Fast Prediction of Photonic Scattering via a Modified Fourier Neural Operator

Publication Date

Spring 4-22-2026

Presentation Length

15 minutes

College

College of Sciences & Mathematics

Department

Chemistry and Physics, Department of

Student Level

Undergraduate

Faculty Mentor

Scott Hawley

Presentation Type

Talk/Oral

Summary

The Fourier Neural Operator (FNO) has shown remarkable success in approximating solutions to partial differential equations in fluid dynamics, yet its potential for nanophotonic systems remains largely unexplored. We present a physics-informed FNO architecture for solving Maxwell’s equations in scattering problems involving nonlinear materials with spatial variations on the order of the source’s central wavelength. Trained on in-house finite-difference time-domain (FDTD) simulations, we develop a model that learns the family of solutions mapping electromagnetic field configurations forward in time across multiple equivalent FDTD time steps. Critically, we evaluate how well the trained models generalize beyond their training domains by measuring prediction errors on unseen permittivity distributions and source parameters. We evaluate accuracy as a function of training data size and time-step size, benchmarking against high-resolution FDTD results, and observe up to a 100× reduction in computational time while obtaining a normalized mean square error (NMSE) as low as 0.24, outperforming interpolation in all cases. These results suggest FNOs could dramatically accelerate the inverse design and optimization workflows for nonlinear photonic interfaces, where conventional solvers remain prohibitively expensive for exploring large parameter spaces.

This document is currently not available here.

Share

COinS