parosol-py 0.1.19

Python package and API wrapper for the ParOSol micro-FE solver

parosol-py

parosol-py is the Python package and runtime wrapper for the ParOSol micro-FE solver. It provides Python helpers for creating solver inputs, running the bundled native executable, reading outputs, and mapping label or density images to material stiffness.

The bundled native ParOSol solver was written by Cyril Flaig and is distributed under the GNU General Public License, version 2 or later. The Python package is therefore distributed as GPL-2.0-or-later.

Install

Prebuilt wheels include the native ParOSol executable and the platform MPI runtime needed to launch multi-process solves:

python -m pip install parosol-py

For local development:

python -m pip install -e .[dev]

Source installs require CMake, an MPI C++ compiler/runtime, HDF5 C++ libraries, and Eigen headers. Release wheels are built for Python 3.11, 3.12, and 3.13. macOS wheels currently target macOS 15 or newer.

Bundled MPI runtimes remain under their own upstream licenses and notices. Windows wheels bundle Microsoft MPI runtime files under Microsoft/MIT terms. macOS and Linux wheels bundle OpenMPI runtime files under OpenMPI's BSD-style terms. These MPI runtime files are not relicensed as GPL by parosol-py.

Local Check

Before relying on GitHub Actions, run the local verification gate from an environment that has the native build tools:

python scripts/local_check.py

Add --smoke-install to install the built wheel into a temporary virtual environment and import it.

Workflows And Cases

ParOSol-py separates reusable recipes from concrete runs:

• A workflow/profile recipe is a reusable .parosol-workflow or future .parosol-profile file. SlicerParOSol creates and edits these recipes.
• A ParOSol case is one materialized run. Shortcut commands write a parosol_case.yaml file with concrete input, output, solver, material, boundary-condition, and postprocessing settings.
• A .parosol bundle is a portable case bundle for solving the same case on another machine.

Built-in reusable workflows are selected by name:

parosol distal-radius.AIM --profile XtremeCTII --output outputs/distal-radius

parosol 10001_QCT.nii.gz \
  --mask 10001_SEG.nii.gz \
  --profile spine-compression \
  --output outputs/10001_spine

parosol 10001_QCT.nii.gz \
  --mask 10001_SEG.nii.gz \
  --profile hip-sideways-fall-left \
  --output outputs/10001_hip_left

Slicer-authored workflows can be replayed headlessly:

parosol new_scan.nii.gz \
  --mask new_mask.nii.gz \
  --profile interactive_custom \
  --template my_workflow.parosol-workflow \
  --output outputs/new_scan

To run a fully materialized case directly:

parosol run outputs/10001_spine/parosol_case.yaml

Python API

import numpy as np

from parosol_py import solve

material = np.ones((10, 10, 10), dtype=np.float32) * 1000.0

result = solve(
    material=material,
    spacing=(0.061, 0.061, 0.061),
    material_unit="MPa",
    test="axial",
    test_axis="z",
    strain=-0.01,
    outputs=("sed",),
    export_dir="outputs/example",
)

print(result.summary)
print(result.exported)

Use dry_run=True to write the ParOSol HDF5 input and command without launching the solver:

result = solve(
    material=material,
    spacing=(0.061, 0.061, 0.061),
    material_unit="MPa",
    test="axial",
    test_axis="z",
    strain=-0.01,
    dry_run=True,
    export_dir="outputs/dry_run",
)

Material Mapping

Label images can be mapped through an explicit material table:

import numpy as np

from parosol_py import LinearIsotropicMaterials, labels_to_material_map

labels = np.array([[[100, 127]]], dtype=np.uint16)
table = LinearIsotropicMaterials(
    youngs_modulus_mpa={100: 8748.0, 127: 8748.0},
    poisson_ratio={100: 0.3, 127: 0.3},
)

mapped = labels_to_material_map(labels, table)

Continuous density images can be converted with one of the supported equations:

from parosol_py import density_to_material_map

mapped = density_to_material_map(
    density_image,
    equation="power",
    coefficient=10000.0,
    exponent=1.7,
    reference_density=1000.0,
    poisson_ratio=0.3,
)

The Mulder grayscale BMD law is available as mulder2007:

mapped = density_to_material_map(
    density_image,
    equation="mulder2007",
    active_mask=outer_contour,
    floor_e_mpa=2.0,
    poisson_ratio=0.3,
)