1: \begin{abstract}

2:   In this paper we investigate the numerical approximation of

3:     an analogue of the Wasserstein distance for optimal transport on

4:     graphs that is defined via a discrete modification of the

5:     Benamou--Brenier formula. This approach involves the logarithmic

6:   mean of measure densities on adjacent nodes of the graph.  For this

7:   model a variational time discretization of the probability densities

8:   on graph nodes and the momenta on graph edges is proposed.  A robust

9:   descent algorithm for the action functional is derived, which in

10:   particular uses a proximal splitting with an edgewise nonlinear

11:   projection on the convex subgraph of the logarithmic mean.  Thereby,

12:   suitable chosen slack variables avoid a global coupling of

13:   probability densities on all graph nodes in the projection step.

14:   For the time discrete action functional $\Gamma$--convergence to the

15:   time continuous action is established.  Numerical results for a

16:   selection of test cases show qualitative and quantitative properties

17:   of the optimal transport on graphs.  Finally, we use our

18:     algorithm to implement a JKO scheme for the gradient flow of the

19:     entropy in the discrete transportation distance, which is known to

20:     coincide with the underlying Markov semigroup, and test our results

21:     against a classical backward Euler discretization of this discrete

22:     heat flow.

23: \end{abstract}

24: