Free probability of type \(B\): analytic interpretation and applications (Q2879629)

Free probability was introduced by Voiculescu in the 1980's. Among his first achievements was the development of analytic tools for computing the associated convolution operations. On the other hand, it was later discovered by Speicher that the combinatorial structure of free probability is described by the lattice of (type ``A'') non-crossing partitions.NEWLINENEWLINEThe lattice of non-crossing partitions can be naturally associated with symmetric groups, which are Weyl groups of type A. \textit{V. Reiner} [Discrete Math. 177, No. 1--3, 195--222 (1997; Zbl 0892.06001)] has shown that one can replace the symmetric group by the hyperoctahedral group (the Weyl group of type B) to obtain a lattice \(NC^{(B)}\) of type \(B\) non-crossing partitions. In [Trans. Am. Math. Soc. 355, No. 6, 2263--2303 (2003; Zbl 1031.46075)], \textit{P. Biane}, \textit{F. Goodman} and \textit{A. Nica} introduced the notions of a type \(\mathrm B\) probability space and type \(B\) free independence, whose combinatorics are governed by the lattice \(NC^{(B)}\).NEWLINENEWLINEIn this article the authors give a new perspective on type B free probability, which they use to develop analytic tools for studying type B free convolutions. They define an infinitesimal law to be a pair \((\mu,\mu')\) of linear functionals on the algebra of noncommutative polynomials \(\mathbb C \langle X_1,\dotsc,X_n \rangle\), such that \(\mu(1) = 1\) and \(\mu'(1) = 0\). The motivating example arises from processes \((X_1(t),\dotsc,X_n(t))_{t \geq 0}\) with distributions \((\mu_t)_{t \geq 0}\), by setting \(\mu = \mu_0\) and \(\mu'(P) = \frac{d}{dt}\bigr|_{t = 0} \bigl[\mu_t(P)\bigr]\). If \((X_1(t),\dotsc,X_k(t))\) is free from \((X_{k+1}(t),\dotsc,X_n(t))\) ``to order \(t\)'' as \(t \to 0\), \((X_1,\dotsc,X_k)\) is said to be infinitesimally free from \((X_{k+1},\dotsc,X_{n})\) with respect to \((\mu,\mu')\).NEWLINENEWLINETo any type B distribution they associate such an infinitesimal law \((\mu,\mu')\), and they show that type B freeness corresponds in this way to infinitesimal freeness. A key result is that if \((\mu,\mu')\) and \((\nu,\nu')\) are the infinitesimal laws associated to the type B distributions \(\mu^B\), \(\nu^B\) then the type B convolution \(\mu^B \boxplus_B \nu^B\) corresponds to the pair \((\eta,\eta')\) with \(\eta = \mu \boxplus \nu\) and \(\eta' = \frac{d}{dt}\bigr|_{t=0} \bigl[\mu_t \boxplus \nu_t\bigr]\), where \(\boxplus\) is the usual free convolution. This allows them to use tools from type A free probability to develop a linearizing transform for type B free additive convolution, which they use to further analyze some type B limit theorems of [\textit{M. Popa}, Colloq. Math. 120, No. 2, 319--329 (2010; Zbl 1213.46063)]. In particular they show that the distribution appearing in the type B central limit theorem corresponds to the the infinitesimal law arising from the process \(X_t = x + ty\) where \(x,y\) are free semicircular random variables. They also study type \(B\) free multiplicative convolution and construct the relevant transform. Finally they consider random matrix models for type \(B\) freeness. It does not appear that type \(B\) freeness arises in any natural way from classical random matrices, however they find that one can obtain type \(B\) freeness through some noncommutative random matrix models.