The minimal time scale after which a Markov chain's distribution is close to its stationary distribution within a specified tolerance, typically measured in total variation distance.

Principle

Convergence speed is governed by spectral properties (eigenvalue gap) and isoperimetric quantities (conductance); larger gaps and higher conductance yield faster mixing.

Demonstration

For a random walk on an n-vertex expander graph, the mixing time grows logarithmically in n because the graph's spectral gap is bounded away from zero, producing rapid convergence to stationarity.

Misapplication

Assuming a short empirical run implies mixing without quantifying distance to stationarity, or conflating mixing time with the time to visit every state (cover time).

Consequence

Provides runtime and burn-in guidelines for Markov chain Monte Carlo and dynamics modeling: guarantees approximate equilibrium sampling after the stated time.

Reversal

A chain with long mixing time exhibits metastability or bottlenecks, so initial distributions persist and sampling remains biased; the inverted case is non-ergodic behavior rather than rapid equilibration.

Boundary

Defined for ergodic (irreducible, aperiodic) Markov chains and depends on the chosen distance and tolerance; different norms (total variation, L2) produce different quantitative mixing times.

Semantic Tension

Versus relaxation time (inverse spectral gap): both quantify convergence but use different norms and constants; relaxation time is algebraic L2-based while mixing time is probabilistic and often stronger.

Synthesis

A quantitative threshold time, derived from spectral and geometric chain properties, after which the chain's law is uniformly close to its stationary distribution within a prescribed error.