Abstract:A machine-learnable variational scheme using Gaussian radial basis functions (GRBFs) is presented and used to approximate linear problems on bounded and unbounded domains. In contrast to standard mesh-free methods, which use GRBFs to discretize strong-form differential equations, this work exploits the relationship between integrals of GRBFs, their derivatives, and polynomial moments to produce exact quadrature formulae which enable weak-form expressions. Combined with trainable GRBF means and covariances, this leads to a flexible, generalized Galerkin variational framework which is applied in the infinite-domain setting where the scheme is conforming, as well as the bounded-domain setting where it is not. Error rates for the proposed GRBF scheme are derived in each case, and examples are presented demonstrating utility of this approach as a surrogate modeling technique.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper attempts to solve the problem of numerically solving partial differential equations (PDEs) on bounded and unbounded domains, especially high - dimensional problems. Specifically, the paper proposes a machine - learning variational method based on Gaussian Radial Basis Functions (GRBFs), aiming at: 1. **Numerical approximation of PDEs**: By using GRBFs as basis functions, a new variational method is proposed to numerically approximate the solutions of linear problems, applicable to both bounded and unbounded domains. 2. **Efficient integration scheme**: Utilizing the properties of GRBFs, an exact integration formula is derived, thereby achieving a weak - form expression without numerical integration. 3. **Adaptive basis functions**: By training the means and covariances of GRBFs, a flexible and generalized Galerkin variational framework is constructed. 4. **Error analysis**: The error rates of the proposed GRBF scheme are derived on bounded and unbounded domains respectively. 5. **Application examples**: Through a series of examples, the effectiveness of this method as a surrogate modeling technique is demonstrated, especially in high - dimensional problems. ### Main contributions 1. **New variational method**: A method for numerically approximating PDEs based on GRBFs is proposed, and the convergence proofs of the Poisson equation on bounded and unbounded domains are provided. 2. **Efficient numerical scheme**: An efficient integration scheme whose computational cost is basically independent of the dimension is developed. 3. **Data - driven adaptive surrogate modeling**: By training the means and covariances of GRBFs, an adaptive method capable of adapting to high - dimensional problems is constructed. 4. **Low - and high - dimensional applications**: The practical value of this method in low - and high - dimensional problems is demonstrated. ### Method overview 1. **Optimization problem**: The problem is formulated as a machine - learning optimization problem, with the objective of minimizing the residual composed of the governing equations and/or the observed data. 2. **Variational framework**: By minimizing the residual, the optimal set of GRBF basis functions is found for solving the solution space of the weak - form PDE. 3. **Integration formula**: Utilizing the properties of GRBFs, an exact integration formula is derived, thereby achieving a weak - form expression without numerical integration. 4. **Boundary condition handling**: For bounded - domain problems, a penalty function method is proposed to handle the boundary conditions. ### Technical challenges 1. **Convergence conditions**: Determine the convergence conditions and their convergence rates of the GRBF - basis - function variational method. 2. **Boundary condition handling**: Accurately handle the boundary conditions and analyze their impact on the convergence of the overall scheme. 3. **Integration error**: When the integration domain is infinite while the problem domain is finite, analyze the impact of the resulting integration error on the scheme. 4. **Parameter calibration**: Effectively construct a surrogate model by calibrating the shape parameters (means and covariances) of GRBFs to match the observed data of similar solutions. ### Conclusion This paper proposes a machine - learning variational method based on GRBFs, which is suitable for the numerical solution of high - dimensional PDEs on bounded and unbounded domains. Through theoretical analysis and experimental verification, the effectiveness and superiority of this method in high - dimensional problems are demonstrated.

Gaussian Variational Schemes on Bounded and Unbounded Domains

Variational Gaussian Processes: A Functional Analysis View

Stable and Accurate Least Squares Radial Basis Function Approximations on Bounded Domains

Uniform approximation of common Gaussian process kernels using equispaced Fourier grids

Variational Gaussian Processes For Linear Inverse Problems

Univariate interpolation by exponential functions and gaussian RBFs for generic sets of nodes

Variational Inference for Uncertainty on the Inputs of Gaussian Process Models

Images of Gaussian and other stochastic processes under closed, densely-defined, unbounded linear operators

Variational Barycentric Coordinates

Flexible and Efficient Inference with Particles for the Variational Gaussian Approximation

Variational formulation based on duality to solve partial differential equations: Use of B-splines and machine learning approximants

Multilevel approximation of Gaussian random fields: Covariance compression, estimation, and spatial prediction

Gaussian kernel expansion with basis functions uniformly bounded in $\mathcal{L}_{\infty}$

Projective Integral Updates for High-Dimensional Variational Inference

Galerkin Neural Networks: A Framework for Approximating Variational Equations with Error Control

Amortized Variational Inference for Deep Gaussian Processes

On quasi-inertial double extragradient-type iterations for variational inequalities and Bregman relatively asymptotically nonexpansive mapping

VARIATIONAL BAYES' METHOD FOR FUNCTIONS WITH APPLICATIONS TO SOME INVERSE PROBLEMS

A Variational Approach for Modeling High-dimensional Spatial Generalized Linear Mixed Models