English

Estimating the Mixing Time of Ergodic Markov Chains

Statistics Theory 2022-08-17 v4 Machine Learning Machine Learning Statistics Theory

Abstract

We address the problem of estimating the mixing time tmixt_{\mathsf{mix}} of an arbitrary ergodic finite-state Markov chain from a single trajectory of length mm. The reversible case was addressed by Hsu et al. [2019], who left the general case as an open problem. In the reversible case, the analysis is greatly facilitated by the fact that the Markov operator is self-adjoint, and Weyl's inequality allows for a dimension-free perturbation analysis of the empirical eigenvalues. As Hsu et al. point out, in the absence of reversibility (which induces asymmetric pair probabilities matrices), the existing perturbation analysis has a worst-case exponential dependence on the number of states dd. Furthermore, even if an eigenvalue perturbation analysis with better dependence on dd were available, in the non-reversible case the connection between the spectral gap and the mixing time is not nearly as straightforward as in the reversible case. Our key insight is to estimate the pseudo-spectral gap γps\gamma_{\mathsf{ps}} instead, which allows us to overcome the loss of symmetry and to achieve a polynomial dependence on the minimal stationary probability π\pi_\star and γps\gamma_{\mathsf{ps}}. Additionally, in the reversible case, we obtain simultaneous nearly (up to logarithmic factors) minimax rates in tmixt_{\mathsf{mix}} and precision ε\varepsilon, closing a gap in Hsu et al., who treated ε\varepsilon as constant in the lower bounds. Finally, we construct fully empirical confidence intervals for γps\gamma_{\mathsf{ps}}, which shrink to zero at a rate of roughly 1/m1/\sqrt{m}, and improve the state of the art in even the reversible case.

Keywords

Cite

@article{arxiv.1902.01224,
  title  = {Estimating the Mixing Time of Ergodic Markov Chains},
  author = {Geoffrey Wolfer and Aryeh Kontorovich},
  journal= {arXiv preprint arXiv:1902.01224},
  year   = {2022}
}

Comments

COLT'19 conference manuscript, with minor fixes