Optimal transport of measures via autonomous vector fields (Q6904082)

The authors are concerned with transporting a probability measure \(\mu_{0}\) to a probability measure \(\mu_{1}\) via a flow approach. Namely, find a time-independent vector field \(v\) such that its flow \((\varphi(t,x))\) satisfies \(\varphi(1,\cdot)_{\sharp} \mu_{0} = \mu_{1}\). This problem relates to the continuity equation : \(\partial_{t} \mu + \mathrm{div}(\mu v) = 0\). Provided the solutions \((\varphi(t,x))\) and \((\mu(t,x))\) of these two problems are unique they can be identified via \(\mu(t, \cdot) = \varphi(t\cdot)_{\sharp} \mu_{0}\) and are related to optimal transportation theory. If one relaxes the constraint and allows the vector field to depend on the time, some solutions are known to exist.\N\NIn this paper, the authors work in a setting that does not guarantee uniqueness, and their main results are that they can study solution through optimal transport tools. More precisely, for \(\mu_{0}\) and \(\mu_{1}\) two probability measures on \(\mathbb{R}\) with convex support and continuous densities on their support, the authors show the existence of a time-independent vector field whose flow transports \(\mu_{0}\) to \(\mu_{1}\), and coincides with the Brenier map at time \(1\). They then show that one can find an arbitrarily close perturbation of \(\mu_{1}\) such that the vector field to transport \(\mu_{0}\) to this perturbation is actually Lipschitz. Some partial generalisations in higher dimensions are given.