Start the tutorial of the aMC@NLO part by typing from the shell:
$ ./bin/mg5_aMC
> tutorial aMCatNLO
and follow the instructions in order to familiarize yourself with the syntax and available commands.

Yes, barring a few limitations described below. On top of these, a factor to be taken into account is that of CPU power. For example, W-boson production can be generated quickly on a single computer. W-boson associated with 1 hard jet already takes a significant amount of time, but for W-boson associated with 2 hard jets you would need a computer cluster.

In order to use MadGraph5_aMC@NLO the following third party codes are required:
• To start the MadGraph5_aMC@NLO python shell and to generate processes, Python version 2.6 or later is needed (however, versions 3.x are not supported).

• To compile the MadGraph5_aMC@NLO processes a (fortran) compiler that supports quadruple precision is needed. We recommend GCC version 4.6 or later. To install the latest fortran compiler, we recommend using the Developer Toolset.

• To shower the parton-level events ((a)MC@NLO parton level events are non-physical without showering) PYTHIA or HERWIG is needed. Pythia6 (Q²-ordered) and Herwig6 are shipped with the MadGraph5_aMC@NLO package.
  Herwig++ and Pythia8 must be installed by the user and the path set by typing from within the MadGraph5_aMC@NLO shell
  > set hwpp_path /<PATH_TO_HERWIG++>/
  and
  > set pythia8_path /<PATH_TO_PYTHIA8>/
  respectively.

• Optionally, LHAPDF can be linked. Set the path to lhapdf-config in the input/mg5_configuration.txt file with the syntax
  lhapdf = /<PATH_TO_LHAPDF>/bin/lhapdf-config
  or type, from within the MadGraph5_aMC@NLO python shell,
  > set lhapdf /<PATH_TO_LHAPDF>/bin/lhapdf-config

Dependencies on Scientific Linux CERN 5 and 6

Running on lxplus or desktops at CERN

CERN machines run on SLC6, where the default gfortran and gcc compilers are 4.4. Since MadGraph5_aMC@NLO requires compilers 4.6 or later the following describes how to use these on CERN machines.
• In the file ./input/mg5_configuration.txt, set the fortran_compiler variable as follows:
  fortran_compiler = /afs/cern.ch/sw/lcg/external/gcc/4.7.0/x86_64-slc6-gcc47-opt/bin/gfortran
  
  An alternative and possibly safer option is that of setting, in each shell where the code is run, the shell variable PATH as follows:
  setenv PATH /afs/cern.ch/sw/lcg/external/gcc/4.7.0/x86_64-slc6-gcc47-opt/bin:${PATH}

• gcc/gfortran4.7 will still need the relevant libraries of the 4.7 version; hence, on each shell where the code is run, one must set:
  $ setenv LD_LIBRARY_PATH /afs/cern.ch/sw/lcg/external/gcc/4.7.0/x86_64-slc6-gcc47-opt/lib64:${LD_LIBRARY_PATH}
  
  The machines of the lxbatch system (LSF) do not use the default library path. Therefore one must also set:
  $ setenv LSF_LD_LIBRARY_PATH ${LD_LIBRARY_PATH}

• In some rare cases, even with the above setting of the PATHs, the gfortran library is not correctly found, because the libgfortran.so is corrupted (and you get many "gfortan_read_write" errors during compilation). The best way to circumvent this problem is to force the code to link the gfortran library statically, so that the libgfortran.a is used instead. Please add the line
  LDFLAGS += --static
  somewhere to the Source/make_opts file.

• Since gfortran/gcc4.7 are not the default in lxplus, dependences such as LHAPDF are best reinstalled (locally). Once this is done, the proper paths must be used in MadGraph5_aMC@NLO.

By default the renormalization and factorization scales are equal to
μ_F = μ_R = H_T/2 = ∑_i E_T(i)/2,
where the sum is over all final state particles and E_T(i) is the transverse energy of particle i. When launching the event generation for a process, these scales can by multiplied by a constant factor or set to fixed values by editing the run_card. If another functional form for the scales is desired, the file
<PROCESS>/SubProcesses/setscales.f
must be edited appropriately (after the generation and output of the process, but before the start (launch) of the event generation).

When launching the event generation, simple generation cuts on charged leptons and jets can be set by editing the run_card. If more involved generation cuts are desired (particularly needed in processes involving photons), the file
<PROCESS>/SubProcesses/cuts.f
must be edited appropriately (after the generation and output of the process, but before the start (launch) of the event generation). Note that generation cuts must be quite a bit looser then analysis cuts, see below.

To remove possible double counting between the parton shower and the NLO computation, in the MC@NLO formalism MC-subtraction terms are included in the parton level results. This makes the parton-level results unphysical: they need to be showered before they can be used in a physical analysis. Before generating the events, the user should specify in the run_card to which MC he/she wants to match (MCs currently supported are HERWIG6, HERWIG++/7, PYTHIA6 (Q²-ordered)) and PYTHIA8.

When generating events for HERWIG or PYTHIA with the command
> launch
the parton-level events will be showered automatically, generating a stdHEP (HERWIG6 and PYTHIA6) or HepMC (HERWIG++/7 and PYTHIA8) event file in the same directory as where the parton-level events are stored, i.e. in the
<PROCESS>/Events/<RUN_NAME>/
directory. These showered events can be used directly for analyses, e.g. by using MadAnalysis5.

If the event generation was launched with the commands
> launch -p
or
> launch --parton
the events won't be showered automatically and the parton level events must be showered by the user instead. This can be done by executing the command
$ ./bin/shower <RUN_NAME>
from within the <PROCESS> directory. This command can also be used to shower the same events more than once using, for example, different shower options. Alternatively, the events can be showered from with the python shell. In order to do this, execute the following commands:
$ ./bin/mg5
MG5> launch <PROCESS_DIRECTORY_NAME> -i
> shower <RUN_NAME>

Before showering events with Herwig++/7, MadGraph5_aMC@NLO must be told where to find this MC, and its dependences ThePEG and HepMC. This can be done by editing the file ./Cards/shower_card.dat, and by setting the following variables:
HWPPPATH = <PATH_TO_HW++_INSTALLATION>
THEPEGPATH = <PATH_TO_THEPEG_INSTALLATION>
HEPMCPATH = <PATH_TO_HEPMC_INSTALLATION>
We point out that such paths are typically the absolute addresses of the directories where one finds the subdirectories bin, lib, and include

Instead of writing the showered events in a StdHep of HepMC event file, events can also be analyzed on-the-fly. For this, the user has to write an analysis file, and copy it with all its companion codes (i.e., the relevant source-file dependences) in the directories:
MCatNLO/HWAnalyzer
MCatNLO/PYAnalyzer
MCatNLO/PY8Analyzer
MCatNLO/HWPPAnalyzer
in the case of Herwig6, Pythia6, Pythia8, and Herwig++/7 respectively. The variable ANALYSE in the file ./Cards/shower_card.dat must be set equal to the list of the files that contain the user's analysis and its dependence.

To get the weights to compute the scale dependence and PDF uncertainty (as described in arXiv:1110.4738 [hep-ph]) you have to set the corresponding parameters in the run_card.dat appropriately. These weights are then automatically written in the parton-level event file (the .lhe event file that still needs to be showered to become physical). In the current set-up of automatically writing a showered event file (the stdHEP or HEPMC files), this information is lost, but it is straightforward to construct oneself a HERWIG or PYTHIA analysis that will retain it (examples are available in the MCatNLO>/*Analyzer/ directories).

MadLoop is the tool embedded in MadGraph5 and used by default by MadGraph5_aMC@NLO to generate the one-loop contributions. However, it can also be used independently to check specific phase-space points. For details on the syntax and usage, please check the MadLoop tutorial,
MG5> tutorial MadLoop

Note that both loops interfered with the Born process as well as loop-induced processes can be generated by MadLoop in standalone mode. The latter can be generated in exactly the same way as for processes with an underlying tree-level contribution. However, at this stage it is not possible to perform any phase-space integration with them.

The same limitations as described below for MadGraph5_aMC@NLO processes also apply for MadLoop standalone. It is typically possible to generate a standalone code for the loops of processes including up to 6 color-charged particles in total. The numerical instabilities likely to occur in such complicated processes are seen to be efficiently cured by the (automatic switch to the) internal use of quadruple precision when necessary. For these large multiplicities, MadLoop's code is too slow at run-time for any kind of integration, but its answer is reliable and can therefore be efficiently used as a cross-check for other one-loop providers.

In the same way as LO MadGraph5. Please refer to the relevant MG5 FAQ's entry: FAQs for MadGraph5

Processes with jets at the Born level require generation cuts on these jets. These generation cuts should be significantly looser than the analysis cuts to prevent biases from occuring in physical observables, see also here for an example for p p > e+ ve j [QCD] (registration is needed). It is recommended to generate at least two event files with different generation cuts and check that the final results after the analysis cuts do not depend on the generation cuts.

As far as virtual corrections are concerned, only QCD corrections in the standard model are currently supported. However, by using Feynrules NLO-validated models (see the FeynRules NLO models database) also QCD corrections in simple BSM models are possible.
All loop diagrams that have internal lines that are not coloured are not included, because these corrections are not considered to be pure QCD corrections: in general, these can also be viewed as EW corrections to a QCD process and perturbative expansion in the EW couplings is not supported yet. It is up to the user to decide whether the exclusion of these contributions gives meaningful (and gauge invariant) results or not.

The decay-chain syntax of MadGraph is not supported. Use the MadSpin package to include spin-correlations for decaying particles.

A consistent SU(2)×U(1) gauge invariant treatment of width effects (such as the complex mass scheme) is still under validation when unstable colored particles appear in the loop. Therefore processes that feature resonances charged in QCD are not yet supported, e.g. p p > W+ W- b b~ [QCD] that has top quark resonant contributions.

Processes in which new resonances open up in the real-emission diagrams give unstable results that are usually wrong. For example p p > W+ W- [QCD] in the five-flavor scheme has top quark resonant contributions in the real emission diagrams that the phase-space integration is not able to handle, leading to a non-converging integral. (Note that for this particular case, p p > W+ W- [QCD], the solution is to use the four-flavor scheme with a diagonal CKM matrix to prevent the top quark resonance from appearing.) It is under investigation what the best solution is in the context of NLO event generation and therefore these processes are not yet supported.

Lepton-hadron collisions (DIS-like) are currently not supported. Lepton-lepton (LEP, ILC, CLIC, etc.) and hadron-hadron (Tevatron, LHC, etc.) collisions are fully supported.

Fully inclusive samples with different jet multiplicities can be obtained automatically at LO. At NLO, the FxFx merging proposed in arXiv:1209.6215 [hep-ph] and is semi-automated. Further information can be found here

During the shower phase some parameters are not arbitrary as they have to be identical to those we have assumed in the Monte Carlo subtraction terms. In particular, the following settings MUST NOT BE CHANGED:

• HERWIG6: SOFTME = .FALSE., HARDME = .FALSE., PTMIN = 0.5D0.

• HERWIGPP (use version 2.4.2 or later and 1.6.1 (or later) for ThePeg): ReconstructionOption=General, InitialInitialBoostOption=LongTransBoost

• PYTHIA6Q: MSTJ(43) = 3, MSTJ(47) = 0, MSTJ(48) = 0, MSTJ(50) = 2, MSTP(67) = 2, MSTP(68) = 0, PARP(67) = 1D0, PARP(71) = 1D0.

• PYTHIA6PT: MSTP(67) = 2, MSTP(68) = 0, PARP(67) = 1D0.

• PYTHIA8 (updated on 2023-08-22):
  
  SpaceShower:pTmaxMatch = 1,
  SpaceShower:pTmaxFudge = 1,
  TimeShower:pTmaxMatch = 1,
  TimeShower:pTmaxFudge = 1,
  TimeShower:globalRecoil = on,
  TimeShower:limitPTmaxGlobal = on,
  TimeShower:nMaxGlobalRecoil = 1,
  TimeShower:globalRecoilMode = 2,
  TimeShower:nMaxGlobalBranch = 1,
  TimeShower:weightGluonToQuark=1,
  SpaceShower:MEcorrections = off,

  For processes with decays generated by MadSpin:
  
  TimeShower:MEcorrections = on,
  TimeShower:MEextended = off.

  For all other processes:
  
  TimeShower:MEcorrections = off.

  If in doubt, please see 2308.06389, or contact the authors.