Who's Your Neighbor? Neighbor Identification for Agent-based Modeling

Agent-based modeling and simulation, based on the cellular automata paradigm, is an approach to modeling complex systems comprised of interacting autonomous agents. Open questions in agent-based simulation focus on scale-up issues encountered in simulating large numbers of agents. Specifically, how many agents can be included in a workable agent-based simulation? One of the basic tenets of agent-based modeling and simulation is that agents only interact and exchange locally available information with other agents located in their immediate proximity or neighborhood of the space in which the agents are situated. Generally, an agent’s set of neighbors changes rapidly as a simulation proceeds through time and as the agents move through space. Depending on the topology defined for agent interactions, proximity may be defined by spatial distance for continuous space, adjacency for grid cells (as in cellular automata), or by connectivity in social networks. Identifying an agent’s neighbors is a particularly time-consuming computational task and can dominate the computational effort in a simulation. Two challenges in agent simulation are (1) efficiently representing an agent’s neighborhood and the neighbors in it and (2) efficiently identifying an agent’s neighbors at any time in the simulation. These problems are addressed differently for different agent interaction topologies. While efficient approaches have been identified for agent neighborhood representation and neighbor identification for agents on a lattice with general neighborhood configurations, other techniques must be used when agents are able to move freely in space. Techniques for the analysis and representation of spatial data are applicable to the agent neighbor identification problem. This paper extends agent neighborhood simulation techniques from the lattice topology to continuous space, specifically R. Algorithms based on hierarchical (quad trees) or non-hierarchical data structures (grid cells) are theoretically efficient. We explore implementing hierarchical and non-hierarchical data structures by using efficient implementations that are designed to address spatial data specifically in the context of agent-based simulation. The algorithms are evaluated and compared according to computation times for neighborhood creation, neighbor identification, and agent updating.