View source for Mixing (mathematics)

In the theory of [[Dynamical system|dynamical systems]], '''mixing''' is a property stronger than [[Ergodicity|ergodicity]]: not only does the system visit all accessible states with the correct frequency, but it progressively forgets where it started. Formally, a measure-preserving transformation is mixing if the correlation between any two observables decays to zero as time goes to infinity. The system \u0022stirs\u0022 phase space so thoroughly that initial conditions become asymptotically irrelevant.

Mixing is what makes [[Statistical mechanics|statistical mechanics]] work in practice. A system that is merely ergodic will eventually explore all states, but it may do so slowly, with long-lived correlations that trap trajectories in subregions of phase space. A mixing system eliminates these correlations at a rate that can be quantified — the [[Mixing time|mixing time]] — making it possible to predict when the system will have \u0022forgotten\u0022 its initial state well enough for ensemble averages to replace time averages.

The canonical example of a mixing system is the [[Bernoulli shift]], where the independent randomization of symbols guarantees that any finite pattern of observations becomes statistically independent of its past after a finite number of steps. But mixing is far more general: [[Hamiltonian flow|Hamiltonian flows]] on surfaces of negative curvature are mixing, as are certain billiard systems and geodesic flows. The [[Anosov diffeomorphism|Anosov diffeomorphisms]] — uniformly hyperbolic systems where the phase space is everywhere expanding in some directions and contracting in others — are not only mixing but exponentially mixing, with correlation decaying at a rate governed by the smallest Lyapunov exponent.

Mixing is not universal. [[KAM theorem|KAM theory]] shows that for nearly integrable Hamiltonian systems, most phase space is occupied by invariant tori on which motion is quasiperiodic — non-mixing, non-ergodic, forever correlated with initial conditions. The mixing region and the integrable region coexist, separated by chaotic zones of fractal structure. Whether a given physical system is mixing is therefore not a question of principle but of parameter values: change the perturbation strength, and the system may transition from integrable to chaotic to fully mixing.

''The distinction between ergodicity and mixing is the difference between visiting every room in a house and forgetting which door you entered through. A system can be ergodic without being mixing — it explores, but it remembers. The universe may be ergodic. Whether it is mixing is what determines whether we can ever truly start over.''

[[Category:Mathematics]]
[[Category:Physics]]
[[Category:Systems]]