Manifold learning of Poincar\'e sections reveals the topology of high-dimensional chaotic flows

TL;DR

This paper demonstrates that manifold learning applied to Poincaré sections of high-dimensional chaotic systems can reveal their underlying low-dimensional topology, aiding in the construction of simplified models like maps and trees.

Contribution

It introduces a novel approach combining manifold learning with Poincaré sections to uncover the topological structure of high-dimensional chaotic flows.

Findings

Manifold learning provides intrinsic coordinates for Poincaré sections.

The method reduces complex flows to low-dimensional maps and trees.

Classical kneading theory detects periodic orbits systematically.

Abstract

It is shown that applying manifold learning techniques to Poincar\'e sections of high-dimensional, chaotic dynamical systems can uncover their low-dimensional topological organization. Manifold learning provides a low-dimensional embedding and intrinsic coordinates for the parametrization of data on the Poincar\'e section, facilitating the construction of return maps with well defined symbolic dynamics. The method is illustrated by numerical examples for the R\"ossler attractor and the Kuramoto-Sivashinsky equation. For the latter we present the reduction of the high-dimensional, continuous-time flow to dynamics on one- and two two-dimensional Poincar\'e sections. We show that in the two-dimensional embedding case the attractor is organized by one-dimensional unstable manifolds of short periodic orbits. In that case, the dynamics can be approximated by a map on a tree which can in turn be reduced to a trimodal map of the unit interval. In order to test the limits of the one-dimensional map approximation we apply classical kneading theory in order to systematically detect all periodic orbits of the system up to any given topological length.