Cumulants for random matrices as convolutions on the symmetric group (Q2502248): Difference between revisions

Motivated by the facts that large random matrices provide asymptotic models for free random variables and free cumulants play a crucial role in the theory of free probability, the authors formulate an analogue of free cumulants for random matrices (calling them ``matricial cumulants'') which approximate the free cumulants under reasonable assumptions. A brief sketch of matricial cumulants is as follows: Let \(S_n\) denote the group of permutations of \(n\) distinct objects and for \(\pi \in S_n\), \(\gamma_n(\pi)=\) number of disjoint cycles in \(\pi\). For \(N=n\), it can be shown that the function \(N^{\gamma_n(\cdot)}\), viewed as an element of the group algebra \(\mathcal{C}S_n\) (the multiplication being the convolution denoted by \(*\)) is invertible. For an \(n\)-tuple \(X=(X_1, \dots X_N)\), of random \(N\times N\) complex matrices, the authors define the matricial cumulants \(\mathcal{C}=\mathcal{C}_X\) by \(\mathcal{C}S_n\ni\mathcal{C}_X:=\mathbb{E}(r_X(\cdot))*(N^{{\gamma(\cdot)}})^{-1}\) where \(r_X(\pi)=\prod_{i=1}^k \mathrm{Tr}( X_{l_1}^{(i)} \ldots X_{n_{i}}^{(i)}):\;\pi=\prod_{i=1}^k(l_{1}^{(i)}\ldots l_{n_{i}}^{(i)})\) being the cyclic decomposition of \(\pi\). Given a noncommutative probability space (\(\mathcal{A}, \phi)\), \(a_1, \ldots a_n \in \mathcal{A}\), and a sequence \(\{X_N^{(N)}\}_N\) of \(N\times N\) random matrices, it is proved that \[ \begin{multlined}\lim_{N\to \infty}\frac{1}{N^{r_n}(\pi)}r_{X(N)}(\pi)=\phi_\pi(a_1\ldots a_n)\text{ for all }\pi\in S_n\\ \Leftrightarrow\lim_{N \to \infty}N^{n-\gamma_n(\pi)}\mathcal{X}_{X(N)}(\pi)=K_\pi(a_1, \ldots a_n)\end{multlined} \] where \(\phi_\pi(a_1\ldots a_n)=\prod_{i=1}^n \phi(a_{l_1^{(i)}}\ldots a_{l_{n_{i}}}^{(i)})\).