Neural Operators for Fast Simulation of Strong Gravitational Lensing

Strong gravitational lensing is a key observational probe for dark matter and cosmology, but traditional ray-tracing simulations are computationally expensive, requiring seconds per image. This project builds a Fourier Neural Operator (FNO) surrogate that learns the continuous mapping from convergence field kappa to lensed image I_lensed, achieving >1000× speedup over traditional solvers. The FNO is resolution-invariant, enabling zero-shot super-resolution and generalization across different discretizations. Deliverables include: (1) a trained FNO model for strong lensing simulation (2) comprehensive benchmarking against U-Net and DeepONet baselines (3) a fully differentiable pipeline for dataset generation (4) integration into the DeepLense ecosystem for LSST-scale analysis.