Formal Riemannian Structure of the Wasserstein metric: Difference between revisions
Line 11: | Line 11: | ||
:<math> \lVert v + w \rVert \ge \lVert v \rVert \; \forall w \in L^2(\rho; X) \; \mathrm{such that} \; \nabla \cdot (w\rho) = 0 </math> | :<math> \lVert v + w \rVert \ge \lVert v \rVert \; \forall w \in L^2(\rho; X) \; \mathrm{such that} \; \nabla \cdot (w\rho) = 0 </math> | ||
where we are taking the <math> L^2(\rho, X) </math> norm. | where we are taking the <math> L^2(\rho, X) </math> norm. Divergence condition implies that our tangent vectors are equivalent up to a vector field with zero divergence. This implies that <math> v </math> is in fact a gradient of some function <math> u </math>, in which case our continuity equation becomes | ||
:<math> \frac{\partial \rho}{\partial t} + \nabla \cdot (\rho nabla u) = 0 </math> | |||
==Riemannian Metric Induced by the Wasserstein Metric== | ==Riemannian Metric Induced by the Wasserstein Metric== |
Revision as of 13:15, 10 June 2020
Given a closed and convex space , two probability measures on the same space, , the 2-Wasserstein metric is defined as
where is a transport plan from to . The Wasserstein metric is indeed a metric in the sense that it satisfies the desired properties of a distance function between probability measures on . Moreover, the Wasserstein metric can be used to define a Riemannian metric on . Such a metric allows one to define angles and lengths of vectors at each point in our ambient space.
Tangent Space Induced by the Wasserstein Metric
A convenient way to formalize tangent vectors in this setting is to consider time derivatives of curves on the manifold. A tangent vector at a point would be the time derivative at 0 of a curve, , where [1]. Since we are dealing with a space of probability measures, additional restrictions need to be added in order to make our tangent space well-defined. For example, we would like our trajectory to satisfy the continuity equation . There are many such vector fields that solve the continuity equation, so we can restrict to a vector field that minimizes kinetic energy, which is defined as . This choice of tangent vectors is justified by the following lemma
- Lemma[2] A vector belongs to the tangent cone at iff
where we are taking the norm. Divergence condition implies that our tangent vectors are equivalent up to a vector field with zero divergence. This implies that is in fact a gradient of some function , in which case our continuity equation becomes
Riemannian Metric Induced by the Wasserstein Metric
References
Cite error: <ref>
tag with name "Villani2" defined in <references>
is not used in prior text.