Abstract:Learning and solving governing equations of a physical system, represented by partial differential equations (PDEs), from data is a central challenge in a variety of areas of science and engineering. Traditional numerical methods for solving PDEs can be computationally expensive for complex systems and require the complete PDEs of the physical system. On the other hand, current data-driven machine learning methods require a large amount of data to learn a surrogate model of the PDE solution operator, which could be impractical. Here, we propose the first solution operator learning method that only requires one PDE solution, i.e., one-shot learning. By leveraging the principle of locality of PDEs, we consider small local domains instead of the entire computational domain and define a local solution operator. The local solution operator is then trained using a neural network, and utilized to predict the solution of a new input function via mesh-based fixed-point iteration (FPI), meshfree local-solution-operator informed neural network (LOINN) or local-solution-operator informed neural network with correction (cLOINN). We test our method on diverse PDEs, including linear or nonlinear PDEs, PDEs defined on complex geometries, and PDE systems, demonstrating the effectiveness and generalization capabilities of our method across these varied scenarios.

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is: **Learning the PDE solution operator from a single partial differential equation (PDE) solution, that is, achieving "one - shot learning"**. Specifically, traditional numerical methods and existing data - driven machine learning methods face the following challenges when solving PDEs: 1. **Limitations of traditional numerical methods**: - For complex systems, traditional numerical methods (such as the finite element method, the finite difference method, and the spectral method) are computationally expensive. - These methods usually require spatial discretization of the complete PDE of the physical system, and interpolation or other reconstruction methods are required when evaluating solutions at other points. - More importantly, these methods require a complete understanding of the underlying PDE. 2. **Limitations of existing data - driven machine learning methods**: - Current data - driven machine learning methods (such as physics - informed neural networks (PINNs)) can generate differentiable PDE solutions, but they require a large amount of training data to learn the PDE solution operator. - Acquiring or storing a large amount of experimental data or computational simulation data can be very expensive or infeasible in practical applications, especially in fields such as geophysics, climate modeling, and fluid dynamics. To solve these problems, the author proposes **a one - shot learning method based on the locality principle**, aiming to train a neural network using only one PDE solution (i.e., one - shot learning) to predict the PDE solution corresponding to a new input function. This method is achieved through the following steps: 1. **Selecting the local domain and the local solution operator**: - Consider a small local domain \(\tilde{\Omega}\) and define the local solution operator \(\tilde{G}\) within this domain. - Learn the local solution operator \(\tilde{G}\) through a neural network, which takes the solution \(u\) at auxiliary points and the forcing term \(f\) within the local domain as inputs and outputs the solution \(u(x^*)\) at the target point. 2. **Training the local solution operator**: - Utilize a single global PDE solution data set \(T=(f_T, u_T)\), place the local domain \(\tilde{\Omega}\) at different positions, and generate multiple input - output data pairs for training the local solution operator \(\tilde{G}\). 3. **Predicting the solution of a new input function**: - For new PDE conditions \(f\), use the pre - trained local solution operator \(\tilde{G}\), and find the global PDE solution through fixed - point iteration (FPI), local solution operator - informed neural network (LOINN), or local solution operator - informed neural network with correction (cLOINN). Through this method, the author demonstrates its effectiveness and generalization ability in multiple PDE types (including linear and nonlinear PDEs, PDEs on complex geometries, and PDE systems).

One-shot learning for solution operators of partial differential equations

Data Generation-based Operator Learning for Solving Partial Differential Equations on Unbounded Domains

Operator Learning Enhanced Physics-informed Neural Networks for Solving Partial Differential Equations Characterized by Sharp Solutions

Learning the solution operator of a nonlinear parabolic equation using physics informed deep operator network

Diffeomorphic Latent Neural Operators for Data-Efficient Learning of Solutions to Partial Differential Equations

Learning Partial Differential Equations with Deep Parallel Neural Operator

Local neural operator for solving transient partial differential equations on varied domains

Pseudo-Differential Neural Operator: Generalized Fourier Neural Operator for Learning Solution Operators of Partial Differential Equations

Learning Only On Boundaries: a Physics-Informed Neural operator for Solving Parametric Partial Differential Equations in Complex Geometries

Finite Operator Learning: Bridging Neural Operators and Numerical Methods for Efficient Parametric Solution and Optimization of PDEs

Learning the solution operator of parametric partial differential equations with physics-informed DeepONets

Data-Efficient Operator Learning via Unsupervised Pretraining and In-Context Learning

Newton Informed Neural Operator for Computing Multiple Solutions of Nonlinear Partials Differential Equations

Solving Partial Differential Equations in Different Domains by Operator Learning method Based on Boundary Integral Equations

Physics-Informed Neural Operator for Learning Partial Differential Equations

LeMON: Learning to Learn Multi-Operator Networks

Variational operator learning: A unified paradigm marrying training neural operators and solving partial differential equations

Physics-Informed Deep Inverse Operator Networks for Solving PDE Inverse Problems

Transformer for Partial Differential Equations' Operator Learning