sisl 0.17.0 Release Notes

Released on 2026-08-28.

New features

Added cifSile for initial support, basic symmetry works

(PR992)
implemented a tensor-IPR method

It retains the complex part as it is not guaranteed to be a real quantity. Note, depending on the size of the states, this can yield a really big matrix.

(PR993)

Fixed Pyparsing deprecation warnings

(PR984)
Sped up Geometry.close and Geometry.within for large auxiliary supercells

The supercell results were accumulated with a np.concatenate on every supercell iteration, copying the whole growing array each time (scaling as the square of the number of supercells). They are now collected and concatenated once, which is noticeably faster for geometries with many supercells (large nsc) or long-range searches.

(PR1001)
Moved fortran import checks to the module itself

Now everything is accessible, but not usable. Instead an error is issued when a fortran module attribute is tried accessed.
Sped up sparse matrix element-wise operations, Rij and eliminate_zeros

SparseAtom.Rij/SparseOrbital.Rij (and the derived rij) now build the distance vectors in a single vectorized call instead of looping over every row in Python. Element-wise operations between two sparse matrices (e.g. H1 + H2) no longer allocate a per-row index array, and SparseCSR.eliminate_zeros now compacts all zero elements in one pass rather than deleting row-by-row.

Fixed sile exploration with specification

E.g. sgeom test{car} has to be changed to sgeom test{startswith=car} to get a singular sile match.

(PR988)
Fixed SparseCSR.eliminate_zeros skipping matrices whose only zero was in row 0

The early-return short-circuit summed row indices rather than counting zero elements, so a matrix whose only explicit zero resided in row 0 was left untouched. The zeros are now detected and removed correctly.
portable complex arithmetic in the fused-type Cython kernels

The k-space matrix kernels are templated on independent fused types for the matrix data and the phases, so Cython instantiates real/complex combinations that never occur at runtime. Those instantiations emitted a cast from a complex to a real value, and in-place += on complex memoryview elements, neither of which compiles unless the C compiler maps complex numbers onto the native C99 _Complex type. Both are now routed through the new cast_add/cast_assign helpers in sisl/_core/_dtypes.pxd, which resolve the real/complex distinction at compile time, so the extensions also build with the struct-based complex fallback (MSVC) and in C++ mode.
reading fhiaims input geometries for molecules

Previously it failed when the lattice was not defined in the file. Now it reads but puts an arbitrary lattice.

Refactor non-standard object identity tests.

Refactors not object is None to object is not None to improve readability and adhere to the standard.

(PR996)
Consolidated the element-deletion + pointer-update idiom in SparseCSR

The three copies of the element deletion logic now share a single SparseCSR._delete_stored helper. This is a pure internal refactor with no change in behavior.
Enabled pyodide builds and more architechtures
enabled windows wheels

Fully compilable wheels for windows. The cython sources has been fixed such that Windows builds are enabled. Note that the shipped wheels do not enable Fortran sources.

A total of 3 people contributed to this release. People with a “+” by their names contributed a patch for the first time.

A total of 5 pull requests were merged for this release.