1: \begin{abstract}

2:  We {study single commodity network flows with suitable robustness and efficiency specs. An original use of a maximum entropy problem for distributions on the paths of the graph turns this problem into a {\em steering problem} for Markov chains with prescribed initial and final marginals. }

3: %We show that maximum entropy problems offer a flexible and powerful tool to handle transport problems over unweighted and weighted graphs. The resulting scheduling consists in selecting transition probabilities for a  Markovian evolution which has prescribed initial and final marginals, as specified by the transportation task.

4: From a computational standpoint, viewing scheduling this way

5: is especially attractive in light of  the existence of an iterative algorithm to compute the solution. 

6: %with linear convergence (in the Hilbert metric) that we proposed recently.

7: The present paper builds on \cite{chen2016networks} by introducing an  index of {\em efficiency} of a transportation plan and points, accordingly, to efficient-robust transport policies. In developing the theory, we establish two  new {\em invariance properties} of the solution (called {\em bridge}) 

8: %on graphs 

9: -- an {\em iterated bridge} invariance property and the invariance of the {\em most probable paths}. These properties, which were tangentially mentioned in our previous work, 

10: %in some special cases, 

11: are fully developed here. We also show that the distribution on paths of the optimal transport policy, which depends on a ``{\em temperature}'' parameter, 

12: %in a manner completely analogous to that of the  Boltzmann distribution, 

13: tends to the solution of the ``most economical'' but possibly less robust {\em optimal mass transport} problem as the temperature goes to zero.  The relevance of all of these properties for transport over networks is illustrated in an example.

14: \end{abstract}

15: