Scrape-Off Layer Surrogate Training, Inference & Coupling Ecosystem for neural-network surrogates of SOLPS-ITER edge plasmas.
SOLSTICE provides trained models and tools for predicting 2D plasma states from operating-point parameters and coupling those predictions to reduced edge-physics workflows such as COTSIM.
pip install "solstice-fusion[models] @ git+https://github.com/abdoudiaw/solstice.git"
from solstice import hub
model = hub.load("pepc-diiid-state-v2")
params = {
"ptot": 6e6,
"chi": 0.7,
"core_fueling": 3e20,
"puff_D2": 1e21,
"dna": 0.5,
}
result = model.predict_batch(params)
fields = {name: values[0] for name, values in result["mean"].items()}
See examples/README.md for runnable plasma-state, source, and REACT COTSIM workflows.
| Model | Description |
|---|---|
| pepc-diiid-state-v2 | DIII-D plasma-state model |
| pepc-diiid-cotsim-677-state | DIII-D plasma-state model with core-boundary flux inputs for the REACT COTSIM coupling |
| pepc-jet-state-v1 | Plasma-state model for a JET-like geometry (Dasbach et al. dataset, major radius scaled 1-10 m) |
Each bundle contains its model definition, normalization, mesh, provenance,
and model card.
Earlier models (e.g. pepc-diiid-state-v1) remain loadable by name; the
release notes say what changed between versions.
Training data is not distributed with this repository. SOLPS converters and training tools are under src/solstice/. SOLPS file parsing follows SOLPS-routines.
This work was supported in part by the U.S. Department of Energy (DOE) Fusion Innovation Research Engine (FIRE) Collaborative Advanced Profile Prediction for Fusion Pilot Plant Design (APP-FPP) via an MIT subcontract under Award No. DE-SC0025853, and by the DOE Office of Science, Office of Fusion Energy Sciences, through the project Enabling Tokamak Pulse Simulation by Machine Learning of Core-Pedestal-Boundary Physics.
Code: Apache-2.0. Released model weights: CC BY 4.0; see the model cards.