ergodicity
/ er-GOD-ic-it-ee /
Ergodicity is the irreducibility condition that makes "time average equals space average" come out as a single clean number. A measure-preserving system is ergodic when it cannot be split into two non-trivial pieces that the dynamics leave separate; the orbit of almost every point eventually visits everywhere the measure allows, so following one typical trajectory reveals the whole space.
Formally, (Omega, F, P, T) is ergodic if every invariant set A (T^(-1) A = A, or A almost-invariant) has P(A) in {0, 1}. Equivalently the invariant sigma-algebra is trivial; equivalently every T-invariant function is constant a.e.; equivalently, for all A, B in F, the Cesaro average (1/n) sum_(k=0)^(n-1) P(T^(-1)^(-k) A intersect B) -> P(A) P(B) (averaged asymptotic independence). The name comes from Boltzmann's ergodic hypothesis in statistical mechanics — the idea that a typical mechanical trajectory spends, in the long run, a fraction of time in each region proportional to that region's measure. Ergodicity is the rigorous form of exactly that hypothesis.
Why it matters: ergodicity is precisely the hypothesis under which Birkhoff's time averages collapse to the constant E[f]. Without it the limit is the genuinely random variable E[ f given I ], differing from one ergodic component to another. Ergodicity is the weakest member of the mixing hierarchy — it is implied by weak mixing, which is implied by mixing — and it is the right notion for laws of large numbers, equidistribution, and the convergence of empirical averages of a stationary process. A caution: ergodicity is a property of the pair (T, P), not of T alone; the same map can be ergodic for one invariant measure and not for another.
The rotation T(x) = x + alpha mod 1 on [0,1) is ergodic exactly when alpha is irrational: an invariant function expanded in Fourier modes c_n e^(2 pi i n x) must satisfy c_n (e^(2 pi i n alpha) - 1) = 0, forcing c_n = 0 for n not 0, so f is constant. If alpha = p/q is rational, orbits are periodic and the system is non-ergodic.
Irrational rotation is the textbook ergodic-but-not-mixing system — a Fourier argument settles it in one line.
Ergodic does not mean random: the irrational rotation is ergodic yet perfectly predictable (zero entropy, not even weakly mixing). Ergodicity only guarantees averaged statistics, not chaos.