Conditional Precision Bottlenecks and Quantized Defect Correction in Fractional Neural Operators

Abstract

Preprint accompanying: "Conditional Precision Bottlenecks and Quantized Defect Correction in Fractional Neural Operators" (2026). This work investigates precision bottlenecks, numerical stability diagnostics, and model compression techniques for Fourier neural operators applied to fractional partial differential equations (PDEs). We introduce Quantized Defect Correction (QDC), an iterative error-correction framework designed to recover high-fidelity PDE solutions from low-precision quantized operator surrogates while preserving computational and memory efficiency across scientific machine learning benchmarks. Official research code and reproducibility artifacts:https://github.com/chaewon-research/qdc-fractional-neural-operators