Mixing and Ergodicity — Mathematics Knowledge System

Overview

Ergodicity and mixing are properties of measure-preserving dynamical systems describing how thoroughly orbits explore phase space. An ergodic system cannot be decomposed into two invariant sets of positive measure -- time averages equal space averages. A mixing system is stronger: correlations between sets A and B decay to zero as time increases (the system 'forgets' its past). The hierarchy runs: mixing implies weak mixing implies ergodicity, with each implication strict. These properties determine the statistical behaviour of the system in the long run.

Intuition

Imagine stirring cream into coffee. 'Ergodic' means the cream eventually reaches every part of the coffee (no region stays pure coffee forever). 'Mixing' means that for any two regions A and B, after enough stirring, the proportion of cream-labelled fluid in A that started in B approaches the overall proportion of B -- the system loses memory of where things started. Weak mixing is between these: there are no persistent cycles, but the convergence of correlations is in a weaker (Cesaro) sense.

Formal Definition

Definition

Let (X, mu, T) be a measure-preserving system. T is ergodic if T^{-1}(A) = A implies mu(A) = 0 or 1. T is (strongly) mixing if for all measurable A, B: mu(T^{-n}(A) cap B) -> mu(A)*mu(B) as n -> inf. T is weakly mixing if (1/N)*sum_{n=0}^{N-1} |mu(T^{-n}A cap B) - mu(A)mu(B)| -> 0. Equivalently: T is weak mixing iff T x T is ergodic.

T \text{ ergodic} \iff \frac{1}{N}\sum_{n=0}^{N-1} f(T^n x) \to \int f\, d\mu \quad \mu\text{-a.e.}

Ergodic theorem

T \text{ mixing} \iff \mu(T^{-n}A \cap B) \to \mu(A)\mu(B) \quad \forall A,B

Strong mixing

T \text{ weak mixing} \iff \frac{1}{N}\sum_{n=0}^{N-1}|\mu(T^{-n}A \cap B) - \mu(A)\mu(B)| \to 0

Weak mixing (Cesaro)

\text{mixing} \Rightarrow \text{weak mixing} \Rightarrow \text{ergodic}

Hierarchy of mixing properties

Notation

Notation	Meaning
$(X, \mu, T)$	Measure-preserving dynamical system
$\mu(A \cap B)$	Measure of intersection of A and B
$T^{-n}(A)$	Pre-image of A under n iterations of T

Theorems

Theorem 1: Birkhoff Pointwise Ergodic Theorem

If (X, mu, T) is an ergodic measure-preserving system and f in L^1(X, mu), then for mu-almost every x: (1/N)*sum_{n=0}^{N-1} f(T^n x) -> integral f dmu as N -> infinity. Time averages equal space averages for a.e. initial point.

Theorem 2: Spectral Characterisation of Mixing

A measure-preserving transformation T is (strongly) mixing iff for every f, g in L^2 with integral f = integral g = 0: <f o T^n, g> -> 0 as n -> infinity. T is weakly mixing iff T has no non-trivial eigenvalues (the only eigenfunctions of the Koopman operator U_T are constants).

Theorem 3: Mixing Implies Ergodicity (But Not Vice Versa)

Every mixing system is ergodic: if mu(T^{-n}A cap A) -> mu(A)^2 and mu(T^{-1}A) = mu(A), taking A with T^{-1}A = A gives mu(A)^2 = mu(A), so mu(A) = 0 or 1. The converse fails: irrational rotations are ergodic but not mixing.

Worked Examples

1
Ergodicity: The Koopman operator has eigenfunctions e^{2*pi*i*n*x} with eigenvalues e^{2*pi*i*n*alpha}. Since alpha is irrational, these eigenvalues are all distinct and not equal to 1 (for n != 0). By the spectral criterion, T is ergodic.
$U_T e^{2\pi inx} = e^{2\pi in\alpha} e^{2\pi inx}$
2
Not mixing: Take A = [0, 1/2). Then mu(T^{-n}A cap A) does NOT converge to 1/4 -- it oscillates. Concretely, if n*alpha is close to 0 mod 1, T^n A nearly overlaps A (mu(T^{-n}A cap A) close to 1/2), while if n*alpha close to 1/2, they nearly miss.
$\mu(T^{-n}A \cap A) \not\to \mu(A)^2 = 1/4$

✓ Answer

Irrational rotations are ergodic (Weyl equidistribution) but not mixing (correlations don't decay -- the spectrum is pure point).

Practice Problems

Mediumfree response

Give an example of a weakly mixing system that is not strongly mixing, and explain the distinction.

Common Mistakes

Common Mistake

Thinking ergodic implies mixing.

Ergodic is the weakest condition: irrational rotations are ergodic but not even weakly mixing. The full hierarchy is: Bernoulli => K-system => mixing => weak mixing => ergodic, with all implications strict.

Quiz

Which implication in the mixing hierarchy is correct?

Ergodic => mixing => weak mixingMixing => weak mixing => ergodicWeak mixing => mixing => ergodicErgodic => weak mixing => mixing

An irrational rotation is NOT mixing because:

It has zero entropyIts Koopman operator has non-trivial eigenvalues (pure point spectrum), so correlations oscillateIt fails to preserve Lebesgue measureIt is not transitive

Historical Background

The ergodic hypothesis was proposed by Boltzmann in the 1870s as a foundation for statistical mechanics: a gas molecule visits all accessible states with equal frequency over long times. The mathematical formulation was given by Birkhoff (pointwise ergodic theorem, 1931) and von Neumann (mean ergodic theorem, 1932). Mixing was introduced by Hopf (1937) in the context of geodesic flows. The modern hierarchy (ergodic -- weak mixing -- mixing -- K-system -- Bernoulli) was developed by Rohlin, Halmos, and the Russian school.

1870s
Boltzmann proposes the ergodic hypothesis for statistical mechanics
Ludwig Boltzmann
1931
Birkhoff proves the pointwise ergodic theorem
George Birkhoff
1932
von Neumann proves the mean ergodic theorem
John von Neumann
1937
Hopf introduces strong mixing for geodesic flows
Eberhard Hopf

Summary

Ergodicity: no non-trivial invariant sets; time averages = space averages for a.e. x (Birkhoff).
Weak mixing: Cesaro averages of correlations decay to zero; T x T is ergodic.
Strong mixing: correlations mu(T^{-n}A cap B) -> mu(A)mu(B) for all A, B.
The hierarchy: mixing => weak mixing => ergodic (all implications strict).

References

BookWalters, P. An Introduction to Ergodic Theory. Springer, 1982.
BookPetersen, K. Ergodic Theory. Cambridge, 1989.
WebsiteWikipedia -- Mixing (mathematics)