Abstract:Ever since deep learning was introduced in the calculation of partial differential equation (PDE), there has been a lot of interests on real time response of system where the kernel function plays an important role. As a popular tool in recent years, physics-informed neural networks (PINNs) was proposed to perform a mesh-free, semi-supervised learning with high flexibility. This paper explores the integration of Lie symmetry groups with deep learning techniques to enhance the numerical solutions of fundamental solution in PDE. We propose a novel approach that combines the strengths of PINN and Lie group theory to address the computational inefficiencies in traditional methods. By incorporating the linearized symmetric condition (LSC) derived from Lie symmetries into PINNs, we introduce a new type of residual loss with lower order of derivative needed to calculate. This integration allows for significant reductions in computational costs and improvements in solution precision. Numerical simulation shows that our method can achieve up to a 50\% reduction in training time while maintaining good accuracy. Additionally, we provide a general theoretical framework to identify invariant infinitesimal generators for arbitrary Cauchy problems. This unsupervised algorithm does not require prior numerical solutions, making it both practical and efficient for various applications. Our contributions demonstrate the potential of combining symmetry analysis with deep learning to advance the field of scientific machine learning.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to improve the computational efficiency and accuracy of fundamental solutions of partial differential equations (PDEs), especially in real - time response systems where kernel functions play an important role**. Specifically, traditional numerical methods have problems such as high computational cost and limited accuracy when solving PDEs. This paper proposes a new method, GsPINN (Green kernel solver based on symmetric Physics - Informed neural networks), which combines Lie symmetry group theory and Physics - Informed Neural Networks (PINN) to address these problems. By introducing the linearized symmetry condition (LSC) and incorporating it into the loss function of PINN, GsPINN can significantly reduce the computational cost and improve the accuracy of the solution. The following are the key points of this paper: 1. **Problem Background**: - The application of deep learning in PDE solving has attracted wide attention in recent years. - PINN, as a mesh - free, semi - supervised learning method, is highly flexible, but traditional methods still have the problem of low computational efficiency when dealing with complex PDEs. 2. **Research Motivation**: - In order to improve the computational efficiency and accuracy of fundamental solutions of PDEs, especially in real - time response systems, the role of kernel functions is crucial. - Traditional PINN methods require additional computational load when dealing with specific boundary and initial conditions, so more efficient algorithms need to be developed. 3. **Proposed Method**: - GsPINN combines the advantages of Lie symmetry group theory and PINN, simplifies and solves complex PDEs by identifying and using the symmetry of PDEs. - Introduce the linearized symmetry condition (LSC) and add it as a new residual loss term to PINN, thereby transferring the calculation of high - order derivatives to low - order derivatives, significantly reducing the computational cost. 4. **Main Contributions**: - Propose a new symmetric residual term based on Lie group and algebra theory. - Provide a general theoretical analysis for finding invariant infinitesimal generators in arbitrary Cauchy problems. - The algorithm is unsupervised and does not require a priori numerical solutions. - Numerical experiments show that this method can reduce the training time by up to 50% while improving the accuracy of the solution. In summary, this paper aims to develop an efficient and accurate method for solving fundamental solutions of PDEs by combining Lie symmetry group theory and deep learning techniques to meet the requirements of real - time response systems.

GsPINN: A novel fast Green kernel solver based on symmetric Physics-Informed neural networks

Enforcing continuous symmetries in physics-informed neural network for solving forward and inverse problems of partial differential equations

Generalized conditional symmetry enhanced physics-informed neural network and application to the forward and inverse problems of nonlinear diffusion equations

A Spatial-Temporally Adaptive PINN Framework for 3D Bi-Ventricular Electrophysiological Simulations and Parameter Inference

Enforcing generalized conditional symmetry in physics-informed neural network for solving the KdV-like equation with Robin initial/boundary conditions

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

General-Kindred Physics-Informed Neural Network to the Solutions of Singularly Perturbed Differential Equations

ST-PINN: A Self-Training Physics-Informed Neural Network for Partial Differential Equations

Invariant deep neural networks under the finite group for solving partial differential equations

SPINN: Sparse, Physics-based, and partially Interpretable Neural Networks for PDEs

Separable Physics-informed Neural Networks for Solving the BGK Model of the Boltzmann Equation

SK-PINN: Accelerated physics-informed deep learning by smoothing kernel gradients

Physics-informed graph neural Galerkin networks: A unified framework for solving PDE-governed forward and inverse problems

Binary structured physics-informed neural networks for solving equations with rapidly changing solutions

RBF-MGN:Solving spatiotemporal PDEs with Physics-informed Graph Neural Network

PhyGNNet: Solving spatiotemporal PDEs with Physics-informed Graph Neural Network

Physics-Informed Neural Networks with Trust-Region Sequential Quadratic Programming

Combining physics-informed graph neural network and finite difference for solving forward and inverse spatiotemporal PDEs

Physics-Informed neural network solver for numerical analysis in geoengineering

Lie Point Symmetry and Physics Informed Networks

Finite-PINN: A Physics-Informed Neural Network Architecture for Solving Solid Mechanics Problems with General Geometries