fourier neural operator (fno)

Fourier Neural Operator (FNO) is a specific highly effective neural operator architecture — that learns resolution-invariant mappings by performing convolutions in the Fourier domain (frequency space) rather than spatial domain.

What Is FNO?

- Mechanism:
1. Fourier Transform (FFT) input to frequency domain.
2. Filter out high frequencies (keep global modes).
3. Linear transform (mixing).
4. Inverse Fourier Transform (iFFT) back to spatial.
- Efficiency: Global convolution in spatial domain is $O(N^2)$; multiplication in Fourier is $O(N \log N)$.

Why FNO Matters

• SOTA: Achieved state-of-the-art in modeling turbulent flows (Navier-Stokes) and weather forecasting (FourCastNet).
• Global Receptive Field: Spectral methods naturally capture global correlations, critical for fluid dynamics.
• Speed: 1000s of times faster than traditional numerical solvers.

Fourier Neural Operator is the speed of light for simulation — solving complex fluid dynamics problems almost instantly by operating in the frequency domain.