A new transportation distance between non-negative measures, with applications to gradients flows with Dirichlet boundary conditions (Q987845): Difference between revisions

It is well known that many evolution equations of the form \[ \partial_t\rho- \text{div}(\nabla\rho- \rho\nabla V)= 0\tag{1} \] can be seen as gradient flows of some integral functionals on the space of probability measures metrized by the Wasserstein distance. However, due to the fact that the mass remains constant along the evolution, this approach only works for Neumann boundary conditions. In order to treat PDEs like (1), with Dirichlet boundary conditions, in the paper under review the authors introduce a new distance on the space of all nonnegative measures (i.e., with no mass constraint); for any \[ \mu_1,\mu_2\in{\mathcal M}_2(\Omega)= \bigg\{\mu\in{\mathcal M}^+(\Omega): \int d^2(x,\partial\Omega)\,d\mu(x)< +\infty\bigg\} \] the distance \(Wb_2(\mu_1,\mu_2)\) is defined by \[ Wb^2_2(\mu_1, \mu_2)= \inf\biggl\{\int_{\overline\Omega\times \overline\Omega}|x- y|^2\,d\gamma(x,y): \pi^i_\#\gamma|_\Omega= \mu_i,\;i= 1,2\biggr\}. \] Notice that the difference between \(Wb_2\) and the usual Wasserstein distance relies on the fact that the admissible transport plans \(\gamma\) are measures on \(\overline\Omega\times \overline\Omega\), while the marginal condition is only imposed on \(\Omega\). This allows to consider Dirichlet boundary conditions in equations like (1) via gradient flows and the minimizing movements method. In the paper several properties of the distance \(Wb_2\) are studied, and in the last section some possible generalizations are indicated, together with some interesting open problems.