Data-driven discovery of Green's functions with human-understandable deep learning

Abstract

There is an opportunity for deep learning to revolutionize science and technology by revealing its findings in a human interpretable manner. To do this, we develop a novel data-driven approach for creating a human-machine partnership to accelerate scientific discovery. By collecting physical system responses under excitations drawn from a Gaussian process, we train rational neural networks to learn Green's functions of hidden linear partial differential equations. These functions reveal human-understandable properties and features, such as linear conservation laws and symmetries, along with shock and singularity locations, boundary effects, and dominant modes. We illustrate the technique on several examples and capture a range of physics, including advection-diffusion, viscous shocks, and Stokes flow in a lid-driven cavity.

54 pages, 23 figures

References in corpus (2)

• Green's matrices of second order elliptic systems with measurable coefficients in two dimensional domains
• Learning elliptic partial differential equations with randomized linear algebra

Cited by in corpus (9)

• Learning elliptic partial differential equations with randomized linear algebra
• A Mathematical Guide to Operator Learning
• Elliptic PDE learning is provably data-efficient
• Learning Domain-Independent Green's Function For Elliptic Partial Differential Equations
• Learning the structure of wind: A data-driven nonlocal turbulence model for the atmospheric boundary layer
• Data-assisted, physics-informed propagators for recurrent flows
• Principled interpolation of Green's functions learned from data
• Neural Green's Operators for Parametric Partial Differential Equations
• Fourier Neural Operators for Time-Periodic Quantum Systems: Learning Floquet Hamiltonians, Observable Dynamics, and Operator Growth