Getting started

Installation

DRAGonFLy depends on fenics-dolfinx 0.11.0 and mpich, which are only distributed through conda-forge. The environment.yml file in the repository root sets these up with Python 3.12 and pip-installs the package with its remaining dependencies, including the LSDOlab packages CSDL, lsdo_geo, lsdo_function_spaces, modOpt and IDWarp-JAX.

For developers

$ git clone https://github.com/LSDOlab/DRAGonFLy.git
$ cd DRAGonFLy
$ conda env create -f environment.yml
$ conda activate dragonfly

This installs the package (dragonfly-sim, imported as dragonfly_sim) in editable mode, with the test and documentation tools.

For users

In an existing conda environment, install the conda-forge dependencies (skip this if they are already present), then the package, from which pip installs the remaining dependencies:

$ conda install -c conda-forge fenics-dolfinx=0.11.0 mpich pyvista networkx
$ pip install git+https://github.com/LSDOlab/DRAGonFLy.git

Dependencies

• From conda-forge: fenics-dolfinx 0.11.0 (not on PyPI), which brings mpi4py, petsc4py, ufl and basix, mpich (not on PyPI), pyvista and networkx.

• From PyPI, installed by pip: numpy, scipy, matplotlib, jax.

• From the main branches of the LSDOlab repositories, installed by pip: CSDL_alpha, lsdo_function_spaces, lsdo_geo, modopt and IDWarp-JAX.

All runtime dependencies are declared in setup.py; pyvista and networkx are listed there too, so pip would fetch them from PyPI if they were missing.

Running an example

From the repository root:

$ OMP_NUM_THREADS=1 mpirun -n 4 python examples/airfoil_opt.py

This builds the 2D airfoil optimization model and verifies its derivatives against finite differences. Examples lists the other cases, including forward and unsteady analyses and the ONERA OAT15A validation case. Output files are written to the working directory.

OMP_NUM_THREADS=1 matters: the conda-forge PETSc links a multithreaded BLAS, which otherwise spawns one thread per core in every MPI rank and stalls the linear solves.

Running the tests

$ OMP_NUM_THREADS=1 pytest

The tests in tests/ cover the boundary/interior integration measures, the shape parameterization and the shape design-variable layers, the default Euler discretization against a stored reference (tests/data/euler_reference.npz), the RANS model (closures, Green-Gauss reconstruction, exact Jacobian and adjoint derivatives against finite differences, a manufactured-solution convergence rate), time integration, checkpoints, the far fields, forward analysis through DG_windtunnel_model, the POD snapshot SVD and the reduced-order model (Euler and RANS), the VTX file writer and the mesh input. They build their meshes in-process and take about a minute, including the first just-in-time compilation of the forms and the re-runs on three MPI ranks. The CGNS test on the OAT15A grid is skipped when the grid is not in meshes/. Some tests re-run their files under mpirun -n 3, so mpirun and pytest must be available in the same environment; without mpirun those tests are skipped.

Building the documentation

$ cd docs
$ make html

and open docs/_build/html/index.html. pip install -r requirements.txt installs the pinned documentation dependencies that the website build uses; pip install .[docs] installs unpinned versions. The website, https://lsdolab.github.io/DRAGonFLy, is rebuilt from main by GitHub Actions.