Principal components selection given extensively many variables (Q2387330)

Recent results in random matrix theory of the Wishart ensemble on the distribution of the largest eigenvalues were derived for sample size \(n\) and number of variables \(p\), both asymptotically large. Principal components analysis being related to the eigenvalue distribution of Wishart matrices, in this paper it is shown that for \(p\) much greater than \(n\) (arbitrary), a simple linear transformation of eigenvalues maps the Wishart eigenvalue statistics to the very well known eigenvalue statistics in the Gaussian orthogonal ensemble. An illustrative example on principal components selection and a discussion complete the approach.