Abstract (eng)
This dissertation presents three interrelated papers on the co-evolution of networks and play. The general structure of these models combines
three elementary events- action adjustment, link creation and link destruction- to one stochastic game dynamics, and focuses on the
asymptotic properties of these processes.
Chapter 2 presents the general mathematical framework of a co-evolutionary model. The players have arbitrary utility functions,
defined on a common set of actions, and employ probabilistic behavioral rules in the above mentioned events. Admissible rules satisfy irreducibility and a large deviations assumption. Beside these technical assumptions, we try to avoid making substantial behavioral assumptions. This generates a well-defined Markov chain, whose long-run properties can be studied analytically by making use of tree-characterization methods due to Freidlin and Wentzell [Random perturbations of dynamical systems, Springer, 1998]. We provide a general technique to compute stochastically stable states in such co-evolutionary models, by defining suitable cost-functions. Making some mild additional assumptions on the structure of the behavioral rules, we demonstrate an interesting and unforeseen connection between the derived ensemble of networks and inhomogeneous random graphs.
The models presented in chapters 4 and 5 particularize the general framework to the class of potential games and behavioral rules of the logit-response form. Under these assumptions, chapter 4 gives a full description of the induced ensemble of networks, and provides additionally closed-form expression for statistics of this ensemble, such as the degree-distribution. The model presented in chapter 5 is more general by allowing the players to have idiosyncratic preferences. This leads us to the definition of structured Bayesian interaction games, following a recent literature of evolutionary game theory studying Bayesian population games.
Chapter 3 establishes a connection between the Markov chain constructed in the general framework of chapter 2 and the continuous-time Markov processes considered in the models of chapters 4 and 5. Chapter 6 recapitulates the results of the thesis and gives an outlook for future research projects.