Abstract:This paper introduces a kernel-based approach for constructing approximations of the Koopman operators for semiflows in discrete time and orbital Koopman operators for continuous time semiflows. The primary advantage of the proposed construction is that the approximations follow certain rates of convergence which are dependent on how data samples fill certain subsets of the state space. In particular, we derive the rate of convergence for two scenarios: (1) the data samples Ξ$$\varXi $$ are dense in a compact state space X, and (2) the data samples Ξ$$\varXi $$ are dense in a limiting set Ω$$\varOmega $$ contained in an Euclidean space. Two general classes of Koopman operator approximations are considered in this paper, referred to as projection-based approximation and data-driven approximation. Projection-based approximations assume that the underlying dynamics governing the discrete or continuous time semiflows is known. On the other hand, data-driven approximations rely samples of the semiflow states to approximate the Koopman operator. In both types of approximations, the regularity of the underlying set and the smoothness of the space of functions on which the Koopman operator acts determine the rates of approximations. In the strongest error bounds derived in the paper, it is shown that the error in approximation of the Koopman operator decays like O(hΩn,Ωp)$$O(h_{\varOmega _n,\varOmega }^p)$$, where hΩn,Ω$$h_{\varOmega _n,\varOmega }$$ is the fill rate of the samples Ωn$$\varOmega _n$$ in the limiting set Ω$$\varOmega $$ and p is an exponent related to the choice of the kernel and the smoothness of functions on which the Koopman operator acts. Such error bounds are obtained when either the limiting subset Ω=X$$\varOmega =X$$, when it is a proper subset Ω⊂X$$\varOmega \subset X$$ that is sufficiently regular, or when it is a type of smooth manifold Ω=M$$\varOmega =M$$.

Efficient Computation of Invariance Proximity: Closed-Form Error Bounds for Finite-Dimensional Koopman-Based Models

A Convex Optimization Approach to Learning Koopman Operators

Approximation of the Koopman operator via Bernstein polynomials

Parallel Learning of Koopman Eigenfunctions and Invariant Subspaces for Accurate Long-Term Prediction

Error bounds for kernel-based approximations of the Koopman operator

Nonlinear Data-Driven Approximation of the Koopman Operator

Modeling Nonlinear Control Systems via Koopman Control Family: Universal Forms and Subspace Invariance Proximity

Physics-informed spectral approximation of Koopman operators

Approximation of discrete and orbital Koopman operators over subsets and manifolds

Learning Invariant Subspaces of Koopman Operators--Part 1: A Methodology for Demonstrating a Dictionary's Approximate Subspace Invariance

Koopman Theory and Linear Approximation Spaces

Generalizing dynamic mode decomposition: Balancing accuracy and expressiveness in Koopman approximations

On Numerical Approximations of the Koopman Operator

Two methods to approximate the Koopman operator with a reservoir computer

Learning Koopman Eigenfunctions and Invariant Subspaces from Data: Symmetric Subspace Decomposition

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

On the approximability of Koopman-based operator Lyapunov equations

Extracting Koopman Operators for Prediction and Control of Non-linear Dynamics Using Two-stage Learning and Oblique Projections

Sharp Spectral Rates for Koopman Operator Learning

Learning Koopman Invariant Subspaces for Dynamic Mode Decomposition

$L^\infty$-error bounds for approximations of the Koopman operator by kernel extended dynamic mode decomposition