Rectangular random matrices, related convolution (Q2391168)

The paper deals with the modeling of asymptotic collective behavior of independent rectangular random matrices. Within such modeling it is possible to compute the limit of the asymptotic normalized traces and thus to give the asymptotic normalized trace of any noncommutative polynomial in these independent random matrices. Since the normalized trace of the \(k\)th power of a matrix is the \(k\)th moment of its spectral law it is possible to give the asymptotic singular law of any polynomial of these random matrices. It is proposed a way to compute the asymptotic normalized trace of products of random matrices taken among a family of independent rectangular random matrices, whose sizes tend to infinity, but with different rates. The modeling of asymptotics of rectangular random matrices enables to define a binary operation for \(\lambda \in [0,1]\) on the set of symmetric probability measures, called free convolution with ratio \(\lambda\). This binary operation allows to determine singular values of a sum of two independent rectangular random matrices, whose dimensions tend to infinity, from their singular laws. Finally, examples of rectangular \(R\)-transforms of symmetric probability measures are given together with examples of computation of rectangular free convolutions.