ABSTRACT: Deterministic dynamics often leads to complex, unpredictable behavior. This randomness or chaos produces information and limits one's ability to predict future events. There are two components to this imposed ignorance. The first arises in a mathematical context from highly convoluted orbit structures in state space. These allow a system to rapidly visit many regions of state space. In a physical context, the second comes from the coupling of the system-under-study to other systems that provide information to it. Extrinsic information sources preclude the exact determination of the system's state. By the mechanism of their complex orbits, chaotic systems amplify this uncertainty into unpredictable macroscopic behavior. The physical study of chaotic dynamical systems is incomplete without an appreciation of how external fluctuations affect their predictability.
Using information theory we describe how to measure the unpredictability of (i) deterministic chaotic systems (without extrinsic noise), and (ii) nondeterministic chaotic systems (coupled to extrinsic noise). Scaling concepts are invaluable tools in this. Scaling reveals that extrinsic noise acts as a disordering field for chaos. Furthermore, even for systems with extrinsic noise, scaling captures fundamental features of chaotic behavior. It provides a unified framework for the topological, metric, and Renyi dimensions and entropies.
The physical relevance of these concepts lies in their ability to analyze noisy chaotic signals. The information theoretic approach to temporally complex behavior is applied to chaotic signals from two hydrodynamic experiments. In addition, the dynamic aspects of pattern evolution and the possible breakdown of (naive) dynamical systems theory is discussed for experiments with an image processing system.
The first appendix contains descriptions of algorithms for dynamical systems studies. The second discusses a movie on the geometric structure of chaotic driven oscillators using animated Poincare sections.