Call the matrix element from MadGraph

[matthewfeickert]

From discussions with @msneubauer a first step is figuring out what needs to be called to be able to call the matrix element from MadGrpah.

[matthewfeickert]

Given discussions with @Sinclert on what was done for the Madminer physics sub-workflow (scailfin/madminer-workflow-ph) I think the cleanest way forward for now is to just containerize MadGraph5_aMC@NLO for the time being (which I've already done in the past with matthewfeickert/madgraph5-amc-nlo:latest).

Though I see that MadGraph5_aMC@NLO py3.0.2 has been released on 2020-02-17 so it is probably worth following up with Olivier Mattelaer on the state of this (I have an old thread open with the dev team on this already as Question "What is the status of Python 3 migration?").

[matthewfeickert]

I now have a Docker image that has MadGraph_aMC@NLO and Python 3.8 working together: matthewfeickert/madgraph5-amc-nlo:mg5_amc2.7.0-python3

[matthewfeickert]

I'm still trying to get to grips with what MadGraph is actually doing (code wise), but after some skimming the 2011 MadGraph paper https://arxiv.org/abs/1106.0522 I might need to take a look at ALOHA (https://arxiv.org/abs/1108.2041).

[matthewfeickert]

I think I'm starting to learn that maybe the way to think about what MadGraph5_aMC@NLO is, is to think of it as an unpackaged Python library (just zipped up and not pip installable) that instead of providing a public Python API or CLI provides a form of an interactive interpreter through the mg5_aMC script (which is just Python afterall)

$ docker run --rm matthewfeickert/madgraph5-amc-nlo:mg5_amc2.7.0-python3 'file $(command -v mg5_aMC)'
/usr/local/MG5_aMC_v2_7_0_py3/bin/mg5_aMC: a /usr/bin/env python3 script, ASCII text executable

So this means that it should be possible to just accept that it isn't packaged at all and then hack-ily (and horribly) set the PYTHONPATH so that it is known and then just write Python code that uses the mg5_aMC modules.

This sounds super gross and annoying...but it will also work.

$ docker run --rm matthewfeickert/madgraph5-amc-nlo:mg5_amc2.7.0-python3 'PYTHONPATH=${PYTHONPATH}:/usr/local/MG5_aMC_v2_7_0_py3; python -c "import madgraph; print(madgraph)"'
<module 'madgraph' from '/usr/local/MG5_aMC_v2_7_0_py3/madgraph/__init__.py'>

[matthewfeickert]

Some follow up email from Fred Olness has proved especially useful as he's pointed me towards this FAQ question from the Wiki: Can I just calculate the amplitudes for a (sub)process and check them point-by-point in phase space points using MadGraph?

[matthewfeickert]

In terms of figuring out smart way to setup and interact with the MadGraph cards, I might want to also look at the https://gitlab.cern.ch/scipp/mario-mapyde project that @kratsg and Mike Hance are writing.