On finding common neighborhoods in massive graphs

We consider the problem of finding pairs of vertices that share large common neighborhoods in massive graphs. We prove lower bounds on the resources needed to solve this problem on resource-bounded models of computation. In streaming models, in which algorithms can access the input only a constant number of times and only sequentially, we show that, even with randomization, any algorithm that determines if there exists any pair of vertices with a large common neighborhood must essentially store and process the input graph off line. In sampling models, in which algorithms can only query an oracle for the common neighborhoods of specified vertex pairs, we show that any algorithm must sample almost every pair of vertices for their respective common neighborhoods.