Dynamical Systems

Ergodic Theory: Mixing and KS Entropy

Ergodic theory asks: when do time observations faithfully represent the whole system? Without this, statistical mechanics, MCMC algorithms, and financial models would lack rigorous foundations. Birkhoff's 1931 theorem vindicated Boltzmann's 1871 hypothesis.

  • Molecular dynamics: ergodicity justifies replacing ensemble averages by time averages when simulating protein folding - saves enormous computational cost
  • MCMC sampling: Markov Chain Monte Carlo works because the chain is ergodic, guaranteeing convergence to the target distribution in Bayesian inference
  • Algorithmic trading: mixing-rate estimates test market randomness; non-ergodic markets invalidate standard portfolio theory assumptions
  • Error-correcting codes: Shannon-McMillan-Breiman theorem connects KS entropy to optimal code rates for ergodic sources

Предварительные знания

  • Measure theory
  • Measure-preserving transformations
  • Lyapunov exponents
  • Previous lesson: ds-25

Ergodic theorem and invariant measures

Boltzmann's statistical mechanics rests on the ergodic hypothesis: time averages equal space averages. George Birkhoff proved this rigorously in 1931. A gas with 10^23 molecules makes about 10^30 collisions per second - a prototypical ergodic system where any observable's time average converges to its ensemble average.

What does Birkhoff's ergodic theorem state?

Birkhoff (1931): for an ergodic system with invariant probability measure mu and integrable f, lim_{T→∞} (1/T) ∫_0^T f(phi_t x) dt = ∫ f dmu for mu-almost every x. Time and space averages coincide.

0

1

Sign In

Mixing and decay of correlations

Pesin's theorem: h_KS(T) = sum of positive Lyapunov exponents for ergodic diffeomorphisms. This connects information-theoretic entropy with exponential orbit divergence.

Ergodicity and mixing. Ergodicity is like a cocktail that, given enough time, visits every region: shake long enough and every molecule will eventually be everywhere. Mixing is stronger: any dye spot spreads uniformly across the whole glass in finite time, regardless of initial shape.

How does strong mixing differ from ergodicity?

Ergodicity: time and space averages coincide. Strong mixing: for any sets A, B we have lim_{t→∞} mu(phi_t(A) ∩ B) = mu(A) mu(B). Mixing is strictly stronger than ergodicity (mixing → ergodic, but not the converse).

Kolmogorov-Sinai entropy

Irrational circle rotation T(x) = x + alpha (mod 1) is ergodic by Weyl's theorem (orbits are dense), but not mixing: the function e^{2*pi*i*x} has autocorrelation e^{2*pi*i*n*alpha} which stays bounded away from zero.

Birkhoff-Khinchin spectral decomposition: every T-invariant measure decomposes uniquely into ergodic components. This is the measure-theoretic analog of decomposing a representation into irreducibles.

What is the physical meaning of positive Kolmogorov-Sinai entropy?

h_KS > 0 is a quantitative signature of chaos. Link to Lyapunov exponents (Pesin's formula): h_KS = sum_{lambda_i > 0} lambda_i. Each unit of entropy per hour means losing one bit of information about the initial condition per hour. The logistic map at r=4: h_KS = ln(2) ≈ 0.693.

Connections to other areas

Ergodic theory unifies dynamical systems, probability theory, and mathematical physics through invariant measures, mixing rates, and entropy.

  • Probability theory — Related topic
  • Information theory — Related topic
  • Quantum mechanics — Related topic
  • Group theory — Related topic

Итоги

  • Birkhoff theorem: time average of f along T-orbits equals space average integral f dmu, for mu-a.e. starting point
  • Ergodicity: T is ergodic iff every T-invariant set has measure 0 or 1
  • Mixing: mu(A cap T^{-n}B) -> mu(A)*mu(B); correlations decay - strictly stronger than ergodicity
  • KS entropy: h_KS = sup_P lim (1/N) H(join T^{-k}P) - rate of information production
  • Pesin's theorem: h_KS = sum of positive Lyapunov exponents for ergodic diffeomorphisms
  • Rohlin hierarchy: Bernoulli => K-system => mixing => ergodic; irrational rotation is only ergodic

Why is irrational circle rotation T: x -> x + alpha (mod 1) ergodic but not mixing?

Weyl's equidistribution theorem: the sequence {n*alpha mod 1} is uniformly distributed for irrational alpha, proving ergodicity. But the Fourier mode e^{2*pi*i*x} has spectral measure concentrated at a single frequency, so its autocorrelation never decays to zero - mixing fails.

Ergodic Theory: Mixing and KS Entropy