Geodesics for a class of distances in the space of probability measures (Q376585)

In [Calc. Var. Partial Differ. Equ. 34, No. 2, 193--231 (2009; Zbl 1157.49042)], \textit{J.~Dolbeault, B.~Nazaret} and \textit{G.~Savaré} introduced a new class of distances between probability measures by introduction of some concave increasing nonnegative nonlinear mobility function \(m\). Assuming that \(\rho_{_0}=\rho_{_0}{\mathcal L}^d\) and \(\rho_{_1}=\rho_{_1}{\mathcal L}^d\), the corresponding squared distance between \(\rho_{_0}\) and \(\rho_{_1}\) is given as the infimum of \[ \int_0^1\int_{\mathbb R^d}\left|v(t,x)\right|^2 m(\rho(t,x))\,dx\,dt \] subject to \(\partial_t\rho+\text{div}(m(\rho)v)=0\), \(\rho(0,.)=\rho_{_0}\), \(\rho(1,.)=\rho_{_1}\). It is natural to view optimal transport theory from a dynamical perspective and to look for geodesics rather than just for transport maps. Similarly, probability measure valued curves governed by the continuity equation play a crucial role in the theory of gradient flows in the Wassertsein space. In this paper, the authors consider the case of a concave power mobility \(m(\rho) =\rho^\alpha\), \(\alpha\in(0,1)\). To avoid both compactness and boundary condition issues, instead of working on a domain or \(\mathbb R^d\) the case of the flat torus \(\mathbb T^d=\mathbb R^d/\mathbb Z^d\) is considered to characterize minimizing geodesics. First, they prove the existence of a potential function and then give necessary and sufficient optimality conditions that take the form of a coupled system of partial differential equations somehow similar to the Mean-Field-Games system of Lasry and Lions. Finally, they present an equivalent formulation posed in a set of probability measures over curves.