apxchol 0.2.2

Metadata-only release: the package author list now matches CITATION.cff (Oleksandr Kulkov, Yves Baumann, Rasmus Kyng). No code changes since 0.2.1.

apxchol 0.2.1

First public release.

Library

• Parallel approximate Cholesky (Kyng–Sachdeva / GKS approximate elimination) preconditioner + tuned PCG for graph-Laplacian and SDDM systems; Laplacian vs SDDM auto-detected, disconnected inputs handled natively.
• Public customization seams: star-elimination rule (eliminator: sample_clique(neighbors, deg, seed, emitter), schedule-independent draws) and IS selection / elimination order (partitioner: uniform find_partition(G, candidates, ctx, selection&) with an optional degree_prepass trait). Lambda adapters, self-verifying examples/, docs/extending.md.
• cpu_solver: factor-once/solve-many, caller-provided output memory, x0; apx_cholesky::set_factor for external factors; CLI apxchol.

Python (wheels published from this release)

• apxchol.factorize(A, *, seed, partitioner, storage, keep_factor, ...) → P/L/D/chol()/factor_nnz/fill_ratio; solve(b, tol=, rtol=, maxiter=, x0=, out=); aspreconditioner().
• manylinux x86_64 wheels, CPython 3.10–3.14.

Also

• Octave/MATLAB MEX bindings; standalone benchmark suite (Hypre BoomerAMG, AMGCL, RCHOL/pRCHOL, ParAC, CMG, Laplacians.jl) with a committed per-cell result store and charts.
• License: BSD 4-Clause.