999thBaam/the-well-fno-demo

FNO Gray-Scott Reaction Diffusion

A Fourier Neural Operator (FNO) trained to predict Gray-Scott reaction-diffusion dynamics — a classic pattern-forming system in physics.

What This Does

The model takes 4 consecutive frames of a Gray-Scott simulation (128x128, 2 chemical fields) and predicts the next frame. It can also roll out multiple steps autoregressively to generate full trajectory predictions.

Model Details

• —Architecture: FNO (4 spectral layers, 128 hidden channels, 16x16 Fourier modes)
• —Parameters: 19M
• —Training data: The Well Gray-Scott Reaction Diffusion dataset
• —Input: 4 timesteps x 2 fields = 8 channels (128x128)
• —Output: 1 timestep x 2 fields = 2 channels (128x128)
• —Best validation MSE: 0.000634

Try It

1. 1.Select a sample trajectory
2. 2.Choose a starting timestep
3. 3.Click Predict for single-step or Rollout for multi-step animation