PATTERN FORMATION IN RANDOM NETWORKS USING GRAPHONS

We study Turing bifurcations on one-dimensional random ring networks where the probability of a connection between two nodes depends on the distance between them. Our approach uses the theory of graphons to approximate the graph Laplacian in the limit as the number of nodes tends to infinity by a nonlocal operator---the graphon Laplacian. For the ring networks considered here, we employ center manifold theory to characterize Turing bifurcations in the continuum limit in a manner similar to the classical partial differential equation case and classify these bifurcations as sub-/sup er-/transcritical. We derive estimates that relate the eigenvalues and eigenvectors of the finite graph Laplacian to those of the graphon Laplacian. We are then able to show that, for a sufficiently large realization of the network, with high probability the bifurcations that occur in the finite graph are well approximated by those in the graphon limit. The number of nodes required depends on the spectral gap between the critical eigenvalue and the remaining ones; the smaller this gap is, the more nodes are required to guarantee that the graphon and graph bifurcations are similar. We demonstrate that if this condition is not satisfied, then the bifurcations that occur in the finite network can differ significantly from those in the graphon limit.