Issue |
ESAIM: PS
Volume 2, 1998
|
|
---|---|---|
Page(s) | 1 - 21 | |
DOI | https://doi.org/10.1051/ps:1998101 | |
Published online | 15 August 2002 |
Une variante de l'inégalité de Cheeger pour les chaînes de Markov finies
A variant of the Cheeger's inequality for finite Markov chains
(miclo@cict.fr)
Sur un ensemble fini, on s'intéresse aux minorations linéaires du trou spectral d'un noyau markovien réversible, en terme de la constante isopérimétrique associée. On montre que la constante optimale est l'inverse du cardinal de l'ensemble moins un, mais on verra aussi comment il est possible de l'améliorer dans certaines situations particulières (arbres pointés radiaux à nombre fini de générations). Une application des inégalités précédentes est de retrouver immédiatement le comportement à basse température des trous spectraux associés aux algorithmes de recuit simulé généralisés finis. Une approche similaire permet également d'obtenir très simplement une caractérisation de l'énergie virtuelle due à Trouvé.
Abstract
On a finite set, we study low bounds on the spectral gap of a reversible Markovian kernel which are linear with respect to the associated isoperimetric constant. We will show that the best constant for that variant of the Cheeger's inequality is the inverse of the cardinal of the set minus 1, but we will also see how to improve it on special settings (e.g. rooted radial trees with a finite number of generations). As an application, the previous bounds enable one to obtain at once the asymptotic behaviour at vanishing temperature of the spectral gaps associated to finite generalized simulated annealing algorithms. In the same spirit, one can also very easily recover the caracterization of the virtual energy given by Trouvé.
Key words: Finite reversible Markov kernels / isoperimetric constant / (modified) spectral gap / (optimal) linear variants of the Cheeger's inequality / generalized simulated annealing algorithms.
© EDP Sciences, SMAI, 1998
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.