Determining Time Delay for Chaotic Time Series Reconstruction Using Autocorrelation Method

In nonlinear time series analysis, phase space reconstruction serves as a fundamental step for studying the dynamic characteristics of chaotic systems. The autocorrelation method represents a classical approach for selecting time delay, which helps determine the optimal time interval when reconstructing the phase space.

The core principle of the autocorrelation method involves calculating the correlation between a time series and its lagged versions to identify an appropriate time delay. When the autocorrelation function first drops to a specific proportion of its initial value (typically 1-1/e), the corresponding lag time becomes the recommended time delay value. This method offers straightforward computation and intuitive interpretation, making it suitable as an initial reference for time delay selection. From an implementation perspective, this typically involves computing the autocorrelation function using methods like Fourier transform-based convolution or direct correlation summation, with common implementations found in libraries such as MATLAB's xcorr function or Python's numpy.correlate.

It's important to note that the autocorrelation method essentially measures linear correlations, while chaotic systems often exhibit nonlinear characteristics. Therefore, in practical applications, it may be necessary to combine other nonlinear methods (such as mutual information) to obtain more accurate reconstruction parameters. The primary advantage of the autocorrelation method lies in its computational efficiency, providing practical value for preliminary analysis and rapid estimation. Algorithmically, this method operates with O(n log n) complexity when using FFT-based implementation, making it significantly faster than mutual information methods for large datasets.

The selection of time delay directly affects the quality of phase space reconstruction - both excessively large and excessively small delays can lead to distortion of dynamic characteristics. The autocorrelation method provides quantitative basis for selecting this critical parameter, establishing itself as one of the fundamental tools in chaotic time series analysis. In code implementation, practitioners often visualize the autocorrelation function's decay to identify the point where it crosses the 1-1/e threshold, facilitating interactive parameter selection.