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
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.