Random volumes under a general matrix-variate model (Q2370789): Difference between revisions

Let the \(p\times n\) matrix \(X\), \(n\geq p\), representing the \(p\) linear independent points in Euclidean \(n\)-space have a general density of the form \[ f(X)=c|XX'|^{\delta}|I-a(1-a)XX'|^{\gamma/(1-\alpha)} \] for \(I-a(1-a)XX'>0\), \(a>0\), \(\gamma>0\), \(-\infty<\alpha<\infty\), where \(a,\gamma,\delta\) and \(\alpha\) are scalar quantities, \(|\cdot|\) denotes the determinant and \(X'\) is the transpose of \(X\). The volume of the convex hull, the \(p\)-parallelotope, is \(\nu=|XX'|\). The author shows that the volume of the \(p\)-parallelotope is connected to the likelihood ratio test statistics for testing various types of hypotheses on the parameters of one or more multivariate Gaussian distributions, to sample generalized variance when the population is Gaussian and to Melin-Barnes type integrals, and that, structurally, the volume of the \(p\)-parallelotope can be viewed as a product of independently distributed real scalar random variables belonging to the categories of type-1, type-2 beta and gamma variables.