An efficient simulation algorithm for continuous-time agent-based linked lives models

The efficient continuous-time simulation of linked lives in demography implies specific challenges. The resulting agent-based models constitute time-inhomogeneous Markov chains which require stochastic simulation algorithms. Each agent is characterized by diverse attributes, including a specific position in a dynamically evolving social network which influences the agent's behavior. This hampers the application of population-based approaches in implementing the stochastic simulation algorithm. However, as events are locally constrained by the social network, many events will happen independently of each other. We develop a stochastic simulation algorithm that maintains a dependency structure to realize lazy re-calculation of events. In case study on a Susceptible-Infected-Recovered-Model with social network and age-dependent susceptibility we evaluate the performance of the algorithm in comparison to an earlier version. The evaluation shows the improved scalability and a significant speedup of up to 150 times that can be achieved by taking dependencies into account when executing linked, continuous-time agent-based models.