Datenanalyse und Stochastische Modellierung
3. Chaos

Chaotic systems

  • deterministic systems with statistical behavior similar to stochastic processes
  • can be discrete or continuous in time
  • yield insights to dynamical origins of statistical behavior measured time series

Maps

\[x_{t+1} = f(x_{t},t)\]

Autonomous if \[f=f(x)\]

Bernoulli map

\[x_{t+1}=2x_{t} \mbox{ mod } 1\] https://upload.wikimedia.org/wikipedia/commons/6/68/Exampleergodicmap.svg
  • stretch-and-fold
  • binary code for numbers \[ z = \sum_{t=0}^\infty \frac{b_t}{2^{t+1}} \]
  • rational numbers: periodic
  • irrational numbers: generally infinite series

Bernoulli map

\[x_{t+1}=2x_{t} \mbox{ mod } 1\]
  • Pseudo-random numbers: see exercise

Measure preserving maps F

\[ \int f(Fx) \rho(x) \mathrm{d}x = \int f(x) \rho(x) \mathrm{d}x \]

Nondecomposable maps F

\[ FA = A \Rightarrow \int_A \rho(x) dx = 0 \mbox{ or } 1 \; \; \forall A\]

Mixing maps F

\[\lim_{n\rightarrow\infty} \rho(F^{n}A\cap B) = \rho(A) \rho(B) \ \ \ \ \forall A,B\]
  • For measure-preserving maps: \[\mbox{Nondecomposable} \Rightarrow \mbox{Mixing}\]

Ergodic maps

A map T is ergodic if

\[\lim_{N\rightarrow \infty} \frac{1}{N} \sum_{n=0}^{N-1} f(F^nx) = \int\mathrm{d}\rho f(x)\]

for almost all orbits and arbitrary measureable f

  • Measure preserving and nondecomposable maps are ergodic

e.g. the Mean Squared Displacement in ergodic systems can be replaced by the time average:

\[ \langle (x(t)-x(0))^2 \rangle = \langle \frac{1}{T-t} \int_{0}^{T-t} \mathrm{d}t^\prime (x(t^\prime+t)-x(t^\prime))^2 \rangle \]

or, in discrete time,

\[ \langle (x(t)-x(0))^2 \rangle = \langle \frac{1}{T-t} \sum_{t^\prime=1}^{T-t} (x(t^\prime+t)-x(t^\prime))^2 \rangle. \]

Lyapunov Exponents

https://en.wikipedia.org/wiki/Lyapunov_exponent#/media/File:Orbital_instability_(Lyapunov_exponent).png
  • Chaotic systems exhibit sensitive dependence on initial conditions
  • Exponential approximation of escape of two nearby points \[ \left|F^t(x_0+\epsilon) - F^t(x_0)\right| \approx \epsilon e^{t\lambda(x_0)} \]
  • The Lyapunov exponent is defined for each initial point \[ \lambda (x_0) = \lim_{t\rightarrow \infty} \frac{1}{t} \sum_{n=1}^{t} \log|F^\prime(x_n)| \]

Fixed Points

\[ x_f=F(x_f) \]

Look at environment

\[ [ x_f-\epsilon, x_f+\epsilon ] \] \[\delta_{t+1} = |x_{t+1}-x_f|=|F(x_f\pm \delta_t)-x_f|=\delta_t \left|\frac{F(x_f\pm\delta_t)-F(x_f)}{\delta_t}\right|=\delta_t|F^\prime(x_f)|\] Stable fixed points \[ 1>\left|\frac{\mathrm{d}F}{\mathrm{d}x}(x_f)\right| \] Unstable fixed points \[ \left|\frac{\mathrm{d}F}{\mathrm{d}x}(x_f)\right|>1 \]

Logistic map

\[x_{t+1}=rx_{t}(1-x_{t}) \]

Fixed points?

\[ x_f = r x_f ( 1 - x_f ) = r x_f - rx_f^2 \] \[ rx_f^2 + (1-r) x_f = 0 \] \[ {x_f}_1 = 0 \; \; \; \; \; {x_f}_2 = \frac{r-1}{r} \]

Stability?

\[ F^\prime (x_f) = r (1-2x_f) = 2-r \]
https://fr.wikipedia.org/wiki/Suite_logistique
  • Number and types of fixed points depend on r
  • \[ 1>r>0 \Rightarrow x \rightarrow 0 \] \[ 3>r>1 \Rightarrow x \rightarrow \frac{r-1}{r} \] \[ 1+\sqrt{6} > r > 3 \Rightarrow x \rightarrow \mbox{ 2-periodic limit cycle } \] \[ 3.54 > r > 1+\sqrt{6} \Rightarrow x \rightarrow \mbox{ 4-periodic limit cycle } \] \[ r>3.54 \Rightarrow x \mbox{ chaotic, with infinite exceptions } \] \[ r>4 \Rightarrow x \mbox{ diverges } \]

Bifurcations

https://upload.wikimedia.org/wikipedia/commons/thumb/5/50/Logistic_Bifurcation_map_High_Resolution.png/800px-Logistic_Bifurcation_map_High_Resolution.png

The pendulum

\[\ddot{\theta} + g \sin(\theta) =0\]

It can be rewritten as

\[\dot{\theta}=p\] \[\dot{p}=-g\sin(\theta)\]

The damped pendulum

\[\ddot{\theta} + \gamma \dot{\theta} + g \sin(\theta) = 0 \]

In phase space

\[ \dot \theta = p \] \[ \dot p = -\gamma p - g \sin(\theta) \]
  • decays to zero

The driven pendulum

\[\ddot{\theta} + \gamma \dot{\theta} + g \sin(\theta) = A\cos(\omega t) \]

In phase space

\[ \dot \theta = p \] \[ \dot p = -\gamma p - g \sin(\theta) + A\cos(\omega t) \] simulation

The kicked rotor

\[ \dot \theta = p \] \[ \dot p = g \sin(\theta) \sum_{n=-\infty}^\infty \delta( n-t/T ) \] video