Neural Operator-enabled Topology-informed Evolutionary Strategy for PDE-Constrained Optimization

2026-07-09 · Source: cs.LG updates on arXiv.org · Field: Science & Research — Mathematics & Computational Sciences, Engineering & Applied Sciences, Research Methodology & Innovation · Depth: Expert, extended

Summary

Neural Operator–enabled Topology-informed Evolutionary Strategy (NOTES) is introduced as a framework for efficient and transferable inverse design of physical systems governed by partial differential equations. It integrates dimensionality reduction, representation learning, and evolutionary optimization. NOTES couples a DeepONet-based neural operator with the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) to optimize in a compact latent space. For nanophotonic beam-deflector inverse design, NOTES reduced dimensionality from 256 to 25, consistently achieving over 95% efficiency. In structural optimization, it discovered designs with compliance down to 246. The method outperforms CMA-ES, topology optimization, and other baselines by learning topology-aware priors from physics-informed training data.

Key takeaway

For Machine Learning Engineers tackling high-dimensional PDE-constrained inverse design, consider implementing NOTES to significantly improve optimization efficiency and design quality. Its combination of DeepONet for nonlinear representation and CMA-ES in a physics-informed latent space consistently yields high-performance, binarized designs, even for unseen operating conditions. This approach offers superior robustness and transferability compared to traditional methods or gradient-based optimizers in complex design spaces.

Key insights

NOTES integrates neural operators and evolutionary strategies for efficient, transferable, and robust PDE-constrained inverse design.

Principles

• Physics-informed data guides latent space search.
• Nonlinear decoders improve complex manifold reconstruction.
• CMA-ES is robust in low-dimensional latent spaces.

Method

NOTES operates in two stages: training a DeepONet-based neural operator on high-performance designs using PCA for latent space construction, then optimizing in this latent space with CMA-ES, evaluating designs via a PDE solver.

In practice

• Design nanophotonic metagratings for high efficiency.
• Optimize structural compliance in MBB beams.
• Generate designs for out-of-distribution wavelengths.

Topics

• Neural Operators
• Evolutionary Strategies
• Inverse Design
• PDE-Constrained Optimization
• Nanophotonics
• Structural Optimization

Code references

• xmh3698/notes-pde-opt

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

Related on AIssential

• Neural Operator-enabled Topology-informed Evolutionary Strategy for PDE-Constrained Optimization — NOTES efficiently optimizes PDE-constrained inverse design by combining neural operators with evolutionary strategies in a compact latent space.
• Neural Operator-enabled Topology-informed Evolutionary Strategy for PDE-Constrained Optimization — NOTES efficiently optimizes PDE-constrained inverse design by combining neural operators with evolutionary strategies in a compact latent space.
• Kernel-based Operator Learning: Error Analysis, Budget Allocation, and a Physics-Informed Extension — Kernel-based operator learning's two-stage framework is enhanced by a budget allocation theory and a physics-informed, no-retraining extension.
• PGD-NO: A Neural Operator with Precomputed Geometry Decomposition for 3D Million-scale Physics Simulations — The PGD-NO neural operator decouples geometry encoding via pre-computation, enabling linear memory scalability for million-scale 3D physics simulations.
• Evolutionary Algorithm-Guided LLMs for Physics-Informed Neural Network Design — Evolutionary algorithms can guide LLMs to iteratively optimize Physics-informed neural network designs.
• 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.

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 cs.LG updates on arXiv.org.