Miscellaneous results related to the Gaussian product inequality conjecture for the joint distribution of traces of Wishart matrices (Q2697684)

The authors study results related to the Gaussian product inequality for multivariate trace-Wishart distributions. A random vector \(\mathbf{X}\) of dimension \(d\geq2\) is said to have a multivariate trace-Wishart distribution if, for all \(\mathbf{t}\in[0,\infty)^d\), \[ \mathbb{E}(e^{-\mathbf{t}^\intercal\mathbf{X}})=|I_p+\mathrm{diag}(t_1I_{p_1},\ldots,t_dI_{p_d})\Sigma|^{-\alpha}, \] for some \(p_1,\ldots,p_d\in\mathbb{N}\) with \(p=p_1+\cdots+p_d\), symmetric positive semidefinite \(p\times p\) matrix \(\Sigma\), and \(2\alpha\in\mathbb{N}\cup(p-1,\infty)\). Here \(|\cdot|\) denotes the matrix determinant and \(I_p\) is the \(p\times p\) identity matrix. For such a random vector \(\mathbf{X}\) and an integer \(1\leq d_1\leq d\), the authors show that \[ \mathbb{E}\left[\prod_{i=1}^d\phi_i(X_i)\right]\geq\mathbb{E}\left[\prod_{i=1}^{d_1}\phi_i(X_i)\right]\mathbb{E}\left[\prod_{i=d_1+1}^d\phi_i(X_i)\right], \] for any completely monotone functions \(\phi_1,\ldots,\phi_d\) for which the expectations exist. As special cases this includes functions of the form \(\phi_i(x)=x^{q_i}\) for \(q_i\in(-\alpha,0]\) (where the lower bound here is to ensure existence of the expectations) and of the form \(\phi_i(x)=e^{-x^{r_i}}\) for \(r_i\in[0,1]\). In the special case \(d=2\), the authors further show that \[ \mathbb{E}[X_1^{q_1}X_2^{q_2}]\geq\mathbb{E}[X_1^{q_1}]\mathbb{E}[X_2^{q_2}], \] for all \(q_1,q_2\in[0,1]\). They conclude with remarks which fit these results within the wider context of the Gaussian product inequality.