What problem does this paper attempt to address?

The problem this paper attempts to address is: how to improve the training accuracy and efficiency of Physics-Informed Neural Networks (PINNs) when solving time-dependent partial differential equations (PDEs) with periodic boundary conditions. ### Background and Motivation Time-dependent partial differential equations are widely present in many scientific and engineering models, used to describe various physical phenomena. When these equations are combined with periodic boundary conditions, they can effectively model phenomena over infinite physical domains. However, solving such equations often faces significant challenges, especially when dealing with stiff PDEs. Traditional data-driven machine learning methods, although successful in some fields, often require a large amount of training data and lack an understanding of physical problems, leading to poor interpretability of the models. Moreover, in many scientific fields, obtaining high-fidelity data is both expensive and time-consuming. ### Research Objectives To address the above challenges, the authors propose a structure-preserving PINN algorithm that embeds initial conditions (IC) and boundary conditions (BC) directly into the neural network structure, thereby reducing dependence on spectral bases, improving training speed, and solution accuracy. Specifically, this method aims to: 1. **Reduce training difficulty**: By embedding IC and BC information into the neural network structure, reduce dependence on data and simplify the training process. 2. **Improve solution accuracy**: Through a structure-preserving approach, enhance the accuracy and stability of PINNs in solving stiff PDEs. 3. **Expand application scope**: This method is applicable not only to PDEs with scalar outputs but also to those with vector and complex outputs. ### Method Overview 1. **Structure-preserving PINN**: By designing specific transformation functions, embed IC and BC information directly into the neural network structure, naturally satisfying these conditions during training. 2. **Handling periodic boundary conditions**: For periodic boundary conditions, design special network structures to ensure that the network output naturally meets periodic conditions. 3. **Training techniques**: Combine mini-batch training and other PINN variants (such as SA-PINN, RBA-PINN, etc.) to further improve training effectiveness. ### Experimental Validation The authors validated the effectiveness of this method on various typical time-dependent PDEs, including: - **Viscous Burger's equation** - **Allen-Cahn equation** - **Cahn-Hilliard equation** - **Kuramoto-Sivashinsky equation** - **Gray-Scott equation** - **Belousov-Zhabotinsky equation** - **Nonlinear Schrödinger equation** Experimental results show that the structure-preserving PINN exhibits higher accuracy and stability in solving these equations, especially when dealing with problems with sharp moving interfaces. ### Conclusion This paper introduces a structure-preserving PINN method that significantly improves the training accuracy and efficiency of solving time-dependent PDEs with periodic boundary conditions by embedding IC and BC information into the neural network structure. Future work will focus on extending this method to 2D or 3D PDEs and further optimizing the design of transformation functions.

Structure Preserving PINN for Solving Time Dependent PDEs with Periodic Boundary

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

Multistep Asymptotic Pre-Training Strategy Based on Pinns for Solving Steep Boundary Singular Perturbation Problems

Enhancing PINNs for solving PDEs via adaptive collocation point movement and adaptive loss weighting

Mass-preserving Spatio-temporal adaptive PINN for Cahn-Hilliard equations with strong nonlinearity and singularity

Pre-training strategy for solving evolution equations based on physics-informed neural networks

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

A Physics Informed Neural Network for Time-Dependent Nonlinear and Higher Order Partial Differential Equations

PPINN: Parareal physics-informed neural network for time-dependent PDEs

A unified scalable framework for causal sweeping strategies for Physics-Informed Neural Networks (PINNs) and their temporal decompositions

Solving free-surface problems for non-shallow water using boundary and initial conditions-free physics-informed neural network (bif-PINN)

Efficient Discrete Physics-informed Neural Networks for Addressing Evolutionary Partial Differential Equations

iPINNs: incremental learning for Physics-informed neural networks

DF-ParPINN: parallel PINN based on velocity potential field division and single time slice focus

NAS-PINN: Neural architecture search-guided physics-informed neural network for solving PDEs

Physics-informed neural networks with parameter asymptotic strategy for learning singularly perturbed convection-dominated problem

TCAS-PINN: Physics-informed neural networks with a novel temporal causality-based adaptive sampling method

Semi-analytic PINN methods for singularly perturbed boundary value problems

Adaptive Learning Rate Residual Network Based on Physics-Informed for Solving Partial Differential Equations

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

Adaptive Task Decomposition Physics-Informed Neural Networks