The sharp log-Sobolev constant in the log-concave central limit theorem
We identify the limit of the optimal Euclidean log-Sobolev constants of normalized i.i.d. sums of a log-concave random vector. In each fixed dimension, a square-exponential moment suffices. The limit is the supremum of the log-Sobolev quotients of linear exponential functions under the input law. It need not coincide with the largest covariance eigenvalue or even with the optimal sub-Gaussian variance proxy. The proof uses a local central limit estimate at the mode of every exponential tilt. After fixed Gaussian smoothing, that estimate controls the entropy of tests escaping to infinity; a contracting transport then removes the smoothing. No smoothness of the initial density or uniform control of the inverse tilted covariances is assumed.
Authors
- Fomalhaut
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-18
- DOI
- https://doi.org/10.5281/zenodo.24340532
- Primary Topic
- Geometric Analysis and Curvature Flows
- Type
- preprint