A group-theoretical approach to conditionally free cumulants (Q776415)

In the theory of free probability, several families of scalars are attached to random variables, leading to the notion of free independence: moments, free cumulants, monotone cumulants and Boolean cumulants. They are related by moments-cumulants relations, involving various families of noncrossing partitions. This paper, part of a series of articles on the relations between combinatorial Hopf algebras and free probability, gives an interpretation of these relations in terms of a group of characters of a Hopf algebra \(H\) of words. Moments, free, monotone and boolean cumulants are seen as characters or infinitesimal characters of \(H\), respectively denoted by \(\Phi\), \(\kappa\), \(\beta\) and \(\rho\). The combinatorial structure of \(H\) induces a structure of unshuffle bialgebra, and dually the algebra \(H^*\) is a shuffle algebra, which means that its associative product \(*\) can be split as the sum of two half-products \(\prec\) and \(\succ\). All these products induce an exponential map, bijection from the Lie algebra of infinitesimal characters to the group of characters. It is shown that that \(\Phi=exp_\succ(\beta)=\exp_\prec(\kappa)=\exp_*(\rho)\). This implies that \(\beta=\Phi^{*-1}\succ \kappa \prec \Phi\), giving the Boolean cumulant-free cumulant relations. The main new result of this paper is stated as follows. If \(\Phi=exp_\succ(\beta)\) and \(\Psi\) are two moments characters, one looks for an infinitesimal character \(R\) such that \(\beta=\Psi^{*-1}\succ R\prec\Psi\). This \(R\) is shown to compute the multivariate conditionally free cumulant. For the entire collection see [Zbl 1437.05003].