Largest entries of sample correlation matrices from equi-correlated normal populations (Q2280559)

The authors proved two important theorems about the limiting distributions of the largest entry outside the diagonal of the sample covariance / correlation matrices generated by a random sample from a high-dimensional normal distribution, assuming that: the normal distribution has the equi-correlation structure (with parameter \(\rho\)), that \(p \to \infty\) and \(\log(p) = o(n^{1/3})\), where \(p\) is the random vector dimension and \(n\) is the number of observations. The theory developed is exemplified by means of an application to a statistical test involving a high-dimensional normal distribution (testing if the covariance matrix is diagonal). Finally, in the ``discussions'' subsection the authors conclude that behaviors of limiting distributions depend on the value of \(\rho\). The limits are the normal distribution if \(\rho\) is reasonably large; the limiting distributions are the extreme-value distribution if \(\rho\) is tiny. Moreover, the authors also figure out the regime to differentiate the two scenarios.