Build

The wheel is built with scikit-build-core. CMake fetches Corrosion 0.6.1 and compiles the capi static library. A C++17 compiler, CMake, Ninja, and Rust are required.

python -m pip install 'rgmin[ase,test]' adds ASE and the test extras.

Options

SolverConfig checks every keyword once. An unknown name raises ValueError and lists the names that are accepted. memory, maxiter, linesearch_maxiter, and extra_updates have to be Python integers. Other numeric options have to be finite.

A Cartesian solver starts with a per-atom cap of 0.2 (ASE maxstep, eOn max_move). maxstep or atom_maxmove replaces it. A non-positive value clears it. A positive maxmove is the Euclidean step limit and leaves the per-atom default off. The cap is the largest per-atom displacement, not the norm of the whole step.

ASE

rgmin.ase.FIRE, LBFGS, CG, and the other method classes are ASE optimizers. Each step is one rgmin iteration. maxstep is the per-atom cap in angstroms. The ASE classes pass it explicitly, and their default is 0.2. ASE calculators are not thread-safe. Each thread that builds an optimizer needs its own calculator. rgmin still guarantees a shareable config, one solver per thread, and an owned copy of the coordinates. The thread-safety page records the details.

from ase.calculators.emt import EMT
from rgmin.ase import FIRE

atoms.calc = EMT()
FIRE(atoms, logfile=None).run(fmax=0.05)

Newton, RFO, and dogleg need hessian(x), a callable that returns the (n, n) matrix. A band or a min-mode run accepts the step rather than the energy. Pass accept= to choose the rule yourself.

restart is refused. read and dump raise NotImplementedError.