Automatically identifying ordinary differential equations from data

Kevin Egan,Weizhen Li,Rui Carvalho

2023-05-03

Abstract:Discovering nonlinear differential equations that describe system dynamics from empirical data is a fundamental challenge in contemporary science. Here, we propose a methodology to identify dynamical laws by integrating denoising techniques to smooth the signal, sparse regression to identify the relevant parameters, and bootstrap confidence intervals to quantify the uncertainty of the estimates. We evaluate our method on well-known ordinary differential equations with an ensemble of random initial conditions, time series of increasing length, and varying signal-to-noise ratios. Our algorithm consistently identifies three-dimensional systems, given moderately-sized time series and high levels of signal quality relative to background noise. By accurately discovering dynamical systems automatically, our methodology has the potential to impact the understanding of complex systems, especially in fields where data are abundant, but developing mathematical models demands considerable effort.

Machine Learning,Dynamical Systems,Computational Physics

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve

This paper aims to address the problem of automatically identifying ordinary differential equations (ODEs) that describe system dynamics from data. Specifically:

1. **Automatic Discovery of Dynamic Models**: The study proposes a method that combines signal denoising, sparse regression, and confidence interval estimation to automatically identify the mathematical model of a dynamic system. This method does not require prior knowledge of the form of the equations.

2. **Handling Noisy Data**: The proposed method can tolerate low to moderate levels of noise, requiring only the assumption of model sparsity and the presence of control terms in the design matrix.

3. **Validating Algorithm Performance**: The authors tested the algorithm on synthetic datasets generated from known ordinary differential equations, exploring its performance under different initial conditions, time series lengths, and noise intensities.

4. **Comparison with Existing Methods**: The paper specifically compares this method with the existing SINDy with AIC method, demonstrating its superiority in 3-dimensional systems, especially with moderately sized datasets and higher signal quality.

In summary, this paper aims to develop an automated tool for extracting mathematical models that describe system dynamics from limited and noisy data, and to validate its effectiveness in various scenarios.