Sharp detection of low-dimensional structure in probability measures via dimensional logarithmic Sobolev inequalities
Name
Sharp detection of low-dimensional structure in probability measures via dimensional logarithmic Sobolev inequalities.pdf
Description
Accepted version
Size
1.75 MB
Format
Adobe PDF
Checksum (MD5)
b4ad6834bb943f0f16c67bd0952f49d8
Author(s)
Li, Matthew TC
Cui, Tiangang
Li, Fengyi
Marzouk, Youssef
Zahm, Olivier
Date Issued
June 27, 2025
Journal
Information and Inference: A Journal of the IMA
Publisher
Oxford University Press (OUP)
Version
Author's final manuscript
Abstract
Abstract
Identifying low-dimensional structure in high-dimensional probability measures is an essential pre-processing step for efficient sampling. To identify this structure, we approximate the target measure as a perturbation of an arbitrary reference measure along a few directions in $\mathbb{R}^{d}$. These directions are determined by minimizing an upper bound on the Kullback–Leibler (KL) divergence between the target and its approximation. Our contribution improves upon previous works by leveraging dimensional logarithmic Sobolev inequalities to refine the bound on the KL divergence. These inequalities lead to a uniformly tighter bound on the KL divergence, thereby enhancing the identification of the most significant perturbation directions. In particular, when the target and reference are both Gaussian, minimizing the resulting bound is equivalent to minimizing the KL divergence. We further demonstrate the applicability of this analysis to the squared Hellinger distance, where analogous reasoning shows that the dimensional Poincaré inequality offers improved bounds.
Identifying low-dimensional structure in high-dimensional probability measures is an essential pre-processing step for efficient sampling. To identify this structure, we approximate the target measure as a perturbation of an arbitrary reference measure along a few directions in $\mathbb{R}^{d}$. These directions are determined by minimizing an upper bound on the Kullback–Leibler (KL) divergence between the target and its approximation. Our contribution improves upon previous works by leveraging dimensional logarithmic Sobolev inequalities to refine the bound on the KL divergence. These inequalities lead to a uniformly tighter bound on the KL divergence, thereby enhancing the identification of the most significant perturbation directions. In particular, when the target and reference are both Gaussian, minimizing the resulting bound is equivalent to minimizing the KL divergence. We further demonstrate the applicability of this analysis to the squared Hellinger distance, where analogous reasoning shows that the dimensional Poincaré inequality offers improved bounds.
Terms of Use
Creative Commons Attribution-NonCommercial-ShareAlike
Persistent DSpace Link
DOI of Published Version
10.1093/imaiai/iaaf021
