Title: The spatial organization of the population density in cities

Abstract: Although the average population density of a city is an extremely simple indicator, it is often used as a determinant factor for describing various aspects of urban phenomena. On the other hand, a plethora of different measures that aim at characterizing the urban form have been introduced in the literature, often with the risk of redundancy. Here, we argue that two measures are enough to capture a wealth of different forms of the population density. First, fluctuations of the local density can be very important and we should distinguish almost homogeneous cities from highly heterogeneous ones. This is easily characterized by an indicator such as the Gini coefficient $G$, or equivalently by the relative standard deviation or the entropy. The second important dimension is the spatial organization of the heterogeneities in population density and we propose a dispersion index $\eta$ that characterizes the degree of localization of highly populated areas. We argue that these two dimensions are enough to characterize the spatial organization of cities, and we discuss this approach using a dataset of about $4,500$ cities belonging to the $10$ largest urban areas in France, for which we have high resolution data. Representing cities in the plane $(G,\eta)$ allows us to construct families of cities. On average, compactness increases with heterogeneity, and we find four large categories of cities (with population $>10,000$ inhabitants): (i) first, homogeneous and dispersed cities with small density fluctuations, (ii) very heterogeneous cities with a compact organization of large densities areas. The last two groups comprise heterogeneous cities with (iii) a monocentric organization or (iv) a more delocalized, polycentric structure. Integrating these two parameters in econometric analysis could improve our understanding of the impact of urban form on various socio-economical aspects.