Gibbs-non-Gibbs transitions via large deviations: computable examples (Q438843)

This work is devoted to the study of dynamical Gibbs-non-Gibbs transitions in several Markovian systems. These are approached with the formalism of large deviations. The method of large deviations focuses on finding the most probable trajectories in the state space. This is accomplished by means of minimizing a suitable cost or action functional, what is done in turn solving the associated Hamilton differential system. In the first part of the article the model under examination is a Langevin equation with a rather general drift and a constant diffusion. Two particular cases of this equation, Brownian motion and the Ornstein-Uhlenbeck process, are examined in more detail. In the last part of the article Markov processes with a discrete state space governed by master equations are studied. The small parameter needed to carry out the large deviation procedure is the noise amplitude in the first part of the article and the distance between neighboring states in the second part. The main goal of the article is detecting the appearance of bad configurations as time evolves. These are characterized as points in state space with a non-unique history, i. e. points for which the most probable trajectories leading to them are not unique. The dynamical appearance of such bad configurations is studied for different initial measures and drift terms for the Langevin equation. Under rather general conditions, the absence of bad configurations for short times is proved. Also, a counterexample of this fact when these conditions are not fulfilled is offered. For longer times, bad configurations and therefore Gibbs-non-Gibbs transitions may take place. This is illustrated with several examples and a general result concerning the Langevin equation with a rather arbitrary drift term. In the last part of the article these results are adapted for master equations.