Recovering smooth dynamics from time series with the aid of recurrence plots

A graphical method based on recurrence plots is used in the reconstruction of the phase space from a time series of measurements. It is demonstrated that if the embedding delay and dimension are correctly chosen, the recurrence plot of a smooth dynamical system has a particularly simple form. It is shown how to use recurrence plots to determine the correct embedding parameters so that reliable quantitative information can be drawn about the system generating the time series. The average line length in the plot is shown to be directly related to the prediction horizon. Furthermore, it is a numerical characteristic of the embedded series independent of the threshold used in the plot. [S1063-651X(99)04206-3].