ComPWA is a project to provide a flexible and modular Partial Wave Analysis framework for various use-cases. It was initiated and is developed by the Panda Collaboration (antiProton ANnihilation at DArmstadt) at Fair (Facility for Antiproton and Ion Research). But ComPWA will not just be used for the Panda physics program, but also for various other experiments to provide a commonly used tool which is stable, efficient and provides comparable results. At the moment there are many PWA-tools on the market, most used just for specific experiments and specific physics cases, some experiments even have multiple tools. But why write the same software again and again? E.g. the model describing physical processes should stay the same independent where and how there was a measurement of the process. Using the actual same implementation of the model does not only save a lot of time, it also ensures that two experiments are comparing the same thing. The same argument holds for optimization-routines and estimation-functions. It might even allow combined fitting of different experiments instead of taking the average of the results!

The natural modularization, following the considerations above, would be to separate into experiment specific information, physics (models, formalisms), estimation how good the model fits the data and optimization of free parameters. The first considerations on this where discussed with experts from different experiments and different technologies where discussed and tested. The result of this process is the first requirement document of the new PWA-Framework. This sketch illustrates the modular concept:

• Computational Back-ends:
  • FunctionTree
  • Interface to Tensorflow
• Physics Models:
  • Helicity Formalism
  • Canonical Formalism
  • K-Matrix Formalism
• Data:
  • Input/Output from/to ROOT files
  • Input/Output from/to Ascii files
  • Direct IO via python (see pycompwa)
  • Data Generation
    • ROOT phase space generator
    • EvtGen phase space generator
    • Model-based hit&miss Monte Carlo data generation
    • Model-based importance sampled data generation
• Estimators:
  • Unbinned Maximum Likelihood
• Optimizers:
  • Interface to Minuit2
  • Interface to Geneva
• User Interfaces/Steering:
  • C++
  • Python, see pycompwa

ComPWA offers the python interface pycompwa which gives access to most functionality of ComPWA and provides an expert-system to generate amplitude models.

ComPWA is supposed to run on most modern unix systems (including MacOS). The following packages are mandatory:

• git (optional, for easier updates and if you want to contribute)
• cmake ( > 3.3 )
• gcc (> 5.1) or clang
• Boost, version >= 1.54

For a more feature rich installation, the following packages are highly recommended:

• python3 (requirement of pycompwa)
• ROOT, version 5.34, 6.08
• Minuit2, version 5.34, 6.08
• Geneva, version 1.8.0

In case that some dependencies are not met on your system use your package manager to install them or use the manual procedures described below. For MacOS you can use e.g. MacPorts as package manager. You can also try to use different (newer) versions than the ones stated above, however those are not tested.

In order to install all dependencies it is probably also useful to have a look on the installation instructions file for TravisCI.

A detailed guide can be found below. The installation basically boils down to:

git clone --recurse-submodules https://github.com/ComPWA/ComPWA-legacy <COMPWA_SOURCE_PATH>
cd <COMPWA_SOURCE_PATH>
mkdir build && cd build
cmake ..
cmake --build .

The repository contains a couple of examples.To learn about more detailed features of ComPWA you also might want to have a look on the examples of the pycompwa package.

Source code documentation via Doxygen is provided here. The main branch is automatically built using TravisCI. Probably, it is interesting to check out the log file and the projects TravisCI configuration file travisCI.yml.

• Boost: to install Boost follow these instructions. The --prefix=path/to/installation/prefix option is useful as you can use the same path to point ComPWA to this specific Boost installation.

• ROOT: to install Root follow these instructions.

• Minuit2: is included in most ROOT installations. In case you want to install a stand-alone version follow these instructions. In addition you should use ./configure --prefix=/your/minuit2/installation/path followed by make install to specify an installation directory which ComPWA later needs to find the library and to install all needed files to this location.

• Geneva: to install Geneva follow these instructions. The DCMAKE_INSTALL_PREFIX="/where/the/library/goes" option is useful as you can use the same path to point ComPWA to this specific Geneva installation:

  cd GENEVA_SOURCE
  mkdir -p build/install
  cd build
  cmake ../ -DCMAKE_INSTALL_PREFIX=./install
  cmake --build .
  make install
  cp install/CMakeModules/FindGeneva.cmake YOUR_COMPWA_PATH/cmake/Modules/
  export GENEVA_ROOT=YOUR_GENEVA_PATH/build/install

  • Note for Fedora 25: The Geneva tests are build by default but might have trouble finding the boost test libraries of the Fedora boost package. A workaround is to disable them within YOUR_GENEVA_PATH/CMakeModules/CommonGenevaBuild.cmake, line 55 (replace the line with SET( GENEVA_BUILD_TESTS FALSE ).
  • Alternatively you can follow the instructions from the Geneva manual.

Get the most recent version:

git clone --recurse-submodules git@github.com:ComPWA/ComPWA-legacy <COMPWA_SOURCE_PATH>

This will clone the repository and its sub-modules to the sub-folder <COMPWA_SOURCE_PATH> within the current directory. For multi-threading ComPWA uses the parallel stl algorithms of c++17. Unfortunately the current compilers do not have any implementations for this. Here ComPWA currently relies on TBB and parallelstl, which are included in ComPWA as git submodules.

ComPWA uses cmake as build system. The usual steps to build all libraries and the test executable are the following:

• Create and enter a build folder (preferably not the ComPWA source folder)

  mkdir build
  cd build

• Set your compiler if you do not use the system default compiler

  export CC=<path_to_your_compiler>
  export CXX=<path_to_your_compiler>

• Build the project. You may leave the DCMAKE_INSTALL_PREFIX empty, but then ComPWA will be installed system wide.

  cmake .. -DCMAKE_INSTALL_PREFIX=<COMPWA_INSTALL_PATH>
  cmake --build .
  make install      # optional
  ctest -C debug    # optional: run test suite

• You might want to create a pre-configured project for an IDE (e.g. vscode, eclipse, Xcode) via:

  cmake -G"Eclipse CDT4 - Unix Makefiles" ../<COMPWA_SOURCE_PATH>

A Dockerfile for ComPWA is provided. You can use it to build an docker image to run ComPWA. Using such an image ComPWA should run on all systems that are supported by docker including several (commercial) cloud computing services. If you are new to docker you can have a look on this tutorial.

Mogon2 is the supercomputer of the Mainz University. If you work on it, you can fulfil the ComPWA installation requirements by loading a series of modules:

module load devel/CMake/3.9.5
module load toolchain/foss/2017a
module load devel/Boost/1.65.1-foss-2017a
module load ROOT/v6.12-foss-2017a-python3
export CC=/cluster/easybuild/broadwell/software/compiler/GCCcore/6.3.0/bin/gcc
export CXX=/cluster/easybuild/broadwell/software/compiler/GCCcore/6.3.0/bin/g++

Now follow: Building ComPWA.