TOWARDS A STATISTICAL MECHANICS OF CITIES

Luís M. A. Bettencourt (Personal webpage)

A grand challenge for complex systems is the construction of a predictive statistical mechanics (and thermodynamics), paralleling and extending those for equilibrium physical systems. Complex systems pose new and difficult problems, including their non-equilibrium dynamics, their internal diversity and connectivity, and their capability for adaptation and evolution. Yet, in the last few years we have come to understand the statistical regularities that characterize several classes of complex systems, from organisms and ecosystems to cities. Simultaneously, we are beginning to understand the statistical theory of learning and adaptation in the context of machine learning and artificial intelligence.

In this talk, I will show how many of the known statistical regularities of cities can be understood and predicted from statistical models that combine motion, interactions, and learning. Such models posit that cities are the result of bottom-up, adaptive advantages from interaction and learning between people, similar to how spin models are used in machine learning. I will finish by discussing the potential of these hybrid (physical and informational) statistical theories in general complex systems.