Percolation is a mathematical framework for modeling how things spread through networks. In classical percolation on an undirected graph, you randomly keep or remove edges with some probability p, and ask whether a giant connected component spanning the infinite graph emerges. This paper studies a directed variant called bidirected percolation: each undirected edge is replaced by two one-way arrows pointing in opposite directions, each kept independently with probability p. Now the relevant question is whether two vertices can reach each other by following directed paths in both directions, a stricter condition called strong connectivity. The paper asks how the threshold probability for an infinite strongly connected cluster, called p-sc, relates to the usual percolation threshold p-c.
The main results draw a sharp dividing line based on a geometric property of graphs called amenability. Loosely, amenable graphs are those where large regions have relatively small boundaries compared to their volume, with the integer lattice being the classic example. Non-amenable graphs, like trees and certain hyperbolic structures, have the opposite property. The paper proves that on every transitive amenable graph, the two thresholds coincide: strong connectivity kicks in exactly when ordinary connectivity does. On non-amenable graphs the situation is different, and suitably modified versions of any such graph satisfy p-sc strictly greater than p-c, meaning there is a range of probabilities where an ordinary infinite cluster exists but no infinite strongly connected cluster does.
The paper also studies what happens right at the critical threshold p-sc in high dimensions. For the integer lattice in 11 or more dimensions, the authors show that the expected size of the strongly connected cluster containing a given vertex is finite exactly at the critical probability. This mirrors a known result for ordinary percolation in high dimensions and suggests that the phase transition for strong connectivity is well-behaved in that setting. Together, these findings reveal that the geometry of the underlying graph fundamentally controls how directional constraints interact with connectivity at large scales.