Abstract:Neural networks are increasingly being used to solve partial differential equations (PDEs), replacing slower numerical solvers. However, a critical issue is that neural PDE solvers require high-quality ground truth data, which usually must come from the very solvers they are designed to replace. Thus, we are presented with a proverbial chicken-and-egg problem. In this paper, we present a method, which can partially alleviate this problem, by improving neural PDE solver sample complexity -- Lie point symmetry data augmentation (LPSDA). In the context of PDEs, it turns out that we are able to quantitatively derive an exhaustive list of data transformations, based on the Lie point symmetry group of the PDEs in question, something not possible in other application areas. We present this framework and demonstrate how it can easily be deployed to improve neural PDE solver sample complexity by an order of magnitude.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the "chicken - or - egg" dilemma faced by neural networks when solving partial differential equations (PDEs). Specifically, neural PDE solvers require high - quality real - data for training, and this data can usually only be generated by traditional numerical solvers. However, these traditional solvers are exactly what the neural solvers aim to replace. Therefore, obtaining sufficient training data becomes a difficult problem. To solve this problem, the author proposes a data augmentation method (LPSDA) based on Lie point symmetry. This method utilizes the Lie point symmetry of PDEs and generates new training samples through a series of continuous transformations, thereby reducing the need for a large amount of real - data and improving the sample complexity of neural PDE solvers, that is, achieving better performance with less data. ### Specific Problem Description 1. **Dilemma in Data Acquisition**: Neural PDE solvers require a large amount of high - quality real - data for training, but this data usually comes from traditional numerical solvers, which forms a circular - dependency problem. 2. **Improving Sample Complexity**: How to train a neural PDE solver with good performance on a limited data set and reduce the dependence on a large amount of real - data. ### Solution The method proposed by the author - Lie point symmetry data augmentation (LPSDA) - solves the above problems in the following ways: - **Utilizing Lie point symmetry**: For a given PDE, a set of data transformations based on its Lie point symmetry group can be quantitatively derived, and these transformations can maintain the invariance of the solution. - **Data Augmentation**: By applying these transformations to the original data, new training samples are generated, thereby increasing the diversity of the data and improving the generalization ability of the model. - **Improving Efficiency**: Compared with directly using traditional solvers to generate a large amount of data, LPSDA can quickly generate new data through mathematical transformations based on the existing data, greatly improving the training efficiency. ### Experimental Verification The author verifies the effectiveness of LPSDA through multiple experiments, including different PDEs (such as the KdV equation, the KS equation, and the Burgers equation), different models (such as ResNet and Fourier Neural Operator), and different training methods (such as the neural operator method and the autoregressive method). The experimental results show that LPSDA can significantly improve the sample complexity of neural PDE solvers, reduce the amount of required training data, and show better stability and accuracy in long - sequence prediction. In conclusion, this paper effectively alleviates the problem of insufficient training data for neural PDE solvers by introducing the Lie point symmetry data augmentation method and improves the generalization ability and training efficiency of the model.

Lie Point Symmetry Data Augmentation for Neural PDE Solvers

Lie Point Symmetry and Physics Informed Networks

Non-equispaced Fourier Neural Solvers for PDEs

Adversarial Learning for Neural PDE Solvers with Sparse Data

Lie Algebra Canonicalization: Equivariant Neural Operators under arbitrary Lie Groups

Solving partial differential equations with sampled neural networks

Physics-guided Data Augmentation for Learning the Solution Operator of Linear Differential Equations

Graph Neural PDE Solvers with Conservation and Similarity-Equivariance

Learning Partial Differential Equations from Data Using Neural Networks

Solving High-Dimensional PDEs with Latent Spectral Models

Solving Partial Differential Equations Using Point-Based Neural Networks.

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

Symmetry group based domain decomposition to enhance physics-informed neural networks for solving partial differential equations

A Natural Primal-Dual Hybrid Gradient Method for Adversarial Neural Network Training on Solving Partial Differential Equations

Neural PDE Solvers for Irregular Domains

Message Passing Neural PDE Solvers

An improved data-free surrogate model for solving partial differential equations using deep neural networks

Generating synthetic data for neural operators

Constrained or Unconstrained? Neural-Network-Based Equation Discovery from Data

Latent Neural PDE Solver: a reduced-order modelling framework for partial differential equations

Solving partial differential equations using large-data models: a literature review