Abstract:Scientific computing has been an indispensable tool in applied sciences and engineering, where traditional numerical methods are often employed due to their superior accuracy guarantees. However, these methods often encounter challenges when dealing with problems involving complex geometries. Machine learning-based methods, on the other hand, are mesh-free, thus providing a promising alternative. In particular, operator learning methods have been proposed to learn the mapping from the input space to the solution space, enabling rapid inference of solutions to partial differential equations (PDEs) once trained. In this work, we address the parametric elliptic interface problem. Building upon the deep operator network (DeepONet), we propose an extended interface deep operator network (XI-DeepONet). XI-DeepONet exhibits three unique features: (1) The interface geometry is incorporated into the neural network as an additional input, enabling the network to infer solutions for new interface geometries once trained; (2) The level set function associated with the interface geometry is treated as the input, on which the solution mapping is continuous and can be effectively approximated by the deep operator network; (3) The network can be trained without any input-output data pairs, thus completely avoiding the need for meshes of any kind, directly or indirectly. We conduct a comprehensive series of numerical experiments to demonstrate the accuracy and robustness of the proposed method.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to address the challenges of numerical simulation in the elliptic interface problem. Specifically: 1. **Problem Background**: - The elliptic interface problem is widely present in fields such as material science and fluid mechanics, where the background domain is composed of different materials and is divided by smooth curves or surfaces, known as interfaces. - The solution to such problems may be discontinuous at the interface, and the global regularity of the solution is low due to discontinuous coefficients and transmission conditions. 2. **Limitations of Existing Methods**: - Traditional numerical methods (such as the finite element method) are highly accurate but face challenges when dealing with complex geometries. - Generating high-quality interface-adapted meshes is extremely expensive for complex interfaces, especially for time-varying problems. 3. **Proposed Method**: - The paper proposes the Extended Interface Deep Operator Network (XI-DeepONet), a deep learning-based operator learning method. - **Three Key Features**: - Incorporates interface geometric information as additional input into the neural network, allowing the trained network to infer solutions under new interface geometries. - Uses a level set function related to the interface geometry as input, making the solution mapping continuous and effectively approximable by the deep operator network. - The network can be trained without any input-output data pairs, completely avoiding the need for any form of mesh. 4. **Experimental Validation**: - A series of numerical experiments demonstrate the accuracy and robustness of the proposed method. In summary, this paper attempts to improve the DeepONet framework by introducing a level set function to solve elliptic interface problems with varying interfaces and demonstrates its effectiveness and generalization capability in handling different interface positions.

XI-DeepONet: An operator learning method for elliptic interface problems

Solving parametric elliptic interface problems via interfaced operator network

Dirichlet-Neumann learning algorithm for solving elliptic interface problems

Deep Interface Alternation Method (DIAM) based on domain decomposition for solving elliptic interface problems

A Mesh-free Method Using Piecewise Deep Neural Network for Elliptic Interface Problems

Interfacing finite elements with deep neural operators for fast multiscale modeling of mechanics problems

R-adaptive DeepONet: Learning Solution Operators for PDEs with Discontinuous Solutions Using an R-adaptive Strategy

Decoupling Numerical Method Based on Deep Neural Network for Nonlinear Degenerate Interface Problems

An Efficient Method for Solving Two-Dimensional Partial Differential Equations with the Deep Operator Network

DeepONet for Solving PDEs: Generalization Analysis in Sobolev Training

Learning the solution operator of a nonlinear parabolic equation using physics informed deep operator network

Tailored Finite Point Operator Networks for Interface problems

Exponential Convergence of Deep Operator Networks for Elliptic Partial Differential Equations

DOSnet as a Non-Black-Box PDE Solver: When Deep Learning Meets Operator Splitting

A Deep Neural Network/Meshfree Method for Solving Dynamic Two-phase Interface Problems

Operator Learning Enhanced Physics-informed Neural Networks for Solving Partial Differential Equations Characterized by Sharp Solutions

Forward and inverse problems for Eikonal equation based on DeepONet

Learning the solution operator of parametric partial differential equations with physics-informed DeepONets

Solving forward and inverse PDE problems on unknown manifolds via physics-informed neural operators

Physics-Informed Tailored Finite Point Operator Network for Parametric Interface Problems