Post-Fordist urban growth is characterised by ever-increasing dispersal and polycentricity. The transformation of urban spatial structure has profound implications for sustainable development. There is now a substantial literature both on confirming the existence of polycentricity through equilibrium and evolutionary theories and on empirical identification of subcentres. However, more research is needed to explore the generically polycentric urban growth. A microscopic simulation approach is thus applied to study how stable subcentres, measured in terms of population density, can be established. In this study, an experiment carried out in an artificial cellular city is reported. One innovative feature of this simulation is that the state of the cellular automaton comprises a quantity variable (population density) as well as a binary state variable (selected or not selected). The two are interlinked through the evolution of the city. The experiment suggests that the combined forces of accumulative population density and local interactions can lead to the formation of stable subcentres. In such a regime, subcentres are first established through stochastic 'errors', for example, clusters of local developments. Thereafter the clusters continue to capture development opportunities through reinforced local interactions. With the evolution, however, disutilities, such as congestion, are accumulated from the concentration of local development. The growth of disutilities changes the relative attractiveness between formed (sub)centres and other areas. When significant disutilities are accumulated, another stochastic change may overturn the dominance of(sub)centres and drive developments to other locations. The subcentre formation is validated under the general discussion of 'goodness of fit' of possible urban automata, and population density surfaces are measured through various 'signatures'.
