Cumulants in noncommutative probability theory. I: Noncommutative exchangeability systems (Q1882590)

This is the first in a series of papers by the author aimed at a unified treatment of cumulants in classical and noncommutative probability. Using a formula of \textit{I. J. Good} [Math. Proc. Camb. Philos. Soc. 78, 333--337 (1975; Zbl 0335.62014)] which expresses classical cumulants as moments of some kind of ``discrete Fourier transform'', the author of the paper under review gives a general definition of cumulants which contains as special cases all of the different notions of cumulant known so far, in classical, free and other noncommutative probability theories (e.g., Boolean cumulants, conditionally free cumulants, etc.). Since every notion of cumulant is naturally associated with a notion of independence, it is expected that there should be a general definition of ``independence'' corresponding to the author's definition of cumulant. Such a notion of independence is formulated in terms of the ``exchangeability system'' \(\mathcal E\) on a noncommutative probability space \(({\mathcal A}, \phi)\) (where \(\mathcal A\) is a complex unital algebra and \(\phi\) is a unital linear functional on \(\mathcal A\)), which consists of a (noncommutative) probability space \(({\mathcal U},\widetilde{\phi})\) and an infinite family \(\eta_k\) of one-to-one homomorphisms from \(\mathcal A\) into \(\mathcal U\) such that \(\widetilde{\phi} \circ \eta_k=\phi\) for all \(k\), and for any \(X_1,\dots ,X_n \in {\mathcal A }\), any choice of \(i_1,\dots,i_n\) and permutation \(\sigma\) of \(\mathbb N\), one has that \[ \widetilde{\phi}(\eta_{i_1}(X_1)\dots \eta_{i_n}(X_n))= \widetilde{\phi}(\eta_{\sigma(i_1)}(X_1)\dots \eta_{\sigma(i_n)}(X_n)). \] Two subalgebras \(\mathcal B\) and \(\mathcal C\) of \(\mathcal A\) are said to be ``\(\mathcal E\)-exchangeable'' or ``\(\mathcal E\)-independent'' if for any choice of \(X_1,\dots,X_n \in {\mathcal B} \cup {\mathcal C}\) and subsets \(I,J \subseteq \{ 1,\dots,n \}\) such that \(I \cap J =\emptyset\), \(I \cup J=\{ 1,\dots,n \}\), \(X_i \in \mathcal B\) for \(i \in I\) and \(X_i \in \mathcal C\) for \(i \in J\), we have the identity \[ \phi\biggl(\prod_{j \in \pi} X_j\biggr)= \phi\biggl(\prod_{j \in \pi^\prime} X_j\biggr) \] whenever \(\pi\) and \(\pi^\prime\) are partitions of \(\{ 1,\dots,n \}\) satisfying \(\pi| _I=\pi^\prime| _I,\) \(\pi| _J=\pi^\prime| _J\). The \(n\)-th cumulant w.r.t. the above exchangeability system \(\mathcal E\) is defined by \[ k_n^{\mathcal E}(X_1,\dots ,X_n):=\frac{1}{n} \widetilde{\phi}(X_1^\omega \dots X_n^\omega), \] where \(\omega\) denotes an \(n\)-th primitive root of unity and \(X_j^\omega:=\omega \eta_1(X_1)+\omega^2 \eta_2(X_2)+\cdots +\omega^n \eta_n(X_n).\) It is shown that \(k_n^{\mathcal E}(X_1,\dots ,X_n)=0\) whenever there is a nonempty subset \(I\) of \(\{ 1,\dots ,n \}\) such that \(J=\{ 1,\dots ,n \}-I\) is also nonempty and the algebras generated by \((X_i)_{i \in I}\) and \((X_j)_{j \in J}\) are \( \mathcal E\)-independent.