Symmetry backend: moyo audit

Verdict: adopt moyo 0.22 with gaps. It covers space-group detection, operations, orbits, Wyckoff letters, standardized and primitive cells, subgroups, the Euclidean normalizer, layer groups and magnetic groups. Three gaps remain: the Niggli/Delaunay/Minkowski reductions are crate-private, the symmetrization module is private (reachable only through MoyoDataset), and Brillouin zones, k-paths and molecular point groups are absent.

xtalforge-symmetry exposes its own types (SpaceGroup, SymOp, SymmetryDataset, WyckoffSite); moyo stays an implementation detail and can be replaced without changing the public API.

Requirements from Crystal-Builder

Source: Crystal-Builder/xtal/core/{spacegroup,symmetry,subgroups}.py, core/molecule.py and analysis/kpath.py (Crystal-Builder commit 78d5476fd9e851fda00cde57b7ab8da3195dcf54).

Crystal-Builder need	Function or class	moyo 0.22 coverage	Status
Space group lookup by Hall symbol	SpaceGroup.from_hall	data::hall_symbol_entry, identify	public
Lookup by Hermann-Mauguin and setting	SpaceGroup.from_name, from_number	data::*, Setting	public
Operations and triplets	SpaceGroup.operations, SymOp.apply	MoyoDataset::operations	public
Inverses and stacked operations	SpaceGroup.inverse_of, stacked	operations field + own arithmetic	port
Detection with tolerance	symmetry.detect	MoyoDataset::new	public
Wyckoff letters and site symmetry	assign_wyckoff	wyckoffs, site_symmetry_symbols	public
P1 expansion and standardization	reduce_to_p1, standardize	std_cell, prim_std_cell, mappings	public
Set space group / reinterpret / impose	set_space_group, _reinterpret, _impose	MoyoDataset + own fit	port
Asymmetrize and snap to special positions	asymmetrize, snap_to_special_positions	MoyoDataset + std_cell	port
Enantiomorph and hand description	enantiomorph, hand_description	identify, operations	port
Duplicate and merge sites	duplicate_groups, merge_duplicates	own soak/merge logic	port
Translationengleiche subgroups	subgroups.subgroups_of	subgroup::enumerate_translationengleiche_subgroups	public
Klassengleiche subgroups	subgroups.maximal_subgroups	subgroup::enumerate_klassengleiche_subgroups	public
Euclidean normalizer	normalizer paths	MoyoDataset::euclidean_normalizer, normalizer_wyckoff_positions	public
Niggli, Delaunay, Minkowski reduction	supercell.niggli_reduce, delaunay_reduce	math::{niggli,delaunay,minkowski}	crate-private
Symmetrization internals	symmetry.standardize	symmetrize module	private
Brillouin zone and k-path	analysis/kpath.py	absent	implement
Molecular point group	core/molecule.py	identify::PointGroup takes rotations	implement adapter

Behavioural comparison against spglib

Input: tests/golden/moyo-audit.json, generated from eight corpus structures (Si, NaCl, graphite, quartz, MOF-5, ZIF-8, and the triclinic and hexagonal edge cases) for symprec in {1e-5, 1e-3, 1e-2, 0.1}. The Rust test crates/xtalforge-symmetry/tests/moyo_audit.rs runs moyo on the same cells and compares:

• space-group number: agrees in all 32 cases;
• Hall number with Setting::Spglib: agrees in all 32 cases; with Setting::default, Si selects Hall 526 instead of spglib's 525 (origin choice 2 instead of 1);
• number of operations: agrees in all 32 cases;
• Wyckoff letters: raw moyo agrees except Si, where moyo reports 8b and spglib reports 8a for the same Hall-525 structure (origin shift). Since the 2026-11-11 audit the wrapper matches the letters on the input coordinates when the input basis is conventional (xtalforge-symmetry/src/wyckoff_origin.rs), so xtalforge reports 8a like spglib; the e2e battery checks the letter counts of all eight samples;
• standardized lattice parameters: agrees for Si, NaCl, MOF-5 and ZIF-8; differs for graphite, quartz, the triclinic edge case and the hexagonal edge case. The differences are conventions, not detection errors: moyo refines the standardized cell and may choose a different setting/orientation, while spglib idealizes to its own standard setting. The test records these four cases as known exceptions.

The comparison uses Setting::Spglib; this is the setting the wrapper exposes by default to match Crystal-Builder's spglib behaviour.

Verification notes

• std_cell refines both the lattice and the positions to the detected symmetry; it does not only apply the transformation. This matches spglib's idealization and is what the Phase 2 API exposes through standardize.
• moyo::base::Cell takes atomic numbers (or integer species ids), not occupancies. Partial occupancy must therefore be encoded as distinct species (for example a split-site species id per component) before detection; the wrapper performs this conversion and documents the species mapping in xtalforge-symmetry::convert.
• Setting::Spglib reproduces spglib's Hall number on the audit corpus; Setting::default chooses origin choice 2 for Si (Hall 526) instead of spglib's 525. The origin choice also explains moyo's 8a/8b Wyckoff difference, which the wrapper now resolves in the input origin.

Gap decisions

Gap	Decision
Niggli, Delaunay, Minkowski private	Open an upstream issue/PR to export them; meanwhile port into xtalforge-symmetry with attribution (Task 2.6.2).
Symmetrization private	Use MoyoDataset::std_cell for Phase 2; if a free symmetrization is needed, port the algorithm with attribution (Task 2.6).
Brillouin zone and k-path	Implement the Wigner-Seitz construction and band paths ourselves (Task 2.10), because Crystal-Builder gets them from ASE Cell.bandpath.
Molecular point groups	Adapt moyo::identify::PointGroup to non-periodic rotations; if the adapter is insufficient, implement the molecular point-group detection (Task 2.8.3).
Origin/setting differences	Expose Setting::Spglib as the default; document the origin choice and Wyckoff-letter difference in the API page.

Build, target and size

• cargo check -p xtalforge-symmetry --target wasm32-unknown-unknown: passes.
• maturin build of the Python extension: passes.
• Dependency tree: moyo 0.22 uses nalgebra 0.35 (the workspace version), serde, strum 0.28, itertools 0.15, log, rustc-hash, union-find; no duplicate nalgebra, serde or thiserror.
• Release compile of xtalforge-symmetry with moyo on this machine: about 45 seconds; the incremental check is much faster.
• WASM package: the full release is 3943 KiB at 1.1.0, including the symmetry tables, force fields, analysis, builders, powder surfaces and the agent session, inside the 3968 KiB budget checked by scripts/check-wasm-size.sh.

Isolation

xtalforge-symmetry defines:

• SpaceGroup: number, Hall number, Hermann-Mauguin symbol, setting and operations, independent of moyo types;
• SymOp: integer rotation and fractional translation;
• SymmetryDataset: standardized cell, primitive standardized cell, orbits, Wyckoff sites and transformations;
• WyckoffSite: letter, multiplicity and site symmetry.

Conversions live in one private module; the public API never names a moyo type. The audit test is the only place that currently imports moyo.