Prefix mi_. Every buffer is caller-owned. A nonzero return means failure; read mi_last_error() on the same thread.

Symbol	Role
mi_cell	Lattice vectors a, b, c and origin
mi_cell_ortho	Header-only diagonal box
mi_cell_from_vectors	Columns a, b, c
mi_cell_from_lammps	xlo xhi ylo yhi zlo zhi plus tilts
mi_cell_from_lammps_bounds	Dump bound spans plus tilts
mi_cell_from_ase	Row-major 3x3
mi_cell_from_con	CON 3x3
mi_cell_from_con_box	Lengths and angles in degrees
mi_cell_from_vesin	vesin 3x3
mi_displacement	Wrapped q - p
mi_displacement_euclidean	Smith test, else the Selling slicer
mi_dist2	Squared minimum-image distance
mi_dist2_many	One source, packed candidates
mi_dist2_pairs	Packed pair lists
mi_dist2_euclidean_many	Euclidean nearest image, one source, packed candidates
mi_dist2_euclidean_pairs	Euclidean nearest image, packed pair lists
mi_dist2_euclidean_pairs_warm	Euclidean pairs from the previous frame's shifts
mi_dist2_ortho_diffs	Highway-shaped ortho wrap, AVX when present
mi_dist2_shifted_many	One lattice shift, many candidates
mi_fixed_many	Positions to 64-bit fixed-point fractions
mi_dist2_fixed	Squared distance between two fixed-point positions
mi_dist2_many_fixed	One fixed-point source, packed candidates
mi_dist2_pairs_fixed	Packed fixed-point pair lists
mi_fixed32_many, mi_dist2_fixed32	32-bit fixed-point fractions and distance
mi_dist2_many_fixed32, mi_dist2_pairs_fixed32	32-bit fixed-point batches
mi_reduce_pairs	Drop self images, unique pairs
mi_last_error	Thread-local error string
mi_version	Process-static version

The C++ header minimage.hpp is a RAII wrap of this ABI. docs.rs/minimage is the Rust reference.

Python

Method	Role
dist2_many(p, qs)	One source, rows of candidates
dist2_pairs(ps, qs)	Paired rows
dist2_euclidean_many(p, qs), dist2_euclidean_pairs(ps, qs)	Euclidean nearest image
dist2_euclidean_pairs_warm(ps, qs, images)	Distances and updated shifts across frames
wrap_many(diffs)	Engine wrap of each row
fixed(r), fixed_many(rs)	Fixed-point fractions, uint64
dist2_fixed(a, b)	Two fixed-point positions
dist2_many_fixed(p, qs)	One fixed-point source, rows
dist2_pairs_fixed(ps, qs)	Paired fixed-point rows
fixed32(r), fixed32_many(rs), dist2_fixed32(a, b)	32-bit fixed point, uint32
dist2_many_fixed32(p, qs), dist2_pairs_fixed32(ps, qs)	32-bit fixed-point batches

An array argument is any DLPack producer on the host, read in place; an array result is a numpy array. A list in gives a list out.

Burn

Item	Role
Lattice::new(cell, device, dtype)	H and Hinv on a device
Lattice::rows	Packed rows to an (n, 3) tensor
Lattice::fractions	Fold on the CPU in double precision, upload
Lattice::fractional	Fold on the device
Lattice::wrap, dist2_pairs, dist2_many	Cartesian rows
Lattice::displacement_fractional, dist2_pairs_fractional, dist2_many_fractional	Fraction rows, exact wrap
to_vec	A rank-one tensor back as doubles