Comparison between variational optimal mass transportation and Lie advection (Q2082840)

The work compares, mainly from the numerical point of view, two different ways to transport a measure \(\mu\) into a measure \(\nu\). The first one is based on the optimal transport map with cost given by the square of the distance, it is related to the highly nonlinear Monge-Amperé equation, and it is notoriously difficult to treat it numerically in real-world situations; the second one is based on the Lie advection method and the transport map takes the form \[ \frac{\log(\mu)-\log(\nu)}{\mu-\nu}\nabla u \] where \(u\) solves the linear Poisson equation \(\Delta u=\mu-\nu\). Even though the latter may not give the optimal map, the authors show with different algorithms and experimental results that it can be computed efficiently and still provides a good approximation of the former.