Summary: Agent-based simulations are generative or computational approaches used for analyzing “complex systems”. What is a “system”? Examples of systems include a collection of molecules in a container, the population in an urban area, and the brokers in a stock market. The entities or agents in these three systems would be molecules, individuals, and stock brokers, respectively. The agents in such systems interact in the sense that molecules collide, individuals come into contact with other individuals, and brokers trade shares. Such systems, often called multiagent systems, are not necessarily complex. The label “complex” is typically attached to a system if the number of agents is large, if the agent interactions are involved, or if there is a large degree of heterogeneity in agent character or their interactions. This is of course not an attempt to define a complex system. Currently, there is no generally agreed upon definition of complex systems. It is not the goal of this entry to provide such a definition – for our purposes, it will be sufficient to think of a complex system as a collection of agents interacting in some manner that involves one or more of the complexity components just mentioned, that is, with a large number of agents, heterogeneity in agent character and interactions, and possibly stochastic aspects to all these parts. The global properties of complex systems, such as their global dynamics, emerge from the totality of local interactions between individual agents over time. While these local interactions are well understood in many cases, little is known about the emerging global behavior arising through interaction. Thus, it is typically difficult to construct global mathematical models such as systems of ordinary or partial differential equations, whose properties one could then analyze. Agent-based simulations are one way to create computational models of complex systems that take their place.
An agent-based simulation, sometimes also called an individual-based or interaction-based simulation (which we prefer), of a complex system is in essence a computer program that realizes some (possibly approximate) model of the system to be studied, incorporating the agents and their rules of interaction. The simulation might be deterministic (i.e., the evolution of agent states is governed by deterministic rules) or stochastic. The typical way in which such simulations are used is to initialize the computer program with a particular assignment of agent states and to run it for some time. The output is a temporal sequence of states for all agents, which is then used to draw conclusions about the complex system one is trying to understand. In other words, the computer program is the model of the complex system, and by running the program repeatedly, one expects to obtain an understanding of the characteristics of the complex system.
For the entire collection see [Zbl 1457.91008].