SEMINAIRE MERCREDI 9 AVRIL 2003 14 heures Salle Séminaire 5 Centre de Physique Théorique Marseille-Luminy Freddy Bouchet Université de Firenze Dipartimento di energetica "Sergio Stecco" Title: Study of Equilibrium Fluctuations and Classification of ensemble inequivalence situations in systems governed by long range interactions. Summary: We will first show that physical systems governed by long range interactions are common, using examples from geophysical fluid dynamics, plasma physics and self gravitating systems. Such systems share peculiar properties, as for instance possible statistical ensemble inequivalence (negative specific heat), or very slow relaxations toward the equilibrium state. We will address these questions using only classical thermodynamic ideas (Gibbs averages). In a first part, we compute explicitely, from the microscopic dynamics, the diffusion coefficient in a simple model with long range interactions. In a second part, we propose a classification of microcanonical and canonical phase transitions and of situations of ensemble inequivalence in such systems. We further describe these two parts in the following. Working with a model system with long range interactions, the HMF model, we will analytically derive the properties of the stochastic process describing the macroscopic variable, and a Fokker-Planck equation describing the statistical evolution of a particle in an equilibrium bath. Classical approaches (projection techniques or kinetic models) of such problems allow to obtain such a Fokker-Planck equation. However, the diffusion coefficient is then expressed in terms of a formal integral on some Liouvillian operator, and can be computed explicitly, only in some particular limits. In our case, combining a perturbative approach for the Hamiltonian dynamics, with microcanonical averages, we explicitly compute the diffusion coefficient. We show that the macroscopic variable is a Gaussian non Markovian process. Its autocorrelation function verify an explicit integral equation, with memory. We explain why we think that such results may be easily generalized to the study of the relaxation toward equilibrium, of out of equilibrium distributions. In a second part of the talk, we will propose a classification of situations of statistical ensemble inequivalence and of phase transitions, for systems with long range interaction. For many of these systems, large deviation techniques allow to prove that microcanonical statistical equilibrium is described by the maximization of an entropy with an energy constraint, whereas the canonical one is described by the associated free energy minimization. The solutions of these two variational problems may be different and lead to ensemble inequivalence. Starting from such variational problems, we will derive the classification, independently of any model. This allow to classify the links between microcanonical and canonical phase transitions. We will also stress that some generic situations for ensemble inequivalence have not yet been observed in any systems.