Say I have a (directed) graph $G$ with an adjacency matrix $A$.
For the sake of the question, let's assume it's normalized column-wise (edge weights are normalized so the sum of out-edge weights per node is equal to 1).
I'd like to calculate the eigenvalue centrality of its nodes using the power method, but I know that it might not converge in some cases (which, if I understand correctly, occurs if and only if it has an eigenvalue of -1).
My question is: Generally, what are the conditions on the graph to avoid this case?
Specifically, if I use page-rank centrality (with damping factor < 1), can I be certain to avoid this case (For every choice of personalization vector?)
[Edited: It appears I can be even more specific: If there is an edge with non-zero weight from each node to each node, can I be sure not have an eigenvalue of -1?]