[Paper Review] $L^{2}$-spectral gaps, weak-reversible and very weak-reversible Markov chains

Achim Wuebker, Zakhar Kabluchko|ArXiv.org|Aug 6, 2009

Markov Chains and Monte Carlo Methods16 references3 citations

TL;DR

This paper establishes necessary and sufficient conditions for the existence of an $L^2$-spectral gap in positive recurrent general state space Markov chains without requiring reversibility. By introducing weak-reversible and very weak-reversible conditions and analyzing isoperimetric constants associated with $P^2$, $P^*P$, and $PP^*$, the authors generalize spectral gap criteria beyond the reversible case, providing a framework applicable to non-reversible chains using $L^2$-norm estimates and spectral theory.

ABSTRACT

The theory of $L^2$-spectral gaps for reversible Markov chains has been studied by many authors. In this paper we consider positive recurrent general state space Markov chains with stationary transition probabilities. Replacing the assumption of reversibility by a less strong one, we still obtain a simple necessary and sufficient condition for the spectral gap property of the associated Markov operator in terms of isoperimetric constant. Moreover, we define a new sequence of isoperimetric constants which provides a necessary and sufficient condition for the existence of a spectral gap in a very general setting. Finally, these results are used to obtain simple sufficient conditions for the existence of a spectral gap in terms of the first and second order transition probabilities.

Motivation & Objective

• To extend $L^2$-spectral gap theory beyond reversible Markov chains, which have been the focus of prior work.
• To address the lack of general conditions for spectral gaps in non-reversible chains due to the failure of Dirichlet form techniques and lack of canonical generalizations.
• To define weaker structural conditions—weak-reversibility and very weak-reversibility—that still allow for spectral gap characterization.
• To provide a necessary and sufficient condition for spectral gaps using a new sequence of isoperimetric constants in a general setting.
• To derive sufficient conditions for spectral gaps based on first- and second-order transition probabilities, enabling practical verification.

Proposed method

• Introduce weak-reversible and very weak-reversible Markov chains via Radon-Nikodym derivatives of joint measures $Q^{(n)}$ and $\tilde{Q}^{(n)}$, generalizing reversibility.
• Define isoperimetric constants $k_n(A)$ and $k_{P^{*n}P^n}(A)$ to characterize spectral gap properties in terms of transition probabilities across sets.
• Use the adjoint operator $P^*$ to define $P^{*n}P^n$ and relate its spectral gap to the original chain’s behavior.
• Establish a representation of the quadratic form $\langle f, (Id - P^*P)f \rangle_\pi$ via a measure $\mu$, enabling spectral analysis.
• Compare the isoperimetric constants of $P^2$, $P^*P$, and $PP^*$ to derive bounds on the spectral gap.
• Apply the spectral mapping theorem and norm estimates to show that $\|Id - P\|_{L^2} \geq \sqrt{\kappa/8} \cdot k_{P+P^* - P^*P}$, linking spectral gap to isoperimetry.

Experimental results

Research questions

• RQ1 Can a necessary and sufficient condition for the $L^2$-spectral gap be established for non-reversible Markov chains without assuming reversibility?
• RQ2 How do weak-reversible and very weak-reversible conditions generalize the classical reversibility assumption in spectral gap theory?
• RQ3 What is the role of isoperimetric constants $k_n$ and $k_{P^{*n}P^n}$ in characterizing spectral gaps in general Markov chains?
• RQ4 Can the spectral gap of $P$ be estimated using the isoperimetric constants of $P^2$, $P^*P$, and $PP^*$?
• RQ5 What sufficient conditions on first- and second-order transition probabilities ensure the existence of a spectral gap?

Key findings

• A necessary and sufficient condition for the $L^2$-spectral gap is established using the isoperimetric constant $k_{P^*P}$, extending results from reversible chains.
• The spectral gap exists if and only if $k_{P^*P} > 0$, which holds under very weak-reversibility when $\operatorname{esssup}_y \left\| \frac{dQ^{(n)}}{d\tilde{Q}^{(n)}}(\cdot,y) \right\|_{L^q(p(y,\cdot))} < \infty$ for some $q \in (1,\infty]$.
• For very weak-reversible chains, the spectral gap is characterized by the positivity of $k_{P^*P}$, which is bounded below by $k^2$, where $k$ is the standard isoperimetric constant.
• The paper proves that $k_{P+P^* - P^*P} \geq k^2$, linking the spectral gap of the symmetrized operator to the original chain’s isoperimetry.
• Sufficient conditions for spectral gaps are derived from the first- and second-order transition kernels, enabling practical verification without full reversibility.
• The example of Häggström shows that even geometric ergodicity does not guarantee a spectral gap, and $P$ lacks a spectral gap when $p^{*n}p^n((3n,2n),(3n,2n)) = 1$ for all $n$, confirming the necessity of the conditions.