Paper 14110-29
Paper 14110-29
Automatic differentiable light scattering with PyMieDiff and TorchGDM (Invited Paper)
16 April 2026 • 15:40 - 16:10 CEST | Madrid 1/Salon 3 (Niveau/Level 0)
Abstract
We introduce two fully differentiable light‑scattering toolkits that leverage open‑source automatic‑differentiation frameworks. PyMieDiff provides end‑to‑end differentiable Mie‑theory calculations, while TorchGDM implements the Green’s Dyadic Method in PyTorch for full‑field 2D/3D multi‑scattering simulations, supporting hybrid volume‑discretization and effective‑model approaches. Both packages enable rapid inverse‑design, optimization, and physics‑informed machine‑learning workflows in nano‑photonics, offering a scalable, open‑source alternative to traditional simulation pipelines.
Presenter
Peter R. Wiecha
Lab. d'Analyse et d'Architecture des Systèmes du CNRS (France)
Peter Wiecha is a permanent researcher (“chargé de recherche”) at LAAS-CNRS in Toulouse, France. His main research interests concern the interaction of light with subwavelength small structures, applications of artificial intelligence in nano-optics and photonics, and the use of deep learning tools for numerical light scattering descriptions.
Peter studied physics at the Technical University of Munich in Germany and wrote a Diploma thesis at the Walter Schottky Institute in Garching. After his PhD from the Université Paul Sabatier Toulouse in 2016, he was postdoc on a DFG research fellowship with Prof. Otto Muskens at the University of Southampton. He joined LAAS-CNRS in 2020.