Abstract:While many recent Physics-Informed Neural Networks (PINNs) variants have had considerable success in solving Partial Differential Equations, the empirical benefits of feature mapping drawn from the broader Neural Representations research have been largely overlooked. We highlight the limitations of widely used Fourier-based feature mapping in certain situations and suggest the use of the conditionally positive definite Radial Basis Function. The empirical findings demonstrate the effectiveness of our approach across a variety of forward and inverse problem cases. Our method can be seamlessly integrated into coordinate-based input neural networks and contribute to the wider field of PINNs research.

What problem does this paper attempt to address?

This paper mainly discusses the use of feature mapping in Physics-Informed Neural Networks (PINNs), particularly addressing the limitations of Fourier-based feature mapping. The authors propose using conditionally positive definite radial basis functions (RBFs) as an alternative method to improve the performance of PINNs in solving partial differential equations (PDEs). The research finds that although Fourier feature mapping is ineffective in certain cases, such as in predicting errors when dealing with jump solutions in the Burgers equation and performing poorly in high-dimensional problems, RBF feature mapping can effectively overcome these issues and demonstrate superior performance in various forward and inverse tasks. The paper compares multiple feature mapping methods through experiments, including basic Fourier features, positional encoding, random Fourier features, quadratic features, and conditionally positive definite RBFs (RBF-P). The results show that RBF-P significantly improves accuracy in solving diffusion, heat conduction, Poisson, Burgers, and Navier-Stokes equations, especially for nonlinear and inverse problems. Additionally, RBF-P exhibits higher robustness in handling noisy data. The paper also introduces the mathematical principles of RBF-P, emphasizing the importance of positive definiteness and symmetry conditions, and verifies the advantages of RBF-P in PINNs through experiments. The study includes comparisons of different RBF types and ablative studies on the number of RBFs and polynomial terms. In summary, this paper unveils the limitations of Fourier feature mapping and presents a new and efficient feature mapping strategy, namely RBF-P, which is crucial for enhancing the performance of PINNs in scientific computing.

RBF-PINN: Non-Fourier Positional Embedding in Physics-Informed Neural Networks

Feature Mapping in Physics-Informed Neural Networks (PINNs)

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

fPINNs: Fractional Physics-Informed Neural Networks

Physics informed neural networks for an inverse problem in peridynamic models

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

Diminishing Spectral Bias in Physics-Informed Neural Networks Using Spatially-Adaptive Fourier Feature Encoding

A Class of Improved Fractional Physics Informed Neural Networks

Deep fuzzy physics-informed neural networks for forward and inverse PDE problems

Physics-informed Radial Basis Network (PIRBN): A Local Approximating Neural Network for Solving Nonlinear Partial Differential Equations

Physics-Informed Kernel Function Neural Networks for Solving Partial Differential Equations

Physics-informed radial basis network (PIRBN): A local approximating neural network for solving nonlinear PDEs

Fourier PINNs: From Strong Boundary Conditions to Adaptive Fourier Bases

ELM-FBPINN: efficient finite-basis physics-informed neural networks

B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data

PINNsFormer: A Transformer-Based Framework For Physics-Informed Neural Networks

Machine Learning Through Physics–Informed Neural Networks: Progress and Challenges

PINNfluence: Influence Functions for Physics-Informed Neural Networks

Bi-orthogonal fPINN: A physics-informed neural network method for solving time-dependent stochastic fractional PDEs