fixed-point-iteration

This paper introduces the Iterative Refinement Neural Operator (IRNO), which augments pretrained neural operators with a learned refinement module applied via fixed-point iteration to mitigate spectral bias. IRNO progressively corrects high-frequency errors, achieving up to 56% improvement on turbulent flow and showing stable extrapolation beyond the trained iteration count.