This document is under active development and discussion!

If you find errors or omissions in this document, please don’t hesitate to submit an issue or open a pull request with a fix. We also encourage you to ask questions and discuss any aspects of the project on the Feel++ Gitter forum. New contributors are always welcome!

1. Discussion Forum

We’re always happy to help out with Feel++ or any other questions you might have. You can ask a question or signal an issue at the Gitter Feel++ salon.

Join the Feel++ chat


This book is available on Github. We use Gitter to discuss the changes in the book.

Join the Feel++ book chat


2. Conventions used in this book

The following typographical conventions are used in the book

Italic indicates new terms

typewriter is used on program listings as well as when referring to programming elements, e.g. functions, variables, statements, data types, environment variables or keywords.

$ typewriter or > typewriter displays commands that the user types literally without the $ or >.

this is a general note.
this is a general warning.
be cautious

3. Mathematical Notations

3.1. Geometry and Meshes

  • \(d=1,2,3\) geometrical dimension

  • \(\Omega \subset \mathbb{R}^d\)

  • \(K\) a cell or element of a mesh

  • \(h\) characteristic mesh size

  • \(k_{\mathrm{geo}}\) polynomial order of the geometrical transformation

  • \(\delta=(h,k_{\mathrm{geo}})\) discretization parameter pair for the geometrical transformation, default value \(k_{\mathrm{geo}}=1\) (straight cells or elements)

  • \(\varphi^K_\delta: \hat{K} \rightarrow K\), geometrical transformation

  • \(\mathcal{T}_{\delta}\) a triangulation, \(\mathcal{T}_\delta = \{ K\; | \; K=\varphi^K_\delta (\hat{K}) \} \)

  • \(\Omega_h \equiv \cup_K {K}\)

3.2. Spaces

  • \(P^k_{c,h} = \{ v_h \in C^0(\bar{\Omega}); \forall K \in \mathcal{T}_h,\ v_h \circ T_K \in \mathbb{P}^k\}\) Space of continuous piecewise polynomial of total degree \(\leq k\).

Introduction to Feel++

Discuss and Contribute
Use Issue 858 to drive development of this section. Your contributions make a difference. No contribution is too small.

4. What is Feel++?

Feel++ is a unified C++ implementation of Galerkin methods (finite and spectral element methods) in 1D, 2D and 3D to solve partial differential equations.

Feel++ is

  1. a versatile mathematical kernel solving easily problems using different techniques thus allowing testing and comparing methods, e.g. cG versus dG.

  2. a small and manageable library which nevertheless encompasses a wide range of numerical methods and techniques and in particular reduced order methods such as the reduced basis method.

  3. a software that follows closely the mathematical abstractions associated with partial differential equations (PDE) and in particular the finite element mathematical framework and variational formulations.

  4. a library that offers solving strategies that scales up to thousands and even tens of thousands of cores.

  5. a library entirely in C++ allowing to create C++ complex and typically non-linear multi-physics applications currently in industry, physics and health-care.

Quick Starts

5. Installation Quick Start

Using Feel++ inside Introduction is the recommended and fastest way to use Feel++. The Docker chapter is dedicated to Docker and using Feel++ in Docker.

We strongly encourage you to follow these steps if you begin with Feel++ in particular as an end-user.

People who would like to develop with and in Feel++ should read through the remaining sections of this chapter.

6. Usage Start

Start the Docker container feelpp/feelpp-base or feelpp/feelpp-toolboxes as follows

> docker run -it -v $HOME/feel:/feel feelpp/feelpp-toolboxes
these steps are explained in the chapter on Feel++ Containers.

Then run e.g. the Quickstart Laplacian that solves the Laplacian problem in Quickstart Laplacian sequential or in Quickstart Laplacian on 4 cores in parallel.

Quickstart Laplacian sequential
> feelpp_qs_laplacian_2d --config-file Testcases/quickstart/laplacian/feelpp2d/feelpp2d.cfg

The results are stored in Docker in


and on your computer


The mesh and solutions can be visualized using e.g. Parariew or Visit.

ParaView (recommended)

is an open-source, multi-platform data analysis and visualization application.


is a distributed, parallel visualization and graphical analysis tool for data defined on two- and three-dimensional (2D and 3D) meshes

Quickstart Laplacian on 4 cores in parallel
> mpirun -np 4 feelpp_qs_laplacian_2d --config-file Testcases/quickstart/laplacian/feelpp2d/feelpp2d.cfg

The results are stored in a simular place as above: just replace np_1 by np_4 in the paths above. The results should look like





7. Syntax Start

Here are some excerpts from Quickstart Laplacian that solves the Laplacian problem. It shows some of the features of Feel++ and in particular the domain specific language for Galerkin methods.

First we load the mesh, define the function space define some expressions

Laplacian problem in an arbitrary geometry, loading mesh and defining spaces
    auto mesh = loadMesh(_mesh=new Mesh<Simplex<FEELPP_DIM,1>>);

    auto Vh = Pch<2>( mesh );
    auto u = Vh->element("u");
    auto mu = expr(soption(_name="")); // diffusion term
    auto f = expr( soption(_name="functions.f"), "f" );
    auto r_1 = expr( soption(_name="functions.a"), "a" ); // Robin left hand side expression
    auto r_2 = expr( soption(_name="functions.b"), "b" ); // Robin right hand side expression
    auto n = expr( soption(_name="functions.c"), "c" ); // Neumann expression
    auto solution = expr( checker().solution(), "solution" );
    auto g = checker().check()?solution:expr( soption(_name="functions.g"), "g" );
    auto v = Vh->element( g, "g" );

Second we define the linear and bilinear forms to solve the problem

Laplacian problem in an arbitrary geometry, defining forms and solving
    auto l = form1( _test=Vh );
    l = integrate(_range=elements(mesh),
    l+=integrate(_range=markedfaces(mesh,"Robin"), _expr=r_2*id(v));
    l+=integrate(_range=markedfaces(mesh,"Neumann"), _expr=n*id(v));

    auto a = form2( _trial=Vh, _test=Vh);
    a = integrate(_range=elements(mesh),
                  _expr=mu*inner(gradt(u),grad(v)) );
    a+=integrate(_range=markedfaces(mesh,"Robin"), _expr=r_1*idt(u)*id(v));
    a+=on(_range=markedfaces(mesh,"Dirichlet"), _rhs=l, _element=u, _expr=g );
    //! if no markers Robin Neumann or Dirichlet are present in the mesh then
    //! impose Dirichlet boundary conditions over the entire boundary
    if ( !mesh->hasAnyMarker({"Robin", "Neumann","Dirichlet"}) )
        a+=on(_range=boundaryfaces(mesh), _rhs=l, _element=u, _expr=g );

More explanations are available in Learning by examples.

Installing Feel++

8. Getting Started

This section describes the available ways to to download, compile and install Feel++.

8.1. Docker

Using Feel++ inside Introduction is the recommended and fastest way to use Feel++. The Docker is dedicated to Docker and Feel++ Containers is dedicated to Feel++ in Docker.

We strongly encourage you to follow these steps if you begin with Feel++ in particular as an end-user.

People who would like to develop with and in Feel++ should read through the remaining sections of this chapter.

8.2. System requirements

8.2.1. Compilers

Feel++ uses C++14 compilers such as GCC6 and Clang. Currently it is not mandatory to have a C++14 stantard library but it will be soon.

There used to be a major compatibility issue between llvm/clang and GCC compilers since GCC5 released the ABI tag which makes it impossible to compile Feel++ using llvm/clang with GCC5 or GCC6 standard libraries for a time. Please see the following table to understand the working C++ compiler / C++ standard library combinations.

Table 1. Table C++ compilers and standard libraries combinations
Compiler Standard Library

clang (3.6, 3.7, 3.8)

libstdc++ 4.9


libc++ (corresponding clang version)

clang (3.8(requires patches), 3.9)

libstdc++ 6


libstdc++ 6

GCC 6.2.1 seems to be problematic on debian/testing — the tests in the testsuite fail. — GCC 6.3.1 or GCC 6.2.0 don’t have any problems.

8.2.2. Required tools and libraries

Other than C++14 compilers, Feel++ requires only a few tools and libraries, namely CMake, Boost C++ libraries and an MPI implementation such as open-mpi or mpich. The table below provides information regarding the minimum and maximum version supported. A — means it has not necessarily been tested with the latest version but we do not expect any issues. Note that for MPI, an implementation with MPI-IO support would be best.

Table 2. Table required tools to compile Feel++
Name Minimum Version Maximum Version Notes







openmpi or mpich




Here is a list of libraries that we recommend to use jointly with Feel++.

Table 3. Table optional external libraries
Library Minimum Version Maximum Version Notes




Enables high performance I/O; Enables MED Support; Be careful on Debian/sid a more recent version of HDF5 breaks MED support




Last is best; a requirement for parallel and high performance computing




last is best; a requirement for eigenvalue problem; depends on PETSc




last is best; a requirement if you want to be able to read many file formats; HDF5 version in Debian/sid currently breaks MED format support.


superlu and superlu_dist


umfpack (colamd,amd)



Uncertainty quantification

Python Libs


Python 3 libraries

Python Sympy


Python 3 module sympy

Here is a list of tools that we recommend to use jointly with Feel++.

Table 4. Table of recommended tools
Tool License Notes

Computer Aided Design


Open Source

Mesh Generation


Open Source





Open Source




Open Source


Open Source

Scripting Languages


Open Source

Python 3 interpreter

Note that all these packages are available under Debian GNU/Linux and Ubuntu. Once you have installed those dependencies, you can go to Compiling.

8.2.5. Suggested tools

Here is a list of tools that we suggest to use jointly with Feel++.

Table 5. Table of suggested tools
Tool License Notes

Computer Aided Design (CAD)


Open Source


Open Source

HDF5 version in Debian/sid currently breaks MED format support.

Modeling, Compilation and Simulation Environment

Open Modelica

Open Source

Debugging and Profiling

Google perftools

Open Source


Open Source

8.3. Feel++ on Linux

We now turn to the installation of the Feel++ dependencies on Linux. Feel++ is currently support on Ubuntu (16.04, 16.10) and Debian (Sid, Testing).

8.3.1. Ubuntu

Ubuntu 16.10 Yaketti Yak

Here is the suggested installation of the Feel++ dependencies on Ubuntu 16.10

$ sudo apt-get -qq update
$ sudo apt-get install automake autoconf libtool libboost-all-dev\
  bash-completion emacs24 gmsh libgmsh-dev libopenturns-dev \
  libbz2-dev libhdf5-openmpi-dev libeigen3-dev libcgal-dev \
  libopenblas-dev libcln-dev libcppunit-dev libopenmpi-dev \
  libann-dev libglpk-dev libpetsc3.7-dev libslepc3.7-dev \
  liblapack-dev libmpfr-dev paraview python-dev libhwloc-dev \
  libvtk6-dev libpcre3-dev python-h5py python-urllib3 xterm tmux \
  screen python-numpy python-vtk6 python-six python-ply wget \
  bison sudo xauth cmake flex gcc-6 g++-6 clang-3.9 \
  clang++-3.9 git ipython openmpi-bin pkg-config
Ubuntu 16.04

Here is the suggested installation of the Feel++ dependencies on Ubuntu LTS 16.04

$ sudo apt-get install autoconf automake bash-completion bison\
 clang++-3.8 clang-3.8 cmake emacs24 flex g++-6 gcc-6 git gmsh\
  ipython libann-dev libbz2-dev libcgal-dev libcln-dev \
  libcppunit-dev libeigen3-dev libglpk-dev libgmsh-dev \
  libhdf5-openmpi-dev libhwloc-dev liblapack-dev libmpfr-dev\
   libopenblas-dev libopenmpi-dev libopenturns-dev libpcre3-dev \
   libpetsc3.6.2-dev libproj-dev libslepc3.6.1-dev libtool \
   libvtk6-dev openmpi-bin paraview pkg-config python-dev \
   python-h5py python-numpy python-ply python-six \
   python-urllib3 python-vtk6 screen sudo tmux wget xauth xterm
We are unfortunately stung by the ABI change in GCC 6 when using clang. You need to recompile the Boost C++ libraries to be able to use clang, see the section in the Annexes on Compiling Boost.

8.3.2. Debian

Debian Sid and Testing

At the time of writing there is little difference between Sid and Testing, here is the recommend dependencies installation command line:

$ apt-get -y install \
    autoconf automake bash-completion bison cmake emacs24 \
    flex git gmsh ipython libann-dev libboost-all-dev \
    libbz2-dev libcgal-dev libcln-dev libcppunit-dev \
    libeigen3-dev libglpk-dev libgmsh-dev \
    libhdf5-openmpi-dev libhwloc-dev liblapack-dev \
    libmpfr-dev libopenblas-dev libopenmpi-dev \
    libopenturns-dev libpcre3-dev libtool libvtk6-dev \
    openmpi-bin paraview petsc-dev pkg-config python-dev \
    python-h5py python-numpy python-ply python-six \
    python-urllib3 python-vtk6 screen slepc-dev sudo \
    tmux wget xauth xterm zsh
Older distributions

Unfortunately the older distributions have the ABI GCC issue with clang, e.g. Debian/jessie, or they are too old to support a simple installation procedure.

8.4. Mac OS X

Feel++ is supported on Mac OSX, starting from OS X 10.9 Mavericks to OS X 10.12 Sierra using Homebrew or MacPorts.

8.4.1. First step

Xcode is required on Mac OSX to install Feel++.

The easiest way to do so is to go through the Apple Store application and to search for Xcode. Xcode will provide the programming environment, e.g clang, for the next steps.

8.4.2. Homebrew

Introduction to HomeBrew

Homebrew is a free/open source software introduced to simplify the installation of other free/open source software on MacOS X. Homebrew is distributed under the BSD 2 Clause (NetBSD) license. For more information, visit their website.


To install the latest version of Homebrew, simply visit their website and follow the instructions. Each new package Homebrew installs is built into an intermediate place called the Cellar (usually /usr/local/Cellar) and then the packages are symlinked into /usr/local (default).

Key commands

Homebrew base command is brew. Here is a list of base available commands:

  • brew doctor: Check if the system has any problem with the current installation of Homebrew;

  • brew install mypackage: This command installs the package mypackage;

  • brew install [--devel|--HEAD] mypackage: These options respectively installs either the development version or the HEAD version of the package mypackage, if such versions are specified in the Formula file;

  • brew uninstall mypackage: This command allows to uninstall the package mypackage.


A Formula is a Ruby script format specific to Homebrew. It allows to describe the installation process of a package. Feel++ uses specific Formulae that you can get in the Feel++ github repository: feelpp/homebrew-feelpp.


This section is aimed at users that do not have Homebrew already installed.
In order to build Feel++ from Homebrew, you have to do the following steps:

First install Homebrew

pass:[\( /usr/bin/ruby -e "\)](curl -fsSL"

then check your Homebrew installation and fix warnings/errors if necessary

$ brew doctor

Install Homebrew-science tap to get the scientific software recommended or suggested for Feel++.

$ brew tap homebrew/homebrew-science

you should see something like

==> Tapping homebrew/science
Cloning into '/usr/local/Homebrew/Library/Taps/homebrew/homebrew-science'...
remote: Counting objects: 661, done.
remote: Compressing objects: 100% (656/656), done.
remote: Total 661 (delta 0), reused 65 (delta 0), pack-reused 0
Receiving objects: 100% (661/661), 591.93 KiB | 0 bytes/s, done.
Tapped 644 formulae (680 files, 1.9M)

Next you install Feel++ tap with

brew tap feelpp/homebrew-feelpp

you should read something like

==> Tapping feelpp/feelpp
Cloning into '/usr/local/Homebrew/Library/Taps/feelpp/homebrew-feelpp'...
remote: Counting objects: 5, done.
remote: Compressing objects: 100% (5/5), done.
remote: Total 5 (delta 0), reused 4 (delta 0), pack-reused 0
Unpacking objects: 100% (5/5), done.
Tapped 1 formula (30 files, 60.7K)

The final step is to either install Feel++

$ brew install feelpp

or just Feel++ dependencies if you plan to build Feel++ from sources yourself

$ brew install --only-dependencies feelpp

Note If you encounter problems, you can fix them using brew doctor. A frequent issue is to force open-mpi with brew link --overwrite open-mpi

Advanced usage

If Homebrew is already installed on your system, you might want to customize your installation for the correct dependencies to be met for Feel++.

Feel++ Dependencies

You can browse Feel++ dependencies using the following command:

$ brew deps feelpp | column

you get the list of formulas Feel++ depends on for its installation

ann		fftw		libtool		slepc
arpack		gcc		metis		suite-sparse
autoconf	glpk		mumps		sundials
automake	gmp		netcdf		superlu
boost		gmsh		open-mpi	superlu_dist
cln		hdf5		parmetis	szip
cmake		hwloc		petsc		tbb
eigen		hypre		scalapack	veclibfort
Customizing builds

If you want to customize the compilation process for a dependency (Set debug mode, Remove checking steps, Remove the link with certain libraries, etc.), you can access to the building options with the info flag. For exemple, with open-mpi:

$ brew info open-mpi

You get various information about the open-mpi formula

open-mpi: stable 2.0.1 (bottled), HEAD
High performance message passing library
Conflicts with: lcdf-typetools, mpich
/usr/local/Cellar/open-mpi/2.0.1 (688 files, 8.6M) *
  Built from source on 2016-09-26 at 10:36:46 with: --c++11 --with-mpi-thread-multiple
==> Dependencies
Required: libevent ✔
==> Requirements
Recommended: fortran ✔
Optional: java ✔
==> Options
	Build using C++11 mode
	Enable C++ MPI bindings (deprecated as of MPI-3.0)
	Build with java support
	Build without fortran support
	Install HEAD version

Then, you then just have to pass the needed flags, when installing the dependency.

Important: boost has to be installed with mpi and c++11 support and mumps needs to be installed with the following scotch5 support.

8.4.3. MacPorts


MacPorts is an open-source community projet which aims to design an easy-to-use system for compiling, installing and upgrading open-source software on Mac OS X operating system. It is distributed under BSD License and facilitate the access to thousands of ports (software) without installing or compiling open-source software. MacPorts provides a single software tree which includes the latest stable releases of approximately 17700 ports targeting the current Mac OS X release (10.9). If you want more information, please visit their website.

MacPorts Installation

To install the latest version of MacPorts, please go to Installing MacPorts page and follow the instructions. The simplest way is to install it with the Mac OS X Installer using the pkg file provided on their website. It is recommended that you install X11 (X Window System) which is normally used to display X11 applications.
If you have installed with the package installer (MacPorts-2.x.x.pkg) that means MacPorts will be installed in /opt/local. From now on, we will suppose that macports has been installed in /opt/local which is the default MacPorts location. Note that from now on, all tools installed by MacPorts will be installed in /opt/local/bin or /opt/local/sbin for example (that’s here you’ll find gcc4.7 or later e.g /opt/local/bin/g++-mp-4.7 once being installed).

Key commands

In your command-line, the software MacPorts is called by the command port. Here is a list of key commands for using MacPorts, if you want more informations please go to MacPorts Commands.

  • sudo port -v selfupdate: This action should be used regularly to update the local tree with the global MacPorts ports. The option -v enables verbose which generates verbose messages.

  • port info mypackage: This action is used to get information about a port. (description, license, maintainer, etc.)

  • sudo port install mypackage: This action install the port mypackage.

  • sudo port uninstall mypackage: This action uninstall the port mypackage.

  • port installed: This action displays all ports installed and their versions, variants and activation status. You can also use the -v option to also display the platform and CPU architecture(s) for which the ports were built, and any variants which were explicitly negated.

  • sudo port upgrade mypackage: This action updgrades installed ports and their dependencies when a Portfile in the repository has been updated. To avoid the upgrade of a port’s dependencies, use the option -n.


A Portfile is a TCL script which usually contains simple keyword values and TCL expressions. Each package/port has a corresponding Portfile but it’s only a part of a port description. Feel++ provides some mandatory Portfiles for its compilation which are either not available in MacPorts or are buggy but Feel++ also provides some Portfiles which are already available in MacPorts such as gmsh or petsc. They usually provide either some fixes to ensure Feel++ works properly or new version not yet available in MacPorts. These Portfiles are installed in ports/macosx/macports.


To be able to install Feel++, add the following line in /opt/local/etc/macports/source.conf at the top of the file before any other sources:

file:///<path to feel top directory>/ports/macosx/macports

Once it’s done, type in a command-line:

 $ cd <your path to feel top directory>/ports/macosx/macports
 $ sudo portindex -f

You should have an output like this:

Reading port index in pass:[\(<\)]your path to feel top directorypass:[\(>\)]/ports/macosx/macports
Adding port science/feel++
Adding port science/gmsh
Adding port science/petsc

Total number of ports parsed:   3
Ports successfully parsed:      3
Ports failed:                   0
Up-to-date ports skipped:       0

Your are now able to type

$ sudo port install feel++

It might take some time (possibly an entire day) to compile all the requirements for Feel++ to compile properly. If you have several cores on your MacBook Pro, iMac or MacBook, we suggest that you configure macports to use all or some of them.

To do that uncomment the following line in the file /opt/local/etc/macports/macports.conf

buildmakejobs	0 pass:[\(\#\)] all the cores

At the end of the sudo port install feel++, you have all dependencies installed. To build all the Makefile, \cmake is automatically launched but can have some libraries may not be found but they are not mandatory for build Feel++, only the features related to the missing libraries will be missing.

Missing ports

cmake can build Makefiles even if some packages are missing (latex2html, VTK …​). It’s not necessary to install them but you can complete the installation with MacPorts, cmake will find them by itself once they have been installed.

9. Feel++ and Python

The recommended version of Python is Python3. Python2 should work but it is no more tested.

9.1. Sympy

Sympy (at least 1.1) is used the computer algebra system for symbolic calculus.

If Sympy 1.1 is not available on your system, install pip (a python package manager) and then sympy. It will install sympy locally (in $HOME/.local) in your $HOME.

Install commands for sympy on Ubuntu/Debian
$ sudo apt-get install python3-pip
$ sudo pip3 install sympy
Install commands for sympy on MACOSX/Homebrew
$ brew install python3
$ python3 -m pip install --upgrade sympy
The Feel++ Homebrew formula takes care of installing python3 and sympy.

9.2. Configuring Feel++

Feel++ requires to find a Python 3 interpreter and the associated Python 3 libraries. To help finding Python, you may have to pass some information to cmake or the helper script configure.

With cmake
cmake ... -DPYTHON_EXECUTABLE=/usr/bin/python3 ...
With configure
$ <path to toplevel feelpp dir>/configure --enable-python=/usr/bin/python3
# or
$ <path to toplevel feelpp dir>/configure --cmakeflags=-DPYTHON_EXECUTABLE=/usr/bin/python3


If you want to use Sympy, you have to setup the PYTHONPATH environment variable to point to some Feel++ directories.

export PYTHONPATH=$PYTHONPATH:<top level feelpp source directory>/feel/feelpython

9.3. Building Feel++

Once the steps to install on Linux or MacOS X has been followed, we explain, in this section, how to download and build Feel++ from source.

9.3.1. For the impatient

First retrieve the source

$ git clone

Create a build directory

$ mkdir build
$ cd build

Configure Feel++

$ CXX=clang++ ../feelpp/configure -r

Compile the Feel++ library

$ make feelpp
you can speed up the make process by passing the option -j<N> where N is the number of concurrent make sub-processes. It compiles N files at a time and respect dependencies. For example -j4 compiles 4 C++ files at a time.
Be aware that Feel++ consumes memory. The Feel++ library compile with 2Go of RAM. But to be more comfortable, 4Go or more would be best. The more, the better.

Compile your first Feel++ applications

$ make quickstart

Execute your first Feel++ application in sequential

$ cd quickstart
$ ./feelpp_qs_laplacian_2d --config-file qs_laplacian_2d.cfg

Execute your first Feel++ application using 4 mpi processes

$ mpirun -np 4 feelpp_qs_laplacian_2d --config-file qs_laplacian_2d.cfg

9.3.2. Downloading sources

Using Tarballs

Feel is distributed as tarballs following each major release. The tarballs are available on the link:[Feel Releases] web page.

Download the latest tarball, then uncompress it with:

$ tar -xzf feelpp-X.YY.0.tar.gz
$ cd feelpp-X.YY.0

You can now move to the section Using cmake.

Using Git

Alternatively, you can download the sources of Feel++ directly from the Git repository.

$ git clone

You should read something like

Cloning into 'feelpp'...
remote: Counting objects: 129304, done.
remote: Compressing objects: 100% (18/18), done.
remote: Total 129304 (delta 6), reused 0 (delta 0), pack-reused 129283
Receiving objects: 100% (129304/129304), 150.52 MiB | 1.69 MiB/s, done.
Resolving deltas: 100% (94184/94184), done.
Checking out files: 100% (7237/7237), done.
$ cd feelpp

The first level directory tree is as follows

$ tree -L 1 -d | column
.			├── databases		├── research
├── applications	├── doc			├── testsuite
├── benchmarks		├── feel		└── tools
├── cmake		├── ports		14 directories
├── contrib		├── projects
├── data		├── quickstart

9.3.3. Configuring Feel++

For now on, we assume that clang++ has been installed in /usr/bin. Yor mileage may vary depending on your installation of course.

It is not allowed to build the library in the top source directory.

It is recommended to have a directory (e.g. FEEL) in which you have both the sources and build directories, as follows

$ ls FEEL
feelpp/ # Sources
feel.opt/ # Build directory

feelpp is the top directory where the source have been downloaded, using git or tarballs.

Using cmake

The configuration step with cmake is as follows

$ cd FEEL/feel.opt
$ cmake ../feelpp -DCMAKE_CXX_COMPILER=/usr/bin/clang++-3.6 -DCMAKE_C_COMPILER=/usr/bin/clang-3.6 -DCMAKE_BUILD_TYPE=RelWithDebInfo

CMake supports different build type that you can set with -DCMAKE_BUILD_TYPE (case insensitive) : * None * Debug : typically -g * Release : typically -O3 -DNDEBUG * MinSizeRel : typically -Os * RelWithDebInfo : typically -g -O2 -DNDEBUG

Using configure

Alternatively you can use the configure script which calls cmake. configure --help will provide the following help.

Listing Configure help
 -b, --build                         build type: Debug, Release, RelWithDebInfo
 -d, --debug                         debug mode
-rd, --relwithdebinfo                relwithdebinfo mode
 -r, --release                       release mode
     --std=c++xx                     c++ standard: c++14, c++1z (default: c++14)
     --stdlib=libxx                  c++ standard library: stdc++(GCC), c++(CLANG) (default: stdc++)
     --max-order=x                   maximum polynomial order to instantiate(default: 3)
     --cxxflags                      override cxxflags
     --cmakeflags                    add extra cmake flags
     --prefix=PATH                   define install path
 -v, --verbose                       enable verbose output
 -h, --help                          help page
     --<package>-dir=PACKAGE_PATH    define <package> install directory
     --disable-<package>             disable <package>
     --generator=GENERATOR           cmake generator

We display below a set of possible configurations:

Compile using Release build type, default c compiler and libstdc

Listing compiling using default compilers
$ ../feelpp/configure -r

Compile using Release build type, clang compiler and libstdc

Listing compiling using clang++
$ CXX=clang++ ../feelpp/configure -r

Compile using Debug build type, clang compiler and libc

Listing compiling using clang/libc in Debug mode
CXX=clang++ ../feelpp/configure -d -stdlib=c++

9.3.4. Compiling Feel++

Once cmake or configure have done their work successfully, you are ready to compile Feel++

$ make

You can speed up the compilation process, if you have a multicore processor by specifying the number of parallel jobs make will be allowed to spawn using the -j flag:

Listing build Feel++ library using 4 concurrent jobs
$ make -j4 feelpp
From now on, all commands should be typed in build directory (e.g feel.opt) or its subdirectories.

9.3.5. Running the Feel++ Testsuite

If you encounter issues with Feel++, you can run the testsuite and send the resulting report. Feel++ has more than 300 tests running daily on our servers. Most of the tests are run both in sequential and in parallel.

The testsuite is in the testsuite directory.

$ cd testsuite

The following command will compile 10 tests at a time

$ make -j10
Listing: Running the Feel++ testsuite
$ ctest -j4 -R .

It will run 4 tests at a time thanks to the option -j4.

10. Docker

Docker is the recommended way if you are beginning using Feel++.

This chapter explains step by step how to get the Feel++ Container System(FCS), how to execute a precompiled application, how to parameter and run models.

10.1. Introduction

Container based technologies are revolutionizing development, deployment and execution of softwares. Containers encapsulate a software and allow to run seamlessly on different platforms — clusters, workstations, laptops — The developer doesn’t have to worry about specific environments and users spend less time in configuring and installing the software. Containers appear to be lightweight virtual machines (VMs) — they are started in a fraction of a second — but they, in fact, have important differences.

One of the differences is the isolation process. The VMs share only the hypervisor, the OS and hardware whereas containers may share, between each other, large parts of filesystems rather than having copies. Another difference is that, unlike in VMs, processes in a container are similar to native processes and they do not incur the overhead due to the VM hypervisor. The figure below illustrates these fundamental differences. We see in particular that the applications 2 and 3 are sharing lib 2 without redundancy.

Figure 1. Figure : VMs vs Containers

Docker is a container technology providing:

  1. an engine to start and stop containers,

  2. a user friendly interface from the creation to the distribution of containers and

  3. a hub  — cloud service for container distribution — that provides publicly a huge number of containers to download and avoid duplicating work.

10.2. Installation

This section covers briefly the installation of Docker. It should be a relatively simple smooth process to install Docker.

10.2.1. Channels

Docker offers two channels: the stable and beta channels.

stable channel

is fully baked and tested software providing a reliable platform to work with. Releases are not frequent.

beta channel

offers cutting edge features and experimental versions of the Docker Engine. This is a continuation of the initial Beta program of Docker to experiment with the latest features in development. It incurs far more instabilities than the stable channel but releases are done frequently — possibly several releases per month.

In the latter we shall consider only installing and using the stable channel.

10.2.2. Installing Docker

At the time of writing this section, Docker is available on Linux, Mac and Windows.

Mac and Windows

The support for Mac and Windows as Host OS was recently released and Docker Inc provides installation processes Docker For Mac and Docker for Windows which are the recommended way of installing Docker on these platforms.


Most Linux distributions have their own packages but they tend to lag behind the stable releases of Docker which could be a serious issue considering the development speed of Docker.

To follow Docker releases, it is probably best to use the packages distributed by Docker.

Installing Binaries

The last possibility is to use Docker Binaries to install Docker. This should be used at the last resort if packages are provided neither by your distribution nor by Docker Inc.

Tested with Docker 1.12

At the time of writing this book, the Docker version we used is Docker 1.12. All commands have been tested with this version.

10.2.3. Running without sudo

On Linux, Docker is a priviledged binary, you need to prefix all your commands with sudo, e.g. on Ubuntu. You need first to belong to the docker group with the following command on Ubuntu

$ sudo usermod -aG docker

It creates the docker group if it doesn’t already exist and adds the current user to the docker group. Then you need to log out and log in again. Similar process is available on other distributions. You need also to restart the docker service

$ sudo service docker restart
From now on, we omit the sudo command when using Docker for the sake of brevity.
Adding a user to the docker group has security implications. On a shared machine, you should consider reading the Docker security page.

10.2.4. Checking Docker

We now check your installation by running docker version To make sure everything is installed correctly and working, try running the docker version command. You should see something like the following on Linux or Mac.

Listing : Output of docker version on Linux
> docker version
 Version:      1.12.1
 API version:  1.24
 Go version:   go1.6.3
 Git commit:   23cf638
 Built:        Mon, 10 Oct 2016 21:38:17 +1300
 OS/Arch:      linux/amd64

 Version:      1.12.1
 API version:  1.24
 Go version:   go1.6.3
 Git commit:   23cf638
 Built:        Mon, 10 Oct 2016 21:38:17 +1300
 OS/Arch:      linux/amd64
Listing : Output of docker version on Mac
> docker version
Version: 1.12.6
API version: 1.24
Go version: go1.6.4
Git commit: 78d1802
Built: Wed Jan 11 00:23:16 2017
OS/Arch: darwin/amd64

Version: 1.12.6
API version: 1.24
Go version: go1.6.4
Git commit: 78d1802
Built: Wed Jan 11 00:23:16 2017
OS/Arch: linux/amd64

If so, you are ready for the next step. If instead you get something like

Listing : Bad response output of docker version on Linux
> docker version
 Version:      1.12.1
 API version:  1.24
 Go version:   go1.6.3
 Git commit:   23cf638
 Built:        Mon, 10 Oct 2016 21:38:17 +1300
 OS/Arch:      linux/amd64
Cannot connect to the Docker daemon. Is the docker daemon running on this host?
Listing : Bad response output of docker version on Mac
> docker version
 Version:      1.12.6
 API version:  1.24
 Go version:   go1.6.4
 Git commit:   78d1802
 Built:        Wed Jan 11 00:23:16 2017
 OS/Arch:      darwin/amd64
Error response from daemon: Bad response from Docker engine

then it means that the Docker daemon is not running or that the client cannot access it.

To investigate the problem you can try running the daemon manually — e.g. sudo docker daemon. This should give you some informations of what might have gone wrong with your installation.

11. Feel++ Containers

Feel++ leverages the power of Docker and provides a stack of container images.

11.1. First steps

To test Docker is installed properly, try

$ docker run feelpp/feelpp-env  echo 'Hello World!'

We have called the docker run command which takes care of executing containers. We passed the argument feelpp/feelpp-env which is a Feel++ Ubuntu 16.10 container with the required programming and execution environment for Feel++.

feelpp/ in feelpp/feelpp-env provides the organization name (or namespace) of the image and feelpp-env is the image name. Note also that Docker specifies a more complete name feelpp/feelpp-env:latest including the tag name :latest. We will see later how we defined the latest tag at the Feel++ organization. See Feel++ Container System for more details.

This may take a while depending on your internet connection but eventually you should see something like

Unable to find image 'feelpp/feelpp-env:latest' locally (1)
latest: Pulling from feelpp/feelpp-env
8e21f82d32cf: Pull complete
0a8dee947f9b: Pull complete
Digest: sha256:457539dbd781594eccd4ddf26a7aefdf08a2fff9dbeb1f601a22d9e7e3761fbc
Status: Downloaded newer image for feelpp/feelpp-env:latest
Hello World!
1 The first line tells us that there is no local copy of this Feel++ image. Docker checks automatically online on the Docker Hub if an image is available.

Once the image is downloaded, Docker launches the container and executes the command we provided echo 'Hello World!' from inside the container. The result of the command is showed on the last line of the output log above.

If you run the command again, you won’t see the download part and the command will be executed very fast.

We can ask Docker to give us a shell using the following command

$ docker run -it feelpp/feelpp-env

It provides a shell prompt from inside the container which is very similar to what you obtain when login with ssh on a remote machine. The flags -i and -t tell Docker to provide an interactive session (-i) with a TTY attached (-t).

11.1.1. Feel++ Container System

The Feel++ Container System (FCS) is organized in layers and provides a set of images.

11.1.2. Naming

The naming convention of the FCS allows the user to know where they come from and where they are stored on the Docker Hub. The name of the images is built as follows



  • feelpp/ is the namespace of the image and organization name

  • feelpp-<component> the image name and Feel++ component

  • [:tag] an optional tag for the image, by default set to :latest

Feel++ images(components) are defined as layers in the FCS in the table below.

Table 6. Table of the current components of the FCS
Component Description Built From


Execution and Programming environment



Feel++ libraries and tools



Feel++ base applications



Feel++ toolboxes


| Note: feelpp-env depends on an operating system image <OS>, the recommended and default <OS> is Ubuntu 16.10. In the future, we will build upon the next Ubuntu LTS or Debian Stable releases.

11.1.3. Tags

By default, the :latest tag is assumed in the name of the images, for example when running

$ docker run -it feelpp/feelpp-base

it is in fact feelpp/feelpp-base:latest which is being launched. The following table displays how the different images depend from one another.

Image Built from


Ubuntu 16.10







11.1.4. Host OS

As we said before the default Host OS is Ubuntu 16.10. However Docker shines in continuous integration. It provides a large set of operating system to build upon and allows to check the software in various contexts. The FCS takes advantage of Docker to build feelpp-libs for several operating systems provided by feelpp-env and with different compilers any time a commit in the Feel++ repository is done.

Table 7. Table providing the list of supported Host OS
Operating system version feelpp-env Tags Compilers



ubuntu-16.10, latest

GCC 6.x, Clang 3.9




GCC 6.x, Clang 3.8




GCC 6.x, Clang 3.9,4.0




GCC 6.x, Clang 3.9

If you are interested in testing Feel++ in these systems, you can run these flavors.

11.1.5. Containers


feelpp-env provides the Host OS and Feel++ dependencies.

$ docker run feelpp/feelpp-env

feelpp-libs builds from feelpp-env and provides:

  1. the Feel++ libraries

  2. the Feel++ mesh partitioner application

$ docker run feelpp/feelpp-libs

feelpp-base builds from feelpp-libs and provides two basic applications:

  1. feelpp_qs_laplacian_*: 2D and 3D laplacian problem

  2. feelpp_qs_stokes_*: 2D stokes problem

$ docker run feelpp/feelpp-base

feelpp-toolboxes builds from feelpp-base and provides

$ docker run feelpp/feelpp-toolboxes

11.2. Running Feel++ Applications

To run Feel++ applications in docker, you need first to create a directory where you will store the Feel++ simulation files. For example, type

> mkdir $HOME/feel

and then type the following docker command

> docker run -it -v $HOME/feel:/feel feelpp/feelpp-libs

The previous command will execute the latest feelpp/feelpp-libs docker image in interactive mode in a terminal (-ti) and mount $HOME/feel in the directory /feel of the docker image.

Running the command df inside the Docker container launched by the previous command

feelpp@4e7b485faf8e:~$ df

will get you this kind of output

Filesystem     1K-blocks      Used Available Use% Mounted on
none           982046716 505681144 426457452  55% /
tmpfs          132020292         0 132020292   0% /dev
tmpfs          132020292         0 132020292   0% /sys/fs/cgroup
/dev/sda2      982046716 505681144 426457452  55% /feel
shm                65536         0     65536   0% /dev/shm

You see on the last but one line the directory $HOME/feel mounted on /feel in the Docker image.

Note that mouting a host sub-directory on /feel is mandatory. If you don’t, the Feel++ applications will exit due to lack of permissions. If you prefer running inside the docker environment you can type unset FEELPP_REPOSITORY and then all results from Feel++ applications will be store in $HOME/feel. But then you will have to use `rsync or ssh to copy your results out of the docker image if needed.

Using Feel++

Unresolved directive in README.adoc - include::04-learning/RBM/README.adoc[leveloffset=+1]

Learning Feel++

12. The Laplacian

12.1. Problem statement

We are interested in this section in the conforming finite element approximation of the following problem:

Laplacian problem

Look for \(u\) such that

\[\left\{\begin{split} -\Delta u &= f \text{ in } \Omega\\ u &= g \text{ on } \partial \Omega_D\\ \frac{\partial u}{\partial n} &=h \text{ on } \partial \Omega_N\\ \frac{\partial u}{\partial n} + u &=l \text{ on } \partial \Omega_R \end{split}\right.\]
\(\partial \Omega_D\), \(\partial \Omega_N\) and \(\partial \Omega_R\) can be empty sets. In the case \(\partial \Omega_D =\partial \Omega_R = \emptyset\), then the solution is known up to a constant.

In the implementation presented later, \(\partial \Omega_D =\partial \Omega_N = \partial \Omega_R = \emptyset\), then we set Dirichlet boundary conditions all over the boundary. The problem then reads like a standard laplacian with inhomogeneous Dirichlet boundary conditions:

Laplacian Problem with inhomogeneous Dirichlet conditions

Look for \(u\) such that

Inhomogeneous Dirichlet Laplacian problem
\[-\Delta u = f\ \text{ in } \Omega,\quad u = g \text{ on } \partial \Omega\]

12.2. Variational formulation

We assume that \(f, h, l \in L^2(\Omega)\). The weak formulation of the problem then reads:

Laplacian problem variational formulation

Look for \(u \in H^1_{g,\Gamma_D}(\Omega)\) such that

Variational formulation
\[\displaystyle\int_\Omega \nabla u \cdot \nabla v +\int_{\Gamma_R} u v = \displaystyle \int_\Omega f\ v+ \int_{\Gamma_N} g\ v + \int_{\Gamma_R} l\ v,\quad \forall v \in H^1_{0,\Gamma_D}(\Omega)\]

12.3. Conforming Approximation

We now turn to the finite element approximation using Lagrange finite element. We assume \(\Omega\) to be a segment in 1D, a polygon in 2D or a polyhedron in 3D. We denote \(V_\delta \subset H^1(\Omega)\) an approximation space such that \(V_{g,\delta} \equiv P^k_{c,\delta}\cap H^1_{g,\Gamma_D}(\Omega)\).

The weak formulation reads:

Laplacian problem weak formulation

Look for \(u_\delta \in V_\delta \) such that

\[\displaystyle\int_{\Omega_\delta} \nabla u_{\delta} \cdot \nabla v_\delta +\int_{\Gamma_{R,\delta}} u_\delta\ v_\delta = \displaystyle \int_{\Omega_\delta} f\ v_\delta+ \int_{\Gamma_{N,\delta}} g\ v_\delta + \int_{\Gamma_{R,\delta}} l\ v_\delta,\quad \forall v_\delta \in V_{0,\delta}\]
from now on, we omit \(\delta\) to lighten the notations. Be careful that it appears both the geometrical and approximation level.

12.4. Feel++ Implementation

In Feel++, \(V_{g,\delta}\) is not built but rather \(P^k_{c,\delta}\).

The Dirichlet boundary conditions can be treated using different techniques and we use from now on the elimination technique.

We start with the mesh

    auto mesh = loadMesh(_mesh=new Mesh<Simplex<FEELPP_DIM,1>>);
the keyword auto enables type inference, for more details see Wikipedia C++11 page.

Next the discretization setting by first defining Vh=Pch<k>(mesh) \(\equiv P^k_{c,h}\), then elements of Vh and expressions f, n and g given by command line options or configuration file.

    auto Vh = Pch<2>( mesh );
    auto u = Vh->element("u");
    auto mu = doption(_name="mu");
    auto f = expr( soption(_name="functions.f"), "f" );
    auto r_1 = expr( soption(_name="functions.a"), "a" ); // Robin left hand side expression
    auto r_2 = expr( soption(_name="functions.b"), "b" ); // Robin right hand side expression
    auto n = expr( soption(_name="functions.c"), "c" ); // Neumann expression
    auto g = expr( soption(_name="functions.g"), "g" );
    auto v = Vh->element( g, "g" );

at the following line

    auto v = Vh->element( g, "g" );

v is set to the expression g, which means more precisely that v is the interpolant of g in Vh.

the variational formulation is implemented below, we define the bilinear form a and linear form l and we set strongly the Dirichlet boundary conditions with the keyword on using elimination. If we don’t find Dirichlet, Neumann or Robin in the list of physical markers in the mesh data structure then we impose Dirichlet boundary conditions all over the boundary.

    auto l = form1( _test=Vh );
    l = integrate(_range=elements(mesh),
    l+=integrate(_range=markedfaces(mesh,"Robin"), _expr=r_2*id(v));
    l+=integrate(_range=markedfaces(mesh,"Neumann"), _expr=n*id(v));

    auto a = form2( _trial=Vh, _test=Vh);
    a = integrate(_range=elements(mesh),
                  _expr=mu*gradt(u)*trans(grad(v)) );
    a+=integrate(_range=markedfaces(mesh,"Robin"), _expr=r_1*idt(u)*id(v));
    a+=on(_range=markedfaces(mesh,"Dirichlet"), _rhs=l, _element=u, _expr=g );
    //! if no markers Robin Neumann or Dirichlet are present in the mesh then
    //! impose Dirichlet boundary conditions over the entire boundary
    if ( !mesh->hasAnyMarker({"Robin", "Neumann","Dirichlet"}) )
        a+=on(_range=boundaryfaces(mesh), _rhs=l, _element=u, _expr=g );

    //! solve the linear system, find u s.t. a(u,v)=l(v) for all v
    if ( !boption( "no-solve" ) )

    cout << "||u_h-g||_L2=" << normL2(_range=elements(mesh), _expr=idv(u)-g) << std::endl;

    auto e = exporter( _mesh=mesh );
    e->add( "u", u );
    e->add( "g", v );

    return 0;


We have the following correspondance:

Element sets Domain




\(\partial \Omega\)







next we solve the algebraic problem

Listing: solve algebraic system
    //! solve the linear system, find u s.t. a(u,v)=l(v) for all v
    if ( !boption( "no-solve" ) )

next we compute the \(L^2\) norm of \(u_\delta-g\), it could serve as an \(L^2\) error if \(g\) was manufactured to be the exact solution of the Laplacian problem.

    cout << "||u_h-g||_L2=" << normL2(_range=elements(mesh), _expr=idv(u)-g) << std::endl;

and finally we export the results, by default it is in the ensight gold format and the files can be read with Paraview and Ensight. We save both \(u\) and \(g\).

Listing: export Laplacian results
    auto e = exporter( _mesh=mesh );
    e->add( "u", u );
    e->add( "g", v );

12.5. Testcases

The Feel++ Implementation comes with testcases in 2D and 3D.

12.5.1. circle

circle is a 2D testcase where \(\Omega\) is a disk whose boundary has been split such that \(\partial \Omega=\partial \Omega_D \cup \partial \Omega_N \cup \partial \Omega_R\).

Here are some results we can observe after use the following command

cd Testcases/quickstart/circle
mpirun -np 4 /usr/local/bin/feelpp_qs_laplacian_2d --config-file circle.cfg

This give us some data such as solution of our problem or the mesh used in the application.



Solution \(u_\delta\)


12.5.2. feelpp2d and feelpp3d

This testcase solves the Laplacian problem in \(\Omega\) an quadrangle or hexadra containing the letters of Feel++


After running the following command

cd Testcases/quickstart/feelpp2d
mpirun -np 4 /usr/local/bin/feelpp_qs_laplacian_2d --config-file feelpp2d.cfg

we obtain the result \(u_\delta\) and also the mesh



Solution \(u_\delta\)



We can launch this application with the current line

cd Testcases/quickstart/feelpp3d
mpirun -np 4 /usr/local/bin/feelpp_qs_laplacian_3d --config-file feelpp3d.cfg

When it’s finish, we can extract some informations



Solution \(u_\delta\)


13. Ressources

13.1. Licenses

Copyright © 2010-2017 by Feel++ Consortium
Copyright © 2005-2015 by Université Joseph Fourier (Grenoble, France)
Copyright © 2005-2015 by University of Coimbra (Portugal)
Copyright © 2011-2015 by Université de Strasbourg (France)
Copyright © 2011-2015 by CNRS (France)
Copyright © 2005-2006 by Ecole Polytechnique Fédérale de Lausanne (EPFL, Switzerland)

Free use of this software is granted under the terms of the L License.

See the LICENSE file for details

This book is part of Feel++ and is licensed under cc LGPL a.

13.2. Authors

There are many other contributors.

Feel++ is currently managed by Christophe Prud’homme, Professor in applied mathematic and scientific computing at the University of Strasbourg, France.

13.3. Funding

Feel++ has been funded by various sources and especially

logo anr logo amies logo irmia logo prace

13.3.1. Current funding

ANR projects
  • Contract (2016-2017)

  • Contract (2016-2017)

  • PEPS Holo3

  • PEPS Solodem

  • PEPS NS2++

  • Hifimagnet (2012-2018)

  • 4fastsim (2016-2017)

13.3.2. Past funding

  • HAMM - (Cosinus call - 2010-2014)

  • OPUS - (TLOG call - 2008-2011)

  • Funding for Cemosis

  • RB4FASTSIM - 2010-2014

PRACE projects
  • HP-FEEL++ 2015-2016

  • HP-FEEL++ 2013-2014

  • HP-PDE{1,2} 2012-2014

Rhônes-Alpes region
  • cluster ISLE [fn:2] and the project CHPID (2009-2011)

13.4. Contributors

Feel++ benefits from the many discussions and close research collaborations with the following persons: Mourad Ismail, Zakaria Belhachmi, Silvia Bertoluzza, Micol Pennacchio, Marcela Szopos, Giovanna Guidoboni, Riccardo Sacco, Gonçalo Pena.

Finally Feel++ also benefits from discussions within collaborative projects with many people (in no particular order):

Yannick Hoarau, Philippe Gilotte, Benjamin Surowiec, Yoann Eulalie, Stephie Edwige, Marion Spreng, Benjamin Vanthong, Thomas Lantz, Mamadou Camara, Camille Boulard, Pierre Gerhard, Frédéric Hecht, Michel Fouquembergh, Denis Barbier, Jean-Marc Gratien, Daniele Di Pietro.

13.5. Consortium

Feel++ was initially developed at École Polytechnique Fédérale de Lausanne(Suisse) and is now a joint effort between Université de Strasbourg, Université Grenoble-Alpes, CNRS, LNCMI and Cemosis.

logo cemosis logo uga logo cnrs logo imati logo uds


Unresolved directive in README.adoc - include::GLOSSARY.adoc[]