Markov chain Monte Carlo estimation of nonlinear dynamics from time series (Q5950434)

The authors show that the Markov chain Monte Carlo (shortly MCMC) technique offers a practical means of analyzing time series that are contaminated with measurement noise. In many cases, it provides markedly superior formance to ordinary least-squares regression. The incorporation of dynamical noise into the estimation, as reported in this paper, allows for a certain amount of mismatch between the model and the dynamics. The authors state that the Bayesian framework is ideal for such estimation, because it allows explicit statement (using the prior probability distributions on parameters) of the confidence in different parts of the model.