The complexity of computing a robust flow

The Maximum Robust Flow problem asks for a flow on the paths of a network maximizing the guaranteed amount of flow surviving the removal of any $k$ arcs. We point out a flaw in a previous publication that claimed $NP$-hardness for this problem when $k=2$. For the case that $k$ is part of the input, we present a new hardness proof. We also discuss the complexity of the integral version of the problem.