# Koopman Operator Dynamical Models: Learning, Analysis and Control

**Authors:** Petar Bevanda, Stefan Sosnowski, Sandra Hirche  
**arXiv:** [2102.02522](https://arxiv.org/abs/2102.02522)  
**Date:** 2021-02-04  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2102.02522v2/

## Key points

- The Koopman operator is a powerful tool for handling nonlinear systems, providing a (globally) linear representation.
- Finite approximations of the Koopman operator are necessary due to its infinite-dimensional nature, but there is currently a lack of an overarching framework for these approximations.
- Existing methods for learning finite approximations of the Koopman operator are often overshadowed by less structured and potentially ill-posed approaches.
- The study by Bevanda, Sosnowski, and Hirche aims to bridge the gap between Koopman operator theory and established system-theoretic concepts through data-driven representations.
- The paper categorizes existing methodologies into unstructured and structured approaches, highlighting key differences between them and providing insights into their potential applications in modeling control systems.

## Summary

, , , , 

The Koopman operator is a powerful tool for handling nonlinear systems, providing a (globally) linear representation. However, due to its infinite-dimensional nature, finite approximations are necessary and currently lack an overarching framework. While principled methods exist for learning these approximations, they are often overshadowed by less structured and potentially ill-posed approaches. Additionally, the deep-rooted connections between Koopman operator theory and established system-theoretic and dynamical system concepts are not universally acknowledged. In their recent study titled "Koopman Operator Dynamical Models: Learning, Analysis and Control," Petar Bevanda, Stefan Sosnowski, and Sandra Hirche aim to bridge this gap by exploring data-driven representations for Koopman operator dynamical models. They categorize existing methodologies into unstructured and structured approaches while highlighting key differences between them. Furthermore, they provide valuable insights into how this paradigm relates to system-theoretic notions and its potential applications in modeling control systems. The paper also addresses current challenges in the field and offers thoughtful commentary on future perspectives for advancing research in this area. By synthesizing theoretical foundations with practical considerations, Bevanda et al. 's work contributes to a deeper understanding of Koopman operator theory and its implications for modeling complex dynamical systems.