Abstract:Data-driven model discovery of complex dynamical systems is typically done using sparse optimization, but it has a fundamental limitation: sparsity in that the underlying governing equations of the system contain only a small number of elementary mathematical terms. Examples where sparse optimization fails abound, such as the classic Ikeda or optical-cavity map in nonlinear dynamics and a large variety of ecosystems. Exploiting the recently articulated Kolmogorov-Arnold networks, we develop a general model-discovery framework for any dynamical systems including those that do not satisfy the sparsity condition. In particular, we demonstrate non-uniqueness in that a large number of approximate models of the system can be found which generate the same invariant set with the correct statistics such as the Lyapunov exponents and Kullback-Leibler divergence. An analogy to shadowing of numerical trajectories in chaotic systems is pointed out.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the limitations of existing data - driven model discovery methods based on sparse optimization when dealing with complex dynamical systems. Specifically, sparse optimization methods assume that the governing equations of the system have a sparse structure, that is, they can be represented by a small number of basic mathematical terms. However, many actual dynamical systems do not satisfy this sparse condition, such as the classical Ikeda map and various ecosystem models. ### Core of the problem 1. **Limitations of sparse optimization**: - Sparse optimization methods are only applicable to systems whose governing equations do indeed have a sparse structure. - For systems that do not satisfy the sparse condition (such as the Ikeda map and certain ecosystems), sparse optimization methods completely fail. 2. **Need for a new method**: - The paper proposes a new method based on the Kolmogorov - Arnold network (KAN) for discovering models of any dynamical system, including those that do not satisfy the sparse condition. - KAN makes use of the Kolmogorov - Arnold theorem, which states that any multivariate function can be decomposed into the sum of univariate functions. ### Key points of the solution - **Application of KAN**: By decomposing complex high - dimensional problems into simpler univariate functions, KAN can be trained more efficiently and provides better interpretability. - **Non - uniqueness**: The paper shows that a large number of approximate models can be found through KAN, and these models can generate the same invariant sets and have correct statistical properties (such as Lyapunov exponents and Kullback - Leibler divergence). - **Example verification**: The paper uses the Ikeda map and a chaotic ecosystem as examples to verify the effectiveness of KAN. ### Conclusion The main contribution of the paper is to propose a brand - new data - driven model discovery framework, which uses KAN to solve the complex dynamical system problems that existing sparse optimization methods cannot handle. This method can not only discover the model of the system, but also generate models with the same dynamic behavior as the real system, even if the mathematical forms of these models may be different or implicitly represented in numerical form. Through this method, researchers can be more widely applied to the modeling and prediction of various complex dynamical systems, thereby promoting the research progress in related fields.

Data-driven model discovery with Kolmogorov-Arnold networks

KAN-ODEs: Kolmogorov-Arnold Network Ordinary Differential Equations for Learning Dynamical Systems and Hidden Physics

A Survey on Kolmogorov-Arnold Network

Kolmogorov-Arnold Networks for Time Series: Bridging Predictive Power and Interpretability

Data-driven modeling of interrelated dynamical systems

A Comprehensive Survey on Kolmogorov Arnold Networks (KAN)

Kolmogorov--Arnold networks in molecular dynamics

Model discovery for nonautonomous translation-invariant problems

Automated Global Analysis of Experimental Dynamics through Low-Dimensional Linear Embeddings

Data-driven discovery of intrinsic dynamics

Discovering governing equations from data by sparse identification of nonlinear dynamical systems

Data-Driven Discovery of Conservation Laws from Trajectories via Neural Deflation

Learning Deep Neural Network Representations for Koopman Operators of Nonlinear Dynamical Systems

Physics Informed Kolmogorov-Arnold Neural Networks for Dynamical Analysis via Efficent-KAN and WAV-KAN

A purely data-driven framework for prediction, optimization, and control of networked processes

Data-driven models of nonautonomous systems

Discovering governing equations from data: Sparse identification of nonlinear dynamical systems

An evolutionary approach for discovering non-Gaussian stochastic dynamical systems based on nonlocal Kramers-Moyal formulas

Automatically discovering ordinary differential equations from data with sparse regression

Evolutionary sparse data-driven discovery of multibody system dynamics

Deep Hidden Physics Models: Deep Learning of Nonlinear Partial Differential Equations