Installation
============

femorph-solver requires Python 3.10 or newer.

.. tab-set::

    .. tab-item:: pip

        .. code-block:: bash

            pip install femorph-solver

    .. tab-item:: uv

        .. code-block:: bash

            uv pip install femorph-solver

Optional extras
---------------

The core install ships with NumPy / SciPy / pyvista only — enough for
the SuperLU direct-solve fallback and all in-process modal / static /
transient analyses.  For the fastest SPD paths and for MAPDL deck
interop, install one or more extras:

.. list-table::
    :widths: 25 15 60
    :header-rows: 1

    * - Extra
      - Install as
      - What it adds
    * - ``mapdl``
      - ``pip install "femorph-solver[mapdl]"``
      - CDB / RST reader (``mapdl-archive``, ``ansys-mapdl-reader``)
    * - ``cholmod``
      - ``pip install "femorph-solver[cholmod]"``
      - SuiteSparse CHOLMOD Cholesky — fastest SPD direct solve for
        sparse problems.  Requires ``libsuitesparse-dev``.
    * - ``pardiso``
      - ``pip install "femorph-solver[pardiso]"``
      - Intel MKL PARDISO — multi-threaded SPD direct solve.
    * - ``umfpack``
      - ``pip install "femorph-solver[umfpack]"``
      - SuiteSparse UMFPACK — sparse LU for non-symmetric systems.
    * - ``pyamg``
      - ``pip install "femorph-solver[pyamg]"``
      - Algebraic multigrid preconditioner.
    * - ``primme``
      - ``pip install "femorph-solver[primme]"``
      - PRIMME block-Davidson eigensolver.
    * - ``solvers``
      - ``pip install "femorph-solver[solvers]"``
      - Convenience: everything except the per-backend quirks.

Every optional backend is *detected at call time* via
:func:`femorph_solver.solvers.linear.list_linear_solvers` and
:func:`femorph_solver.solvers.eigen.list_eigen_solvers` — if an extra
isn't installed, the corresponding name drops out of the registry and
``linear_solver="auto"`` falls through to the next best option.