2
$\begingroup$

Setup:

Let $X$ be a complete pointed metric space.
Let us briefly recall that the Wasserstein space $W_1(X)$ is identifiable with a subset of the Arens-Eells (or Lipschitz-Free) space $\operatorname{AE}(X)$ (i.e.: the pre-dual of $\operatorname{Lip}_0(X)$), (via the identification of the (dense) set of finitely-supported probability measures $\{\sum_{i=1}^I k_i\delta_{x_i}\}\subseteq W_1(X)$ with its corresponding elements $\{\sum_{i=1}^I k_iE_{x_i}\}\subseteq W_1(X)$ in the pre-dual of $\operatorname{Lip}_0(X)$); where $E_x:\operatorname{Lip}_0(X):\ni f \mapsto f(x)\in \mathbb{R}$ is the point-evaluation map.


I'm having trouble relating the following two, "seemingly interchangeable" notions.

  • (B1) In the theory of Lipschitz-spaces: a barycenter map is a continuous left-inverse of the map $\delta:X\rightarrow \operatorname{AE}(X)$ sending $x$ to the point-evaluation map $x\mapsto E_x\in \operatorname{AE}(X)$.

  • (B2) In Optimal Transport/Probability Theory: a barycentre map refers to a continuous left-inverse of the map $x\mapsto \delta_x\in W_1(X)$.


Question:

Clearly, via the isometric identification of $W_1(X)$ with a subset of $\operatorname{AE}(X)$, if (B1) holds, then (B2) must also. However, is the converse true? That is, if (B2) holds, then can the map $x\mapsto \delta_x$ be extended to all of $\operatorname{AE}(X)$?

Esp. when this map is Lipschitz, then can the extension also be concluded to be Lipschitz?

$\endgroup$
2
  • 1
    $\begingroup$ Deleted my previous comment, which was nonsense. I'll try to think about this more ... $\endgroup$
    – Nik Weaver
    Mar 27, 2022 at 1:31
  • $\begingroup$ @NikWeaver No worries, as you can see from the date posted to the date bountied...I think its nontrivial $\endgroup$
    – ABIM
    Mar 27, 2022 at 9:25

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.