Magnetism

Whether the calculation is spin-polarised, how the spins are arranged, and what to start the moments at.

Not a default model

is_magnetic is published, but external validation on an independent dataset has not been run yet. See the model card's "How good it is" section.

Quantity	What it is	Model
is_magnetic	Whether the ground state is spin-polarised	mMACE+MLP
ordering	Non-magnetic, ferro-, antiferro-, ferrimagnetic	none
magnetic_moments	Starting moment per site	none

These are three questions, not one, and two of them do not fit the current prediction shape: an ordering is a label and per-site moments are one value per site. An antiferromagnetic ordering can also need a magnetic supercell, which changes the structure every later recommendation is made for — including the mesh.

That prediction-shape gap is closed now, not still open: ContractSpec.labels (a legal label set, for ordering) and ContractSpec.index_convention (a per-site or per-species position convention, for magnetic_moments) both exist in goldilocks_ml.inference. What is missing for either quantity is a model, not a way to declare one — antiferromagnetic sublattice detection in particular has no implementation anywhere upstream of this package to build on yet (see src/goldilocks_ml/models/magnetism/magnetic_moments/'s module docstring), only an FM/ferrimagnetic-only initial-moment guess and a ported moment relaxation.