DYTURBO - fast Drell-Yan Monte Carlo and quadrature integrator

DYTURBO is a fast Drell-Yan integrator, based on the DYRES and DYNNLO programs. Numerical integration is performed using both the Vegas Monte Carlo integration and quadrature rules.

How to setup USER

1. ask for guest rights and get tarbal from webpage

    https://gitlab.cern.ch/DYdevel/DYTURBO/wikis/home
    tar xzvf dyturbo-VERSION.tar.gz
    cd dyturbo-VERSION
   

2. setup and compile

    (optional PATH=<lhapdfdir>:$PATH if you want to use your version of lhapdf previously installed in <lhapdfdir>)
    (optional setup root to enable root output)
    ./configure [--enable-root]
    make
    make install
   

3. have fun

    ./bin/dyturbo input/test.in
   

How to setup DEVEL

1. ask for developer access and checkout the repository

git clone ssh://[email protected]:7999/DYdevel/DYTURBO.git
git co devel

2. setup/compile and test

autoreconf -i
(optional PATH=<lhapdfdir>:$PATH if you want to use your version of lhapdf previously installed in <lhapdfdir>)
./configure --enable-test --enable-Ofast [--enable-root]
make && make install && make check

3. branch your repository add your code and dont forget to test ;)

   make check
   

Description of project

• autotools are used for building project

• on first configuration LHAPDF and Cuba will be downloaded and installed (if not found already)

• additional configuration options:

  • --enable-Ofast Use unsafe compiler optimization flags -Ofast.
  • --enable-root Use root for histograming. (default=no)

• debug options:

  • --enable-debug Add debug flags to include gdb debug symbols.
  • --enable-trapFPE Stop of floating point errors.
  • --enable-checkBounds add -fbounds-check flag for compilation.

Directory structure

• cernlib contains code from CERNLIB libraries
• dynnlo contains code from DYNNLO
• dyres contains code from DYRES
• mcfm contains code from MCFM
• src contains DYTURBO C++ code which is steering the calculations and running the Fortran procedures
• scripts contains scripts for plotting and submitting to
• input files can be found in input folder (where else :) )

Customization

• user customization can be done in folder user
• three files can be change to customize output of the calculation:
  • user/user_kinem.cxx for defining new kinematic observable
  • user/user_book.cxx for booking histograms
  • user/user_cuts.cxx for define user cuts

Merging results

1. Merging of jobs

   • we recommend to run DYTURBO per each term (RES,CT,FO,REAL,VIRT) separately

   • to obtain correct normalization from run of several jobs with different random seeds, please, use our merger

     ./bin/merger -X merged_file.root result_file_1.root result_file_2.root ...

   • Note: in case of real term for NNLO prediction it is better to use median results (object with suffix median in output file of merger program). It will remove ouliers from your distribution, which are caused by color dipole cancellations

2. Merging of terms
   • if you want to obtain finite order prediction you should sum up counter term you should sum up fixed order (i.e. FO for NLO and REAL+VIRT for NNLO) to counter term (simply with hadd)
   • to obtain final prediction you need to finite order to resummation part (simply with hadd)

Merge program description

• To print basic usage just run with help argument

    [user@host DYTURBO]$ ./bin/merger -h
    usage: -h [-h] [-v] [-x]  <output> <input list>
     Please keep separated switches!!! Its on my todolist! 
        -x    Normalize histograms to Xsection. 
        -v    Increase verbosity (very chatty). 
        -h    Print this help message. 
  
     For more info read README.md
     ./bin/merger --help
     usage: --help [-h] [-v] [-x]  <output> <input list>
      Please keep separated switches!!! Its on my todolist! 
         -x    Normalize histograms to Xsection. 
         -v    Increase verbosity (very chatty). 
         -2d   Make 2d projections and outliers for 2D. 
         -h    Print this help message. 
  
     For more info read README.md
  

Detailed description

• basic naming convention: object could be TH1x, TH2x, TProfile, TProfile2D. When I say average or median I mean per each bin.

• merger program produces several stuff on one merge:

  • for TH1x: average (no suffix), median (suffix _median)

  • for TH1x with name h_qt and pt: it creates also rebinned version (suffix _rebin) with Zpt measurement binning and their mean and median version

  • for TH2x: mean and median same as TH1x + creates projection to each axis (_px, _py) and their mean and median version

  • for TProfiles: average, i.e. TProfile::Add(o,1/N) of all object (no suffix) and outlier removal (suffix "outlier"). More details bellow.

  • for TProfiles2D: average + my plan is to add projections so we can fill only 2D during dyturbo run.

• There is option -X, which turns on "Scaling to final cross section": At the end of dyturbo I am saving result of integration into extra 2D histogram per each term and qt-y bin. This information is then used to correct integral of TH1x, TH2x.

Outlier removal for profiles

• Right now, outliers are removed for TProfiles and preparing also for 2D (for other objects there is median version).

• each bin of profile consist of two terms: nominator sum(w*y) ( the first moment y of cros section) and denominator sum(w) (cross section), where y is studied observable, w is weight of event and we are suming per all events.

• in first step we create denominator median sumhat(w_i) = median ( sum(w_ij) from all merged files j, per each bin i

• we calculate chi-square from denominator

                (sum(w_ij)-sumhat(w_i))**2
      chi2_ij = --------------------------
                 (delta (sum(w_ij)) )**2
  

• then we decide about outlier by Prob(chi2_ij,1) > 7 sigma throw away

• from the rest we get files with indexes l

• final outlier profile values terms are:

      sum(w_i*y_i) = sum(w_il*y_il) / N_i
      sum(w_i) = sum(w_il) / N_i
  

  where N_i is number of jobs contributing to bin i.

• the TProfile is then created with hard-set of above values

Submission script

Basic script for submission is submit_DYTURBO.sh. It could be customize to submit jobs to any batch system. If you need help, please run:

 ./scripts/submit_DYTURBO.sh --help

Or contact me [[email protected]]. An example for grid is here

./scripts/submit_DYTURBO.sh --grid --proc wp,wm,z0 --collider lhc7 --order 2 --term RES,CT,REAL,VIRT --pdfvar all

Dont forget to configure and compile before submiting

./configure --enable-Ofast --enable-root
make -j
make install

There are several grid related scripts in scripts folder, which you can edit for your need

• run_prun.sh: function for submit jobs with prun command
• compile_grid.sh: compilation on grid side
• run_grid.sh: running on grid side

LICENCE AND REFERENCES

This project consist of several packages and each has its own licence.

@todo We did not decided yet what licence should be there for our contributions.

Code used: * Gonsalves, Pawlowski, Wai calculation of dsigma/dqt Phys. Rev. D40, 2245 (1989) Code from http://www.physics.buffalo.edu/gonsalves/ewbqt/public.tar.gz

Papers used: * [Collins-Soper-Sterman]