Home Statistical Dictionary Aperiodic Markov Chain

Aperiodic Markov Chain

A Markov chain in which no state is forced to return only at multiples of some fixed cycle length greater than 1, ruling out mechanical, clockwork-like periodic behavior.

Stochastic ProcessesIntermediateFree

In Plain English

Some Markov chains march to a beat: a state might only ever be revisited every 2 steps, or every 3, like a strict schedule. That's periodicity, and a chain that avoids it entirely, where returns can happen at essentially any timing rather than a fixed rhythm, is called aperiodic. Aperiodicity matters because a chain stuck in lockstep never actually settles into a single stable long-run distribution, it just keeps cycling between a few distributions in sync with its period.

Definition

The period of a state $i$ is $d (i) = g cd {n \geq 1 : p_{ii}^{(n)} > 0}$ , the greatest common divisor of all the return-time lengths with positive probability. A state is aperiodic if $d (i) = 1$ ; a Markov chain is aperiodic if every state has period 1 (in an irreducible chain, period is a class property, so it suffices to check one state). A chain that instead has period $d > 1$ cycles deterministically through $d$ disjoint groups of states, one per time step, and consequently $p_{ii}^{(n)} = 0$ whenever $n$ is not a multiple of $d$ , meaning $P^{n}$ never converges to a fixed limiting matrix as $n \to \infty$ . A simple example of a periodic chain is a random walk on a bipartite graph, since every step alternates between the two parts, giving period 2.

Formula

d (i) = g cd {n \geq 1 : p_{ii}^{(n)} > 0}

Notation

d (i)

—Period of state $i$: the gcd of all return-time lengths with positive probability

Properties

A periodic chain doesn't fail to have a stationary distribution, it can still have one, but the chain's actual state distribution at time $n$ keeps oscillating around that stationary distribution forever rather than converging to it, unlike an aperiodic chain.
The most common way a chain becomes periodic is structural: any bipartite-like alternation, or any state machine that must pass through a fixed sequence of 'phases' before repeating, introduces a period greater than 1 by construction.
In MCMC algorithm design, ensuring the constructed Markov chain is aperiodic (often by allowing a small self-transition probability, $p_{ii} > 0$ , which automatically forces $d (i) = 1$ ) is a standard, easy fix to guarantee the chain converges to its target stationary distribution.

At a Glance

Defining conditionevery state has period

d (i) = 1

: no forced cyclical return pattern

Periodic examplea random walk on a bipartite graph has period 2 (alternates between two sides)

Class propertyin an irreducible chain, all states share the same period

Why it mattersaperiodicity (with irreducibility, recurrence) ensures

P^{n}

converges to a fixed limit

Last updated September 8, 2026← Back to the Dictionary