From PyPi

Warning

Make sure that the installation folder name and any parent folder names do not contain white spaces since this will break the installation at various points. In general white spaces in filenames and folder names should be avoided.

By far the easiest way to get access to Dune is by installing into a virtual environment using pip. In some folder e.g. Dune setup a virtual environment and activate it

python3 -m venv dune-env
source dune-env/bin/activate

Then download and build Dune (note that this takes some time to have coffee/tea at hand):

pip install --pre dune-fem

Note

This tutorial is based on the upcoming 2.10 version of Dune. A 2.9 release version is available in the Python package index and can be obtained by removing the –pre from the above install command. Note that a few of the features described in this tutorial will not be available and that the API has changed in some places. See the description in the changelog for details.

Additional external packages

To use additional external packages like petsc or umfpack which are not in a default location, the cmake prefix variable can be set during installation process:

CMAKE_FLAGS="-DCMAKE_PREFIX_PATH=PATHLIST" pip install --pre dune-fem

where PATHLIST is a semicolon separated list of path containing external packages.

Testing the installation

Important

The current installation requires MPI to be available and requires the additional installation of mpi4py:

pip install mpi4py

Note

The first time you construct an object of a specific realization of one of the Dune interfaces (e.g. here a structured grid), the just in time compiler needs to be invoked. This can take quite some time - especially for grid realizations. This needs to be done only once so rerunning the above code a second time (even using other parameters in the structuredGrid function) should execute almost instantaniously.

To test that everything works you can download all the scripts described in this tutorial and try them out

python -m dune.fem
cd fem_tutorial
python concepts.py

All examples are available as Python scripts or IPython notebooks - for the later jupyter is needed

pip install jupyterlab
jupyter lab

This has been tested with different Linux distributions, MacOS, and using the Windows Subsystem for Linux.

Hints for Windows

Some hints on getting Dune to work using the Windows Subsystem for Linux (tested on Windows 11). Installation in three steps:

  1. First we need to install the wsl (here an Ubuntu version):

    Open PowerShell as administrator and run wsl --install (find Windows PowerShell and right click; the first entry should be Run as administrator). This step takes quite some time (‘get a coffee’ long). Close the PowerShell again (enter exit). Possibly one needs to restart after this step (second coffee).

  2. We need to add some packages and setup a Python virtual environment. Open the wsl (Pinguin icon) - again some installation is done. Enter a new username and password. Then run the following commands:

    sudo apt update  # enter the password you used above
sudo apt install --reinstall ca-certificates
sudo apt install python3-dev python3-pip python3-venv cmake
sudo apt install jupyter-core
python3 -m venv dune-env
source dune-env/bin/activate
pip install jupyterlab

3. We have reached the Dune specific part of the installation as desribed above in a lot more detail (I would suggest some tea at this stage)

pip install --pre dune-fem
python -m dune.fem

The last step downloads the tutorial scripts into the folder fem_tutorial.

Each time to open the Linux terminal (wsl) again you will need to run the following commands:

source ~/dune-env/bin/activate

To work on one of the scripts from the tutorial you can either use jupyter-lab

cd ~/fem_tutorial/
jupyter lab &

then open the given link in your favourite web browser.

Instead of using the notebooks you can also run the python scripts from the command line, e.g., run

python concepts.py

in the fem_tutorial folder.

From Source

Note

We strongly encourage the use of a python virtual environment and the following instructions are written assuming that a virtual environment is activated.

Requirements

The following dependencies are needed for Dune-Fem python binding:

The optional Dune modules are only need for the parts of the tutorial discussing extension modules.

Building the Required Dune Modules

Read the instructions on how to build Dune with Python support which also links to general instructions on how to build Dune modules.

The dune-fem-dg module offers an example script to build all required modules from source.

Test your completed installation by opening a Python terminal and running

import math
from dune.grid import structuredGrid
from dune.fem.function import gridFunction
grid = structuredGrid([0,0],[1,1],[10,10])
@gridFunction(grid,name="test",order=2)
def f(x):
   return math.sin(x.two_norm*2*math.pi)
f.plot()

If you have everything set up correctly (and have matplotlib) you should get a colored figure and are hopefully ready to go…

Troubleshooting

  • Issue with C++ compiler: If the gnu compiler is used, version needs to be 7 or later. This can be checked in terminal with

    $ g++ --version

    If your version is out of date, you will need to upgrade your system to use Dune

  • Python version: It is possible that the python version may be an issue. The scripts require python3 including the development package being installed. If during the Dune installation you get the error

    fatal error: pyconfig.h: No such file or directory

    This can probably be fixed by installing additional python3 libraries with e.g.

    $ sudo apt-get install libpython3-dev

  • MPI not found: One other problem is that a default version of Open MPI may already be installed. This will lead to errors where Dune appears to be looking in the wrong directory for Open MPI (e.g. usr/lib/openmpi instead of the home directory where the script installs it). This can be solved by running

    $ make uninstall

    in the original MPI install directory, followed by removing the folder. It will then be necessary to reinstall Open MPI and Dune. It may also be necessary to direct mpi4py to the new MPI installation. It is possible to check whether this is a problem by running python and trying out

    from mpi4py import MPI

    If it comes up with an error, this can be fixed by installing mpi4py manually using the following commands

    $ git clone https://bitbucket.org/mpi4py/mpi4py.git
$ cd mpi4py
$ python setup.py build --mpicc=/path/to/openmpi/bin/mpicc
$ python setup.py install --user

  • User warning from numpy:

    UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.

    This is caused by some shared library using the compiler flag fast-math. Check for example that you are not using this flag in your cmake setup for Dune. See here for a detailed description.

  • Using petsc and petsc4py: It is important the all dune modules and petsc4py use the same version of petsc. An runtime exception will be thrown if some incompatibility is detected. The simplest way to avoid any issues is to set the PETSC_DIR environment variable during the building, e.g., pip installation of the dune packages and running the scripts. The following gives an example of how to first install petsc and petsc4py into a virtual environment and then install dune-fem. Assuming that the virtual environment with python version X.Y is in $HOME/dune-env and we want a few extra petsc packages we could run: ` pip install mpi4py PETSC_CONFIGURE_OPTIONS="--download-hypre --download-mumps --download-parmetis --download-ml --download-metis --download-scalapack" pip install petsc export PETSC_DIR=$HOME/dune-env/lib/pythonX.Y/site-packages/petsc pip install petsc4py pip install dune-fem ` After this the PETSC_DIR environment variable should not be required anymore. This is all a bit experimental so let us know if it is not working. Of course an installed petsc or build from source should work in the same way, without the need for the pip install petsc line.

  • Newly installed software is not used: After for example adding petsc to your system one needs to remove an existing dune-py module which contains the jit compiled modules. New software components are not automatically picked up. One can run

    python -m dune info

    to find the location of the dune-py folder. That folder needs to be removed before the new component can be used.

    If issues remain reinstalling all dune packages might be required - including the cached wheels:

    pip uninstall -y `pip freeze | grep dune`
pip cache remove "dune*"

    When running pip install for the dune packages setting prefix path for cmake as described above might be necessary if the new package is not in a default location.

  • Output to terminal seems a bit random: the issue is (probably) that Python buffers its print output and C++ does not. So if in a mixed program both are writing to the terminal (or piped into a file) the C++ output often appears before the Python output. This can be fixed by (i) adding flush=True to the Python print statements or setting the environment variable PYTHONUNBUFFERED to some non zero value.

https://zenodo.org/badge/DOI/10.5281/zenodo.3706994.svg