Euclidean Fourier Neural Operators Enhance Domain Transferability

Nathanael Bosch, Niklas Frederik Schmitz, Michael F. Herbst· August 31, 2026 View original

Key takeaways

  • Traditional FNOs struggle with domain transfer due to discrete spectral weights.
  • EFNOs introduce a continuous spectral kernel for domain independence.
  • EFNOs generalize effectively to unseen grid sizes and varying domains.
  • This advancement is crucial for scientific machine learning applications.

Who benefits

Materials SciencePhysicsEngineering SimulationScientific ComputingDrug Discovery

Summary

This paper introduces Euclidean Fourier Neural Operators (EFNOs) as a domain-independent alternative to traditional FNOs, addressing their limitation in transferring across different periodic domains. EFNOs achieve this by parameterizing the spectral kernel as a continuous function of the physical wavevector, enabling consistent operator learning across varying domain shapes and sizes.

Fourier Neural Operators (FNOs) are recognized for their efficiency in learning mappings between function spaces, largely due to their inherent independence from the grid resolution used for training and evaluation. However, a significant limitation of FNOs is their dependence on the specific periodic domain they are applied to. Their discrete spectral weights are tied to integer Fourier mode numbers, meaning that when applied to a different domain, the same trained weights represent a different underlying operator, hindering transferability. To overcome this challenge, researchers propose Euclidean Fourier Neural Operators (EFNOs). The core innovation of EFNOs lies in parameterizing the spectral kernel as a continuous function of the physical wavevector. This design allows EFNOs to learn operators that act consistently across periodic domains of diverse shapes and sizes, making them truly domain-independent. The effectiveness of EFNOs was demonstrated through experiments on a simple heat equation and a complex materials science task involving learning exchange-correlation potentials across various crystal structures. Results indicate that EFNOs successfully generalize to unseen grid sizes and different domains, offering a robust solution for tasks requiring transfer across varying physical contexts.

Why it matters

EFNOs significantly improve the generalizability of neural operators across different physical domains, which is critical for scientific machine learning applications where data often comes from varied experimental setups.

How to implement this in your domain

  1. 1Investigate EFNOs for scientific machine learning tasks requiring generalization across different physical scales or geometries.
  2. 2Evaluate EFNOs as a potential replacement for traditional FNOs in simulations or modeling where domain transfer is crucial.
  3. 3Collaborate with research teams to adapt EFNO architectures for specific materials science or fluid dynamics problems.
  4. 4Explore the open-source code (if available) to understand implementation details and integrate into existing frameworks.

Original post by Nathanael Bosch, Niklas Frederik Schmitz, Michael F. Herbst

"arXiv:2608.28425v1 Announce Type: new Abstract: Fourier neural operators (FNOs) provide an efficient framework for learning mappings between function spaces as they are, by construction, independent of the grid resolution at which they are trained and evaluated. However, FNOs are…"

View on X

Originally posted by Nathanael Bosch, Niklas Frederik Schmitz, Michael F. Herbst on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

New Optimizer Accelerates LLM Pretraining with Curvature-Conditioned Momentum

This research proposes a curvature-conditioned multiscale momentum method with sphere constraints to accelerate large language model pretraining. It addresses challenges from noise-dominant gradients and ill-conditioned loss landscapes by enhancing progress along flat directions, significantly improving upon existing adaptive optimizers like AdamW and Muon.

Shuchen Zhu, Yuxin Fang, Mingze Wang, Kun YuanAug 31, 2026
AI Engineering & DevToolsAI Research

SymboLLM-FE Boosts Feature Engineering with LLMs and Symbolic Regression

This paper introduces SymboLLM-FE, a novel approach combining symbolic regression and large language models for automated feature engineering on tabular data. It aims to generate highly interpretable and performant features while overcoming the limitations of traditional AutoFE and LLM-based methods.

Zi-Jian Cheng, Zi-Yi Jia, Zhi Zhou, Yu-Feng Li, Lan-Zhe GuoAug 31, 2026
AI ResearchAI Engineering & DevTools

OpenEuroLLM Scaling Laws: Optimizing Learning Rates and Batch Sizes

This research investigates the scaling behavior of learning rates and batch sizes in pretraining large language models, developing a model to capture their relationships and evaluating learning rate schedules. It also characterizes the dependence of loss on model capacity and dataset size, establishing a baseline for future OpenEuroLLM development.

Niccol\`o Ajroldi, Diana Alexandra Onutu, Haider Al-Tahan, J\"org Franke, Sampo Pyysalo, Jenia Jitsev, Aaron KleinAug 31, 2026