Spaces:

---

proxima-fusion

/

constellaration-bench

like 22

Running

Hi,

I noticed that the evaluation on the leaderboard server rejects solutions with negative
edge_rotational_transform_over_n_field_periods, even when the absolute value satisfies the constraint (e.g.,
|iota/NFP| = 0.26 >= 0.25).

This appears to be because _normalized_constraint_violations in problems.py computes:

self._edge_rotational_transform_over_n_field_periods_lower_bound

• metrics.edge_rotational_transform_over_n_field_periods

without taking the absolute value. For stellarators optimized with codes like DESC (which uses a left-handed
coordinate convention), the rotational transform is naturally negative. The physical quantity is the same —
only the sign convention differs.

Could np.abs() be added to this constraint, i.e.:

self._edge_rotational_transform_over_n_field_periods_lower_bound

• np.abs(metrics.edge_rotational_transform_over_n_field_periods)

This would make the benchmark convention-agnostic and accept valid solutions regardless of the coordinate
system used during optimization.

Thank you!

Like：https://github.com/proximafusion/constellaration/pull/93