Skew-Elliptical Distribution by Arellano-Valle).

What about the general case ($\boldsymbol{\Sigma} \ne \boldsymbol{I}$)?

$\endgroup$
0
$\begingroup$

To get the differential entropy in the general case, we draw on two properties:

  • If $\boldsymbol{x}$ is a standard Student-t random vector, then $\boldsymbol{y} = \boldsymbol{\mu} + \boldsymbol{L} \boldsymbol{x}$ is a Student-t random vector with mean $\boldsymbol{\mu}$ and covariance matrix $\boldsymbol{\Sigma} = \boldsymbol{L}\boldsymbol{L}'$.
  • It is known that for a vector valued random variable $\boldsymbol{x}$ and a matrix $\boldsymbol{A}$ we have $h(\boldsymbol{A} \boldsymbol{x}) = h(\boldsymbol{x}) + \log |\boldsymbol{A}|$ (see https://en.wikipedia.org/wiki/Differential_entropy#Properties_of_differential_entropy).

As a result, we can write for a Student-t random vector $\boldsymbol{x}$ with mean $\boldsymbol{\mu}$ and covariance matrix $\boldsymbol{\Sigma}$ $$ h(\boldsymbol{x}) = h_{\boldsymbol{\Sigma}=\boldsymbol{I}} + \frac{1}{2} \log | \boldsymbol{\Sigma} | $$ where $$ h_{\boldsymbol{\Sigma}=\boldsymbol{I}} = - \log \frac{ \Gamma \left( \frac{\nu+d}{2} \right) } { \Gamma \left( \frac{\nu}{2} \right) (\nu \pi)^{\frac{d}{2}} } + \left( \frac{\nu+d}{2} \right) \left( \psi \left( \frac{\nu+d}{2} \right) - \psi \left( \frac{\nu}{2} \right) \right) $$

$\endgroup$

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.