Natural statistics for spectral samples (Q355128)

Let \(X\) denote a real diagonal \(n \times n\) matrix and \(U\) a random unitary \(n \times n\) matrix, distributed according to Haar measure. Consider the eigenvalues of an \(m \times m\) left upper sub-block \(Y\) of the random matrix \(U X U^*\). While for \(m = n\) the spectrum of \(Y\) is concentrated on the diagonal elements of \(X\), this is no longer the case for \(m < n\). In the article under review, the authors call the spectrum of \(Y\) a spectral sample of \(X\). This should be compared to the case that \(U\) is a uniformly distributed permutation matrix, which leads to the classical notion of random sampling. One goal of the paper is to explicitly determine spectral analogues of \(k\)-statistics and polykays, which is achieved using symbolic umbral techniques. Another goal is to provide an interpretation of free cumulants (in the sense of free probability theory) in terms of spectral sampling. In particular, the approximation of free cumulants by matricial cumulants, as developed by \textit{M. Capitaine} and \textit{M. Casalis} [Probab. Theory Relat. Fields 136, No. 1, 19--36 (2006; Zbl 1146.15015)], fits into the spectral sampling framework.