Recent research on the non-stationary nature of the dynamics of complex systems is reviewed through three specific models. The long time dynamics consists of a slow, decelerating but spasmodic release of generalized intrinsic strain. These events are denoted quakes. Between the quakes weak fluctuations occur but no essential change in properties are induced. The accumulated effect of the quakes, however, is to induce a direct change in the probability density functions characterising the system. We discuss how the log-Poisson statistics of record dynamics may be an effective description of the long time evolution and describe how an analysis of the times at which the quakes occur enables one to check the applicability of record dynamics.
PACS numbers:Out of equilibrium systems are often treated as being in a stationary state characterised by time independent statistical measures. Although this is probably the case in some situations there are many instances where this is not so and where one may miss essential aspects of the behaviour if attempts are made to treat the phenomena as stationary or nearly stationary.Complex systems often display evolving macroscopic properties. The most important task of a theoretical treatment is then to understand the link between the microscopic fluctuations, which will often exhibit an approximate time reversal symmetry and the macroscopic directed evolution. The description should as well explain the nature of the emergent macroscopic dynamics.Here we review how the concept of record dynamics, developed by Sibani and Littlewood[1], has successfully served as a paradigm for the description of the evolution of three very different models: the relaxation of a spin glass following an initial temperature quench, the penetration of an external magnetic field into a disordered type II superconductor and a model of evolutionary ecology. In all three cases macroscopic variables, which exhibit a degree of intermittent dynamics, can be identified. Furthermore, the sequence of transitions between metastable configurations can be analysed in terms of the record statistics.The work reviewed here is a result of collaboration with