Reachability-Based Robustness of Network Controllability under Node and Edge Attacks

In real world applications, control is always performed without perfect knowledge, perfect models, and often, under changing conditions. Such circumstances are particularly true of complex systems. As a result, application of control theory to complex systems requires the development and implementation of control policies that are robust to unexpected and potentially malicious changes to the underlying network. This paper makes three important contributions along this direction. First, we introduce a new definition of robustness which captures realistic constraints imposed by many control problems. Second, we develop a novel algorithm for computing this robustness measure. Third, we conduct a thorough assessment of the control robustness of different synthetic networks to a wide array of attacks/network perturbations. We find that our robustness measure is behaviorally different from other robustness measurements in the literature and that the attacks considered highlight a number of ways in which network properties correlate with control robustness.