A coupling of two probability measures μ and ν on measurable spaces is a joint probability measure π on the product space whose marginals are μ and ν. More generally, a coupling of a family of measures is any joint law with prescribed marginals.

Principle

Coupling constructs a joint realization that makes two marginal distributions comparable pathwise; it turns abstract distance or divergence questions into probabilistic events about paired random variables.

Demonstration

To bound total variation between μ and ν, produce a coupling (X,Y) with marginals μ, ν and note TV(μ,ν) ≤ P(X ≠ Y); for finite sets an optimal coupling attains equality.

Misapplication

Assuming a particular coupling is canonical or unique without optimization; for example, using an arbitrary product coupling when a transport-optimal coupling is required will give weaker bounds.

Consequence

Appropriate couplings yield probabilistic proofs of convergence, bounds on distances (Wasserstein, TV), and constructions of monotone or co-adapted couplings that imply stochastic domination or contractivity properties.

Reversal

The opposite construction is forming the product measure (independent coupling) which minimizes dependence; reversing a coupling problem yields marginal decomposition or disintegration rather than joint synchronization.

Boundary

Couplings exist for probability measures on standard measurable spaces by extension theorems, but optimality questions require additional structure (metric, cost function); couplings do not prescribe uniqueness or regularity by themselves.

Semantic Tension

Coupling versus transport plan: both are joint measures with given marginals, but in optimal transport the coupling is evaluated by a cost functional, while in probabilistic coupling one often focuses on eventwise relations like equality or order.

Synthesis

A coupling is any joint law with specified marginals that translates abstract measure comparisons into concrete paired random variables, enabling pathwise comparisons and quantitative bounds.