The study of metric and geometric structures induced by optimal transport costs (notably Wasserstein distances) on spaces of probability measures, linking mass displacement problems to analysis, partial differential equations, and Riemannian-type geometry on measure spaces.

Principle

Represent probability measures as points in a metric space where cost-minimizing transport plans define distances; interpret geodesics as displacement interpolations, and use gradient-flow and convexity notions (e.g., displacement convexity) to relate transport geometry to evolution equations and curvature notions on measure spaces.

Demonstration

On the real line or in Euclidean space, the L^2 Wasserstein distance between two absolutely continuous measures is realized by a monotone transport map; geodesics correspond to pushforwards along linear interpolation of maps, and the heat equation can be seen as the gradient flow of entropy in Wasserstein space.

Misapplication

Using Wasserstein distances without verifying absolute continuity or ignoring boundary conditions can produce non-existent optimal maps; treating discrete empirical measures as smooth objects and applying continuous transport formulas can mislead computations and asymptotics.

Consequence

Optimal transport geometry provides a unifying formalism for comparing measures, establishes functional inequalities, informs PDE regularity and long-time behavior via convexity in transport space, and supplies metrics useful in statistics and machine learning.

Reversal

Reverse by considering only weak-* topologies on measures or integral probability metrics that ignore transport structure; this loses geometric interpolation and the connection to dynamics governed by transport-driven gradient flows.

Boundary

Results depend on cost functions, regularity of measures, and the underlying space (Euclidean vs. manifolds); many smooth Riemannian analogues require absolutely continuous measures and convexity assumptions that fail for singular or atomic measures.

Semantic Tension

Tension exists between transport-based metrics (Wasserstein) that capture spatial rearrangement and other divergences (Kullback–Leibler, total variation) that measure pointwise discrepancy; each emphasizes different aspects of similarity between measures.

Synthesis

Optimal transport geometry recasts comparison of measures as a geometric problem where transport costs define distances, enabling geometric analysis of measure evolution, convexity, and PDEs while providing practical metrics for applications.