On the central limit question for strictly stationary, reversible Markov chains

Richard C. Bradley · 2025

Read On the central limit question for strictly stationary, reversible Markov chains by Richard C. Bradley (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-12951.txt.

Introduction

Richard C Bradley wrote a 2025 mathematics paper probing when the central limit theorem holds for reversible Markov chains, the random processes that step between states with no memory. The theorem, which promises bell-shaped averages, can fail even for well-behaved stationary chains. Reversibility, where the process looks the same run backward, sometimes rescues it and sometimes does not.

Bradley constructs several families of chains that are stationary, reversible, and tame by standard measures, yet break the theorem. Known results showed reversibility helps a lot when mixing is fast. His counterexamples sharpen the boundary: with slow, power-type mixing, the extra leverage nearly vanishes. The paper maps where theory can still hope for help.

Catalog record

CREATORRichard C. Bradley
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE138 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-12951

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

On the central limit question for strictly stationary, reversible Markov chains Richard C. Bradley Abstract This paper will provide several classes of strictly stationary, countable-state, irreducible, aperiodic Markov chains that are reversible and have finite second moments, such that the central limit theorem fails to hold. The main purpose is to examine the extent to which, for the development of central limit theory for strictly stationary Markov chains (and functions of them) under the strong mixing and absolute regularity conditions, the property of reversibility (if it holds) can provide extra leverage. It is known, partly as a by-product of research done by Roberts, Rosenthal, and Tweedie in two papers in 1997 and 2001, that for the case of exponential mixing rates, reversibility provides notable extra leverage of that kind. In contrast, a class of counterexamples in a paper of Doukhan, Massart, and Rio in 1994 showed (implicitly) that for the case of power-type mixing rates, reversibility apparently provides almost no such extra leverage. Further perspective on that latter fact will be provided by some counterexamples in this paper. Other counterexamples here will (indirectly) provide some tentative, uncertain evidence for the possibility that for mixing rates that are “between” power-type and exponential (for example, sub-exponential), reversibility may in fact provide some small but nontrivial extra leverage. Keywords: strictly stationary, reversible Markov chain; strong mixing; absolute regularity; central limit question AMS 2020 Mathematics Subject Classifications: 60J10, 60G10 1 Introduction In this paper, for simplicity, every Markov chain (i) is indexed by the set \mathbb{Z} of all integers, (ii) is strictly stationary, and (iii) has as its state space either the set \mathbb{R} of all real numbers (accompanied by its Borel \sigma -field \mathcal{R} ) or a finite or countably infinite subset of \mathbb{R} . Over twenty years ago, in the papers of Roberts and Rosenthal [ 27 ] and Roberts and Tweedie [ 28 ] together, some special theory for Markov chains was developed which included (among other things) some striking results for strictly stationary Markov chains that are reversible. Of course the term “reversible” simply means that the distribution of the Markov chain as a whole is invariant under a reversal of the “direction of time”. (More on that in Section 2 .) It is well known that some key aspects of the theory in those two papers [ 27 ] and [ 28 ] can be reformulated in the terminology of the dependence coefficients associated with certain strong mixing conditions; see e.g. [ 20 ] , [ 23 ] , [ 24 ] , or the exposition in [ 7 ] . It is well known that as a consequence of the aforementioned special theory developed in the papers [ 27 ] and [ 28 ] , together with classic central limit theorems (such as in Rosenblatt [ 30 ] or Lifshits [ 21 ] ) for (functions of) strictly stationary Markov chains that satisfy the \rho -mixing condition, the following holds: For a (function of a) strictly stationary, reversible Markov chain, if the second moments are finite and the variances of the partial sums diverge to \infty , then \beta -mixing (absolute regularity) with a mixing rate at least exponentially fast, is sufficient to ensure that the central limit theorem holds. It is also known, from counterexamples in [ 3 ] , [ 12 ] , [ 15 ] , and (after a very careful examination of certain hidden details) [ 16 ] , that if the assumption of reversibility were omitted altogether in that context, then the central limit theorem may fail to hold. Thus for (functions of) strictly stationary Markov chains that have finite second moments and satisfy “exponential \beta -mixing”, the extra property of reversibility, if satisfied, provides extra leverage for the central limit theorem to hold. Cuny and Lin [ 12 ] examined that general theme in much detail, under “exponential \alpha -mixing” and with the assumption of reversibility replaced by related weaker assumptions. This paper here will focus on the construction of counterexamples intended to illustrate “what can go wrong” with respect to the central limit theorem under the seemingly “nicest” (i.e. most favorable) assumptions. Accordingly, we shall retain our focus here on reversibility and (in the counterexamples) on \beta -mixing (which is stronger than \alpha -mixing). More on all that in Section 2 . In this paper, the following question will be examined: Suppose that for (a function of) a strictly stationary Markov chain, (i) absolute regularity holds with a mixing rate that is slower than exponential, and (ii) the (invariant) marginal distribution is assumed to satisfy a “moment” condition (or a more general “tail” condition) that is stronger than finite second moment. To what extent does the extra property of reversibility, if satisfied, provide extra leverage for the central limit theorem to hold? In this paper, some counterexamples will be constructed, leading to the following tentative assessments regarding that question: For the case where the random variables in (the function of) the Markov chain are bounded or at least have finite absolute moments of some order higher than second moments, the answer is in some sense “practically no extra leverage”. That was shown to hold in some sense by a class of counterexamples in the paper of Doukhan et al. [ 14 ] . Further nuances on that will be provided by counterexamples constructed here in this paper. For the case where the second moments are finite but absolute moments of order 2+\delta are infinite for all \delta>0 , the answer remains uncertain, but from counterexamples constructed in this paper, it seems plausible that the answer might be something like “some small but nontrivial extra leverage”. That musing will be put in a more precise form in Remark 5.9 in Section 5 in this paper, in connection with some counterexamples described there. The rest of this paper is organized as follows: Section 2 will give a review of certain mixing conditions for Markov chains, along with further relevant technical details on the discussion above. Sections 3 and 4 will describe counterexamples that respectively deal with two different aspects of the question above for the case of strictly stationary, reversible Markov chains that are bounded. Section 5 will describe some counterexamples that deal broadly with the question above for the case of strictly stationary, reversible Markov chains that are not bounded. Section 6 will provide some preliminary work for subsequent proofs. Sections 7 , 8 , and 9 will then give proofs of the properties of the counterexamples described in Sections 3 , 4 , and 5 respectively. 2 Mixing conditions and Markov chains In this section, we shall lay out some notations and review some background facts regarding mixing conditions and Markov chains. Most of the exposition in this section will be taken practically verbatim from the introductions of recent papers by the author [ 8 ] [ 10 ] . Notations 2.1 . First, here are a few really basic notations: \mathbb{N} denotes the set of all positive integers, \mathbb{Z} denotes the set of all integers, \mathbb{R} denotes the set of all real numbers, and \mathcal{R} denotes the Borel \sigma -field on \mathbb{R} . The usual notations such as \mathbb{R}^{\mathbb{Z}} and

Read the complete text