On asymptotics for the spectrum of the product of two random rectangular matrices (Q642087)

Consider an \(n \times p\) random matrix \(X\) and a \(p \times n\) random matrix \(Y\) with jointly independent centered entries of variance \(1/n\) and \(1/p\), respectively. Then, assuming that \(\lim_{n \to \infty} n/p = q > 0\), the author proves that as \(n \to \infty\), the mean empirical spectral distribution of the product \(W = XY\) converges to a distribution on the unit disk with density \[ \frac{1}{\pi \sqrt{ (1-q)^2 + 4 q (x^2 + y^2)}}\;1_{\{ x^2 + y^2 \leq 1\}}, \] which in the case of quadratic matrices is nothing else than the square of the uniform distribution on the unit circle. The methods are based on those that were used by \textit{F.\ Götze} and the author in their work on the circular law [Ann. Probab. 38, No. 4, 1444--1491 (2010; Zbl 1203.60010)].