The following is work done for a class on Swarm Intelligence. 

In disconnected networks, like the one shown, agents have limited ability to communicate across the sub-networks. This is often solved using "data mules". Choosing the best data mule is inherently a distributed optimization problem, that often requires a lot of communication to come to a consensus. 

We proposed a method to solve this optimization with little communication, with analytical bounds on the frequency of communication between the networks.