Make sure the following tools are installed and running:
- MadGraph (we’ve tested our setup with MG5_aMC v2.6.2 and v2.6.5). See https://launchpad.net/mg5amcnlo for installation instructions. Note that MadGraph requires a Fortran compiler as well as Python 2.6 or 2.7. (Note that you can still run most MadMiner analysis steps with Python 3.)
- For the analysis of systematic uncertainties, LHAPDF6 has to be installed with Python support (see also the documentation of MadGraph’s systematics tool).
For the detector simulation part, there are different options. For simple parton-level analyses, we provide a bare-bones option to calculate truth-level observables which do not require any additional packages.
We have also implemented a fast detector simulation based on Delphes with a flexible framework to calculate observables. Using this adds additional requirements:
- Pythia8 and the MG-Pythia interface, installed from within the MadGraph command line interface: execute
<MadGraph5_directory>/bin/mg5_aMC, and then inside the MadGraph interface, run
- Delphes. Again, you can (but this time you don’t have to) install it from the MadGraph command line interface with
(These tools currently have a bug: the MG-Pythia interface and Delphes currently do not keep track of additional weights that are in the LHE file. This is not a big deal, MadMiner now offers an option to extract these weights from the LHE file. Alternatively, there is a unofficial patch for these tools that solves these issues. It is available upon request.)
Finally, Delphes can be replaced with another detector simulation, for instance a full detector simulation based
with Geant4. In this case, the user has to implement code that runs the detector simulation, calculates the observables,
and stores the observables and weights in the HDF5 file. The
LHEProcessor classes might provide
some guidance for this.
We’re currently working on a reference Docker image that has all these dependencies and the needed patches installed.