On limiting spectral distribution of product of two random matrices when the underlying distribution is isotropic (Q1110954)

In multivariate analysis the distributions of functions of the eigenvalues of the product of random matrices are useful. For instance, in multivariate analysis of variance, discriminant analysis and tests for the equality of two covariance matrices it is of interest to consider the eigen-values of the multivariate F matrix which is defined as \(F=n S1 S2^{(-1)}/m\) where S1 and S2 are distributed independently of each other as central Wishart matrices with m and n degrees of freedom, respectively, and \(E(S1/m)=E(S2/n).\) It has been shown that the limiting spectral distribution of ST/n exists when \(S=XX'\) is distributed independently of a nonnegative definite symmetric random matrix T, \(X=[x1\),..., xn] and x1,..., xn is a sample from a multivariate normal population. This result is extended when x1,..., xn is a sample from an isotropic population and the moments of any order of xi exist for \(i=1\),..., n. A random vector \(x'=(x1\),..., x) is said to have isotropic distribution if the distribution of x is the same as that of Px, where P is an orthogonal matrix of order p. The authors prove that the limit of the spectral distribution of ST/n exists when x1,..., xn is a sample from an isotropic population and the second moments of xi \((i=1\),..., n) exist.