Optimal Transport & Wasserstein Distance

Monotone rearrangement theorem · W₁ and W₂ distances · Sinkhorn algorithm for discrete OT

Optimal transport finds the cheapest way to move mass from distribution μ to ν. W₁ = ∫|F_μ⁻¹(t)−F_ν⁻¹(t)|dt (quantile coupling — monotone rearrangement, cheapest for convex costs on ℝ). W₂² = ∫(T(x)−x)² dμ(x) where T=∇φ is a gradient of a convex function (Brenier theorem). Sinkhorn: add entropic regularization ε·H(P), solve via alternating row/column normalization. Right: discrete OT coupling matrix between 8-point distributions.