MuScleFit.cc

Go to the documentation of this file.

//  \class MuScleFit
//  Fitter of momentum scale and resolution from resonance decays to muon track pairs
//
//  $Date: 2012/12/20 16:09:21 $
//  $Revision: 1.112 $
//  \author R. Bellan, C.Mariotti, S.Bolognesi - INFN Torino / T.Dorigo, M.De Mattia - INFN Padova
//
//  Recent additions:
//  - several parameters allow a more flexible use, tests, and control handles
//    for the likelihood. In particular, a set of integers controls the order
//    with which parameters are released; another controls which parameters are
//    fixed. A function allows to smear momenta and angles of muons from the
//    resonance before any correction, using a set of random numbers generated
//    once and for all in the constructor (this way the smearing remains the same
//    for any given muon no matter how many times one loops and what corrections
//    one applies).
//    For a correct use of these flags, please see the function minimizeLikelihood() in
//    MuScleFitUtils.cc
//  - the fit now allows to extract resolution functions simultaneously with the
//    momentum scale. So far a simple parametrization
//    of muon momentum resolution and angle resolution has been implemented, but
//    extensions are straightforward.
//  - It is however advisable to fit separately resolution and scale. The suggested
//    course of action is:
//    1) fit the scale with a simple parametrization
//    2) check results, fit with more complicated forms
//    3) verify which is a sufficiently accurate description of the data
//    4) fix scale parameters and fit for the resolution
//    5) go back to fitting the scale with resolution parameters fixed to fitted values
//  - Also note that resolution fits may fail to converge due to instability of the
//    probability distribution to the limit of large widths. Improvements here are
//    advisable.
//  - The treatment of signal windows in the Y region
//    has to be refined because of overlaps. More work is needed here, assigning a different
//    weight to different hypothesis of the resonance producing a given mass, if there are
//    multiple candidates.
//  - Also, larger windows are to be allowed for fits to SA muons.
//  - File Probs_1000.root contains the probability distribution of lorentzians convoluted
//    with gaussian smearing functions, for the six resonances. A 1000x1000 grid
//    in mass,sigma has been computed (using root macro Probs.C).
//    A wider interval of masses for each resonance should be computed, to be used for standalone muons
//
//
//  Notes on additions, TD 31/3/08
//
//  - background model: at least a couple of different models, with two parameters,
//    should be included in the fitting procedure such that the function massprob(),
//    which produces the probability in the likelihood computation, incorporates the
//    probability that the event is from background. That is, if the fitting function
//    knows the shape of the mass spectrum ( say, a falling exponential plus a gaussian
//    signal) it becomes possible to fit the scale together with the background shape
//    and normalization parameters. Of course, one should do one thing at a time: first
//    a scale fit, then a shape fit with scale parameters fixed, and then a combined
//    fit. Convergence issues should be handled case by case.
//  - The correct implementation of the above idea requires a reorganization of pass
//    parameters (in the cfg) and fit parameters. The user has to be able to smear,
//    bias, fix parameters, choose scale fitting functions, resolution fitting functions,
//    and background functions. It should be possible to separate the fit functions from
//    the biasing ones, which would allow a more thorough testing.
//  - all the above can be obtained by making the .cfg instructions heavier. Since this
//    is a routine operated by experts only, it is a sensible choice.
//  - One should thus envision the following:
//      1) a set of parameters controlling the biasing function: parBias()
//      2) a set of parameters controlling the smearing function: parSmear()
//      3) a set of parameters to define resolution modeling and initial values: parResol()
//      3b) parResol() gets fix and order bits by parResolFix() and parResolOrder()
//      4) a set of parameters to define scale modeling and initial values: parScale()
//      4b) parScale() gets fix and order bits by parScaleFix() and parScaleOrder()
//      5) a set of parameters controlling the background shape and normalization: parNorm()
//      5b) parNorm() gets fix and order bits by parNormFix() and parNormOrder()
//    The likelihood parameters then become a vector which is dynamically composed of
//    sets 3), 4), and 5): parval() = parResol()+parScale()+parNorm()
//  - In order to study better the likelihood behavior it would be advisable to introduce
//    some histogram filling on the last iteration of the likelihood. It is not clear
//    how best to achieve that: probably the simplest way is to make a histogram filling
//    function run just after the likelihood computation, such that the best value of the
//    fit parameters is used.
//  - The muon pair which we call our resonance must be chosen in a way which does not
//    bias our likelihood: we cannot just choose the pair closest to a resonance.
//
//
//    Notes on additions, T.Dorigo 22/12/2008
//    ---------------------------------------
//
//  - File Probs_new_1000_CTEQ.root now contains a set of 24 additional two-dim histograms,
//    defining the probability distribution of Z boson decays as a function of measured mass
//    and expected sigma in 24 different bins of Z rapidity, extracted from CTEQ 6 PDF (at
//    Leading Order) from the convolution in the factorization integral. See programs CTEQ.cpp
//    and Fits.C.
//  - The probability for Z boson events now thus depends on the measured rapidity of the dimuon
//    system. All functions in file MuScleFitUtils.cc have been suitably changed.
//
// ----------------------------------------------------------------------------------
//    Modifications by M. De Mattia 13/3/2009
//    ---------------------------------------
//  - The histograms map was moved to a base class (MuScleFitBase) from which this one inherits.
//
//    Modifications by M. De Mattia 20/7/2009
//    ---------------------------------------
//  - Reworked background fit based on ranges. See comments in the code for more details.
// ---------------------------------------------------------------------------------------------
// Base Class Headers
// ------------------
#include "FWCore/Framework/interface/EDLooper.h"
#include "FWCore/Framework/interface/LooperFactory.h"
#include "FWCore/Utilities/interface/InputTag.h"
#include "FWCore/Framework/interface/Frameworkfwd.h"
#include "FWCore/Framework/interface/Event.h"
#include "FWCore/ParameterSet/interface/ParameterSet.h"

#include <CLHEP/Vector/LorentzVector.h>
#include <vector>

#include "FWCore/Framework/interface/EventSetup.h"
#include "FWCore/Framework/interface/ESHandle.h"
#include "FWCore/ServiceRegistry/interface/Service.h"

#include "MuScleFitBase.h"
#include "Histograms.h"
#include "MuScleFitPlotter.h"
#include "MuonAnalysis/MomentumScaleCalibration/interface/Functions.h"
#include "MuonAnalysis/MomentumScaleCalibration/interface/RootTreeHandler.h"
#include "MuScleFitMuonSelector.h"

#include "DataFormats/TrackReco/interface/Track.h"
#include "DataFormats/MuonReco/interface/Muon.h"
#include "DataFormats/MuonReco/interface/CaloMuon.h"
#include "DataFormats/MuonReco/interface/MuonFwd.h"

#include "RecoMuon/TrackingTools/interface/MuonPatternRecoDumper.h"
#include "RecoMuon/TrackingTools/interface/MuonServiceProxy.h"

#include "DataFormats/PatCandidates/interface/Muon.h"
#include "DataFormats/PatCandidates/interface/CompositeCandidate.h"
#include "DataFormats/Candidate/interface/LeafCandidate.h"
#include "DataFormats/Common/interface/TriggerResults.h"
#include "FWCore/Common/interface/TriggerNames.h"

#include "HLTrigger/HLTcore/interface/HLTConfigProvider.h"

#include "HepPDT/defs.h"
#include "HepPDT/TableBuilder.hh"
#include "HepPDT/ParticleDataTable.hh"

#include "HepMC/GenParticle.h"
#include "HepMC/GenEvent.h"

#include "SimGeneral/HepPDTRecord/interface/ParticleDataTable.h"
#include "SimDataFormats/Track/interface/SimTrackContainer.h"
#include "SimDataFormats/Vertex/interface/SimVertexContainer.h"
#include "SimDataFormats/GeneratorProducts/interface/HepMCProduct.h"

#include "TFile.h"
#include "TTree.h"
#include "TMinuit.h"


// To use callgrind for code profiling uncomment also the following define.
// #define USE_CALLGRIND

#ifdef USE_CALLGRIND
#include "valgrind/callgrind.h"
#endif


// To read likelihood distributions from the database.
    //#include "CondFormats/RecoMuonObjects/interface/MuScleFitLikelihoodPdf.h"
    //#include "CondFormats/DataRecord/interface/MuScleFitLikelihoodPdfRcd.h"

namespace edm {
  class ParameterSet;
  class Event;
  class EventSetup;
}


class MuScleFit: public edm::EDLooper, MuScleFitBase
{
 public:
  // Constructor
  // -----------
  MuScleFit( const edm::ParameterSet& pset );

  // Destructor
  // ----------
  virtual ~MuScleFit();

  // Operations
  // ----------
  void beginOfJobInConstructor();
  // void beginOfJob( const edm::EventSetup& eventSetup );
  // virtual void beginOfJob();
  virtual void endOfJob();

  virtual void startingNewLoop( unsigned int iLoop );

  virtual edm::EDLooper::Status endOfLoop( const edm::EventSetup& eventSetup, unsigned int iLoop );
  virtual void endOfFastLoop( const unsigned int iLoop );

  virtual edm::EDLooper::Status duringLoop( const edm::Event & event, const edm::EventSetup& eventSetup );
  virtual void duringFastLoop();

  template<typename T>
  std::vector<reco::LeafCandidate> fillMuonCollection( const std::vector<T>& tracks );
 private:

 protected:
  void selectMuons(const edm::Event & event);
  void selectMuons(const int maxEvents, const TString & treeFileName);

  template<typename T>
  void takeSelectedMuonType(const T & muon, std::vector<reco::Track> & tracks);
  bool selGlobalMuon(const pat::Muon* aMuon);
  bool selTrackerMuon(const pat::Muon* aMuon);  

  bool checkDeltaR( reco::Particle::LorentzVector& genMu, reco::Particle::LorentzVector& recMu );
  void fillComparisonHistograms( const reco::Particle::LorentzVector & genMu, const reco::Particle::LorentzVector & recoMu, const std::string & inputName, const int charge );

  void applySmearing( reco::Particle::LorentzVector & mu );
  void applyBias( reco::Particle::LorentzVector & mu, const int charge );

  void checkParameters();

  MuonServiceProxy *theService;

  // Counters
  // --------
  int numberOfSimTracks;
  int numberOfSimMuons;
  int numberOfSimVertices;
  int numberOfEwkZ;

  bool ifHepMC;
  bool ifGenPart;

  // Constants
  // ---------
  double minResMass_hwindow[6];
  double maxResMass_hwindow[6];

  // Total number of loops
  // ---------------------
  unsigned int maxLoopNumber;
  unsigned int loopCounter;

  bool fastLoop;

  MuScleFitPlotter *plotter;

  // The reconstructed muon 4-momenta to be put in the tree
  // ------------------------------------------------------
  reco::Particle::LorentzVector recMu1, recMu2;
  int iev;
  int totalEvents_;

  bool compareToSimTracks_;
  edm::InputTag simTracksCollection_;
  bool PATmuons_;
  std::string genParticlesName_;

  // Input Root Tree file name. If empty events are read from the edm root file.
  std::string inputRootTreeFileName_;
  // Output Root Tree file name. If not empty events are dumped to this file at the end of the last iteration.
  std::string outputRootTreeFileName_;
  // Maximum number of events from root tree. It works in the same way as the maxEvents to configure a input source.
  int maxEventsFromRootTree_;

  std::string triggerResultsLabel_;
  std::string triggerResultsProcess_;
  std::vector<std::string> triggerPath_;
  bool negateTrigger_;
  bool saveAllToTree_;

  std::auto_ptr<MuScleFitMuonSelector> muonSelector_;
};

template<typename T>
std::vector<reco::LeafCandidate> MuScleFit::fillMuonCollection( const std::vector<T>& tracks )
{
  std::vector<reco::LeafCandidate> muons;
  typename std::vector<T>::const_iterator track;
  for( track = tracks.begin(); track != tracks.end(); ++track ) {
    reco::Particle::LorentzVector mu;
    mu = reco::Particle::LorentzVector(track->px(),track->py(),track->pz(),
                                       sqrt(track->p()*track->p() + MuScleFitUtils::mMu2));
    // Apply smearing if needed, and then bias
    // ---------------------------------------
    MuScleFitUtils::goodmuon++;
    if (debug_>0) 
      std::cout <<std::setprecision(9)<< "Muon #" << MuScleFitUtils::goodmuon
                       << ": initial value   Pt = " << mu.Pt() << std::endl;

    applySmearing(mu);
    applyBias(mu, track->charge());

    reco::LeafCandidate muon(track->charge(),mu);
    // Store modified muon
    // -------------------
    muons.push_back (muon);
  }
  return muons;
}

template<typename T>
void MuScleFit::takeSelectedMuonType(const T & muon, std::vector<reco::Track> & tracks)
{
  // std::cout<<"muon "<<muon->isGlobalMuon()<<muon->isStandAloneMuon()<<muon->isTrackerMuon()<<std::endl;
  //NNBB: one muon can be of many kinds at once but with the theMuonType_ we are sure
  // to avoid double counting of the same muon
  if(muon->isGlobalMuon() && theMuonType_==1)
    tracks.push_back(*(muon->globalTrack()));
  else if(muon->isStandAloneMuon() && theMuonType_==2)
    tracks.push_back(*(muon->outerTrack()));
  else if(muon->isTrackerMuon() && theMuonType_==3)
    tracks.push_back(*(muon->innerTrack()));

  else if( theMuonType_ == 10 && !(muon->isStandAloneMuon()) ) //particular case!!
    tracks.push_back(*(muon->innerTrack()));
  else if( theMuonType_ == 11 && muon->isGlobalMuon() )
    tracks.push_back(*(muon->innerTrack()));
  else if( theMuonType_ == 13 && muon->isTrackerMuon() )
    tracks.push_back(*(muon->innerTrack()));
}


// Constructor
// -----------
MuScleFit::MuScleFit( const edm::ParameterSet& pset ) :
  MuScleFitBase( pset ),
  totalEvents_(0)
{
  MuScleFitUtils::debug = debug_;
  if (debug_>0) std::cout << "[MuScleFit]: Constructor" << std::endl;

  if ((theMuonType_<-4 || theMuonType_>5) && theMuonType_<10) {
    std::cout << "[MuScleFit]: Unknown muon type! Aborting." << std::endl;
    abort();
  }

  loopCounter = 0;

  // Boundaries for h-function computation (to be improved!)
  // -------------------------------------------------------
  minResMass_hwindow[0] = 71.1876; // 76.;
  maxResMass_hwindow[0] = 111.188; // 106.;
  minResMass_hwindow[1] = 10.15;
  maxResMass_hwindow[1] = 10.55;
  minResMass_hwindow[2] = 9.8;
  maxResMass_hwindow[2] = 10.2;
  minResMass_hwindow[3] = 9.25;
  maxResMass_hwindow[3] = 9.65;
  minResMass_hwindow[4] = 3.58;
  maxResMass_hwindow[4] = 3.78;
  minResMass_hwindow[5] = 3.0;
  maxResMass_hwindow[5] = 3.2;

  // Max number of loops (if > 2 then try to minimize likelihood more than once)
  // ---------------------------------------------------------------------------
  maxLoopNumber = pset.getUntrackedParameter<int>("maxLoopNumber", 2);
  fastLoop = pset.getUntrackedParameter<bool>("FastLoop", true);

  // Selection of fits according to loop
  MuScleFitUtils::doResolFit = pset.getParameter<std::vector<int> >("doResolFit");
  MuScleFitUtils::doScaleFit = pset.getParameter<std::vector<int> >("doScaleFit");
  MuScleFitUtils::doCrossSectionFit = pset.getParameter<std::vector<int> >("doCrossSectionFit");
  MuScleFitUtils::doBackgroundFit = pset.getParameter<std::vector<int> >("doBackgroundFit");

  // Bias and smear types
  // --------------------
  int biasType = pset.getParameter<int>("BiasType");
  MuScleFitUtils::BiasType = biasType;
  // No error, the scale functions are used also for the bias
  MuScleFitUtils::biasFunction = scaleFunctionVecService( biasType );
  int smearType = pset.getParameter<int>("SmearType");
  MuScleFitUtils::SmearType = smearType;
  MuScleFitUtils::smearFunction = smearFunctionService( smearType );

  // Fit types
  // ---------
  int resolFitType = pset.getParameter<int>("ResolFitType");
  MuScleFitUtils::ResolFitType = resolFitType;
  MuScleFitUtils::resolutionFunction = resolutionFunctionService( resolFitType );
  MuScleFitUtils::resolutionFunctionForVec = resolutionFunctionVecService( resolFitType );
  int scaleType = pset.getParameter<int>("ScaleFitType");
  MuScleFitUtils::ScaleFitType = scaleType;
  MuScleFitUtils::scaleFunction = scaleFunctionService( scaleType );
  MuScleFitUtils::scaleFunctionForVec = scaleFunctionVecService( scaleType );

  // Initial parameters values
  // -------------------------
  MuScleFitUtils::parBias         = pset.getParameter<std::vector<double> >("parBias");
  MuScleFitUtils::parSmear        = pset.getParameter<std::vector<double> >("parSmear");
  MuScleFitUtils::parResol        = pset.getParameter<std::vector<double> >("parResol");
  MuScleFitUtils::parResolStep    = pset.getUntrackedParameter<std::vector<double> >("parResolStep", std::vector<double>());
  MuScleFitUtils::parResolMin     = pset.getUntrackedParameter<std::vector<double> >("parResolMin", std::vector<double>());
  MuScleFitUtils::parResolMax     = pset.getUntrackedParameter<std::vector<double> >("parResolMax", std::vector<double>());
  MuScleFitUtils::parScale        = pset.getParameter<std::vector<double> >("parScale");
  MuScleFitUtils::parScaleStep    = pset.getUntrackedParameter<std::vector<double> >("parScaleStep", std::vector<double>());
  MuScleFitUtils::parScaleMin     = pset.getUntrackedParameter<std::vector<double> >("parScaleMin", std::vector<double>());
  MuScleFitUtils::parScaleMax     = pset.getUntrackedParameter<std::vector<double> >("parScaleMax", std::vector<double>());
  MuScleFitUtils::parCrossSection = pset.getParameter<std::vector<double> >("parCrossSection");
  MuScleFitUtils::parBgr          = pset.getParameter<std::vector<double> >("parBgr");
  MuScleFitUtils::parResolFix        = pset.getParameter<std::vector<int> >("parResolFix");
  MuScleFitUtils::parScaleFix        = pset.getParameter<std::vector<int> >("parScaleFix");
  MuScleFitUtils::parCrossSectionFix = pset.getParameter<std::vector<int> >("parCrossSectionFix");
  MuScleFitUtils::parBgrFix          = pset.getParameter<std::vector<int> >("parBgrFix");
  MuScleFitUtils::parResolOrder        = pset.getParameter<std::vector<int> >("parResolOrder");
  MuScleFitUtils::parScaleOrder        = pset.getParameter<std::vector<int> >("parScaleOrder");
  MuScleFitUtils::parCrossSectionOrder = pset.getParameter<std::vector<int> >("parCrossSectionOrder");
  MuScleFitUtils::parBgrOrder          = pset.getParameter<std::vector<int> >("parBgrOrder");

  MuScleFitUtils::resfind     = pset.getParameter<std::vector<int> >("resfind");
  MuScleFitUtils::FitStrategy = pset.getParameter<int>("FitStrategy");

  // Option to skip unnecessary stuff
  // --------------------------------
  MuScleFitUtils::speedup = pset.getParameter<bool>("speedup");

  // Option to skip simTracks comparison
  compareToSimTracks_ = pset.getParameter<bool>("compareToSimTracks");
  simTracksCollection_ = pset.getUntrackedParameter<edm::InputTag>("SimTracksCollection", edm::InputTag("g4SimHits"));

  triggerResultsLabel_ = pset.getUntrackedParameter<std::string>("TriggerResultsLabel");
  triggerResultsProcess_ = pset.getUntrackedParameter<std::string>("TriggerResultsProcess");
  triggerPath_ = pset.getUntrackedParameter<std::vector<std::string> >("TriggerPath");
  negateTrigger_ = pset.getUntrackedParameter<bool>("NegateTrigger", false);
  saveAllToTree_ = pset.getUntrackedParameter<bool>("SaveAllToTree", false);

  PATmuons_ = pset.getUntrackedParameter<bool>("PATmuons", false);
  genParticlesName_ = pset.getUntrackedParameter<std::string>("GenParticlesName", "genParticles");

  // Use the probability file or not. If not it will perform a simpler selection taking the muon pair with
  // invariant mass closer to the pdf value and will crash if some fit is attempted.
  MuScleFitUtils::useProbsFile_ = pset.getUntrackedParameter<bool>("UseProbsFile", true);

  // This must be set to true if using events generated with Sherpa
  MuScleFitUtils::sherpa_ = pset.getUntrackedParameter<bool>("Sherpa", false);

  MuScleFitUtils::rapidityBinsForZ_ = pset.getUntrackedParameter<bool>("RapidityBinsForZ", true);

  // Set the cuts on muons to be used in the fit
  MuScleFitUtils::separateRanges_ = pset.getUntrackedParameter<bool>("SeparateRanges", true);
  MuScleFitUtils::maxMuonPt_ = pset.getUntrackedParameter<double>("MaxMuonPt", 100000000.);
  MuScleFitUtils::minMuonPt_ = pset.getUntrackedParameter<double>("MinMuonPt", 0.);
  MuScleFitUtils::minMuonEtaFirstRange_ = pset.getUntrackedParameter<double>("MinMuonEtaFirstRange", -6.);
  MuScleFitUtils::maxMuonEtaFirstRange_ = pset.getUntrackedParameter<double>("MaxMuonEtaFirstRange", 6.);
  MuScleFitUtils::minMuonEtaSecondRange_ = pset.getUntrackedParameter<double>("MinMuonEtaSecondRange", -100.);
  MuScleFitUtils::maxMuonEtaSecondRange_ = pset.getUntrackedParameter<double>("MaxMuonEtaSecondRange", 100.);
  MuScleFitUtils::deltaPhiMinCut_ = pset.getUntrackedParameter<double>("DeltaPhiMinCut", -100.);
  MuScleFitUtils::deltaPhiMaxCut_ = pset.getUntrackedParameter<double>("DeltaPhiMaxCut", 100.);

  MuScleFitUtils::debugMassResol_ = pset.getUntrackedParameter<bool>("DebugMassResol", false);
  // MuScleFitUtils::massResolComponentsStruct MuScleFitUtils::massResolComponents;

  // Check for parameters consistency
  // it will abort in case of errors.
  checkParameters();

  // Generate array of gaussian-distributed numbers for smearing
  // -----------------------------------------------------------
  if (MuScleFitUtils::SmearType>0) {
    std::cout << "[MuScleFit-Constructor]: Generating random values for smearing" << std::endl;
    TF1 G("G", "[0]*exp(-0.5*pow(x,2))", -5., 5.);
    double norm = 1/sqrt(2*TMath::Pi());
    G.SetParameter (0,norm);
    for (int i=0; i<10000; i++) {
      for (int j=0; j<7; j++) {
    MuScleFitUtils::x[j][i] = G.GetRandom();
      }
    }
  }
  MuScleFitUtils::goodmuon = 0;

  if(theMuonType_ > 0 &&  theMuonType_ < 4) {
    MuScleFitUtils::MuonTypeForCheckMassWindow = theMuonType_-1;
    MuScleFitUtils::MuonType = theMuonType_-1;
  }
  else if(theMuonType_ == 0 || theMuonType_ == 4 || theMuonType_ == 5 || theMuonType_ >= 10 || theMuonType_==-1 || theMuonType_==-2 || theMuonType_==-3 || theMuonType_==-4) {
    MuScleFitUtils::MuonTypeForCheckMassWindow = 2;
    MuScleFitUtils::MuonType = 2;
  }
  else{
    std::cout<<"Wrong muon type "<<theMuonType_<<std::endl;
    exit(1);
  }

  // When using standalone muons switch to the single Z pdf
  if( theMuonType_ == 2 ) {
    MuScleFitUtils::rapidityBinsForZ_ = false;
  }

  // Initialize ResMaxSigma And ResHalfWidth - 0 = global, 1 = SM, 2 = tracker
  // -------------------------------------------------------------------------
  MuScleFitUtils::massWindowHalfWidth[0][0] = 20.;
  MuScleFitUtils::massWindowHalfWidth[0][1] = 0.35;
  MuScleFitUtils::massWindowHalfWidth[0][2] = 0.35;
  MuScleFitUtils::massWindowHalfWidth[0][3] = 0.35;
  MuScleFitUtils::massWindowHalfWidth[0][4] = 0.2;
  MuScleFitUtils::massWindowHalfWidth[0][5] = 0.2;
  MuScleFitUtils::massWindowHalfWidth[1][0] = 50.;
  MuScleFitUtils::massWindowHalfWidth[1][1] = 2.5;
  MuScleFitUtils::massWindowHalfWidth[1][2] = 2.5;
  MuScleFitUtils::massWindowHalfWidth[1][3] = 2.5;
  MuScleFitUtils::massWindowHalfWidth[1][4] = 1.5;
  MuScleFitUtils::massWindowHalfWidth[1][5] = 1.5;
  MuScleFitUtils::massWindowHalfWidth[2][0] = 20.;
  MuScleFitUtils::massWindowHalfWidth[2][1] = 0.35;
  MuScleFitUtils::massWindowHalfWidth[2][2] = 0.35;
  MuScleFitUtils::massWindowHalfWidth[2][3] = 0.35;
  MuScleFitUtils::massWindowHalfWidth[2][4] = 0.2;
  MuScleFitUtils::massWindowHalfWidth[2][5] = 0.2;

  muonSelector_.reset(new MuScleFitMuonSelector(theMuonLabel_, theMuonType_, PATmuons_,
                        MuScleFitUtils::resfind,
                        MuScleFitUtils::speedup, genParticlesName_,
                        compareToSimTracks_, simTracksCollection_,
                        MuScleFitUtils::sherpa_, debug_));

  MuScleFitUtils::backgroundHandler = new BackgroundHandler( pset.getParameter<std::vector<int> >("BgrFitType"),
                                                             pset.getParameter<std::vector<double> >("LeftWindowBorder"),
                                                             pset.getParameter<std::vector<double> >("RightWindowBorder"),
                                                             MuScleFitUtils::ResMass,
                                                             MuScleFitUtils::massWindowHalfWidth[MuScleFitUtils::MuonTypeForCheckMassWindow] );

  MuScleFitUtils::crossSectionHandler = new CrossSectionHandler( MuScleFitUtils::parCrossSection, MuScleFitUtils::resfind );

  // Build cross section scale factors
  // MuScleFitUtils::resfind

  MuScleFitUtils::normalizeLikelihoodByEventNumber_ = pset.getUntrackedParameter<bool>("NormalizeLikelihoodByEventNumber", true);
  if(debug_>0) std::cout << "End of MuScleFit constructor" << std::endl;

  inputRootTreeFileName_ = pset.getParameter<std::string>("InputRootTreeFileName");
  outputRootTreeFileName_ = pset.getParameter<std::string>("OutputRootTreeFileName");
  maxEventsFromRootTree_ = pset.getParameter<int>("MaxEventsFromRootTree");

  MuScleFitUtils::startWithSimplex_ = pset.getParameter<bool>("StartWithSimplex");
  MuScleFitUtils::computeMinosErrors_ = pset.getParameter<bool>("ComputeMinosErrors");
  MuScleFitUtils::minimumShapePlots_ = pset.getParameter<bool>("MinimumShapePlots");

  beginOfJobInConstructor();
}

// Destructor
// ----------
MuScleFit::~MuScleFit () {
  if (debug_>0) std::cout << "[MuScleFit]: Destructor" << std::endl;
  std::cout << "Total number of analyzed events = " << totalEvents_ << std::endl;

  if( !(outputRootTreeFileName_.empty()) ) {
    // Save the events to a root tree unless we are reading from the edm root file and the SavedPair size is different from the totalEvents_
    if( !(inputRootTreeFileName_.empty() && (int(MuScleFitUtils::SavedPair.size()) != totalEvents_)) ) {
      std::cout << "Saving muon pairs to root tree" << std::endl;
      RootTreeHandler rootTreeHandler;
      if( MuScleFitUtils::speedup ) {
        // rootTreeHandler.writeTree(outputRootTreeFileName_, &(MuScleFitUtils::SavedPair), theMuonType_, 0, saveAllToTree_);
        rootTreeHandler.writeTree(outputRootTreeFileName_, &(muonPairs_), theMuonType_, 0, saveAllToTree_);
      }
      else {
        // rootTreeHandler.writeTree(outputRootTreeFileName_, &(MuScleFitUtils::SavedPair), theMuonType_, &(MuScleFitUtils::genPair), saveAllToTree_ );
        rootTreeHandler.writeTree(outputRootTreeFileName_, &(muonPairs_), theMuonType_, &(genMuonPairs_), saveAllToTree_ );
      }
    }
    else {
      std::cout << "ERROR: events in the vector = " << MuScleFitUtils::SavedPair.size() << " != totalEvents = " << totalEvents_ << std::endl;
    }
  }
}

// Begin job
// ---------
void MuScleFit::beginOfJobInConstructor()
// void MuScleFit::beginOfJob ()
// void MuScleFit::beginOfJob( const edm::EventSetup& eventSetup )
{
  if (debug_>0) std::cout << "[MuScleFit]: beginOfJob" << std::endl;
  //if(maxLoopNumber>1)
  if( MuScleFitUtils::useProbsFile_ ) {
    readProbabilityDistributionsFromFile();
  }

  if (debug_>0) std::cout << "[MuScleFit]: beginOfJob" << std::endl;

  // Create the root file
  // --------------------
  for (unsigned int i=0; i<(maxLoopNumber); i++) {
    std::stringstream ss;
    ss << i;
    std::string rootFileName = ss.str() + "_" + theRootFileName_;
    theFiles_.push_back (new TFile(rootFileName.c_str(), "RECREATE"));
  }
  if (debug_>0) std::cout << "[MuScleFit]: Root file created" << std::endl;

  std::cout << "creating plotter" << std::endl;
  plotter = new MuScleFitPlotter(theGenInfoRootFileName_);
  plotter->debug = debug_;
}

// End of job method
// -----------------
void MuScleFit::endOfJob () {
  if (debug_>0) std::cout << "[MuScleFit]: endOfJob" << std::endl;
}

// New loop
// --------
void MuScleFit::startingNewLoop( unsigned int iLoop )
{
  if (debug_>0) std::cout << "[MuScleFit]: Starting loop # " << iLoop << std::endl;

  // Number of muons used
  // --------------------
  MuScleFitUtils::goodmuon = 0;

  // Counters for problem std::cout-ing
  // -----------------------------
  MuScleFitUtils::counter_resprob = 0;

  // Create the root file
  // --------------------
  fillHistoMap(theFiles_[iLoop], iLoop);

  loopCounter = iLoop;
  MuScleFitUtils::loopCounter = loopCounter;

  iev = 0;
  MuScleFitUtils::iev_ = 0;

  MuScleFitUtils::oldNormalization_ = 0;
}

// End of loop routine
// -------------------
edm::EDLooper::Status MuScleFit::endOfLoop( const edm::EventSetup& eventSetup, unsigned int iLoop )
{
  unsigned int iFastLoop = 1;

  // Read the events from the root tree if requested
  if( !(inputRootTreeFileName_.empty()) ) {
    selectMuons(maxEventsFromRootTree_, inputRootTreeFileName_);
    // When reading from local file all the loops are done here
    totalEvents_ = MuScleFitUtils::SavedPair.size();
    iFastLoop = 0;
  }
  else {
    endOfFastLoop(iLoop);
  }

  // If a fastLoop is required we do all the remaining iterations here
  if( fastLoop == true ) {
    for( ; iFastLoop<maxLoopNumber; ++iFastLoop ) {

      std::cout << "Starting fast loop number " << iFastLoop << std::endl;

      // In the first loop is called by the framework
      // if( iFastLoop > 0 ) {
      startingNewLoop(iFastLoop);
      // }

      // std::vector<std::pair<lorentzVector,lorentzVector> >::const_iterator it = MuScleFitUtils::SavedPair.begin();
      // for( ; it != SavedPair.end(); ++it ) {
      while( iev<totalEvents_ ) {
    if( iev%50000 == 0 ) {
      std::cout << "Fast looping on event number " << iev << std::endl;
    }
    // This reads muons from SavedPair using iev to keep track of the event
        duringFastLoop();
      }
      std::cout << "End of fast loop number " << iFastLoop << ". Ran on " << iev << " events" << std::endl;
      endOfFastLoop(iFastLoop);
    }
  }

  if (iFastLoop>=maxLoopNumber-1) {
    return kStop;
  } else {
    return kContinue;
  }
}

void MuScleFit::endOfFastLoop( const unsigned int iLoop )
{
  // std::cout<< "Inside endOfFastLoop, iLoop = " << iLoop << " and loopCounter = " << loopCounter << std::endl;

  if( loopCounter == 0 ) {
    // plotter->writeHistoMap();
    // The destructor will call the writeHistoMap after the cd to the output file
    delete plotter;
  }

  std::cout << "Ending loop # " << iLoop << std::endl;

  // Write the histos to file
  // ------------------------
  // theFiles_[iLoop]->cd();
  writeHistoMap(iLoop);

  // Likelihood minimization to compute corrections
  // ----------------------------------------------
  // theFiles_[iLoop]->cd();
  TDirectory * likelihoodDir = theFiles_[iLoop]->mkdir("likelihood");
  likelihoodDir->cd();
  MuScleFitUtils::minimizeLikelihood();

  // ATTENTION, this was put BEFORE the minimizeLikelihood. Check for problems.
  theFiles_[iLoop]->Close();
  // ATTENTION: Check that this delete does not give any problem
  delete theFiles_[iLoop];

  // Clear the histos
  // ----------------
  clearHistoMap();
}

// Stuff to do during loop
// -----------------------
edm::EDLooper::Status MuScleFit::duringLoop( const edm::Event & event, const edm::EventSetup& eventSetup )
{
  edm::Handle<edm::TriggerResults> triggerResults;
  event.getByLabel(edm::InputTag(triggerResultsLabel_.c_str(), "", triggerResultsProcess_.c_str()), triggerResults);
  //event.getByLabel(InputTag(triggerResultsLabel_),triggerResults);
  bool isFired = false;

  if(triggerPath_[0] == "")
    isFired = true;
  else if(triggerPath_[0] == "All"){
    isFired =triggerResults->accept();
    if(debug_>0)
      std::cout<<"Trigger "<<isFired<<std::endl;
  }
  else{
    bool changed;
    HLTConfigProvider hltConfig;
    hltConfig.init(event.getRun(), eventSetup, triggerResultsProcess_, changed);


    const edm::TriggerNames triggerNames = event.triggerNames(*triggerResults); 

    for (unsigned i=0; i<triggerNames.size(); i++) { 
      std::string hltName = triggerNames.triggerName(i);

      // match the path in the pset with the true name of the trigger  
      for ( unsigned int ipath=0; ipath<triggerPath_.size(); ipath++ ) {
    if ( hltName.find(triggerPath_[ipath]) != std::string::npos ) {
        unsigned int triggerIndex( hltConfig.triggerIndex(hltName) );
      
      // triggerIndex must be less than the size of HLTR or you get a CMSException: _M_range_check
        if (triggerIndex < triggerResults->size()) {
          isFired = triggerResults->accept(triggerIndex);
          if(debug_>0) 
        std::cout << triggerPath_[ipath] <<" "<< hltName << " " << isFired<<std::endl;
        }       
    } // end if (matching the path in the pset with the true trigger name      
      }
    }

  }

  if( negateTrigger_ && isFired ) return kContinue;
  else if( !(negateTrigger_) && !isFired ) return kContinue;

#ifdef USE_CALLGRIND
  CALLGRIND_START_INSTRUMENTATION;
#endif

  if (debug_>0) {
    std::cout << "[MuScleFit-duringLoop]: loopCounter = " << loopCounter
     << " Run: " << event.id().run() << " Event: " << event.id().event() << std::endl;
  }

  // On the first iteration we read the bank, otherwise we fetch the information from the muon tree
  // ------------------------------------ Important Note --------------------------------------- //
  // The fillMuonCollection method applies any smearing or bias to the muons, so we NEVER use
  // unbiased muons.
  // ----------------------------------------------------------------------------------------------
  if( loopCounter == 0 ) {

    if( !fastLoop || inputRootTreeFileName_.empty() ) {
      if( debug_ > 0 ) std::cout << "Reading from edm event" << std::endl;
      selectMuons(event);
      duringFastLoop();
      ++totalEvents_;
    }
  }

  return kContinue;

#ifdef USE_CALLGRIND
  CALLGRIND_STOP_INSTRUMENTATION;
  CALLGRIND_DUMP_STATS;
#endif
}

void MuScleFit::selectMuons(const edm::Event & event)
{
  recMu1 = reco::Particle::LorentzVector(0,0,0,0);
  recMu2 = reco::Particle::LorentzVector(0,0,0,0);

  std::vector<reco::LeafCandidate> muons;
  muonSelector_->selectMuons(event, muons, genMuonPairs_, MuScleFitUtils::simPair, plotter);
  //  plotter->fillRec(muons); // @EM method already invoked inside MuScleFitMuonSelector::selectMuons()

  // Find the two muons from the resonance, and set ResFound bool
  // ------------------------------------------------------------
  std::pair<reco::Particle::LorentzVector, reco::Particle::LorentzVector> recMuFromBestRes =
    MuScleFitUtils::findBestRecoRes(muons);

  if (MuScleFitUtils::ResFound) {
    if (debug_>0) {
      std::cout <<std::setprecision(9)<< "Pt after findbestrecores: " << (recMuFromBestRes.first).Pt() << " "
           << (recMuFromBestRes.second).Pt() << std::endl;
      std::cout << "recMu1 = " << recMu1 << std::endl;
      std::cout << "recMu2 = " << recMu2 << std::endl;
    }
    recMu1 = recMuFromBestRes.first;
    recMu2 = recMuFromBestRes.second;
    if (debug_>0) {
      std::cout << "after recMu1 = " << recMu1 << std::endl;
      std::cout << "after recMu2 = " << recMu2 << std::endl;
      std::cout << "mu1.pt = " << recMu1.Pt() << std::endl;
      std::cout << "mu2.pt = " << recMu2.Pt() << std::endl;
    }
    MuScleFitUtils::SavedPair.push_back( std::make_pair( recMu1, recMu2 ) );
  } else {
    MuScleFitUtils::SavedPair.push_back( std::make_pair( lorentzVector(0.,0.,0.,0.), lorentzVector(0.,0.,0.,0.) ) );
  }
  // Save the events also in the external tree so that it can be saved late

  // std::cout << "SavedPair->size() " << MuScleFitUtils::SavedPair.size() << std::endl;
  muonPairs_.push_back(MuonPair(MuScleFitUtils::SavedPair.back().first,
                MuScleFitUtils::SavedPair.back().second,
                event.run(), event.id().event()));
  // Fill the internal genPair tree from the external one
  if( MuScleFitUtils::speedup == false ) {
    MuScleFitUtils::genPair.push_back(std::make_pair( genMuonPairs_.back().mu1, genMuonPairs_.back().mu2 ));
  }
}

void MuScleFit::selectMuons(const int maxEvents, const TString & treeFileName)
{
  std::cout << "Reading muon pairs from Root Tree in " << treeFileName << std::endl;
  RootTreeHandler rootTreeHandler;
  std::vector<std::pair<int, int> > evtRun;
  if( MuScleFitUtils::speedup ) {
    rootTreeHandler.readTree(maxEvents, inputRootTreeFileName_, &(MuScleFitUtils::SavedPair), theMuonType_, &evtRun);
  }
  else {
    rootTreeHandler.readTree(maxEvents, inputRootTreeFileName_, &(MuScleFitUtils::SavedPair), theMuonType_, &evtRun, &(MuScleFitUtils::genPair));
  }
  // Now loop on all the pairs and apply any smearing and bias if needed
  std::vector<std::pair<int, int> >::iterator evtRunIt = evtRun.begin();
  std::vector<std::pair<lorentzVector,lorentzVector> >::iterator it = MuScleFitUtils::SavedPair.begin();
  std::vector<std::pair<lorentzVector,lorentzVector> >::iterator genIt;
  if(MuScleFitUtils::speedup == false) genIt = MuScleFitUtils::genPair.begin();
  for( ; it != MuScleFitUtils::SavedPair.end(); ++it, ++evtRunIt ) {

    // Apply any cut if requested
    // Note that cuts here are only applied to already selected muons. They should not be used unless
    // you are sure that the difference is negligible (e.g. the number of events with > 2 muons is negligible).
    double pt1 = it->first.pt();
    // std::cout << "pt1 = " << pt1 << std::endl;
    double pt2 = it->second.pt();
    // std::cout << "pt2 = " << pt2 << std::endl;
    double eta1 = it->first.eta();
    // std::cout << "eta1 = " << eta1 << std::endl;
    double eta2 = it->second.eta();
    // std::cout << "eta2 = " << eta2 << std::endl;
    // If they don't pass the cuts set to null vectors
    bool dontPass = false;
    bool eta1InFirstRange; 
    bool eta2InFirstRange; 
    bool eta1InSecondRange; 
    bool eta2InSecondRange; 

    if( MuScleFitUtils::separateRanges_ ) {
      eta1InFirstRange = eta1 >= MuScleFitUtils::minMuonEtaFirstRange_ && eta1 < MuScleFitUtils::maxMuonEtaFirstRange_;
      eta2InFirstRange = eta2 >= MuScleFitUtils::minMuonEtaFirstRange_ && eta2 < MuScleFitUtils::maxMuonEtaFirstRange_;
      eta1InSecondRange = eta1 >= MuScleFitUtils::minMuonEtaSecondRange_ && eta1 < MuScleFitUtils::maxMuonEtaSecondRange_;
      eta2InSecondRange = eta2 >= MuScleFitUtils::minMuonEtaSecondRange_ && eta2 < MuScleFitUtils::maxMuonEtaSecondRange_;

      // This is my logic, which should be erroneous, but certainly simpler...
      if( !(pt1 >= MuScleFitUtils::minMuonPt_ && pt1 < MuScleFitUtils::maxMuonPt_ &&
        pt2 >= MuScleFitUtils::minMuonPt_ && pt2 < MuScleFitUtils::maxMuonPt_ &&
        eta1InFirstRange && eta2InSecondRange ) ) {
    dontPass = true;
      }
    }
    else {
      eta1 = fabs(eta1);
      eta2 = fabs(eta2);
      eta1InFirstRange = eta1 >= MuScleFitUtils::minMuonEtaFirstRange_ && eta1 < MuScleFitUtils::maxMuonEtaFirstRange_;
      eta2InFirstRange = eta2 >= MuScleFitUtils::minMuonEtaFirstRange_ && eta2 < MuScleFitUtils::maxMuonEtaFirstRange_;
      eta1InSecondRange = eta1 >= MuScleFitUtils::minMuonEtaSecondRange_ && eta1 < MuScleFitUtils::maxMuonEtaSecondRange_;
      eta2InSecondRange = eta2 >= MuScleFitUtils::minMuonEtaSecondRange_ && eta2 < MuScleFitUtils::maxMuonEtaSecondRange_;
      if( !(pt1 >= MuScleFitUtils::minMuonPt_ && pt1 < MuScleFitUtils::maxMuonPt_ &&
        pt2 >= MuScleFitUtils::minMuonPt_ && pt2 < MuScleFitUtils::maxMuonPt_ &&
        ( ((eta1InFirstRange && !eta2InFirstRange) && (eta2InSecondRange && !eta1InSecondRange)) ||
          ((eta2InFirstRange && !eta1InFirstRange) && (eta1InSecondRange && !eta2InSecondRange)) )) ) {
    dontPass = true;
      }
    }
    
    // Additional check on deltaPhi
    double deltaPhi = MuScleFitUtils::deltaPhi(it->first.phi(), it->second.phi());
    if( (deltaPhi <= MuScleFitUtils::deltaPhiMinCut_) || (deltaPhi >= MuScleFitUtils::deltaPhiMaxCut_) ) dontPass = true;
    
    if( dontPass ) {
      // std::cout << "removing muons not passing cuts" << std::endl;
      it->first = reco::Particle::LorentzVector(0,0,0,0);
      it->second = reco::Particle::LorentzVector(0,0,0,0);
    }

    // First is always mu-, second mu+
    if( (MuScleFitUtils::SmearType != 0) || (MuScleFitUtils::BiasType != 0) ) {
      applySmearing(it->first);
      applyBias(it->first, -1);
      applySmearing(it->second);
      applyBias(it->second, 1);
    }
    muonPairs_.push_back(MuonPair(it->first, it->second,
                 evtRunIt->second, evtRunIt->first));

    // Fill the internal genPair tree from the external one
    if( MuScleFitUtils::speedup == false ) {
      genMuonPairs_.push_back(GenMuonPair(genIt->first, genIt->second, 0));
      ++genIt;
    }
  }
  plotter->fillTreeRec(MuScleFitUtils::SavedPair);
  if( !(MuScleFitUtils::speedup) ) {
    plotter->fillTreeGen(MuScleFitUtils::genPair);
  }
}

void MuScleFit::duringFastLoop()
{
  // On loops>0 the two muons are directly obtained from the SavedMuon array
  // -----------------------------------------------------------------------
  MuScleFitUtils::ResFound = false;
  recMu1 = (MuScleFitUtils::SavedPair[iev].first);
  recMu2 = (MuScleFitUtils::SavedPair[iev].second);
  if (recMu1.Pt()>0 && recMu2.Pt()>0) {
    MuScleFitUtils::ResFound = true;
    if (debug_>0) std::cout << "Ev = " << iev << ": found muons in tree with Pt = "
                       << recMu1.Pt() << " " << recMu2.Pt() << std::endl;
  }

  if( debug_>0 ) std::cout << "About to start lik par correction and histo filling; ResFound is "
            << MuScleFitUtils::ResFound << std::endl;
  // If resonance found, do the hard work
  // ------------------------------------
  if( MuScleFitUtils::ResFound ) {

    // Find weight and reference mass for this muon pair
    // -------------------------------------------------
    // The last parameter = true means that we want to use always the background window to compute the weight,
    // otherwise the probability will be filled only for the resonance region.
    double weight = MuScleFitUtils::computeWeight( (recMu1+recMu2).mass(), iev, true );
    if (debug_>0) {
      std::cout << "Loop #" << loopCounter << "Event #" << iev << ": before correction     Pt1 = "
       << recMu1.Pt() << " Pt2 = " << recMu2.Pt() << std::endl;
    }
    // For successive iterations, correct the muons only if the previous iteration was a scale fit.
    // --------------------------------------------------------------------------------------------
    if ( loopCounter>0 ) {
      if ( MuScleFitUtils::doScaleFit[loopCounter-1] ) {
        recMu1 = (MuScleFitUtils::applyScale(recMu1, MuScleFitUtils::parvalue[loopCounter-1], -1));
        recMu2 = (MuScleFitUtils::applyScale(recMu2, MuScleFitUtils::parvalue[loopCounter-1],  1));
      }
    }
    if (debug_>0) {
      std::cout << "Loop #" << loopCounter << "Event #" << iev << ": after correction      Pt1 = "
       << recMu1.Pt() << " Pt2 = " << recMu2.Pt() << std::endl;
    }

    reco::Particle::LorentzVector bestRecRes( recMu1+recMu2 );

    //Fill histograms
    //------------------
   
    mapHisto_["hRecBestMu"]->Fill(recMu1, -1,weight);
    mapHisto_["hRecBestMuVSEta"]->Fill(recMu1);
    mapHisto_["hRecBestMu"]->Fill(recMu2, +1,weight);
    mapHisto_["hRecBestMuVSEta"]->Fill(recMu2);
    mapHisto_["hDeltaRecBestMu"]->Fill(recMu1, recMu2);
    // Reconstructed resonance
    mapHisto_["hRecBestRes"]->Fill(bestRecRes,+1, weight);
    mapHisto_["hRecBestResAllEvents"]->Fill(bestRecRes,+1, 1.);
//     // Fill histogram of Res mass vs muon variables
//     mapHisto_["hRecBestResVSMu"]->Fill (recMu1, bestRecRes, -1);
//     mapHisto_["hRecBestResVSMu"]->Fill (recMu2, bestRecRes, +1);
//     // Fill also the mass mu+/mu- comparisons
//     mapHisto_["hRecBestResVSMu"]->Fill(recMu1, recMu2, bestRecRes);

    mapHisto_["hRecBestResVSMu"]->Fill (recMu1, bestRecRes, -1, weight);
    mapHisto_["hRecBestResVSMu"]->Fill (recMu2, bestRecRes, +1, weight);
    // Fill also the mass mu+/mu- comparisons
    mapHisto_["hRecBestResVSMu"]->Fill(recMu1, recMu2, bestRecRes, weight);
    
    //-- rc 2010 filling histograms for mu+ /mu- ------
    //  mapHisto_["hRecBestResVSMuMinus"]->Fill (recMu1, bestRecRes, -1);
    // mapHisto_["hRecBestResVSMuPlus"]->Fill (recMu2, bestRecRes, +1);
  
    //-- rc 2010 filling histograms MassVsMuEtaPhi------
    //  mapHisto_["hRecBestResVSMuEtaPhi"]->Fill (recMu1, bestRecRes,-1);
    //  mapHisto_["hRecBestResVSMuEtaPhi"]->Fill (recMu2, bestRecRes,+1);

    // Fill histogram of Res mass vs Res variables
    // mapHisto_["hRecBestResVSRes"]->Fill (bestRecRes, bestRecRes, +1);
    mapHisto_["hRecBestResVSRes"]->Fill (bestRecRes, bestRecRes, +1, weight);






    std::vector<double> * parval;
    std::vector<double> initpar;
    // Store a pointer to the vector of parameters of the last iteration, or the initial
    // parameters if this is the first iteration
    if (loopCounter==0) {
      initpar = MuScleFitUtils::parResol;
      initpar.insert( initpar.end(), MuScleFitUtils::parScale.begin(), MuScleFitUtils::parScale.end() );
      initpar.insert( initpar.end(), MuScleFitUtils::parCrossSection.begin(), MuScleFitUtils::parCrossSection.end() );
      initpar.insert( initpar.end(), MuScleFitUtils::parBgr.begin(), MuScleFitUtils::parBgr.end() );
      parval = &initpar;
    } else {
      parval = &(MuScleFitUtils::parvalue[loopCounter-1]);
    }

    //Compute pt resolution w.r.t generated and simulated muons
    //--------------------------------------------------------
    if( !MuScleFitUtils::speedup ) {

      //first is always mu-, second is always mu+
      if(checkDeltaR(MuScleFitUtils::genPair[iev].first,recMu1)) {
        fillComparisonHistograms( MuScleFitUtils::genPair[iev].first, recMu1, "Gen", -1 );
      }
      if(checkDeltaR(MuScleFitUtils::genPair[iev].second,recMu2)){
        fillComparisonHistograms( MuScleFitUtils::genPair[iev].second, recMu2, "Gen", +1 );
      }
      if( compareToSimTracks_ ) {
        //first is always mu-, second is always mu+
        if(checkDeltaR(MuScleFitUtils::simPair[iev].first,recMu1)){
          fillComparisonHistograms( MuScleFitUtils::simPair[iev].first, recMu1, "Sim", -1 );
        }
        if(checkDeltaR(MuScleFitUtils::simPair[iev].second,recMu2)){
          fillComparisonHistograms( MuScleFitUtils::simPair[iev].second, recMu2, "Sim", +1 );
        }
      }
    }

    // ATTENTION: this was done only when a matching was found. Moved it outside because, genInfo or not, we still want to see the resolution function
    // Fill also the resolution histogramsm using the resolution functions:
    // the parameters are those from the last iteration, as the muons up to this point have also the corrections from the same iteration.
    // Need to use a different array (ForVec), containing functors able to operate on std::vector<double>
    mapHisto_["hFunctionResolPt"]->Fill( recMu1, MuScleFitUtils::resolutionFunctionForVec->sigmaPt(recMu1.Pt(), recMu1.Eta(), *parval ), -1 );
    mapHisto_["hFunctionResolCotgTheta"]->Fill( recMu1, MuScleFitUtils::resolutionFunctionForVec->sigmaCotgTh(recMu1.Pt(), recMu1.Eta(), *parval ), -1 );
    mapHisto_["hFunctionResolPhi"]->Fill( recMu1, MuScleFitUtils::resolutionFunctionForVec->sigmaPhi(recMu1.Pt(), recMu1.Eta(), *parval ), -1 );
    mapHisto_["hFunctionResolPt"]->Fill( recMu2, MuScleFitUtils::resolutionFunctionForVec->sigmaPt(recMu2.Pt(), recMu2.Eta(), *parval ), +1 );
    mapHisto_["hFunctionResolCotgTheta"]->Fill( recMu2, MuScleFitUtils::resolutionFunctionForVec->sigmaCotgTh(recMu2.Pt(), recMu2.Eta(), *parval ), +1 );
    mapHisto_["hFunctionResolPhi"]->Fill( recMu2, MuScleFitUtils::resolutionFunctionForVec->sigmaPhi(recMu2.Pt(), recMu2.Eta(), *parval ), +1 );

    // Compute likelihood histograms
    // -----------------------------
    if( debug_ > 0 ) std::cout << "mass = " << bestRecRes.mass() << std::endl;
    if (weight!=0.) {
      double massResol;
      double prob;
      double deltalike;
      if (loopCounter==0) {
    std::vector<double> initpar;
    for (int i=0; i<(int)(MuScleFitUtils::parResol.size()); i++) {
      initpar.push_back(MuScleFitUtils::parResol[i]);
    }
    for (int i=0; i<(int)(MuScleFitUtils::parScale.size()); i++) {
      initpar.push_back(MuScleFitUtils::parScale[i]);
    }
//  for (int i=0; i<(int)(MuScleFitUtils::parCrossSection.size()); i++) {
//    initpar.push_back(MuScleFitUtils::parCrossSection[i]);
//  }
        MuScleFitUtils::crossSectionHandler->addParameters(initpar);

    for (int i=0; i<(int)(MuScleFitUtils::parBgr.size()); i++) {
      initpar.push_back(MuScleFitUtils::parBgr[i]);
    }
    massResol = MuScleFitUtils::massResolution( recMu1, recMu2, initpar );
    // prob      = MuScleFitUtils::massProb( bestRecRes.mass(), bestRecRes.Eta(), bestRecRes.Rapidity(), massResol, initpar, true );
    prob      = MuScleFitUtils::massProb( bestRecRes.mass(), bestRecRes.Eta(), bestRecRes.Rapidity(), massResol,
                          initpar, true, recMu1.eta(), recMu2.eta() );
      } else {
    massResol = MuScleFitUtils::massResolution( recMu1, recMu2,
                                                    MuScleFitUtils::parvalue[loopCounter-1] );
    // prob      = MuScleFitUtils::massProb( bestRecRes.mass(), bestRecRes.Eta(), bestRecRes.Rapidity(),
        //                                       massResol, MuScleFitUtils::parvalue[loopCounter-1], true );
    prob      = MuScleFitUtils::massProb( bestRecRes.mass(), bestRecRes.Eta(), bestRecRes.Rapidity(),
                                              massResol, MuScleFitUtils::parvalue[loopCounter-1], true,
                          recMu1.eta(), recMu2.eta() );
      }
      if( debug_ > 0 ) std::cout << "inside weight: mass = " << bestRecRes.mass() << ", prob = " << prob << std::endl;
      if (prob>0) {
        if( debug_ > 0 ) std::cout << "inside prob: mass = " << bestRecRes.mass() << ", prob = " << prob << std::endl;

    deltalike = log(prob)*weight; // NB maximum likelihood --> deltalike is maximized
    mapHisto_["hLikeVSMu"]->Fill(recMu1, deltalike);
    mapHisto_["hLikeVSMu"]->Fill(recMu2, deltalike);
    mapHisto_["hLikeVSMuMinus"]->Fill(recMu1, deltalike);
    mapHisto_["hLikeVSMuPlus"]->Fill(recMu2, deltalike);

        double recoMass = (recMu1+recMu2).mass();
        if( recoMass != 0 ) {
          // IMPORTANT: massResol is not a relative resolution
          mapHisto_["hResolMassVSMu"]->Fill(recMu1, massResol, -1);
          mapHisto_["hResolMassVSMu"]->Fill(recMu2, massResol, +1);
          mapHisto_["hFunctionResolMassVSMu"]->Fill(recMu1, massResol/recoMass, -1);
          mapHisto_["hFunctionResolMassVSMu"]->Fill(recMu2, massResol/recoMass, +1);
        }

        if( MuScleFitUtils::debugMassResol_ ) {
          mapHisto_["hdMdPt1"]->Fill(recMu1, MuScleFitUtils::massResolComponents.dmdpt1, -1);
          mapHisto_["hdMdPt2"]->Fill(recMu2, MuScleFitUtils::massResolComponents.dmdpt2, +1);
          mapHisto_["hdMdPhi1"]->Fill(recMu1, MuScleFitUtils::massResolComponents.dmdphi1, -1);
          mapHisto_["hdMdPhi2"]->Fill(recMu2, MuScleFitUtils::massResolComponents.dmdphi2, +1);
          mapHisto_["hdMdCotgTh1"]->Fill(recMu1, MuScleFitUtils::massResolComponents.dmdcotgth1, -1);
          mapHisto_["hdMdCotgTh2"]->Fill(recMu2, MuScleFitUtils::massResolComponents.dmdcotgth2, +1);
        }

        if( !MuScleFitUtils::speedup ) {
          double genMass = (MuScleFitUtils::genPair[iev].first + MuScleFitUtils::genPair[iev].second).mass();
          // Fill the mass resolution (computed from MC), we use the covariance class to compute the variance
          if( genMass != 0 ) {
        mapHisto_["hGenResVSMu"]->Fill((MuScleFitUtils::genPair[iev].first), (MuScleFitUtils::genPair[iev].first + MuScleFitUtils::genPair[iev].second), -1);
        mapHisto_["hGenResVSMu"]->Fill((MuScleFitUtils::genPair[iev].second), (MuScleFitUtils::genPair[iev].first + MuScleFitUtils::genPair[iev].second), +1);
            double diffMass = (recoMass - genMass)/genMass;
            // double diffMass = recoMass - genMass;
            // Fill if for both muons
            double pt1 = recMu1.pt();
            double eta1 = recMu1.eta();
            double pt2 = recMu2.pt();
            double eta2 = recMu2.eta();
            // This is to avoid nan
            if( diffMass == diffMass ) {
              // Mass relative difference vs Pt and Eta. To be used to extract the true mass resolution
              mapHisto_["hDeltaMassOverGenMassVsPt"]->Fill(pt1, diffMass);
              mapHisto_["hDeltaMassOverGenMassVsPt"]->Fill(pt2, diffMass);
              mapHisto_["hDeltaMassOverGenMassVsEta"]->Fill(eta1, diffMass);
              mapHisto_["hDeltaMassOverGenMassVsEta"]->Fill(eta2, diffMass);
              // This is used for the covariance comparison
              mapHisto_["hMassResolutionVsPtEta"]->Fill(pt1, eta1, diffMass, diffMass);
              mapHisto_["hMassResolutionVsPtEta"]->Fill(pt2, eta2, diffMass, diffMass);
            }
            else {
              std::cout << "Error, there is a nan: recoMass = " << recoMass << ", genMass = " << genMass << std::endl;
            }
          }
          // Fill with mass resolution from resolution function
          double massRes = MuScleFitUtils::massResolution(recMu1, recMu2, MuScleFitUtils::parResol);
          mapHisto_["hFunctionResolMass"]->Fill( recMu1, std::pow(massRes,2), -1 );
          mapHisto_["hFunctionResolMass"]->Fill( recMu2, std::pow(massRes,2), +1 );
        }

        mapHisto_["hMass_P"]->Fill(bestRecRes.mass(), prob);
        if( debug_ > 0 ) std::cout << "mass = " << bestRecRes.mass() << ", prob = " << prob << std::endl;
        mapHisto_["hMass_fine_P"]->Fill(bestRecRes.mass(), prob);

        mapHisto_["hMassProbVsRes"]->Fill(bestRecRes, bestRecRes, +1, prob);
        mapHisto_["hMassProbVsMu"]->Fill(recMu1, bestRecRes, -1, prob);
        mapHisto_["hMassProbVsMu"]->Fill(recMu2, bestRecRes, +1, prob);
        mapHisto_["hMassProbVsRes_fine"]->Fill(bestRecRes, bestRecRes, +1, prob);
        mapHisto_["hMassProbVsMu_fine"]->Fill(recMu1, bestRecRes, -1, prob);
        mapHisto_["hMassProbVsMu_fine"]->Fill(recMu2, bestRecRes, +1, prob);
      }
    }
  } // end if ResFound

  // Fill the pair
  // -------------
  if (loopCounter>0) {
    if (debug_>0) std::cout << "[MuScleFit]: filling the pair" << std::endl;
    MuScleFitUtils::SavedPair[iev] = std::make_pair( recMu1, recMu2 );
  }

  iev++;
  MuScleFitUtils::iev_++;

  // return kContinue;
}

bool MuScleFit::checkDeltaR(reco::Particle::LorentzVector& genMu, reco::Particle::LorentzVector& recMu){
  //first is always mu-, second is always mu+
  double deltaR = sqrt(MuScleFitUtils::deltaPhi(recMu.Phi(),genMu.Phi()) * MuScleFitUtils::deltaPhi(recMu.Phi(),genMu.Phi()) +
                       ((recMu.Eta()-genMu.Eta()) * (recMu.Eta()-genMu.Eta())));
  if(deltaR<0.01)
    return true;
  else if( debug_ > 0 ) {
    std::cout<<"Reco muon "<<recMu<<" with eta "<<recMu.Eta()<<" and phi "<<recMu.Phi()<<std::endl
    <<" DOES NOT MATCH with generated muon from resonance: "<<std::endl
    <<genMu<<" with eta "<<genMu.Eta()<<" and phi "<<genMu.Phi()<<std::endl;
  }
  return false;
}

void MuScleFit::fillComparisonHistograms( const reco::Particle::LorentzVector & genMu, const reco::Particle::LorentzVector & recMu,
                      const std::string & inputName, const int charge )
{
  std::string name(inputName + "VSMu");
  mapHisto_["hResolPt"+name]->Fill(recMu, (-genMu.Pt()+recMu.Pt())/genMu.Pt(), charge);
  mapHisto_["hResolTheta"+name]->Fill(recMu, (-genMu.Theta()+recMu.Theta()), charge);
  mapHisto_["hResolCotgTheta"+name]->Fill(recMu,(-cos(genMu.Theta())/sin(genMu.Theta())
                                                 +cos(recMu.Theta())/sin(recMu.Theta())), charge);
  mapHisto_["hResolEta"+name]->Fill(recMu, (-genMu.Eta()+recMu.Eta()),charge);
  mapHisto_["hResolPhi"+name]->Fill(recMu, MuScleFitUtils::deltaPhiNoFabs(recMu.Phi(), genMu.Phi()), charge);

  // Fill only if it was matched to a genMu and this muon is valid
  if( (genMu.Pt() != 0) && (recMu.Pt() != 0) ) {
    mapHisto_["hPtRecoVsPt"+inputName]->Fill(genMu.Pt(), recMu.Pt());
  }
}

void MuScleFit::applySmearing( reco::Particle::LorentzVector & mu )
{
  if( MuScleFitUtils::SmearType>0 ) {
    mu = MuScleFitUtils::applySmearing( mu );
    if (debug_>0) std::cout << "Muon #" << MuScleFitUtils::goodmuon
                       << ": after smearing  Pt = " << mu.Pt() << std::endl;
  }
}

void MuScleFit::applyBias( reco::Particle::LorentzVector & mu, const int charge )
{
  if( MuScleFitUtils::BiasType>0 ) {
    mu = MuScleFitUtils::applyBias( mu, charge );
    if (debug_>0) std::cout << "Muon #" << MuScleFitUtils::goodmuon
                       << ": after bias      Pt = " << mu.Pt() << std::endl;
  }
}

// Simple method to check parameters consistency. It aborts the job if the parameters
// are not consistent.
void MuScleFit::checkParameters() {

  // Fits selection dimension check
  if( MuScleFitUtils::doResolFit.size() != maxLoopNumber ) {
    std::cout << "[MuScleFit-Constructor]: wrong size of resolution fits selector = " << MuScleFitUtils::doResolFit.size() << std::endl;
    std::cout << "it must have as many values as the number of loops, which is = " << maxLoopNumber << std::endl;
    abort();
  }
  if( MuScleFitUtils::doScaleFit.size() != maxLoopNumber ) {
    std::cout << "[MuScleFit-Constructor]: wrong size of scale fits selector = " << MuScleFitUtils::doScaleFit.size() << std::endl;
    std::cout << "it must have as many values as the number of loops, which is = " << maxLoopNumber << std::endl;
    abort();
  }
  if( MuScleFitUtils::doCrossSectionFit.size() != maxLoopNumber ) {
    std::cout << "[MuScleFit-Constructor]: wrong size of cross section fits selector = " << MuScleFitUtils::doCrossSectionFit.size() << std::endl;
    std::cout << "it must have as many values as the number of loops, which is = " << maxLoopNumber << std::endl;
    abort();
  }
  if( MuScleFitUtils::doBackgroundFit.size() != maxLoopNumber ) {
    std::cout << "[MuScleFit-Constructor]: wrong size of background fits selector = " << MuScleFitUtils::doBackgroundFit.size() << std::endl;
    std::cout << "it must have as many values as the number of loops, which is = " << maxLoopNumber << std::endl;
    abort();
  }

  // Bias parameters: dimension check
  // --------------------------------
  if ((MuScleFitUtils::BiasType==1  && MuScleFitUtils::parBias.size()!=2) || // linear in pt
      (MuScleFitUtils::BiasType==2  && MuScleFitUtils::parBias.size()!=2) || // linear in |eta|
      (MuScleFitUtils::BiasType==3  && MuScleFitUtils::parBias.size()!=4) || // sinusoidal in phi
      (MuScleFitUtils::BiasType==4  && MuScleFitUtils::parBias.size()!=3) || // linear in pt and |eta|
      (MuScleFitUtils::BiasType==5  && MuScleFitUtils::parBias.size()!=3) || // linear in pt and sinusoidal in phi
      (MuScleFitUtils::BiasType==6  && MuScleFitUtils::parBias.size()!=3) || // linear in |eta| and sinusoidal in phi
      (MuScleFitUtils::BiasType==7  && MuScleFitUtils::parBias.size()!=4) || // linear in pt and |eta| and sinusoidal in phi
      (MuScleFitUtils::BiasType==8  && MuScleFitUtils::parBias.size()!=4) || // linear in pt and parabolic in |eta|
      (MuScleFitUtils::BiasType==9  && MuScleFitUtils::parBias.size()!=2) || // exponential in pt
      (MuScleFitUtils::BiasType==10 && MuScleFitUtils::parBias.size()!=3) || // parabolic in pt
      (MuScleFitUtils::BiasType==11 && MuScleFitUtils::parBias.size()!=4) || // linear in pt and sin in phi with chg
      (MuScleFitUtils::BiasType==12 && MuScleFitUtils::parBias.size()!=6) || // linear in pt and para in plus sin in phi with chg
      (MuScleFitUtils::BiasType==13 && MuScleFitUtils::parBias.size()!=8) || // linear in pt and para in plus sin in phi with chg
      MuScleFitUtils::BiasType<0 || MuScleFitUtils::BiasType>13) {
    std::cout << "[MuScleFit-Constructor]: Wrong bias type or number of parameters: aborting!" << std::endl;
    abort();
  }
  // Smear parameters: dimension check
  // ---------------------------------
  if ((MuScleFitUtils::SmearType==1 && MuScleFitUtils::parSmear.size()!=3) ||
      (MuScleFitUtils::SmearType==2 && MuScleFitUtils::parSmear.size()!=4) ||
      (MuScleFitUtils::SmearType==3 && MuScleFitUtils::parSmear.size()!=5) ||
      (MuScleFitUtils::SmearType==4 && MuScleFitUtils::parSmear.size()!=6) ||
      (MuScleFitUtils::SmearType==5 && MuScleFitUtils::parSmear.size()!=7) ||
      (MuScleFitUtils::SmearType==6 && MuScleFitUtils::parSmear.size()!=16) ||
      (MuScleFitUtils::SmearType==7 && MuScleFitUtils::parSmear.size()!=0) ||
      MuScleFitUtils::SmearType<0 || MuScleFitUtils::SmearType>7) {
    std::cout << "[MuScleFit-Constructor]: Wrong smear type or number of parameters: aborting!" << std::endl;
    abort();
  }
  // Protect against bad size of parameters
  // --------------------------------------
  if (MuScleFitUtils::parResol.size()!=MuScleFitUtils::parResolFix.size() ||
      MuScleFitUtils::parResol.size()!=MuScleFitUtils::parResolOrder.size()) {
    std::cout << "[MuScleFit-Constructor]: Mismatch in number of parameters for Resol: aborting!" << std::endl;
    abort();
  }
  if (MuScleFitUtils::parScale.size()!=MuScleFitUtils::parScaleFix.size() ||
      MuScleFitUtils::parScale.size()!=MuScleFitUtils::parScaleOrder.size()) {
    std::cout << "[MuScleFit-Constructor]: Mismatch in number of parameters for Scale: aborting!" << std::endl;
    abort();
  }
  if (MuScleFitUtils::parCrossSection.size()!=MuScleFitUtils::parCrossSectionFix.size() ||
      MuScleFitUtils::parCrossSection.size()!=MuScleFitUtils::parCrossSectionOrder.size()) {
    std::cout << "[MuScleFit-Constructor]: Mismatch in number of parameters for Bgr: aborting!" << std::endl;
    abort();
  }
  if (MuScleFitUtils::parBgr.size()!=MuScleFitUtils::parBgrFix.size() ||
      MuScleFitUtils::parBgr.size()!=MuScleFitUtils::parBgrOrder.size()) {
    std::cout << "[MuScleFit-Constructor]: Mismatch in number of parameters for Bgr: aborting!" << std::endl;
    abort();
  }

  // Protect against an incorrect number of resonances
  // -------------------------------------------------
  if (MuScleFitUtils::resfind.size()!=6) {
    std::cout << "[MuScleFit-Constructor]: resfind must have 6 elements (1 Z, 3 Y, 2 Psi): aborting!" << std::endl;
    abort();
  }
}

bool MuScleFit::selGlobalMuon(const pat::Muon* aMuon) {

  reco::TrackRef iTrack = aMuon->innerTrack();
  const reco::HitPattern& p = iTrack->hitPattern();

  reco::TrackRef gTrack = aMuon->globalTrack();
  const reco::HitPattern& q = gTrack->hitPattern();

  return (//isMuonInAccept(aMuon) &&// no acceptance cuts!
      iTrack->found() > 11 &&
      gTrack->chi2()/gTrack->ndof() < 20.0 &&
          q.numberOfValidMuonHits() > 0 &&
          iTrack->chi2()/iTrack->ndof() < 4.0 &&
      aMuon->muonID("TrackerMuonArbitrated") &&
      aMuon->muonID("TMLastStationAngTight") &&
          p.pixelLayersWithMeasurement() > 1 &&
      fabs(iTrack->dxy()) < 3.0 &&  //should be done w.r.t. PV!
          fabs(iTrack->dz()) < 15.0 );//should be done w.r.t. PV!
}
 

bool MuScleFit::selTrackerMuon(const pat::Muon* aMuon) {
  
  reco::TrackRef iTrack = aMuon->innerTrack();
  const reco::HitPattern& p = iTrack->hitPattern();

    return (//isMuonInAccept(aMuon) // no acceptance cuts!
      iTrack->found() > 11 &&
      iTrack->chi2()/iTrack->ndof() < 4.0 &&
      aMuon->muonID("TrackerMuonArbitrated") &&
      aMuon->muonID("TMLastStationAngTight") &&
          p.pixelLayersWithMeasurement() > 1 &&
      fabs(iTrack->dxy()) < 3.0 && //should be done w.r.t. PV!
          fabs(iTrack->dz()) < 15.0 );//should be done w.r.t. PV!
 
}


#include "FWCore/Framework/interface/MakerMacros.h"
DEFINE_FWK_LOOPER(MuScleFit);