A Khintchine decomposition for free probability (Q690879): Difference between revisions

Let \((A, \tau)\) be a \(W^*\) probability space, and consider a random variable \(x \in A\) with mean zero and finite variance. A probability measure \(\mu\) on the real line is said to be \(\boxplus\)-infinitely divisible if, for every \(n \in {\mathbb N}\), there exists a measure \(\mu_{1/n}\) such that \(\mu\) \(=\) \(\mu_{1/n} \boxplus\dotsb\boxplus \mu_{1/n}\), where the measure on the right-hand side is the \(n\)-fold free convolution. Moreover, a measure \(\mu\) is said to be indecomposable if it has no nontrivial decomposition. The author proves a Khintchine type decomposition for the additive free convolution \(\boxplus\), which just corresponds to the classical decomposition result of Khintchine for the usual convolution \(*\) in probability theory. \noindent Theorem 1. For a probability measure \(\mu\), there exist measures \(\mu_i\) \((i=0,1,2, \dots)\) such that \(\mu_0\) is \(\boxplus\)-infinitely divisible, \(\mu_i\) is indecomposable for \(i=1, 2, \dots\) and \(\mu\) \(=\) \(\mu_0 \boxplus\mu_1\boxplus \mu_2 \boxplus \dotsm\). This decomposition is not unique. Additionally, he proves that, for \(\mu\) \(=\) \(\mu_{1,k} \boxplus \mu_{2,k}\) (for all \(k \in {\mathbb N}\)), there exist translations \(\{ \hat{\mu}_{i,k} \}\) so that \(\mu\) \(=\) \(\hat{\mu}_{1,k} \boxplus \hat{\mu}_{2,k}\) and the family of measures \(\{ \hat{\mu}_{i,k} \}\) is tight for \(i=1, 2\). This indicates that the family of all \(\boxplus\)-divisors of a measure \(\mu\) is compact up to translation. Analogous results are true even in the case of multiplicative convolution. For other related works, see, e.g., [\textit{S. T. Belinschi}, Ann. Inst. Henri Poincaré, Probab. Stat. 42, No. 5, 635--648, (2006; Zbl 1107.46043)] for the Lebesgue decomposition of a free additive convolution; [\textit{S. T. Belinschi} and \textit{H. Bercovici}, Can.\ Math.\ Bull.\ 51, No. 1, 26--31 (2008; Zbl 1144.46056)] for the Khintchine theorem for a multiplicative free convolution; and [\textit{H. Bercovici} and \textit{J.-C. Wang}, Indiana Univ.\ Math.\ J. 57, No. 6, 2601--2610 (2008; Zbl 1171.46043)] for freely indecomposable measures.