Likelihood and Bayesian Prediction of Chaotic Systems
The ideas of mathematical chaos are demonstrated by using a nonlinear dynamical system \(x_{t+1}=f(x_ t;a)\), \(f\) is the logistic map \(x_{t+1}=ax_ t(1-x_ t)\), \(t=0,1,2,\dots\), with \(x_ 0\in[0,1]\), \(a\in[0,4]\). The aim is the prediction of \(x_ t\), \(t>n\), when \(y_ t=x_ t+\sigma e_ t\), \(t=1,\dots,n\), is observed. The parameters \(a\) or \(x_ 0\) are unknown. The random variables \(e_ i\) are iid \(N(0,1)\) distributed and \(\sigma\) is known. For large \(a\) the likelihood function behaves extremely irregular. Bayesian prediction gives high errors for long-time forecasts. ML- forecasts show well qualitative features of the process. The connection of ergodic theory and chaos is also discussed. An ergodic theorem of Hopf for dynamic systems is illustrated by using \[ f(x_ t; b)=\begin{cases} bx_ t &\text{if } 0\leq x_ t<0.5\\ b(1-x_ t) &\text{if } 0.5<x_ t\leq 1\end{cases} \] with \(0<b\leq 1\) (especially for \(b=2\)). Time discrete dynamical systems arise also by discretization of differential equations. An example is the Duffing equation for a pendulum that is attracted between two magnets. For this case interesting numerical results of Bayesian prediction are presented. It seems that short time prediction is possible in some cases in presence of chaos for models with unknown parameters.
- Sequentially constrained Monte Carlo
- Indistinguishable states. II: The imperfect model scenario
- Asymptotic inference for dynamical systems observed with error
- Bayesian learning when chaos looms large
- Approximate importance sampling Monte Carlo for data assimilation
- Prediction and predictability in population biology: noise and chaos
- Maximum likelihood estimation of stochastic chaos representations from experimental data
- Exploiting dynamical coherence: a geometric approach to parameter estimation in nonlinear models
- Projected Shadowing-Based Data Assimilation
- Shadowing-based data assimilation method for partially observed models
- Rising Above Chaotic Likelihoods
- Why preferring parametric forecasting to nonparametric methods?
- On prediction and chaos in stochastic systems
- Indistinguishable states. I: Pefect model scenario.
- Some models are useful, but how do we know which ones? Towards a unified Bayesian model taxonomy
- Examples of inconsistent Bayes procedures based on observations on dynamical systems
- A guide to using shadowing filters for forecasting and state estimation
- Statistical inference for dynamical systems: a review
- Chaotic secure communication based on nonlinear autoregressive filter with changeable parameters
- Sequential reconstruction of driving-forces from nonlinear nonstationary dynamics
This page was built for publication: Likelihood and Bayesian Prediction of Chaotic Systems
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4031116)
