Analysis · Concept hub

Ergodic Theory

1931 CE20th-century Hungary and the United States (von Neumann and Birkhoff)

Concept

Time average equals space average. Von Neumann and Birkhoff (1931) — foundational in statistical physics, chaos, and dynamical systems.

Understand it in one breath

"Can a long-time average along one trajectory reveal the average over the whole space?" For an ergodic measure-preserving system and a suitable observable, the time average equals the space average for almost every starting point. Not every dynamical system is ergodic, and MCMC convergence likewise requires conditions such as irreducibility, aperiodicity, and a stationary distribution.

At a glance

System	Time average	Space average	Ergodic?
Fair coin tosses	Heads ≈ 1/2	P(heads) = 1/2	✓
Irreducible, aperiodic Markov chain	Same proportions along a long trajectory	Stationary distribution	✓ (under conditions)
Collisions among gas molecules	Average motion of one molecule ≈	Average over all molecules	Ergodic hypothesis
Invariant subspace preserved	Trapped in one region	Full-space measure	✗ (an invariant exists)
MCMC convergence	Sample average → true average	Bayesian inference	✓ (for that algorithm)

Under ergodic conditions, long-time averages connect to space averages. Each MCMC algorithm still requires its own convergence conditions to be checked.

Key formula

\lim_{T \to \infty} \frac{1}{T} \int_0^T f(\phi_t x)\, dt = \int f\, d\mu

Modern applications

MCMC Bayesian inference, simulations in statistical physics, and the analysis of chaotic systems.

Beyond MathVoyage

Loading…