Statistical properties of chaotic dynamical systems

Matthew Nicol
Department of Mathematics, University of Houston

Abstract: In many situations a physical, biological or financial system is modeled by a transformation T acting on a phase space X where time evolution corresponds to the application of the map, moving an initial point in phase space to a subsequent state. For example if x is the number of animals in a population at time n=0 then a common population model predicts the next population level at time n=1 to be T(x)=Kx(1-Lx), where K,L are parameters. The point T(x) then moves to T(T(x)):=T2(x). The sequence {Tn(x)} corresponds to population levels at subsequent times n. More generally if φ: X→R is a function or measurement on the system then {φ⋅Tn} is a time-series of measurements on the underlying system. If the map T has an invariant probability measure then this time-series {φ⋅Tn} is a stochastic process: that is, a sequence of random variables on a probability space. Although the time-series of measurements on such deterministic systems are not independent (in fact usually highly correlated) the standard limit theorems such as the strong law of large numbers, the central limit theorem and Brownian motion-like behavior of the stochastic process often hold if the system has a slight degree of hyperbolicity. We show how ideas and techniques from probability and ergodic theory can be applied to understand the statistical behavior of time-series of measurements on `chaotic' dynamical systems.