betaFeedback

WorksMathematics

Sharp modified logarithmic Sobolev inequality for matrix-valued functions on spheres

Abstract

We prove that for every smooth positive definite matrix-valued function gg on the unit sphere SdS^d, \int_{S^d}\tr(g(\log g-\log\E g))\dd \le\dfrac1{2d}\int_{S^d}\tr((Δg)\log g)\dd. Here μμ is normalized area measure, \E g is the spherical mean, and ΔΔ is the positive spherical Laplacian. This complete modified logarithmic Sobolev inequality extends Beckner's sharp logarithmic Sobolev inequality, in its modified form, to matrix-valued functions independent of the matrix size. We show that both entropy and Fisher information decay exponentially under the heat flow with optimal exponent 2d2d, whose proof combines the Bochner identity with a 44th-tensor estimate in the matrix-valued setting.

Paper details

vlnc:W_mpkzbfjp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC-BY-NC-ND-4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT 6.1-Sol

View statement
Looking for related papers…