Response Theory via Generative Score Modeling

Area of Science:

• Complex Systems Analysis
• Statistical Physics
• Machine Learning

Background:

• Dynamical systems are susceptible to external perturbations.
• Understanding system responses is crucial for prediction and control.
• Conventional methods struggle with non-Gaussian statistics.

Purpose of the Study:

• To develop a novel approach for analyzing dynamical system responses to perturbations.
• To accurately estimate system responses, including those with non-Gaussian statistics.
• To provide a versatile tool for predicting the behavior of complex dynamical systems.

Main Methods:

• Combining score-based generative modeling with the generalized fluctuation-dissipation theorem.
• Numerical validation using time-series data.
• Application to stochastic partial differential equations (SPDEs) of increasing complexity.

Main Results:

• Accurate estimation of system responses, even with non-Gaussian statistics.
• Demonstrated improved accuracy over conventional methods.
• Successful application to Ornstein-Uhlenbeck process, stochastic Allen-Cahn equation, and 2D Navier-Stokes equations.

Conclusions:

• The proposed methodology offers a powerful and versatile tool for analyzing complex dynamical systems.
• It enables accurate prediction of statistical behavior under perturbations.
• Potential applications span various fields involving complex system dynamics.