Abstract:In this paper, a multi-scale Fourier neural operator (MscaleFNO) is proposed to reduce the spectral bias of the FNO in learning the mapping between highly oscillatory functions, with application to the nonlinear mapping between the coefficient of the Helmholtz equation and its solution. The MscaleFNO consists of a series of parallel normal FNOs with scaled input of the function and the spatial variable, and their outputs are shown to be able to capture various high-frequency components of the mapping's image. Numerical methods demonstrate the substantial improvement of the MscaleFNO for the problem of wave scattering in the high-frequency regime over the normal FNO with a similar number of network parameters.

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is **to reduce the spectral bias of the Fourier neural operator (FNO) when learning highly oscillatory function mappings**, especially for wave scattering problems in the high - frequency regime. Specifically, the paper proposes a multi - scale Fourier neural operator (MscaleFNO) to improve the ability of the traditional FNO in capturing high - frequency components. The following is a summary of the key points that the paper attempts to solve: 1. **Spectral bias problem** Deep neural networks (DNN) tend to preferentially learn low - frequency components during the training process, while having a weaker ability to learn high - frequency components. This phenomenon is called spectral bias. For physical systems involving high - frequency content (such as wave scattering problems), this bias will lead to a decline in model performance. 2. **Learning challenges of high - frequency mappings** In many practical applications, the input and output functions may contain a large amount of high - frequency information. For example, in the non - linear mapping of the Helmholtz equation, the relationship between the coefficients and the solutions may contain complex high - frequency components. The traditional FNO performs poorly when dealing with such problems. 3. **Application of multi - scale methods** To address the above challenges, the paper introduces multi - scale methods into FNO and proposes MscaleFNO. By processing the input functions and spatial variables in parallel at different scales, MscaleFNO can more effectively capture various high - frequency components in the mapping image. 4. **Application in wave scattering problems** The paper pays special attention to wave scattering problems in the high - frequency regime and verifies the superior performance of MscaleFNO in this field. Experimental results show that, with a similar number of parameters, MscaleFNO is significantly better than the traditional FNO. ### Formula summary - Frequency - domain decomposition formula: $$ f(x)=\sum_{i = 1}^M f_i(x),\quad f_i(x)=\int_{A_i}\hat{f}(k)e^{ik\cdot x}dk $$ - Scaled spectral support range: $$ \text{supp }\hat{f}^{\text{(scale)}}_i(k)\subset\left\{k\in\mathbb{R}^d,\frac{K_{i - 1}}{\alpha_i}\leq|k|\leq\frac{K_i}{\alpha_i}\right\} $$ - Mathematical expression of MscaleFNO: $$ u(x)=\sum_{i = 1}^N\gamma_i\text{FNO}_{\theta_m}[c_ix,c_ia(x)](x) $$ Through these methods and formulas, the paper successfully solves the problem of the poor performance of the traditional FNO in the high - frequency regime, providing a new tool for the modeling of complex physical systems.

MscaleFNO: Multi-scale Fourier Neural Operator Learning for Oscillatory Function Spaces

Incremental Spatial and Spectral Learning of Neural Operators for Solving Large-Scale PDEs

Multi-Resolution Active Learning of Fourier Neural Operators

MgFNO: Multi-grid Architecture Fourier Neural Operator for Parametric Partial Differential Equations

Toward a Better Understanding of Fourier Neural Operators from a Spectral Perspective

Discretization Error of Fourier Neural Operators

Spectral Neural Operators

Learning the boundary-to-domain mapping using Lifting Product Fourier Neural Operators for partial differential equations

Fourier neural operator based fluid–structure interaction for predicting the vesicle dynamics

Mitigating spectral bias for the multiscale operator learning

Invertible Fourier Neural Operators for Tackling Both Forward and Inverse Problems

Fourier neural operator based fluid-structure interaction for predicting the vesicle dynamics

DiffFNO: Diffusion Fourier Neural Operator

Spherical Fourier Neural Operators: Learning Stable Dynamics on the Sphere

Spectral-Refiner: Fine-Tuning of Accurate Spatiotemporal Neural Operator for Turbulent Flows

Dilated convolution neural operator for multiscale partial differential equations

Model-Parallel Fourier Neural Operators as Learned Surrogates for Large-Scale Parametric PDEs

LNO: Laplace Neural Operator for Solving Differential Equations

NSNO: Neumann Series Neural Operator for Solving Helmholtz Equations in Inhomogeneous Medium

A Born Fourier Neural Operator for Solving Poisson’s Equation with Limited Data and Arbitrary Domain Deformation