Limit theorems in free probability theory. I (Q2468421)

\textit{H.\,Bercovici} and \textit{V.\,Pata} [Ann.\ Math.\ (2) 149, No.\,3, 1023--1060 (1999; Zbl 0945.46046)] established the distributional behavior of sums of free identically distributed random variables and described explicitly the correspondence between limit laws for free and classical additive convolutions. In the present paper, the authors generalize the above results to the case of free nonidentically distributed random variables, by using an analytical approach to the definition of the additive free convolution which is based on the authors' previous article [\textit{G.\,P.\thinspace Chistyakov} and \textit{F. Götze}, ``The arithmetic of distributions in free probability theory'' (2005; arXiv:math.OA/0508245)]. More precisely, the authors show that the parallelism between limit laws for additive free and classical convolutions (found by Bercovici and Pata [loc.\,cit.])\ holds in the general case of free nonidentically distributed random variables. In fact, they prove the semicircle approximation theorem (an analogue of the Berry-Esséen inequality), the law of large numbers, based upon their own analytical approach to the definition of the additive free convolution which allows one to obtain estimates of the rate of convergence of distribution functions of free sums. The authors describe Lévy's class \({\mathcal L}_{\boxplus}\) of limiting distributions of normed sums of free random variables obeying infinitesimal conditions, and also derive, as in the classical case, a canonical representation of the measures in the class \({\mathcal L}_{\boxplus}\). Furthermore, they provide a characterization of the class \({\mathcal L}_{\boxplus}\) by means of the property of self-decomposability, extending results by \textit{O.\,E.\thinspace Barndorff-Nielsen} and \textit{S.\,Thorbjørnsen} [Proc.\ Natl.\ Acad.\ Sci.\ USA 99, No.\,26, 16576--16580 (2002; Zbl 1063.46052)]. The essence of remarkable features in this paper consists in the restatement of additive free convolution on probability measures on the real line, via the authors' analytical approach. Let \({\mathcal M}\) be the family of all Borel \(p\)-measures defined on the real line \({\mathbb R}\). For \(\mu_1, \mu_2 \in {\mathcal M}\), let the \(p\)-measure \(\mu_1 * \mu_2\) denote the classical convolution of \(\mu_1\) and \(\mu_2\), and let the \(p\)-measure \(\mu_1 \boxplus \mu_2\) denote the free additive convolution of \(\mu_1\) and \(\mu_2\). The latter one was originally introduced first for compactly supported \(p\)-measures by \textit{D.\,V.\thinspace Voiculescu} [J.\ Funct.\ Anal.\ 66, 323--346 (1986; Zbl 0651.46063)], and was extended to \(p\)-measures with finite variance by \textit{H. Maassen} [J. Funct. Anal. 106, No.~2, 409--438 (1992; Zbl 0784.46047)] and finally to the class \({\mathcal M}\) by \textit{H.\,Bercovici} and \textit{D.\,Voiculescu} [Indiana Univ.\ Math.\ J.\ 42, No.\,3, 733--773 (1993; Zbl 0806.46070)]. For the Nevanlinna class \({\mathcal N}\) of analytic functions \(F : {\mathbb C}^+ \to {\mathbb C}^+ \cup {\mathbb R}\), let the class \({\mathcal F}\) be the subclass of Nevanlinna functions \(F_{\mu}\) for which \(F_{\mu}(z) / z \to 1\) as \(z \to \infty\) nontangentially to \({\mathbb R}\) such that \(\text{Re}\, z / \text{Im} \, z\) stays bounded, where \(F_{\mu}\) denotes the reciprocal Cauchy transform for \(\mu \in {\mathcal M}\), with \(F_{\mu}(z) = G_{\mu}(z)^{-1}\) and \[ G_{\mu}(z) = \int_{- \infty}^{\infty} \frac{ \mu( d t)}{z - t}, \quad z \in {\mathbb C}^+. \] The authors define the free convolution \(\mu_1 \boxplus \mu_2\) of \(p\)-measures \(\mu_1\) and \(\mu_2\), in an alternative manner, by purely complex analytic methods. Since there are unique functions \(Z_1(z)\) and \(Z_2(z)\) in the class \({\mathcal F}\) such that, for \(z \in {\mathbb C}^+\), \[ z = Z_1(z) + Z_2(z) - F_{\mu_1}( Z_1(z) ) \quad \text{and} \] \[ F_{\mu_1} ( Z_1( z) ) = F_{\mu_2}( Z_2(z)), \] the function \(F_{\mu_1}( Z_1 (z))\) belongs again to \({\mathcal F}\) and hence there exists a \(p\)-measure \(\mu\) such that \(F_{\mu_1}( Z_1(z)) = F_{\mu}(z)\). Then the free convolution \(\mu_1 \boxplus \mu_2\) is defined as \(\mu_1 \boxplus \mu_2 = \mu\), and the measure \(\mu\) depends on \(\mu_1\) and \(\mu_2\) only.