Multi-marginal optimal transport on the Heisenberg group (Q820872): Difference between revisions

The classic Kantorovich problem of optimal transport considers two probability spaces \((X_1,\mu_1)\) and \((X_2,\mu_2)\) and a lower semi-continuous cost function \(c:X_1 \times X_2 \to \mathbb{R}\) and seeks the most economical transport between the measures. More precisely, it tries to find the coupling \(\gamma\) which minimizes the integral \[ \min _{\gamma \in \Gamma(\mu_1, \mu_2)} \int_{X_1 \times X_2} c(x_1, x_2)\, d \gamma, \] where \(\Gamma(\mu_1,\mu_2)\) denotes the set of all couplings between \(\mu_1\) and \(\mu_2\). One can then recover the transport plan from the disintegration of \(\gamma\). In this paper, the authors consider a generalization of this problem, known as the ``multi-marginal problem.'' The multi-marginal problem considers the case where there are \(n\) probability spaces \((X_1,\mu_1) \ldots (X_n,\mu_n)\). In other words, one has a cost function \(c:X_1 \times \ldots \times X_n \to \mathbb{R}\) and wants to minimize the integral \[ \min _{\gamma \in \Gamma(\mu_1, \ldots, \mu_n)} \int_{X_1 \times \ldots \times X_n} c(x_1, \ldots, x_n) \, d \gamma. \] For fairly general costs and measures, it is a known that the solution to the Kantorovich optimal transport problem exists and is induced by a transport map \(T:X_1 \to X_2\) and so also solves the Monge problem of optimal transport [\textit{W. Gangbo} and \textit{A. Świȩch}, Commun. Pure Appl. Math. 51, No. 1, 23--45 (1998; Zbl 0889.49030)]. In other words, the optimal coupling is one where each point in \(X_1\) is sent to a unique point in \(X_2\). However, this is not necessarily the case for the multi-marginal problem and the theory of the Monge problem in this setting is an active area of research (see [\textit{B. Pass}, ESAIM, Math. Model. Numer. Anal. 49, No. 6, 1771--1790 (2015; Zbl 1330.49050)] for further details). In this paper, the authors consider the case where the spaces \(X_i\) are the Heisenberg group \(\mathbb{H}^n\) and the cost function is given by \[ c_{d}\left(x_{1}, \ldots, x_{m}\right) = \inf _{y \in \mathbb{H}^{n}}\left\{\sum_{i=1}^{m} d^{2}\left(x_{i}, y\right)\right\}, \] where \(d\) is the Carnot-Carathéodory distance between points. By adapting techniques of \textit{L. Ambrosio} and \textit{S. Rigot} [J. Funct. Anal. 208, No. 2, 261--301 (2004; Zbl 1076.49023)] and \textit{Y.-H. Kim} and \textit{B. Pass} [Am. J. Math. 137, No. 4, 1045--1060 (2015; Zbl 1328.49044)], the authors show the existence of unique solutions to the Monge problem of multi-marginal optimal transport in this setting. To do so, they require two additional technical assumptions, which arise due to several properties of the geometry that are not present in the Riemannian case.