[Paper Review] Free Stein kernels and an improvement of the free logarithmic Sobolev inequality

Max Fathi, Brent Nelson|arXiv (Cornell University)|May 19, 2016

Random Matrices and Applications43 references18 citations

TL;DR

This paper introduces a free Stein kernel relative to the semicircular law, enabling a free analog of the HSI inequality that improves upon the free logarithmic Sobolev inequality of Biane and Speicher. The method yields a rate of convergence in the multivariate entropic free central limit theorem and computes free Stein kernels for key operator families, advancing non-commutative functional inequalities via free probability tools.

ABSTRACT

We introduce a free version of the Stein kernel, relative to a semicircular law. We use it to obtain a free counterpart of the HSI inequality of Ledoux, Peccatti and Nourdin, which is an improvement of the free logarithmic Sobolev inequality of Biane and Speicher, as well as a rate of convergence in the (multivariate) entropic free Central Limit Theorem. We also compute the free Stein kernels for several relevant families of self-adjoint operators.

Motivation & Objective

• To develop a free probability counterpart to the classical Stein kernel for semicircular laws.
• To refine the free logarithmic Sobolev inequality of Biane and Speicher using a free HSI-type inequality.
• To establish a rate of convergence in the multivariate entropic free central limit theorem.
• To compute free Stein kernels for important families of self-adjoint operators.
• To provide a framework for non-microstates free entropy and Fisher information via semigroup and convexity arguments.

Proposed method

• Introduces a free Stein kernel as a non-commutative analog of the classical Stein kernel, tailored to the semicircular law.
• Uses the free Stein kernel to derive a free HSI inequality, improving the free logarithmic Sobolev inequality.
• Applies semigroup techniques and convexity properties of potentials to derive functional inequalities for non-microstates free entropy.
• Employs the conjugate variable method and Jacobian calculus in free probability to relate Fisher information and entropy.
• Uses the inverse function theorem and Jacobian determinant bounds to control entropy changes under free transformations.
• Applies the inequality $ t - 1 - \log t \geq 0 $ to establish positivity of entropy differences under free transformations.

Experimental results

Research questions

• RQ1 Can a free Stein kernel be defined relative to the semicircular law to extend classical Stein kernel methods to free probability?
• RQ2 How can the free logarithmic Sobolev inequality be improved using a free HSI-type inequality?
• RQ3 What is the rate of convergence in the multivariate entropic free central limit theorem under this improved inequality?
• RQ4 Which families of self-adjoint operators admit explicit free Stein kernel computations?
• RQ5 How do free Fisher information and entropy behave under free transformations with positive Jacobian?

Key findings

• The free HSI inequality improves the free logarithmic Sobolev inequality of Biane and Speicher, providing a tighter bound on free entropy in terms of free Fisher information.
• A rate of convergence in the multivariate entropic free central limit theorem is established via the improved inequality, quantifying the speed of convergence to the free semicircular law.
• The free Stein kernel is explicitly computed for several relevant families of self-adjoint operators, including those with convex potentials.
• The inequality $ \chi(F(Y)|V) \geq \chi(Y|V) $ holds for free Gibbs states under convexity and Jacobian conditions, confirming entropy monotonicity.
• The method yields a sharp bound $ \chi(F(Y)|V) - \chi(Y|V) \leq \frac{1}{2\rho} \Phi^*(F(Y)|V) $ under uniform convexity of the potential, linking entropy difference to Fisher information.
• The proof relies on the identity $ \tau \otimes \tau^{op} \circ \text{Tr}(JG - 1 - \log|JG|) \geq 0 $, which ensures positivity of entropy changes under free transformations.