Testing topological conjugacy of time series

TL;DR

This paper develops statistical tests to determine if two dynamical systems are topologically conjugate based on finite samples, with proven convergence and practical numerical examples demonstrating effectiveness.

Contribution

It introduces new tests for topological conjugacy from finite data, including methods for Takens' embedding and observable time series comparison.

Findings

Tests are close to zero for conjugated systems

Tests diverge for non-conjugated systems

Methods are scalable and robust

Abstract

This paper considers a problem of testing, from a finite sample, a topological conjugacy of two dynamical systems $(X,f)$ and $(Y,g)$. More precisely, given $x_1,\ldots, x_n \subset X$ and $y_1,\ldots,y_n \subset Y$ such that $x_{i+1} = f(x_i)$ and $y_{i+1} = g(y_i)$ as well as $h: X \rightarrow Y$, we deliver a number of tests to check if $f$ and $g$ are topologically conjugated via $h$. The values of the tests are close to zero for conjugated systems and large for systems that are not conjugated. Convergence of the test values, in case when sample size goes to infinity, is established. A number of numerical examples indicating scalability and robustness of the methods are given. In addition, we show how the presented method specialize to a test of sufficient embedding dimension in Takens' embedding theorem. Our methods also apply to the situation when we are given two observables of deterministic processes, of a form of one or higher dimensional time-series. In this case, their similarity can be accessed by comparing the dynamics of their Takens' reconstructions.