Complex systems are commonly modeled as networks of units that integrate external input and emit brief pulses of activity upon reaching a threshold. Such pulse-coupled interactions can give rise to cascades of activity propagating through the network in the form of avalanches. While the statistics of these collective events are known to depend sensitively on network topology, rigorous analytical descriptions are usually restricted to homogeneous networks or mean-field approximations which are limited to particular aspects of generated avalanches like size and duration. Here we develop a mathematically rigorous theory of avalanche formation in recurrent networks of pulse-coupled units without assuming homogeneity. For arbitrary nonnegative coupling matrices, the framework provides a closed form for the avalanche assembly distribution, which quantifies the probability that a given subset of units is activated during an avalanche. This approach represents a substantial conceptual advance over existing theories allowing analysis of the structure-function relation between network topology and detailed statistics of generated avalanches. This is particularly relevant for the irregular, highly structured, and asymmetric coupling topologies ubiquitous in natural systems, such as brain networks with pronounced columnar, modular, and hierarchical organization.

Close

PrevNext

View figure in article

• Biological neural networks
• Collective dynamics
• Complex systems