Paper detail

A Dynamic Model of Cascades on Random Networks with a Threshold Rule

Cascades on random networks are typically analyzed by assuming they map onto percolation processes and then are solved using generating function formulations. This approach assumes that the network is infinite and weakly connected, yet furthermore approximates a dynamic cascading process as a static percolation event. In this paper we propose a dynamic Markov model formulation that assumes a finite network with arbitrary average nodal degree. We apply it to the case where cascades follow a threshold rule, that is, that a node will change state ("flip") only if a fraction, exceeding a given threshold, of its neighbors has changed state previously. The corresponding state transition matrix, recalculated after each step, records the probability that a node of degree k has i flipped neighbors after j steps in the cascade's evolution. This theoretical model reproduces a number of behaviors observed in simulations but not yet reported in the literature. These include the ability to predict cascades in a domain previously predicted to forbid cascades without assuming that the network is locally tree-like, and, due to the dynamic nature of the model, a "near death" behavior in which cascades initially appear about to die but later explode. Cascades in the "no cascades" region require a sufficiently large seed of initially flipped nodes whose size scales with the size of the network or else the cascade will die out. Our theory also predicts the well known properties of cascades, for instance that a single node seed can start a global cascade in the appropriate regime regardless of the (finite) size of the network. The theory and simulations developed here are compared with a foundational paper by Watts which used generating function theory.