Dirichlet partition on symmetric matrices (Q5963333)

An \(r\times r\) symmetric random matrix \(X\) is said to have a Wishart distribution \(W_{\theta}\) with parameter \(\theta > \frac{r-1}{2}\) if its probability density function is given by \((\Gamma_{\Omega} (\theta))^{-1}\exp(-tr X)(\det X)^{\theta- \frac{r+1}{2}} I_{\Omega} (X),\) where \(\Gamma_{\Omega} (\theta)= \pi^{ \frac{r(r-1)}{4}} \prod_{k=1}^{r} \Gamma (\theta- \frac{k-1}{2})\) is the multivariate gamma function. The concept of Dirichlet distribution on symmetric matrices relies on the Wishart distribution and was introduced in [\textit{H. Massam}, Ann. Stat. 22, No. 1, 369--394 (1994; Zbl 0811.62023)], in the absolutely continuous case. In the contribution under review, some properties concerning the residual allocation model structure as well as the scaling property of the Dirichlet partition are given. Therein, the gamma distribution is replaced by the Wishart distribution. On the other hand, a result by \textit{T. Huillet} and \textit{S. Martinez} [Methodol. Comput. Appl. Probab. 5, No. 4, 467--492 (2003; Zbl 1038.60007)], that gives a representation of the mean expected value of many functionals of Dirichlet distributions in terms of the mean of independent gamma random variables in the scalar case, is extended to the case of positive definite symmetric matrices by using the notion of generalized power of a matrix. In such a way, the computation of spacing functionals of matrix-Dirichlet distributions in terms of more simple functionals of independent Wishart random variables is presented. As a consequence, the computation of the Laplace transform of the Dirichlet distribution is given. Furthermore, the authors are able to compute any homogeneous functional of Dirichlet partition from one using independent Wishart variables.