Deep Learning-based Surrogate Modelling of the LOD Method for Multiscale Problems

2026-07-14 · Source: Artificial Intelligence · Field: Science & Research — Mathematics & Computational Sciences, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

Multiscale problems, particularly elliptic PDEs with rough and high-contrast inputs, pose significant challenges for traditional numerical methods due to the need for fine discretizations. While neural operator models offer a data-driven alternative, they often struggle with accuracy in strongly heterogeneous or oscillatory conditions. This work introduces LOD-MSNO (LOD-Multiscale Neural Operator), a novel hybrid approach designed to address these limitations. LOD-MSNO integrates the Localized Orthogonal Decomposition (LOD) method, known for its accuracy in multiscale problems but high computational cost, as a strong multiscale prior. It leverages LOD's representation of solutions via problem-adapted basis functions and employs data-driven operator learning to mitigate LOD's computational bottlenecks. The proposed method includes theoretical error estimates and demonstrates potential to surpass existing neural operator baselines in accuracy for complex multiscale inputs, largely preserving neural operator models' computational efficiency.

Key takeaway

For Research Scientists and Machine Learning Engineers developing solutions for multiscale problems with rough or high-contrast inputs, you should investigate LOD-MSNO. This hybrid approach offers a path to achieve higher accuracy than current neural operator baselines by integrating the Localized Orthogonal Decomposition method as a multiscale prior, while largely maintaining computational efficiency. Consider exploring its theoretical error estimates and applying it to complex elliptic PDEs to improve model performance and reduce simulation times.

Key insights

LOD-MSNO combines the accuracy of Localized Orthogonal Decomposition with the efficiency of neural operators for multiscale problems.

Principles

• Multiscale priors enhance neural operator accuracy for heterogeneous inputs.
• Hybrid numerical-data-driven methods can mitigate computational bottlenecks.
• Problem-adapted basis functions improve solution representation.

Method

LOD-MSNO leverages LOD's solution representation as a multiscale prior, learning coefficients through data-driven operator learning to reduce computational cost.

In practice

• Apply LOD-MSNO to elliptic PDEs with rough coefficients.
• Use hybrid models for computationally intensive numerical methods.

Topics

• Deep Learning
• Neural Operators
• Localized Orthogonal Decomposition
• Multiscale Problems
• Elliptic PDEs
• Surrogate Modelling

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• LLT: Local Linear Transformer for PDE Operator Learning — LLT combines linear global attention with explicit local spatial mixing to efficiently learn PDE operators across diverse mesh types.
• Subgrid-Scale Parameterization in Burgers' Equation Using Structure-Preserving Neural Networks and Entropy Variables — Structure-preserving neural networks and entropy variables enable robust subgrid-scale parameterization for PDEs.
• Neural Operator-Based Surrogate Model for CFD:Helical Coil Steam Generator in Small Modular Reactor — Neural operator-based surrogates, combined with ROMs and multi-scale techniques, enable real-time CFD for SMR digital twins.
• Reformulating Neural Operators in $d+1$ Dimensions for Embedding Evolution — Neural operators redefined in d+1 dimensions using Schrödingerisation better capture system evolution and achieve superior performance.
• PGD-NO: A Neural Operator with Precomputed Geometry Decomposition for 3D Million-scale Physics Simulations — PGD-NO decouples geometry encoding from solution querying to enable scalable neural PDE simulations on massive meshes.
• Neural Operator-enabled Topology-informed Evolutionary Strategy for PDE-Constrained Optimization — NOTES integrates neural operators and evolutionary strategies for efficient, transferable, and robust PDE-constrained inverse design.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.