sisl 0.16.4 Release Notes

Released on 2026-03-19.

Changes and improvements

Change trs default in MonkhorstPack depending on spin

Now for non-collinear or higher spin-configurations it defaults to trs=False.

(PR927)
Full implementation of Bloch 3D expansion

It now fully utilizes the Toeplitz matrix structure for huge performance improvements.

(PR951)
Improved performance of real-space self-energies

Now there is only a single copy at the end of the integration, so it should be much faster.

(PR951)
Improved Atoms.index usability

Now one gets an empty list if no atoms are found. One can also get the combined indices of several atom types, e.g., atoms.index([“C”, “H”]).

(PR954)
Changed spherical_harm argument order to match scipy 1.17

Since this is generally not really used, it should be safe to do. Sorry for the inconvenience.

(PR969)
Dropped support for Python 3.9
Enabled simpler rotation argument to Geometry.rotate

This makes rotation arguments more versatile, and also easier to combine.

One can now do lists of rotations that then gets parsed directly.
Implemented dependency-groups in pyproject.toml

This affects developers where the docs build specification is now hidden from end-users.
added oxyz to extract coordinates of orbitals

This allows easier access to orbital coordinates rather than having to convert indices to atoms and then to supercell coordinates.

Basically equivalent to: `python atoms = geom.o2a(orbitals) geom.axyz(atoms) == geom.oxyz(orbitals) `

Fixed spin-rotations

Now one can denote rotations in a much more generic way.

Removed rotation_matrix and superseeded by parse_rotation which returns a Quaternion which can be used to create a rotation-matrix.

(PR924)
Reading the basis-enthalpy for Siesta 5.4 changed

(PR929)
Fixed finding coordinates within elliptical cylinders

(PR947)
Fixed import error for scipy deprecations

(PR968)
Fixed order in ORB_INDX files when species are not linear

(PR971)
Fixed AtomCategory usage when sisl.geom hasn’t been imported

(PR982)
bug fix for pandas >= 3 migration

(PR983)
DIIS mixing now uses a scaling factor for inverse

This provides a better stability for the inversion algorithm when the metrics are much bigger than 1.
Fixed reshape calls, removed from ndarray in later versions
Fixed skewed cell neighbor search

For the linear neighbor search mechanism it needed to account for skewed cells when the binsize is very close to the inter-atomic distances.

(reported by @ialcon on Discord, thanks!)

Fixed wording in Geometry.scale documentation

It is now clearer that both lattice vectors and atomic coordinates are scaled.

(PR949)