Characterizations of noncentral chi-squared-generating covariance structures for a normally distributed random vector (Q505485)

The authors analyze the characterization of covariance structures for which some quadratic forms are distributed as non-central independent chi-squared random variables. More precisely, let \(y \sim N_n(\mu,V)\), where \(y\) denotes an \(n \times 1\) random vector, \(\mu\) is the \(n \times 1\) null vector and \(V\) is a \(n \times n\) covariance matrix. Let \(\{ A_1, A_2, \dots, A_k \}\) be a set of \(k \leq n\) different \(n \times n\) nonnegative definite matrices and \(\{ y^TA_iy \}_{i=1}^k\) denote a family of quadratic forms. The aim of this work is to characterize the sets of nonnegative definite and positive definite covariance matrices \(V\) such that \(\{ y^TA_iy \}_{i=1}^k\) are mutually independent chi-squared random variables. To precisely formulate the problem of interest, let \(y \sim N_n(\mu,V)\), the set of nonnegative definite commutative matrices \[ {\mathcal A}_c \equiv \{A_i \in \mathbb{R}_n^{\geq} : A_iA_j=A_jA_i, \;i,j \in \{1,2,\dots,k\}, i\neq j \} \] and the set of covariance matrices \[ \bar{V}({\mathcal A}_c) \equiv \{V \in \mathbb{R}_n^{\geq} : A_iVA_i=\alpha_i A_i \;\text{and} \;A_iVA_j=0, \;A_i,A_j \in {\mathcal A}_c, \alpha_i>0, i,j \in \{1,2,\dots,k\}, i\neq j \} \] where \(\mathbb{R}_n^{\geq}\) denotes the cone consisting of all symmetric nonnegative definite matrices in \(\mathbb{R}_{n \times n}\). In the literature, it refers to the covariance matrices in \(\bar{V}({\mathcal A}_c) \) as independent chi-squared-generating (ICSG) covariance structures for the set of matrices \({\mathcal A}_c\). The authors derive two ICSG covariance structure characterization results. In the first one, they obtain an explicit expression for members of \(\bar{V}({\mathcal A}_c) \) in terms of \(A_i \in {\mathcal A}_c \) and corresponding Moore-Penrose pseudo-inverse \(A_i^{-} \in {\mathcal B}_c\), where \(i \in \{1,2,\dots,k\}\) and \[ {\mathcal B}_c \equiv \{A_i^{-} : A_i \in {\mathcal A}_c \;\text{and for each} \;A_j \in {\mathcal A}_c, A_jA_i^{-}=0, \;i,j \in \{1,2,\dots,k\}, i\neq j \}. \] In the second result, the authors characterize the set of positive definite ICSG covariance structures corresponding to the set \(\{A_1,A_2, \dots,A_k\}\) of real \(n \times n\) symmetric matrices. Finally, an example application of the proposed explicit ICSG covariance-structure characterization results fora finite cardinality \(k\) is provided.