No label defined (Q4249095)

The paper presents, without proofs, several recent results of the author, some of them figuring in a preprint, together with some corollaries and discussions. Let \(g_{i j}^{k}\), \(1 \leq i, j \leq n\), \(1 \leq k \leq r\), be independent centered complex gaussian random variables, with \(E(g_{i j}^{k} \overline{g}_{i j}^{k}) =(1 + \delta_{i j}) / n\), \(g_{i j}^{k} = \overline {g}_{j i}^{k}\), and let \(\sigma_{n}(i, j, k) = \sigma_{n}(j, i, k) \geq 0\). Consider \(G_{n}(k) =(\sigma_{n}(i, j, k) g_{i j}^{k})_{i, j} \in A_{n}\) -- the *-algebra of \(n \times n\)-matrices with entries in the *-algebra generated by all \(g_{i j}^{k}\), \(1 \leq 1, j \leq n\), \(1 \leq k \leq r\). Let \(\Delta_{n}\) be the *-subalgebra of \(A_{n}\) consisting of all diagonal matrices with scalar entries and consider the conditional expectation \(E_{n} : A_{n} \rightarrow \Delta_{n}\) taking a matrix into its diagonal part and then into the corresponding expectations; \(( \Delta_{n} \subset A_{n}, E : A_{n} \rightarrow \Delta_{n})\) is a so called non-commutative \(\Delta_{n}\)-valued probability space. The joint \(\Delta_{n}\)-distribution of the ``\(\Delta_{n}\)-valued random variables'' \(G_{n}(1),\dots{}, G_{n}(k)\) (\(\in A_{n}\)) is the \(\Delta_{n}\)-linear map from the polynomials in \(k\) noncommuting \(X_{i}\)'s, with ``coefficients'' in \(\Delta_{n}\), taking \(b_{0} X_{i_{1}} b_{1} \cdots X_{i_ {p}} b_{p}\) into \(E(b_{0} G_{n}(i_{1}) b_{1} \cdots G_{n}(i_{p}) b_{p})\). Let \(\eta_{n i j} : \Delta_{n} \rightarrow \Delta_{n}\) be defined as \(d \rightarrow E_{n}(G_{n}(i) d G_{n}(j))\) (for \(i \neq j\) it is null) and consider \(\Delta_{n} \subset L^{\infty} =\) L\(^{\infty}(0, 1)\) via the partition \([ k / n ,(k + 1) / n ]\), \(0 \leq k \leq n-1\). Suppose that, for some \(\eta_{i j} : L^{\infty} \rightarrow L^{\infty}\), the fact that \(d_{n} \rightarrow d\) in norm implies \(\eta_{n i j}(d_{n}) \rightarrow \eta_{i j}(d)\) in norm. Then \(G_{n}(1),\dots{}, G_{n}(k)\) are asymptotically free with amalgamation over \(L^{\infty}\), for \(n \rightarrow \infty\). This means that their joint \(\Delta_{n}\)-distribution converges to the joint L\(^{\infty}\)-distribution of a family \(Z_{1},\dots{}, Z_{k}\) such that \(E(x_{1} \cdots x_{p}) = 0\) if \(x_{j}\) is in the algebra generated by \(L^{\infty}\) and \(Z_{i(j)}\), \(E(x_{j}) = 0\) and \(i(j) \neq i(j + 1)\). Moreover, \(Z_{i}\) is a ``semicircular \(\Delta_{n}\)-family'' having covariance \(\eta\) (\(E(Z_{i} d Z_{j}) = \eta_{i j}(d)\)). In fact, the corollary follows from a more general result of the author, in which \(\eta_{i j} \neq 0\). Generally, if \(Z(1), \dots ,Z(r)\) is a semicircular \(L^{\infty}\)-family with covariance \(\eta\) and \(\eta_{i j}(a)= c_{i j} \varphi(a)\), \(\varphi(a)= \int a(t)d t\), \(c_{i j}\) constants, then \(C^*(Z(i)\), \(1 \leq i \leq r)\) and \(L^{\infty}\) are free with respect to \(\varphi \circ E\). Other related results are mentioned. Let \(d_{n} \in \Delta_{n}\), \(d_{n} \rightarrow d\) in \(L^{\infty}\). Then the distribution of \(d_{n}+ G_{n}(1)\) with respect to \(( t r \circ E_{n})/ n\) converges to that of \(d+ Z(1)\) with respect to \(\varphi \circ E\). Using the \(R\)-transform, the author obtains a formula for \(z(1+ \sum_{k \geq 1} E(( d+ Z(1))^{k})z^{k})\), \(z \in \mathbb{C}\), enabling to determine the distribution of \(d+ Z(1)\). The last topic concerns factors \(L(F(n))\) corresponding to free groups with \(n \leq \infty\) generators. If \([ M:N ] < \infty\), \(M , N\) \(\text{II}_{1}\)-factors, then for all \(n\) there exists a subfactor \(Q\) of \(P= M \ast L(F(n))\) with \([ P:Q ]= [ M:N ]\). A connection between the free Araki-Woods factors and the above \(G_{n}(i)\) is discussed. For the entire collection see [Zbl 0903.00097].