Reconstructing dynamical systems as zero-noise limits

TL;DR

This paper introduces a data-driven method to approximate deterministic dynamical systems as zero-noise limits of Markov processes, enabling the approximation of invariant sets and asymptotic properties.

Contribution

It presents a novel approach to reconstruct dynamical systems through zero-noise Markov process limits, capturing invariant sets and asymptotic behaviors.

Findings

The method approximates invariant sets with arbitrary topology.

It provides a low-noise approximation of the dynamics.

Under certain conditions, it converges statistically to true orbits.

Abstract

A dynamical system may be defined by a simple transition law - such as a map or a vector field. The objective of most learning techniques is to reconstruct this dynamic transition law. This is a major shortcoming, as most dynamic properties of interest are asymptotic properties such as an attractor or invariant measure. Thus approximating the dynamical law may not be sufficient to approximate these asymptotic properties. This article presents a method of representing a discrete-time deterministic dynamical system as the zero-noise limit of a Markov process. The Markov process approximation is completely data-driven. Besides proving a low-noise approximation of the dynamics the process also approximates the invariant set, via the support of its stationary measures. Thus invariant sets of arbitrary dynamical systems, even with complicated non-smooth topology, can be approximated by this technique. Under further assumptions, we show that the technique performs a convergent statistical approximation as well as approximations of true orbits.