Data-driven approximation of Koopman operators and generators: Convergence rates and error bounds

Liam Llamazares-Elias,Samir Llamazares-Elias,Jonas Latz,Stefan Klus
2024-06-19
Abstract:Global information about dynamical systems can be extracted by analysing associated infinite-dimensional transfer operators, such as Perron-Frobenius and Koopman operators as well as their infinitesimal generators. In practice, these operators typically need to be approximated from data. Popular approximation methods are extended dynamic mode decomposition (EDMD) and generator extended mode decomposition (gEDMD). We propose a unified framework that leverages Monte Carlo sampling to approximate the operator of interest on a finite-dimensional space spanned by a set of basis functions. Our framework contains EDMD and gEDMD as special cases, but can also be used to approximate more general operators. Our key contributions are proofs of the convergence of the approximating operator and its spectrum under non-restrictive conditions. Moreover, we derive explicit convergence rates and account for the presence of noise in the observations. Whilst all these results are broadly applicable, they also refine previous analyses of EDMD and gEDMD. We verify the analytical results with the aid of several numerical experiments.
Numerical Analysis,Dynamical Systems
What problem does this paper attempt to address?
The paper aims to address the problem of extracting global information related to the Koopman operator and its generator in dynamical systems. Specifically, the authors propose a unified framework that utilizes Monte Carlo sampling techniques to approximate operators of interest (such as the Koopman operator or its generator) in a finite-dimensional space. The main contributions of the paper include: 1. **Unified Framework**: Introduces a general framework for studying data-driven dynamical system analysis methods (e.g., Extended Dynamic Mode Decomposition (EDMD) and Generator Extended Dynamic Mode Decomposition (gEDMD)), and uses Monte Carlo approximation to estimate the linear operator A within this framework. 2. **Convergence Proof**: Proves the almost everywhere convergence of the proposed approximate operator and its spectrum, and shows that they converge to the eigenvalues and eigenfunctions of the original operator along a subsequence. 3. **Error Bound**: Derives bounds for the approximation error and extends these results to scenarios with observational noise. 4. **Minimal Assumptions**: The above results are derived under minimal assumptions, which are described at the end of Section 2. Through these efforts, the paper not only improves the analysis of existing methods (such as EDMD and gEDMD) but also provides a more general approach to handling a broader range of operator approximation problems. Additionally, the theoretical results are validated through numerical experiments. Overall, the paper aims to address the limitations of existing methods in practical applications and provide a more robust theoretical foundation.