phanerozoic/coilstellaration-direct

Direct coil predictors for Coilstellaration

Networks for the Coilstellaration benchmark, which asks for a stellarator coilset given a plasma boundary and four engineering targets. Each file maps the boundary's Fourier coefficients and the targets to the Fourier coefficients and currents of the coils in one forward pass, and none holds a loop.

File	Tracks	What it is
model.onnx	fixed_shape	a residual multilayer perceptron
corrected.onnx	fixed_shape	that network followed by a corrector network, which reads the field error and clearances of the first network's coilset and returns a change to it
variable.onnx	fixed_shape, variable_shape	one residual multilayer perceptron for the dataset's 17 shapes
corrected_variable.onnx	fixed_shape, variable_shape	a network of that kind followed by a corrector network, which also reads the gradient of the requirement penalties

The networks were trained on the training split of proxima-fusion/coilstellaration, on the benchmark's four requirements computed for their own outputs by a differentiable model of the metrics, with boundaries blended from pairs of training boundaries beside the split's own. A twentieth of the training boundaries were held out for model selection, and the benchmark's eval split was not read. The Python files are the training and export code.

Interface

The graphs follow the benchmark's ONNX contract. Their inputs are boundary_r_cos and boundary_z_sin (float32, [5, 9]), requirement_metrics (float32, [4]), and n_field_periods and n_coils_per_half_period (int64 scalars). Their outputs are coil_x_n, coil_y_n and coil_z_n (float32, [n_coils, 15], Fourier order 7 in DESC's [sin(7) .. sin(1), const, cos(1) .. cos(7)] layout) and currents (float32, [n_coils], amperes). model.onnx and corrected.onnx are for three field periods and five coils; the other two take any of the 17 shapes.

import numpy as np
import onnxruntime as ort
from huggingface_hub import hf_hub_download

session = ort.InferenceSession(hf_hub_download("phanerozoic/coilstellaration-direct", "variable.onnx"))
r_cos = np.zeros((5, 9), np.float32)
z_sin = np.zeros((5, 9), np.float32)
r_cos[0, 4], r_cos[1, 4], z_sin[1, 4] = 1.0, 0.15, 0.15
coil_x_n, coil_y_n, coil_z_n, currents = session.run(
    ["coil_x_n", "coil_y_n", "coil_z_n", "currents"],
    {
        "boundary_r_cos": r_cos,
        "boundary_z_sin": z_sin,
        "requirement_metrics": np.array([0.4, 0.8, 3.0, 0.05], np.float32),
        "n_field_periods": np.asarray(4, np.int64),
        "n_coils_per_half_period": np.asarray(6, np.int64),
    },
)