Second order freeness and fluctuations of random matrices. I: Gaussian and Wishart matrices and cyclic Fock spaces (Q2491599)

The authors aim at extending the relation between random matrices and free probability theory from the level of expectations to the level of fluctuations. To do so, they introduce the concept of ``second order freeness'' and interpret the global fluctuations of Gaussian and Wishart random matrices by applying a general limit theorem (Theorem 6.9 in \S 6) for second order freeness. With the introduction of the cyclic Fock space, an operator algebraic model is first given for fluctuation analysis and then the fluctuations of random matrices are also formulated in terms of the usual creation, annihilation and preservation operators. On the other hand, the authors introduce an abstract freeness property for bilinear tracial functionals, which plays an important and crucial role in providing a conceptual understanding of the introduced model as well as in proving the assertions on the fluctuations. The goal of the paper is to show that orthogonal families of Gaussian and Wishart random matrices are asymptotically free of second order. Note that the second order freeness is stronger than the freeness of Voiculescu, however that it appears to be a central feature of ensembles of random matrices. The first main theorem (Theorem 5.3 in \S 5.1) is to show that semi-circular families on the full Fock space are free of second order. For \(f\) in a Hilbert space \({\mathcal H}\), \(\ell(f)\) (resp., \(\ell^*(f)\)) denotes the creation (resp., annihilation) operator on the full Fock space \({\mathcal F}({\mathcal H})\). Let \(X_N(f)\) \((f \in {\mathcal H})\) be a family of Hermitian Gaussian random matrices. Note that they converge, for \(N \to \infty\), in distribution to a semi-circular family, given by \(\omega(f) =\) \(\ell(f) + \ell^*(f)\) realized on \({\mathcal F}({\mathcal H})\), cf.\ \textit{D.~Voiculescu} [Proceedings of the International Congress of Mathematicians (ICM '94), Aug. 3--11, 1994, Zürich, Switzerland, Vol.~I, 227--241 (1995; Zbl 0839.46059)]. \(\langle \cdot, \cdot \rangle_{ cyc }\) is an inner product of the cyclic Fock space \[ {\mathcal F}_{cyc}({\mathcal H}) = \bigoplus_{n=0}^{\infty} {\mathcal H}_{cyc}^{ \otimes n}, \] where \({\mathcal H}_{cyc}^{\otimes n}\) is the corresponding \(n\)-th particle space of the circular tensors \([ f_1 \otimes \cdots \otimes f_n ]\). The mapping \(\text{ \textbf{c} }\) between the linear Fock spaces \({\mathcal F}_{lin}({\mathcal H})\) and \({\mathcal F}_{cyc}({\mathcal H})\) is given recursively by \(\text{\textbf{c}} \Omega\) \(=0\) (\(\Omega\): the vacuum), \(\text{\textbf{c}} (f) = [f]\), and \(\text{\textbf{c}} ( f_1 \otimes \cdots \otimes f_n)=[ f_1 \otimes \cdots \otimes f_n]\) \(+\) \(\langle f_1, \bar{f}_n \rangle \cdot \text{\textbf{c}} (f_2 \otimes \cdots \otimes f_{n-1})\). Let \(( k_r)_{r \in {\mathbb N} }\) denote the classical cumulants and \(\text{Tr}\) the non-normalized trace. Theorem. For all \(n, m \in {\mathbb N}\) and all \(f_1, \dots, f_n\), \(g_1, \dots, g_m \in {\mathcal H}\), \[ \begin{aligned} \lim_{N \to \infty} k_2 & \left\{ \text{Tr} [ X_N(f_1) \cdots X_N(f_n) ], \,\, \text{Tr} [ X_N(g_1) \cdots X_N(g_m) ] \right\} \\ & = \langle \text{\textbf{c}} \omega(f_1) \cdots \omega(f_n) \Omega, \text{\textbf{c}} \omega(g_m) \cdots \omega(g_1) \Omega \rangle_{cyc}. \end{aligned} \] The key point of the above theorem is that we can express the fluctuations of the Gaussian matrices via the operators \(\omega(f)\). Therefore, the purpose of the second main theorem (Theorem 5.7 in \S 5.2) is to express with a similar type of formula the limit fluctuations of compound Wishart matrices \(\{ P_N ( D_i ) \}\) with the help of the corresponding operators \(p(d)\), giving a realization of compound Poisson elements (instead of \(\omega(f)\) this time). Moreover, the diagonalization of the fluctuations in the Gaussian and Wishart cases is given in Section 8, thus recovering and extending results of \textit{T.~Cabanal--Duvillard} [Ann.\ Inst.\ Henri Poincaré, Probab.\ Stat.\ 37, No.~3, 373--402 (2001; Zbl 1016.15020)]. Lastly, the authors prove asymptotic freeness of second order (Theorem 9.4 in \S 9) for orthogonal families of Gaussian and Wishart random matrices.