POWHEG event generator for QCD and EW corrections. A detailed description of the algorithms and methods implemented in this code, as well as, results obtained by using this code can be found in DOI: 10.18154/RWTH-2017-04899. The results are also published in DOI: 10.1007/JHEP05(2017)090.
The program requires an MPI implementation. It is only tested with openmpi.
LHAPDF 6 is used for pdf evaluation. There is also an interface to
LHAPDF 5 available but it is not tested extensively. lhapdf-config is
used to find your LHAPDF installation. Another requirement is the boost C++
library.
requirement list:
- MPI (openmpi)
- LHAPDF
- Collier (downloaded during compilation)
- boost
The source code consists of five parts:
- mypowheg contains a FKS and POWHEG implemenation. The code is in the src directory.
- vegas contains a VEGAS implementation written in C. It uses MPI to parallelize the integration.
- powheggen is a POWHEG event generator. It uses vegas and mypowheg.
- external contains external code for matrix elements etc.
- processes contains POWHEG event generators for processes. Each individual process uses powheggen.
Currently three processes are implemented:
- w: charged current Drell-Yan: p p -> W+ -> mu+ nu
- z: neutral current Drell-Yan: p p -> Z/a -> mu+ mu-
- wj: p p -> W+ j -> mu+ nu j (under development, do not use!)
Earlier versions of w and z were used to make the plots in arXiv:1612.04292.
All files with a name ending with "_test" contain unit tests implemented with googletest.
CMake is used to generate build files. To create build files, you have to create a separate build directory. Run
mkdir build
cd build
cmake PATH_TO_SOURCE
By default the code is compiled optimized and with debugging symbols.
If you use the release mode in cmake (-DCMAKE_BUILD_TYPE=Release), the most
aggressive optimizations are used and the program is tuned to the CPU
architecture. This mode is not tested. The debug mode in
cmake (-DCMAKE_BUILD_TYPE=Debug) disables all optimizations.
After calling cmake the build directory contains all required build files. We assume that Makefiles are used. Other options can be chosen with the -G option of cmake. You can now use
make Collier
make
to compile the code. If you run make test all tests are called. The different
POWHEG generator executables are located in the processes directory. You can run them with
cd processes/w
mpirun -np 2 ./w
An example config file config.txt is copied to each process directory.
Each process generates the files vegas_grid.dat and run.dat. These files
contain the vegas state after the Btilde integration.
If the program is called with the -load flag, vegas_grid.dat and run.dat
is loaded and the program can generate events.
Furthermore, the file norm.txt, maxima.txt, and vegas_max.txt are created.
norm.txt contains cumulative histograms for the upperbounding function norms.
They can be plotted with gnuplot, e.g.
gnuplot> set terminal pdfcairo
gnuplot> set output 'norm.pdf'
gnuplot> load 'norm.txt'maxima.txt contains a histogram for the Maximum found during the B~ integration.
It can be used to modify the maximum in run.dat in order to make the event generation
more efficient. vegas_max.txt contains the maxima found during the vegas integration all
VEGAS bins. The idea is to use those maxima for event generation to make it more efficient.
This feature is not fully implemented so far. One would determine the current VEGAS bin
in every dimension and the maximum. The maximum of all maxima could then be used as
unweighting maximum.
Every process expects a configuration file config.txt in its working directory.
The configuration files contains global options for the event generator and can
also contain process specific options (e.g. cuts, scales etc.).
The config format is inspired by the ini-format.
The global flags are listed below:
[recombination]
dR = 0.00000001 # recombination parameter for NLO calculations. Should be set
# tiny values for event generation.
[config]
SqrtS = 13000.0 # center of mass energy in GeV
Verbose = 0 # either 0 or 1 (optional, default: 0)
CalculateXSec = 0 # calculate the cross section separately for born, virtual,
# real etc. This calculation is only needed for a precise
# cross section determination. (optional, default: 0)
# grid setup for Btilde integration
[integration.setup]
iterations = 5 # number of VEGAS iterations for the grid setup
nevents = 1000000 # number of points in each VEGAS iteration
IgnoreVirtual = 0 # Ignore virtual corrections during grid setup
# (optional, default:0)
# Btilde integration
[integration]
iterations = 5 # number of VEGAS iterations
nevents = 1000000 # number of points in each VEGAS iteration
# grid setup for cross section calculation. Only used if config.CalculateXSec = 1.
[xsec.setup]
iterations = 5 # number of VEGAS iterations
nevents = 1000000 # number of points in each VEGAS iteration
# cross section calculation. Only used if config.CalculateXSec = 1.
[xsec]
iterations = 5 # number of VEGAS iterations
nevents = 1000000 # number of points in each VEGAS iteration
# random seeds for the VEGAS integrations
[random]
seed1 = 1 # random seed for grid setups (0 = random seed)
seed2 = 2 # random seed for the integration (0 = random seed)
# folding of real part in Btilde (reduces number of negative events, increases
# the run time).
[RadiationSampling]
RealXi = 1 # folding of xi
RealY = 1 # folding of y
RealPhi = 1 # folding of phi
RemnantXi = 1 # folding of pdf remnant xi
[RadiationType]
QCD = 1 # enable/disable QCD corrections
EW = 1 # enable/disable EW corrections
RadiationQCD = 1 # generate QCD radiation (optional, default: 1)
RadiationEW = 1 # generate EW radiation (optional, default: 1)
SplitME = 0 # NC Drell-Yan (z specific): split matrix element in ISR/FSR
ModS = 0 # use modified FKS S functions. (0: default,
# 1: Breit-Wigner improved, 2: Charge & Breit-Wigner improved)
NoInterference = 0 # NC Drell-Yan (z specific): disable interference terms
QCDPDFScheme = MSbar # PDF renormalization scheme
EWPDFScheme = DIS # PDF renormalization scheme
# Approximations = NoPhotonRadiation # different approximations. deprecated!
[EventGeneration]
GenerateEvents = 1 # enable POWHEG event generation (optional, default: 1)
nevents = 100000 # total number of events
nperiteration = 10000 # number of events per iteration
seed = 0 # random seed (0 = random seed)
NegativeEvents = 0 # use negative events or discard them. (optional, default: 0)
BornUnweighting = 0 # born unweighting. Do not use! (optional, default: 0)
BornOnly = 0 # use born instead of B~ (optional, default: 0)
[PDF]
LHAID = 244600 # select pdfs
[cuts]
CutReal = 0 # cut on real phase space. Enables real NLO calculation.
# Do not use for event generation. (optional, default:0)
[Unweighting]
GuessVirtual = 1 # guess virtual in unweighting, only calculate if necessary
MaxMultiple = 3 # write events with Btilde > Max multiple times to event file
[Radiation]
kT2min = 0.8 # minimal radiation kT2It is possible to generate fixed-order NLO distribution with the code. You have to
enable the option CutReal (see above) and you have to implement a process-specific
histogramming class.
The histogramming class has to inherit from the Histograms class in src/process/histograms.h.
The only method that has to be reimplemented is the Fill() function.
A simple example is
class MyHistograms : public Histograms {
public:
explicit MyHistograms(int n) : Histograms(n) {}
virtual ~MyHistograms() {}
protected:
virtual bool Fill(Phasespace::Phasespace const *const ps,
double wgt) override {
const auto &momenta = ps->Momenta;
// add to histograms
GetHist(0)->AddValue(momenta[4].PT(), wgt);
return true;
}
};An instance of MyHistograms has to be used as hists variable in UserProcess::Data, i.e.
// use one histogram
hists = new MyHistograms(1);
hists->Init1D(0, 1000, 0.0, 1000.0);
hists->SetName(0, "some description");The histograms are written to histogram_%d.hist files, where %d is the number of the histogram.
If CutReal is not set, only a cut on the underlying born phase space is performed.
Doxygen can be used to generate code documentation. Go to the doc directory and run
doxygen
to generate html/index.html.