MULTIPLE SPECIES EFFECTS AND TRANSITIONS IN SCHELLING SEGREGATION AGENT-BASED MODEL SIMULATIONS Ken A. Hawick Computer Science, Massey University, North Shore 102-904, Auckland, New Zealand Email: k.a.hawick@massey.ac.nz; Tel: +64 9 414 0800 Fax: +64 9 441 8181 ABSTRACT Multiple species effects in agent-based models can re- solve stand-offs and obstacles to model systems that are evolving towards a long term static or dynamic equilib- rium. The Schelling model of segregation is simulated us- ing a range of different individual agent species. The agents are found to form complex spatial structures as they self- organise to maximise their segregation. We implement a custom simulation code to explore a large range of num- bers of agent species. We describe various measures of complexity such as the bulk agent satisfaction fraction, the agent membership of the largest community, and the num- ber of separate community clusters, that can all be used to locate the phase transition in the satisfaction parameter for various geometric configurations of the model. We report on systems of up to ten separate agent species on square, triangular and hexagonal lattices and also discuss impli- cations of reversing the sense of the agent satisfaction to promote anti-ferromagnetic structures. KEY WORDS Schelling segregation; agent-based model; neighbourhood effects; multiple species; agent simulation. 1 Introduction The Schelling model of segregation [36] is one of a number of quantifiable sociological phenomenological models [37] such as the voter and opinion systems [16, 21, 38] and the Axelrod cultural dissemination [17] that have generated recent interest in applications of agent- based modelling (ABM). The Schelling model allows a semi quantitative inves- tigation of racial and population segregatory systems and has attracted increasing interest since it has become feasi- ble in recent years to model realistically sized agent sys- tems through computer simulations of social systems [6]. Phenomena such as political compromises [29], promotion of particular ideas [30], corporate promotions [31], policy interventions [26] and more general social dynamics [39] are all amenable to such models as well as the spatial segre- gation [34] effects modelled by the original Schelling sys- tem. There are various Schelling-like models [40] reported in the literature and which have been used to model and study effects including various social aspects [9] such as Figure 1. Snapshots of Schelling segregation model with 2 to 9 different agent species and τ values of 0.575 after 2000 model time-steps from random initialisation. the housing system [8] and market [23] in the USA [12] where ghetto-ization is common [10]. Generally these sociological systems are complex systems [28] with complex network or geometries in- volved. However the Schelling system has been most widely studied on simple square lattice systems. In such a regime it can be compared to other models that exhibit phase transitions such as the Ising [22] or Potts [20, 33] models [27]. Work has been reported on the effect of geom- etry and neighbourhood size on the Schelling model [18]; on the role of finite system size effects [7]; on the system dimensionality [5] and geometry [32]; as well as the scal- ing properties [11]; and of hierarchical system-of-system effects [3] within Schelling systems. Attempts have been made to unify of the various Schelling-like system models [1, 14] within a mathematical framework [35] but is has proved hard to establish com- pletely the phase diagram [13]. In practice much work has compared visual observations from small to medium sized simulations of the Schelling system with real world socio- logical phenomena [4]. In this present work we introduce multiple agent species into the Schelling system. Figure 1 shows some model system snapshots of agents segregating under the Schelling model dynamics after 2000 model time steps since a random initialisation. The agents are coloured and a range of different number of included agents are shown from 2 to 9 separate species. Work is reported on the effect of agent vision dis- tance in the Schelling model [25] but although like most Proceedings of the IASTED International Conference Intelligent Systems and Control (ISC 2013) November 11 - 13, 2013 Marina del Rey, USA DOI: 10.2316/P.2013.807-027 189