Logarithmic Sobolev inequalities, matrix models and free entropy (Q1412940): Difference between revisions

Suggested by some results due to \textit{D. Bakry} and \textit{M. Émery} [Séminaire de Probabilités, XIX, 1983/84, Berlin, Springer, Lect. Notes Math. 1123, 177--206 (1985; Zbl 0561.60080)] and \textit{M. Ledoux} [Séminaire de Probabilités, XXXIII, Berlin, Springer, Lect. Notes Math. 1709, 120--216 (1999; Zbl 0957.60016)], the author establishes a generalization of Voiculescu's inequality between the free entropy \(\chi(\mu)\) and the free Fisher information \(\Phi(\mu)\) of probability measures \(\mu\) on the real line, cf. \textit{D. V. Voiculescu} [Invent. Math. 132, No. 1, 189--227 (1998; Zbl 0930.46053)], by making use of the logarithmic Sobolev inequality for Gibbs measures on matrix spaces with a uniformly convex potential \(V\). Moreover, the author also discusses the construction of non-commutative random variables with prescribed conjugate variables by using again logarithmic Sobolev inequalities through the Bakry-Émery criterion, this time on unitary groups. Finally, the author computes the second derivative (Hessian) of the free entropy functional on the trace state space at a point where the free Fisher information is finite, and shows that it is the complete Dirichlet form \(Q\) on the \(L^2\) space of the trace state. Then the author proves that semi-circular systems \(X\) are characterized by a Poincaré inequality with respect to this Dirichlet form \(Q\).