An expectation formula for the multivariate Dirichlet distribution (Q5935558)

Let \((X_1,\dots, X_n)\) be a vector of random variables following a Dirichlet distribution on \(R^n_+\) with the parameter \((p_1,\dots, p_n)\in R^n_+\). For \(f_1,\dots, f_n>0\), there exists the well-known expectation formula: NEWLINE\[NEWLINEE(f_1X_1+ \cdots+ f_nX_n)^{-(p_1 +\cdots+ p_n)}= \prod^n_{i=1} f_i^{-p_i}.\tag{EF}NEWLINE\]NEWLINE The aim of this paper is to obtain a strong generalization of the (EF) expectation formula to the singular and non-singular multivariate Dirichlet distributions. The new, (EF)-type generalized formula is derived within the context of symmetric cones. The fundamental basis on which the proof of the main result is relying is that the cone of \(n\times n\) positive-definite symmetric matrices (and, more generally, the symmetric cone associated with a simple Jordan algebra) has the structure of a Gelfand pair. Using the framework of symmetric cones, the obtained results are still valid for multivariate Dirichlet distributions on all symmetric cones. Several particular cases of the (EF)-type generalized formula are discussed, and a connection between the obtained results and the theory of multiple hyper-geometric functions is pointed out.