The size distribution of 'cities' delineated with a network theory-based method and mobile phone GPS data

We delineate cities independent of administrative boundaries in Japan by using a network theory-based method and GPS-based human mobility data. We divide the country into approximately 1 x 1 km(2) cells and detect the partition of cells that is optimal from the perspective of information theory. The resulting groups of cells are specified as cities. We find that the combination of two lognormal distributions better fits the city-size distribution than a distribution with a Pareto upper tail. Moreover, we show that a jump diffusion process is the stochastic process of the city population underlying such a distribution.