Cumulants for random matrices as convolutions on the symmetric group. II (Q2385610): Difference between revisions

This is a continuation of authors' previous work [Probab. Theory Relat. Fields 136, No. 1, 19--36 (2006; Zbl 1146.15015)], where they defined and studied `U-cumulants' for matrices. For an \(n\)-tuple of random matrices, say \(X=(X_1,\dots,X_n)\), the \(U\)-cumulant is a complex-valued map \(C_X(\cdot)\) on the group \(S_n\) of permutations of \(n\) objects. These cumulants converge in a suitable sense to the free cumulants of the limiting noncommutative random variables when the size of the matrices tend to infinity, and they also enjoy some nice properties like the so-called `linearization' for matrix models whose law is invariant under unitary conjugation. However, it turns out that the above nice properties fail to hold for matrix models which are not unitary invariant. For this reason, the authors have introduced in the present article a new variant of cumulants called `\(O\)-cumulants' which will play a similar role for orthogonally invariant matrix models. These are defined as maps \(C^{O+}_X, C^{O-}_X \) from \(S_{2n}\) (instead of \(S_n\) as in case of \(U\)-cumulants) to \(C\), given by: \[ C^{O+}=\frac{1}{|H_n|}{\mathbb{M}}^+_X \ast W_\Lambda, \] \[ C^{O-}_X=\frac{1}{|H_n|} W_\Lambda \ast {\mathbb{M}}^-_X . \] Here \(W_\Lambda\) is closely related to the so-called `orthogonal Weingarten function' introduced by \textit{B. Collins} and \textit{P. Śnidy} [''Integration with respect to the Haar measure on unitary, orthogonal and symplectic group,'' Commun. Math. Phys. 264, No. 3, 773--795 (2006; Zbl 1108.60004)]. To explain \({\mathbb{M}}^{\pm}_X\), we need some notation. For \(\pi \in S_n\), we denote by \(s_\pi\) the element of \(S_{2n}\) which sends \(i\) to \(\pi(i)\) and \(n+i\) to \(n+\pi(i)\) for each \(i=1,\dots,n\). For an \(n\)-tuple \(\epsilon=(\epsilon_1,\dots,\epsilon_n)\) of numbers from \(\{ \pm \}\), we consider the element \(\tau_\epsilon\) of \(S_{2n}\) which is the product of the transpositions \((i,n+i)\) over those \(i\) for which \(\epsilon_i=-1\). Let \(H_n\) be the subgroup of \(S_{2n}\) consisting of permutations of the form \(s_{\pi} \tau_{\epsilon}\) where \(\pi, \epsilon\) are allowed to vary, and let \(\{ g_1, \dots, g_k\) be elements of \(S_{2n}\) such that \(S_{2n}/H_n=\bigcup g_l H_n=\bigcup H_ng_l\). The elements \(g_l\) are chosen to be of the form \(g_l=\tau_{\epsilon^l} s_{\pi_l} \tau_{\epsilon^l}\) for some \(\pi_l, \epsilon^l\). With this notation, \[ {\mathbb{M}^+_X}(g):={\mathbb{E}}(r_{\pi_l}(X_1^{\epsilon^l_1},\dots,X_n^{\epsilon_n^l})), \] for \(g \in g_lH_n\), where \(r_\pi(Z_1,\dots,Z_n):=\prod_{C \in C(\pi)} Tr \left(\prod_{j \in C} Z_j \right),\) \(C(\pi)\) being the set of disjoint cycle of \(\pi\).