Neural Operators - Learning PDE Solutions with Sobolev Loss

Train FNO models , run a grid search across PDEs, and produce plots per PDE. Automatically uses CUDA if available. Settings come from runner.sh Includes resolution generalization: train on --res and auto-evaluate on a higher resolution (--res_hi, defaults to 2×).

Quick start

python -m venv .venv && source .venv/bin/activate   # (Windows: .venv\Scripts\activate)
pip install --upgrade pip wheel
pip install torch numpy scipy matplotlib pyyaml
pip install neuraloperator || echo "neuraloperator optional"
bash runner.sh
python plot_all_pdes.py --runs_root runs_best --out figs/val_loss_all_pdes_log.png --log