Learning dynamical systems from data: Gradient-based dictionary optimization

Mohammad Tabish, Neil K. Chada, Stefan Klus

arXiv:2411.04775·math.DS·Published 2024-11-07·Updated 2025-07-01

The Koopman operator plays a crucial role in analyzing the global behavior of dynamical systems. Existing data-driven methods for approximating the Koopman operator or discovering the governing equations of the underlying system typically require a fixed set of basis functions, also called dictionary. The optimal choice of basis functions is highly problem-dependent and often requires domain knowledge. We present a novel gradient descent-based optimization framework for learning suitable and interpretable basis functions from data and show how it can be used in combination with EDMD, SINDy, and PDE-FIND. We illustrate the efficacy of the proposed approach with the aid of various benchmark problems such as the Ornstein-Uhlenbeck process, Chua's circuit, a nonlinear heat equation, as well as protein-folding data.

TopicsDynamical Systems & PDE Learning

Tagsdynamical-systems koopman-operator

arXiv categoriesmath.DS, cs.LG, stat.ML

arXiv abstract pagePDF