Abstract:Elliptic partial differential equations (PDEs) are a major class of time-independent PDEs that play a key role in many scientific and engineering domains such as fluid dynamics, plasma physics, and solid mechanics. Recently, neural operators have emerged as a promising technique to solve elliptic PDEs more efficiently by directly mapping the input to solutions. However, existing networks typically cannot handle complex geometries and inhomogeneous boundary values present in the real world. Here we introduce Boundary-Embedded Neural Operators (BENO), a novel neural operator architecture that embeds the complex geometries and inhomogeneous boundary values into the solving of elliptic PDEs. Inspired by classical Green's function, BENO consists of two branches of Graph Neural Networks (GNNs) for interior source term and boundary values, respectively. Furthermore, a Transformer encoder maps the global boundary geometry into a latent vector which influences each message passing layer of the GNNs. We test our model extensively in elliptic PDEs with various boundary conditions. We show that all existing baseline methods fail to learn the solution operator. In contrast, our model, endowed with boundary-embedded architecture, outperforms state-of-the-art neural operators and strong baselines by an average of 60.96\%. Our source code can be found

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the difficulties encountered by existing neural operators when dealing with elliptic partial differential equations (Elliptic PDEs) with complex geometries and inhomogeneous boundary conditions. Specifically, existing neural network methods are usually unable to effectively handle complex geometric structures and non - uniform boundary values in the real world, resulting in poor performance in solving elliptic PDEs. To solve this problem, the authors propose Boundary - Embedded Neural Operators (BENO), a new type of neural operator architecture. BENO significantly improves the ability to solve elliptic PDEs by embedding complex geometries and inhomogeneous boundary conditions into the solution process. The following are the specific improvements of BENO: 1. **Two - branch Graph Neural Networks (GNNs)**: BENO contains two GNN branches, which are respectively used to model the influence of internal source terms and boundary values. This helps to better capture the interaction between the boundary and the internal region. 2. **Transformer Encoder**: In order to model the influence of the boundary on the global solution, BENO uses a Transformer encoder to encode the entire boundary information into a latent vector and input it into each message - passing layer. This enables the model to capture the complex influence of the boundary geometry and values on the interaction between internal nodes. 3. **Boundary - embedded Message - passing Mechanism**: BENO enhances the adaptability to complex boundary conditions by embedding boundary information into the message - passing process. This design enables the model to perform well when dealing with boundary conditions of different shapes and values. Experimental results show that BENO significantly outperforms existing baseline methods when dealing with elliptic PDEs with complex geometries and inhomogeneous boundary conditions, with an average improvement of 60.96%. ### Summary This paper aims to address the shortcomings of existing neural operators in dealing with elliptic PDEs with complex geometries and inhomogeneous boundary conditions, and proposes the innovative architecture of BENO. By introducing two - branch GNNs, Transformer encoders, and boundary - embedded message - passing mechanisms, BENO can more effectively handle these challenges and demonstrates excellent performance and generalization ability in experiments.

BENO: Boundary-embedded Neural Operators for Elliptic PDEs

Learning Only On Boundaries: a Physics-Informed Neural operator for Solving Parametric Partial Differential Equations in Complex Geometries

Numerical solutions of boundary problems in partial differential equations: A deep learning framework with Green's function

Neural Green's Operators for Parametric Partial Differential Equations

SPFNO: Spectral operator learning for PDEs with Dirichlet and Neumann boundary conditions

Data Generation-based Operator Learning for Solving Partial Differential Equations on Unbounded Domains

Solving Partial Differential Equations in Different Domains by Operator Learning method Based on Boundary Integral Equations

BINet: Learning to Solve Partial Differential Equations with Boundary Integral Networks

Convolutional Neural Operators for robust and accurate learning of PDEs

Boundary integrated neural networks (BINNs) for 2D elastostatic and piezoelectric problems: Theory and MATLAB code

Physics-Informed Geometry-Aware Neural Operator

DeepGreen: deep learning of Green's functions for nonlinear boundary value problems

BelNet: basis enhanced learning, a mesh-free neural operator

Neural Operator: Graph Kernel Network for Partial Differential Equations

Learning Partial Differential Equations with Deep Parallel Neural Operator

Neural Operator Approximations for Boundary Stabilization of Cascaded Parabolic PDEs

Positional Embeddings for Solving PDEs with Evolutional Deep Neural Networks

Solving parametric elliptic interface problems via interfaced operator network

Diffeomorphic Latent Neural Operators for Data-Efficient Learning of Solutions to Partial Differential Equations