Abstract:Our recent intensive study has found that physics-informed neural networks (PINN) tend to be local approximators after training. This observation leads to this novel physics-informed radial basis network (PIRBN), which can maintain the local property throughout the entire training process. Compared to deep neural networks, a PIRBN comprises of only one hidden layer and a radial basis "activation" function. Under appropriate conditions, we demonstrated that the training of PIRBNs using gradient descendent methods can converge to Gaussian processes. Besides, we studied the training dynamics of PIRBN via the neural tangent kernel (NTK) theory. In addition, comprehensive investigations regarding the initialisation strategies of PIRBN were conducted. Based on numerical examples, PIRBN has been demonstrated to be more effective and efficient than PINN in solving PDEs with high-frequency features and ill-posed computational domains. Moreover, the existing PINN numerical techniques, such as adaptive learning, decomposition and different types of loss functions, are applicable to PIRBN. The programs that can regenerate all numerical results can be found at <a class="link-external link-https" href="https://github.com/JinshuaiBai/PIRBN" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the difficulties encountered by existing Physics - Informed Neural Networks (PINNs) when solving Partial Differential Equations (PDEs) with high - frequency features or ill - posed computational domains. Specifically: 1. **Limitations of PINNs**: - When dealing with problems of high - frequency features or ill - posed computational domains, PINNs are more difficult to train and have poor prediction effects. - During the training process, PINNs tend to become local approximators, but for some problems, this characteristic is difficult to achieve. 2. **Proposed new method**: - The paper proposes a new neural network structure - the Physics - Informed Radial Basis Network (PIRBN), which naturally has the local approximation property. - PIRBN contains only one hidden layer and uses the radial basis function as the activation function, and can maintain the local approximation property throughout the training process. 3. **Theoretical analysis and verification**: - Through the Neural Tangent Kernel (NTK) theory, it is proved that the training process of PIRBN can converge to a Gaussian process. - The initialization strategy, the number of sample points, and the selection of the radial basis activation function of PIRBN are studied to ensure its effectiveness and stability under different conditions. - Through numerical experiments, the superior performance of PIRBN in solving PDE problems with high - frequency features and ill - posed computational domains is verified. ### Formula summary - **PDE and boundary conditions**: \[ G[u](x) = 0, \quad x \in \Omega \] \[ B[u](x) = 0, \quad x \in \partial\Omega \] - **PINN's loss function**: \[ \mathcal{L}(\theta) = \mathcal{L}_g(\theta) + \mathcal{L}_b(\theta) \] where, \[ \mathcal{L}_g(\theta) = \frac{1}{n_g} \sum_{i = 1}^{n_g} \left( G[u_\theta](x_i) \right)^2 \] \[ \mathcal{L}_b(\theta) = \frac{1}{n_b} \sum_{i = 1}^{n_b} \left( B[u_\theta](x_i) \right)^2 \] - **Covariance matrix in NTK theory**: \[ \Sigma_g(x, x') = \mathbb{E}_{\theta} \left[ \nabla_\theta G[u_\theta](x) \cdot \nabla_\theta G[u_\theta](x') \right] \] \[ \Sigma_b(x, x') = \mathbb{E}_{\theta} \left[ \nabla_\theta B[u_\theta](x) \cdot \nabla_\theta B[u_\theta](x') \right] \] - **Output formula of PIRBN**: \[ y(x) = \sum_{i = 1}^d a_i \phi(\| x - c_i \|_2) \] where $\phi(r) = e^{-b r^2}$ is the Gaussian radial basis function. ### Conclusion By introducing PIRBN, the paper solves the limitations of PINNs in solving complex PDEs, especially when dealing with high - frequency features and ill - posed computational domains.

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

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

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

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

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

Physics-informed attention-based neural network for solving non-linear partial differential equations

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

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

MRF-PINN: A Multi-Receptive-Field convolutional physics-informed neural network for solving partial differential equations

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

Recipes for when physics fails: recovering robust learning of physics informed neural networks

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

Physics-informed neural networks combined with polynomial interpolation to solve nonlinear partial differential equations

RANG: A Residual-based Adaptive Node Generation Method for Physics-Informed Neural Networks

Physics-informed attention-based neural network for hyperbolic partial differential equations: application to the Buckley-Leverett problem

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

Deep NURBS -- Admissible Physics-informed Neural Networks

Physics-Informed Boundary Integral Networks (PIBI-Nets): A Data-Driven Approach for Solving Partial Differential Equations

iPINNs: incremental learning for Physics-informed neural networks

NeuralPDE: Automating Physics-Informed Neural Networks (PINNs) with Error Approximations