Current browse context:
math.PR
Change to browse by:
References & Citations
Mathematics > Probability
Title: Spectral Telescope: Convergence Rate Bounds for Random-Scan Gibbs Samplers Based on a Hierarchical Structure
(Submitted on 24 Aug 2022 (v1), last revised 14 Oct 2022 (this version, v3))
Abstract: Random-scan Gibbs samplers possess a natural hierarchical structure. The structure connects Gibbs samplers targeting higher dimensional distributions to those targeting lower dimensional ones. This leads to a quasi-telescoping property of their spectral gaps. Based on this property, we derive three new bounds on the spectral gaps and convergence rates of Gibbs samplers on general domains. The three bounds relate a chain's spectral gap to, respectively, the correlation structure of the target distribution, a class of random walk chains, and a collection of influence matrices. Notably, one of our results generalizes the technique of spectral independence, which has received considerable attention for its success on finite domains, to general state spaces. We illustrate our methods through a sampler targeting the uniform distribution on a corner of an $n$-cube.
Submission history
From: Qian Qin [view email][v1] Wed, 24 Aug 2022 04:53:02 GMT (147kb,D)
[v2] Fri, 26 Aug 2022 03:37:21 GMT (147kb,D)
[v3] Fri, 14 Oct 2022 02:27:38 GMT (28kb)
Link back to: arXiv, form interface, contact.