Multiplicative Dynamic Mode Decomposition

Nicolas Boullé,Matthew J. Colbrook
2024-05-09
Abstract:Koopman operators are infinite-dimensional operators that linearize nonlinear dynamical systems, facilitating the study of their spectral properties and enabling the prediction of the time evolution of observable quantities. Recent methods have aimed to approximate Koopman operators while preserving key structures. However, approximating Koopman operators typically requires a dictionary of observables to capture the system's behavior in a finite-dimensional subspace. The selection of these functions is often heuristic, may result in the loss of spectral information, and can severely complicate structure preservation. This paper introduces Multiplicative Dynamic Mode Decomposition (MultDMD), which enforces the multiplicative structure inherent in the Koopman operator within its finite-dimensional approximation. Leveraging this multiplicative property, we guide the selection of observables and define a constrained optimization problem for the matrix approximation, which can be efficiently solved. MultDMD presents a structured approach to finite-dimensional approximations and can more accurately reflect the spectral properties of the Koopman operator. We elaborate on the theoretical framework of MultDMD, detailing its formulation, optimization strategy, and convergence properties. The efficacy of MultDMD is demonstrated through several examples, including the nonlinear pendulum, the Lorenz system, and fluid dynamics data, where we demonstrate its remarkable robustness to noise.
Dynamical Systems,Machine Learning,Numerical Analysis,Optimization and Control,Spectral Theory
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to approximate the Koopman operator more accurately in nonlinear dynamical systems while maintaining its inherent multiplicative structure. The Koopman operator is an infinite - dimensional operator that can linearize nonlinear dynamical systems, thus facilitating the study of the spectral properties of the system and predicting the time evolution of observables. However, existing approximation methods usually need to select a set of observables to capture the finite - dimensional behavior of the system. This selection is often heuristic, which may lead to the loss of spectral information and may seriously complicate the structure - preservation problem. To this end, the paper introduces a new method - Multiplicative Dynamic Mode Decomposition (MultDMD). MultDMD guides the selection of observables by enforcing the multiplicative structure of the Koopman operator in the finite - dimensional approximation and defines a constrained optimization problem to solve the matrix approximation, which can be solved efficiently. This method provides a structured finite - dimensional approximation method that can more accurately reflect the spectral properties of the Koopman operator. Specifically, the main contributions of the paper include: 1. **Preservation of multiplicative structure**: By enforcing the multiplicative property of the Koopman operator in the finite - dimensional approximation, the structural consistency of the approximation method is ensured. 2. **Selection of observation functions**: The multiplicative property is used to guide the selection of observation functions, avoiding the problems caused by heuristic selection. 3. **Definition of optimization problem**: A specific constrained optimization problem for solving the matrix approximation is proposed, which can be solved efficiently. 4. **Theoretical framework**: The theoretical framework of MultDMD is elaborated in detail, including its formulation, optimization strategy and convergence properties. 5. **Numerical experiments**: The effectiveness of MultDMD is verified through multiple numerical experiments, including the nonlinear pendulum, Lorenz system and fluid dynamics data, showing its robustness in noisy environments. In summary, this paper aims to solve the structure - preservation and observation - function - selection problems in existing Koopman - operator approximation methods by introducing the MultDMD method, thereby providing a more accurate and more robust analysis tool for dynamical systems.