CMS 3D CMS Logo

Classes | Public Member Functions | Private Member Functions | Private Attributes

TtFullLepKinSolver Class Reference

#include <TtFullLepKinSolver.h>

List of all members.

Classes

struct  NeutrinoSolution

Public Member Functions

TtDilepEvtSolution addKinSolInfo (TtDilepEvtSolution *asol)
NeutrinoSolution getNuSolution (TLorentzVector LV_l, TLorentzVector LV_l_, TLorentzVector LV_b, TLorentzVector LV_b_)
void SetConstraints (double xx=0, double yy=0)
 TtFullLepKinSolver (double, double, double, std::vector< double >)
 TtFullLepKinSolver ()
void useWeightFromMC (bool useMC)
 ~TtFullLepKinSolver ()

Private Member Functions

int cubic (double *c_coeff, double *c_sol)
void FindCoeff (const TLorentzVector al, const TLorentzVector l, const TLorentzVector b_al, const TLorentzVector b_l, double mt, double mat, double pxboost, double pyboost, double *q_coeff)
int quartic (double *q_coeff, double *q_sol)
double sqr (double x)
void SWAP (double &realone, double &realtwo)
void TopRec (const TLorentzVector al, const TLorentzVector l, const TLorentzVector b_al, const TLorentzVector b_l, double sol)
double WeightSolfromMC ()
double WeightSolfromShape ()

Private Attributes

double C
double D
TF2 * EventShape_
double F
TLorentzVector genLV_n
TLorentzVector genLV_n_
double k16
double k26
double k36
double k46
double k51
double k56
double k61
TLorentzVector LV_n
TLorentzVector LV_n_
TLorentzVector LV_t
TLorentzVector LV_t_
TLorentzVector LV_tt_t
TLorentzVector LV_tt_t_
double m1
double m2
double m3
double mab
double maw
double mb
double mw
double n1
double n2
double n3
double pom
double pxmiss_
double pymiss_
double topmass_begin
double topmass_end
double topmass_step
bool useMCforBest_

Detailed Description

Definition at line 13 of file TtFullLepKinSolver.h.


Constructor & Destructor Documentation

TtFullLepKinSolver::TtFullLepKinSolver ( )

Definition at line 5 of file TtFullLepKinSolver.cc.

References EventShape_.

{
  // That crude parametrisation has been obtained from a fit of O(1000) pythia events.
  // It is normalized to 1.
  EventShape_ = new TF2("landau2D","[0]*TMath::Landau(x,[1],[2],0)*TMath::Landau(y,[3],[4],0)",0,500,0,500);
  EventShape_->SetParameters(30.7137,56.2880,23.0744,59.1015,24.9145);
}
TtFullLepKinSolver::TtFullLepKinSolver ( double  b,
double  e,
double  s,
std::vector< double >  nupars 
)

Definition at line 13 of file TtFullLepKinSolver.cc.

References b, ExpressReco_HICollisions_FallBack::e, EventShape_, mab, maw, mb, mw, pxmiss_, pymiss_, asciidump::s, topmass_begin, topmass_end, and topmass_step.

{
  topmass_begin = b;
  topmass_end = e;
  topmass_step = s;
  pxmiss_ = 0;
  pymiss_ = 0;
  mw = 80.22;
  maw = 80.22;
  mab = 4.800;
  mb = 4.800;
  // That crude parametrisation has been obtained from a fit of O(1000) pythia events.
  // It is normalized to 1.
  EventShape_ = new TF2("landau2D","[0]*TMath::Landau(x,[1],[2],0)*TMath::Landau(y,[3],[4],0)",0,500,0,500);
  EventShape_->SetParameters(nupars[0],nupars[1],nupars[2],nupars[3],nupars[4]);
}
TtFullLepKinSolver::~TtFullLepKinSolver ( )

Definition at line 33 of file TtFullLepKinSolver.cc.

References EventShape_.

{
  delete EventShape_;
}

Member Function Documentation

TtDilepEvtSolution TtFullLepKinSolver::addKinSolInfo ( TtDilepEvtSolution asol)

Definition at line 40 of file TtFullLepKinSolver.cc.

References reco::LeafCandidate::energy(), FindCoeff(), genLV_n, genLV_n_, TtDilepEvtSolution::getCalJetB(), TtDilepEvtSolution::getCalJetBbar(), TtDilepEvtSolution::getElectronm(), TtDilepEvtSolution::getElectronp(), TtDilepEvtSolution::getGenN(), TtDilepEvtSolution::getGenNbar(), TtDilepEvtSolution::getMuonm(), TtDilepEvtSolution::getMuonp(), TtDilepEvtSolution::getTaum(), TtDilepEvtSolution::getTaup(), TtDilepEvtSolution::getWmDecay(), TtDilepEvtSolution::getWpDecay(), reco::LeafCandidate::px(), pxmiss_, reco::LeafCandidate::py(), pymiss_, reco::LeafCandidate::pz(), quartic(), TtDilepEvtSolution::setRecTopMass(), TtDilepEvtSolution::setRecWeightMax(), topmass_begin, topmass_end, topmass_step, TopRec(), useMCforBest_, CommonMethods::weight(), WeightSolfromMC(), and WeightSolfromShape().

Referenced by TtDilepEvtSolutionMaker::produce().

{
  TtDilepEvtSolution fitsol(*asol);
  
  //antilepton and lepton
  TLorentzVector LV_e, LV_e_;
  //b and bbar quark
  TLorentzVector LV_b, LV_b_;
  
  bool hasMCinfo = true;
  if(fitsol.getGenN()) { // protect against non-dilept genevents
    genLV_n = TLorentzVector(fitsol.getGenN()->px(), fitsol.getGenN()->py(),
                             fitsol.getGenN()->pz(), fitsol.getGenN()->energy());
  } else hasMCinfo = false;

  if(fitsol.getGenNbar()) { // protect against non-dilept genevents
    genLV_n_ = TLorentzVector(fitsol.getGenNbar()->px(), fitsol.getGenNbar()->py(),
                              fitsol.getGenNbar()->pz(), fitsol.getGenNbar()->energy());
  } else hasMCinfo = false;
  // if MC is to be used to select the best top mass and is not available,
  // then nothing can be done. Stop here.
  if(useMCforBest_&&!hasMCinfo) return fitsol;

  // first lepton
  if (fitsol.getWpDecay() == "muon") {
    LV_e = TLorentzVector(fitsol.getMuonp().px(), fitsol.getMuonp().py(),
                          fitsol.getMuonp().pz(), fitsol.getMuonp().energy());
  } else if (fitsol.getWpDecay() == "electron") {
    LV_e = TLorentzVector(fitsol.getElectronp().px(), fitsol.getElectronp().py(),
                          fitsol.getElectronp().pz(), fitsol.getElectronp().energy());
  } else if (fitsol.getWpDecay() == "tau") {
    LV_e = TLorentzVector(fitsol.getTaup().px(), fitsol.getTaup().py(),
                          fitsol.getTaup().pz(), fitsol.getTaup().energy());
  }
    
  // second lepton
  if (fitsol.getWmDecay() == "muon") {
    LV_e_ = TLorentzVector(fitsol.getMuonm().px(), fitsol.getMuonm().py(),
                          fitsol.getMuonm().pz(), fitsol.getMuonm().energy());
  } else if (fitsol.getWmDecay() == "electron") {
    LV_e_ = TLorentzVector(fitsol.getElectronm().px(), fitsol.getElectronm().py(),
                           fitsol.getElectronm().pz(), fitsol.getElectronm().energy());
  } else if (fitsol.getWmDecay() == "tau") {
    LV_e_ = TLorentzVector(fitsol.getTaum().px(), fitsol.getTaum().py(),
                           fitsol.getTaum().pz(), fitsol.getTaum().energy());
  }

  // first jet
  LV_b = TLorentzVector(fitsol.getCalJetB().px(), fitsol.getCalJetB().py(),
                        fitsol.getCalJetB().pz(), fitsol.getCalJetB().energy());

  // second jet
  LV_b_ = TLorentzVector(fitsol.getCalJetBbar().px(), fitsol.getCalJetBbar().py(),
                         fitsol.getCalJetBbar().pz(), fitsol.getCalJetBbar().energy());
  
  //loop on top mass parameter
  double weightmax = -1e30;
  double mtmax = 0;
  for (double mt = topmass_begin; 
              mt < topmass_end + 0.5*topmass_step; 
              mt += topmass_step) {
    //cout << "mt = " << mt << endl;
    double q_coeff[5], q_sol[4];
    FindCoeff(LV_e, LV_e_, LV_b, LV_b_, mt, mt, pxmiss_, pymiss_, q_coeff);
    int NSol = quartic(q_coeff, q_sol);
    
    //loop on all solutions
    for (int isol = 0; isol < NSol; isol++) {
      TopRec(LV_e, LV_e_, LV_b, LV_b_, q_sol[isol]);
      double weight = useMCforBest_ ? WeightSolfromMC() : WeightSolfromShape();
      if (weight > weightmax) {
        weightmax =weight;
        mtmax = mt;
      }
    }
    
    //for (int i=0;i<5;i++) cout << " q_coeff["<<i<< "]= " << q_coeff[i];
    //cout << endl;
    
    //for (int i=0;i<4;i++) cout << " q_sol["<<i<< "]= " << q_sol[i];
    //cout << endl;
    //cout << "NSol_" << NSol << endl;
  }
  
  fitsol.setRecTopMass(mtmax);
  fitsol.setRecWeightMax(weightmax);
  
  return fitsol;
}
int TtFullLepKinSolver::cubic ( double *  c_coeff,
double *  c_sol 
) [private]

Definition at line 378 of file TtFullLepKinSolver.cc.

References funct::cos(), h, i, L1TEmulatorMonitor_cff::p, phi, Pi, lumiQueryAPI::q, sqr(), mathSSE::sqrt(), and SWAP().

Referenced by quartic().

                                                            {
  int i, nreal;
  double w, p, q, dis, h, phi;
  
  if (coeffs[3]!=0.0) {
    /* cubic problem? */
    w = coeffs[2]/(3*coeffs[3]);
    p = sqr(coeffs[1]/(3*coeffs[3])-sqr(w))*(coeffs[1]/(3*coeffs[3])-sqr(w));
    q = -0.5*(2*sqr(w)*w-(coeffs[1]*w-coeffs[0])/coeffs[3]);
    dis = sqr(q)+p;
    /* discriminant */
    if (dis<0.0) {
      /* 3 real solutions */
      h = q/sqrt(-p);
      if (h>1.0) h = 1.0;
      /* confine the argument of */
      if (h<-1.0) h = -1.0;
      /* acos to [-1;+1] */
      phi = acos(h);
      p = 2*TMath::Power(-p, 1.0/6.0);
      for (i=0; i<3; i++) 
        koreny[i] = p*cos((phi+2*i*TMath::Pi())/3.0) - w;
      if (koreny[1]<koreny[0]) SWAP(koreny[0], koreny[1]);
      /* sort results */
      if (koreny[2]<koreny[1]) SWAP(koreny[1], koreny[2]);
      if (koreny[1]<koreny[0]) SWAP(koreny[0], koreny[1]);
      nreal = 3;
    }
    else {
      /* only one real solution */
      dis = sqrt(dis);
      h = TMath::Power(fabs(q+dis), 1.0/3.0);
      p = TMath::Power(fabs(q-dis), 1.0/3.0);
      koreny[0] = ((q+dis>0.0)? h : -h) + ((q-dis>0.0)? p : -p) -  w;
      nreal = 1;
    }

    /* Perform one step of a Newton iteration in order to minimize
       round-off errors */
    for (i=0; i<nreal; i++) {
      h = coeffs[1] + koreny[i] * (2 * coeffs[2] + 3 * koreny[i] * coeffs[3]);
      if (h != 0.0)
        koreny[i] -= (coeffs[0] + koreny[i] * (coeffs[1] + koreny[i] * (coeffs[2] + koreny[i] * coeffs[3])))/h;
    }
  }

  else if (coeffs[2]!=0.0) {
    /* quadratic problem? */
    p = 0.5*coeffs[1]/coeffs[2];
    dis = sqr(p) - coeffs[0]/coeffs[2];
    if (dis>=0.0) {
      /* two real solutions */
      dis = sqrt(dis);
      koreny[0] = -p - dis;
      koreny[1] = -p + dis;
      nreal = 2;
    }
    else
      /* no real solution */
      nreal = 0;
  }

  else if (coeffs[1]!=0.0) {
    /* linear problem? */
    koreny[0] = -coeffs[0]/coeffs[1];
    nreal = 1;
  }

  else
    /* no equation */
    nreal = 0;
  
  return nreal;
}
void TtFullLepKinSolver::FindCoeff ( const TLorentzVector  al,
const TLorentzVector  l,
const TLorentzVector  b_al,
const TLorentzVector  b_l,
double  mt,
double  mat,
double  pxboost,
double  pyboost,
double *  q_coeff 
) [private]

Definition at line 171 of file TtFullLepKinSolver.cc.

References C, D, F, k16, k26, k36, k46, k51, k56, k61, m1, m2, m3, mab, maw, mb, mw, n1, n2, n3, pom, funct::pow(), and sqr().

Referenced by addKinSolInfo(), and getNuSolution().

                                                        {

double E, apom1, apom2, apom3;
double k11, k21, k31, k41,  cpom1, cpom2, cpom3, l11, l21, l31, l41, l51, l61, k1, k2, k3, k4, k5,k6;
double l1, l2, l3, l4, l5, l6, k15, k25, k35, k45;

  C = -al.Px()-b_al.Px()-l.Px()-b_l.Px() + px_miss;
  D = -al.Py()-b_al.Py()-l.Py()-b_l.Py() + py_miss;

  // right side offirst two linear equations - missing pT

  
  E = (sqr(mt)-sqr(mw)-sqr(mb))/(2*b_al.E())-sqr(mw)/(2*al.E())-al.E()+al.Px()*b_al.Px()/b_al.E()+al.Py()*b_al.Py()/b_al.E()+al.Pz()*b_al.Pz()/b_al.E();
  F = (sqr(mat)-sqr(maw)-sqr(mab))/(2*b_l.E())-sqr(maw)/(2*l.E())-l.E()+l.Px()*b_l.Px()/b_l.E()+l.Py()*b_l.Py()/b_l.E()+l.Pz()*b_l.Pz()/b_l.E();
  
  m1 = al.Px()/al.E()-b_al.Px()/b_al.E();
  m2 = al.Py()/al.E()-b_al.Py()/b_al.E();
  m3 = al.Pz()/al.E()-b_al.Pz()/b_al.E();
  
  n1 = l.Px()/l.E()-b_l.Px()/b_l.E();
  n2 = l.Py()/l.E()-b_l.Py()/b_l.E();
  n3 = l.Pz()/l.E()-b_l.Pz()/b_l.E();
  
  pom = E-m1*C-m2*D;
  apom1 = sqr(al.Px())-sqr(al.E());
  apom2 = sqr(al.Py())-sqr(al.E());
  apom3 = sqr(al.Pz())-sqr(al.E());
  
  k11 = 1/sqr(al.E())*(pow(mw,4)/4+sqr(C)*apom1+sqr(D)*apom2+apom3*sqr(pom)/sqr(m3)+sqr(mw)*(al.Px()*C+al.Py()*D+al.Pz()*pom/m3)+2*al.Px()*al.Py()*C*D+2*al.Px()*al.Pz()*C*pom/m3+2*al.Py()*al.Pz()*D*pom/m3);
  k21 = 1/sqr(al.E())*(-2*C*m3*n3*apom1+2*apom3*n3*m1*pom/m3-sqr(mw)*m3*n3*al.Px()+sqr(mw)*m1*n3*al.Pz()-2*al.Px()*al.Py()*D*m3*n3+2*al.Px()*al.Pz()*C*m1*n3-2*al.Px()*al.Pz()*n3*pom+2*al.Py()*al.Pz()*D*m1*n3);
  k31 = 1/sqr(al.E())*(-2*D*m3*n3*apom2+2*apom3*n3*m2*pom/m3-sqr(mw)*m3*n3*al.Py()+sqr(mw)*m2*n3*al.Pz()-2*al.Px()*al.Py()*C*m3*n3+2*al.Px()*al.Pz()*C*m2*n3-2*al.Py()*al.Pz()*n3*pom+2*al.Py()*al.Pz()*D*m2*n3);
  k41 = 1/sqr(al.E())*(2*apom3*m1*m2*sqr(n3)+2*al.Px()*al.Py()*sqr(m3)*sqr(n3)-2*al.Px()*al.Pz()*m2*m3*sqr(n3)-2*al.Py()*al.Pz()*m1*m3*sqr(n3));
  k51 = 1/sqr(al.E())*(apom1*sqr(m3)*sqr(n3)+apom3*sqr(m1)*sqr(n3)-2*al.Px()*al.Pz()*m1*m3*sqr(n3));
  k61 = 1/sqr(al.E())*(apom2*sqr(m3)*sqr(n3)+apom3*sqr(m2)*sqr(n3)-2*al.Py()*al.Pz()*m2*m3*sqr(n3));
  
  cpom1 = sqr(l.Px())-sqr(l.E());
  cpom2 = sqr(l.Py())-sqr(l.E());
  cpom3 = sqr(l.Pz())-sqr(l.E());
  
  l11 = 1/sqr(l.E())*(pow(maw,4)/4+cpom3*sqr(F)/sqr(n3)+sqr(maw)*l.Pz()*F/n3);
  l21 = 1/sqr(l.E())*(-2*cpom3*F*m3*n1/n3+sqr(maw)*(l.Px()*m3*n3-l.Pz()*n1*m3)+2*l.Px()*l.Pz()*F*m3);
  l31 = 1/sqr(l.E())*(-2*cpom3*F*m3*n2/n3+sqr(maw)*(l.Py()*m3*n3-l.Pz()*n2*m3)+2*l.Py()*l.Pz()*F*m3);
  l41 = 1/sqr(l.E())*(2*cpom3*n1*n2*sqr(m3)+2*l.Px()*l.Py()*sqr(m3)*sqr(n3)-2*l.Px()*l.Pz()*n2*n3*sqr(m3)-2*l.Py()*l.Pz()*n1*n3*sqr(m3));
  l51 = 1/sqr(l.E())*(cpom1*sqr(m3)*sqr(n3)+cpom3*sqr(n1)*sqr(m3)-2*l.Px()*l.Pz()*n1*n3*sqr(m3));
  l61 = 1/sqr(l.E())*(cpom2*sqr(m3)*sqr(n3)+cpom3*sqr(n2)*sqr(m3)-2*l.Py()*l.Pz()*n2*n3*sqr(m3));
  
  k1 = k11*k61;
  k2 = k61*k21/k51;
  k3 = k31;
  k4 = k41/k51;
  k5 = k61/k51;
  k6 = 1;
  
  l1 = l11*k61;
  l2 = l21*k61/k51;
  l3 = l31;
  l4 = l41/k51;
  l5 = l51*k61/(sqr(k51));
  l6 = l61/k61;
  
  k15 = k1*l5-l1*k5;
  k25 = k2*l5-l2*k5;
  k35 = k3*l5-l3*k5;
  k45 = k4*l5-l4*k5;
  
  k16 = k1*l6-l1*k6;
  k26 = k2*l6-l2*k6;
  k36 = k3*l6-l3*k6;
  k46 = k4*l6-l4*k6;
  k56 = k5*l6-l5*k6;
  

  koeficienty[0] = k15*sqr(k36)-k35*k36*k16-k56*sqr(k16);
  koeficienty[1] = 2*k15*k36*k46+k25*sqr(k36)+k35*(-k46*k16-k36*k26)-k45*k36*k16-2*k56*k26*k16;
  koeficienty[2] = k15*sqr(k46)+2*k25*k36*k46+k35*(-k46*k26-k36*k56)-k56*(sqr(k26)+2*k56*k16)-k45*(k46*k16+k36*k26);
  koeficienty[3] = k25*sqr(k46)-k35*k46*k56-k45*(k46*k26+k36*k56)-2*sqr(k56)*k26;
  koeficienty[4] = -k45*k46*k56-pow(k56,3);
  
  // normalization of coefficients
  int moc=(int(log10(fabs(koeficienty[0])))+int(log10(fabs(koeficienty[4]))))/2;
  
  koeficienty[0]=koeficienty[0]/TMath::Power(10,moc);
  koeficienty[1]=koeficienty[1]/TMath::Power(10,moc);
  koeficienty[2]=koeficienty[2]/TMath::Power(10,moc);
  koeficienty[3]=koeficienty[3]/TMath::Power(10,moc);
  koeficienty[4]=koeficienty[4]/TMath::Power(10,moc);
}
TtFullLepKinSolver::NeutrinoSolution TtFullLepKinSolver::getNuSolution ( TLorentzVector  LV_l,
TLorentzVector  LV_l_,
TLorentzVector  LV_b,
TLorentzVector  LV_b_ 
)

Definition at line 136 of file TtFullLepKinSolver.cc.

References FindCoeff(), LV_n, LV_n_, TtFullLepKinSolver::NeutrinoSolution::neutrino, TtFullLepKinSolver::NeutrinoSolution::neutrinoBar, pxmiss_, pymiss_, quartic(), topmass_begin, topmass_end, topmass_step, TopRec(), CommonMethods::weight(), TtFullLepKinSolver::NeutrinoSolution::weight, and WeightSolfromShape().

Referenced by TtFullLepKinSolutionProducer::produce().

                                                                                             {                                                                                                                      
    
  math::XYZTLorentzVector maxLV_n  = math::XYZTLorentzVector(0,0,0,0); 
  math::XYZTLorentzVector maxLV_n_ = math::XYZTLorentzVector(0,0,0,0);   

  //loop on top mass parameter
  double weightmax = -1;
  for(double mt = topmass_begin; 
             mt < topmass_end + 0.5*topmass_step; 
             mt += topmass_step) {
    double q_coeff[5], q_sol[4];
    FindCoeff(LV_l, LV_l_, LV_b, LV_b_, mt, mt, pxmiss_, pymiss_, q_coeff);
    int NSol = quartic(q_coeff, q_sol);
    
    //loop on all solutions
    for (int isol = 0; isol < NSol; isol++) {
      TopRec(LV_l, LV_l_, LV_b, LV_b_, q_sol[isol]);
      double weight = WeightSolfromShape();
      if (weight > weightmax) {
        weightmax =weight;
        maxLV_n.SetPxPyPzE(LV_n.Px(), LV_n.Py(), LV_n.Pz(), LV_n.E());
        maxLV_n_.SetPxPyPzE(LV_n_.Px(), LV_n_.Py(), LV_n_.Pz(), LV_n_.E());
      }
    }
  }
  TtFullLepKinSolver::NeutrinoSolution nuSol;
  nuSol.neutrino    = reco::LeafCandidate(0, maxLV_n  );
  nuSol.neutrinoBar = reco::LeafCandidate(0, maxLV_n_ ); 
  nuSol.weight = weightmax; 
  return nuSol;
}
int TtFullLepKinSolver::quartic ( double *  q_coeff,
double *  q_sol 
) [private]

Definition at line 317 of file TtFullLepKinSolver.cc.

References trackerHits::c, cubic(), debug_cff::d0, debug_cff::d1, h, i, sqr(), mathSSE::sqrt(), matplotRender::t, and z.

Referenced by addKinSolInfo(), and getNuSolution().

                                                                   {
  int i, nreal;
  double w, b0, b1, b2;
  double c[4];
  double d0, d1, h, t, z;
  double *px;

  
  if (koeficienty[4]==0.0) 
    return cubic(koeficienty, koreny);
  /* quartic problem? */
  w = koeficienty[3]/(4*koeficienty[4]);
  /* offset */
  b2 = -6*sqr(w) + koeficienty[2]/koeficienty[4];
  /* koeficienty. of shifted polynomial */
  b1 = (8*sqr(w) - 2*koeficienty[2]/koeficienty[4])*w + koeficienty[1]/koeficienty[4];
  b0 = ((-3*sqr(w) + koeficienty[2]/koeficienty[4])*w - koeficienty[1]/koeficienty[4])*w + koeficienty[0]/koeficienty[4];

  c[3] = 1.0;
  /* cubic resolvent */
  c[2] = b2;
  c[1] = -4*b0;
  c[0] = sqr(b1) - 4*b0*b2;
  
  
  i = cubic(c, koreny);
  z = koreny[0];
  //double z1=1.0,z2=2.0,z3=3.0;
  //TMath::RootsCubic(c,z1,z2,z3);
  //if (z2 !=0) z = z2;
  //if (z1 !=0) z = z1;
  /* only lowermost root needed */

  nreal = 0;
  px = koreny;
  t = sqrt(0.25*sqr(z) - b0);
  for (i=-1; i<=1; i+=2) {
    d0 = -0.5*z + i*t;
    /* coeffs. of quadratic factor */
    d1 = (t!=0.0)? -i*0.5*b1/t : i*sqrt(-z - b2);
    h = 0.25*sqr(d1) - d0;
    if (h>=0.0) {
      h = sqrt(h);
      nreal += 2;
      *px++ = -0.5*d1 - h - w;
      *px++ = -0.5*d1 + h - w;
    }
  }

  if (nreal==4) {
    /* sort results */
//    if (koreny[2]<koreny[0]) SWAP(koreny[0], koreny[2]);
//    if (koreny[3]<koreny[1]) SWAP(koreny[1], koreny[3]);
//    if (koreny[1]<koreny[0]) SWAP(koreny[0], koreny[1]);
//    if (koreny[3]<koreny[2]) SWAP(koreny[2], koreny[3]);
//    if (koreny[2]<koreny[1]) SWAP(koreny[1], koreny[2]);
  }
  return nreal;

}
void TtFullLepKinSolver::SetConstraints ( double  xx = 0,
double  yy = 0 
)

Definition at line 130 of file TtFullLepKinSolver.cc.

References pxmiss_, and pymiss_.

Referenced by TtDilepEvtSolutionMaker::produce(), and TtFullLepKinSolutionProducer::produce().

{
  pxmiss_ = xx;
  pymiss_ = yy;
}
double TtFullLepKinSolver::sqr ( double  x) [inline, private]

Definition at line 55 of file TtFullLepKinSolver.h.

Referenced by cubic(), FindCoeff(), and quartic().

{return (x*x);}
void TtFullLepKinSolver::SWAP ( double &  realone,
double &  realtwo 
) [private]

Definition at line 454 of file TtFullLepKinSolver.cc.

References printConversionInfo::aux.

Referenced by cubic().

                                                               {
  if (realtwo < realone) {
    double aux = realtwo;
    realtwo = realone;
    realone = aux;
  }
}
void TtFullLepKinSolver::TopRec ( const TLorentzVector  al,
const TLorentzVector  l,
const TLorentzVector  b_al,
const TLorentzVector  b_l,
double  sol 
) [private]

Definition at line 267 of file TtFullLepKinSolver.cc.

References printConversionInfo::aux, C, D, F, k16, k26, k36, k46, k51, k56, k61, LV_n, LV_n_, LV_t, LV_t_, LV_tt_t, LV_tt_t_, m1, m2, m3, n1, n2, n3, pom, and funct::pow().

Referenced by addKinSolInfo(), and getNuSolution().

                                            {

  TVector3 t_ttboost;
  TLorentzVector aux;
  double pxp, pyp, pzp, pup, pvp, pwp;
  
    
  pxp = sol*(m3*n3/k51);   
  pyp = -(m3*n3/k61)*(k56*pow(sol,2) + k26*sol + k16)/(k36 + k46*sol);
  pzp = -1/n3*(n1*pxp + n2*pyp - F);
  pwp = 1/m3*(m1*pxp + m2*pyp + pom);
  pup = C - pxp;
  pvp = D - pyp;
     
  LV_n_.SetXYZM(pxp, pyp, pzp, 0.0);
  LV_n.SetXYZM(pup, pvp, pwp, 0.0);
      
  
  LV_t_ = b_l + l + LV_n_;
  LV_t = b_al + al + LV_n;  
  
  
  aux = (LV_t_ + LV_t);
  t_ttboost = -aux.BoostVector();
  LV_tt_t_ = LV_t_;
  LV_tt_t = LV_t;
  LV_tt_t_.Boost(t_ttboost);
  LV_tt_t.Boost(t_ttboost); 
}
void TtFullLepKinSolver::useWeightFromMC ( bool  useMC) [inline]
double TtFullLepKinSolver::WeightSolfromMC ( ) [private]

Definition at line 301 of file TtFullLepKinSolver.cc.

References genLV_n, genLV_n_, LV_n, LV_n_, and CommonMethods::weight().

Referenced by addKinSolInfo().

                                           {

  double weight = 1;
  weight = ((LV_n.E() > genLV_n.E())? genLV_n.E()/LV_n.E(): LV_n.E()/genLV_n.E())
           *((LV_n_.E() > genLV_n_.E())? genLV_n_.E()/LV_n_.E(): LV_n_.E()/genLV_n_.E());
  return weight;

}
double TtFullLepKinSolver::WeightSolfromShape ( ) [private]

Definition at line 310 of file TtFullLepKinSolver.cc.

References EventShape_, LV_n, and LV_n_.

Referenced by addKinSolInfo(), and getNuSolution().

                                              {
  
  // Use the parametrized event shape to obtain the solution weight.
  return EventShape_->Eval(LV_n.E(),LV_n_.E());

}

Member Data Documentation

double TtFullLepKinSolver::C [private]

Definition at line 66 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::D [private]

Definition at line 67 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::F [private]

Definition at line 68 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

TLorentzVector TtFullLepKinSolver::genLV_n [private]

Definition at line 86 of file TtFullLepKinSolver.h.

Referenced by addKinSolInfo(), and WeightSolfromMC().

TLorentzVector TtFullLepKinSolver::genLV_n_ [private]

Definition at line 86 of file TtFullLepKinSolver.h.

Referenced by addKinSolInfo(), and WeightSolfromMC().

double TtFullLepKinSolver::k16 [private]

Definition at line 70 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::k26 [private]

Definition at line 71 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::k36 [private]

Definition at line 72 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::k46 [private]

Definition at line 73 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::k51 [private]

Definition at line 75 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::k56 [private]

Definition at line 74 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::k61 [private]

Definition at line 76 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

TLorentzVector TtFullLepKinSolver::LV_n [private]

Definition at line 84 of file TtFullLepKinSolver.h.

Referenced by getNuSolution(), TopRec(), WeightSolfromMC(), and WeightSolfromShape().

TLorentzVector TtFullLepKinSolver::LV_n_ [private]

Definition at line 84 of file TtFullLepKinSolver.h.

Referenced by getNuSolution(), TopRec(), WeightSolfromMC(), and WeightSolfromShape().

TLorentzVector TtFullLepKinSolver::LV_t [private]

Definition at line 84 of file TtFullLepKinSolver.h.

Referenced by TopRec().

TLorentzVector TtFullLepKinSolver::LV_t_ [private]

Definition at line 84 of file TtFullLepKinSolver.h.

Referenced by TopRec().

TLorentzVector TtFullLepKinSolver::LV_tt_t [private]

Definition at line 84 of file TtFullLepKinSolver.h.

Referenced by TopRec().

TLorentzVector TtFullLepKinSolver::LV_tt_t_ [private]

Definition at line 84 of file TtFullLepKinSolver.h.

Referenced by TopRec().

double TtFullLepKinSolver::m1 [private]

Definition at line 77 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::m2 [private]

Definition at line 78 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::m3 [private]

Definition at line 79 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::mab [private]

Definition at line 64 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TtFullLepKinSolver().

double TtFullLepKinSolver::maw [private]

Definition at line 64 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TtFullLepKinSolver().

double TtFullLepKinSolver::mb [private]

Definition at line 64 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TtFullLepKinSolver().

double TtFullLepKinSolver::mw [private]

Definition at line 64 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TtFullLepKinSolver().

double TtFullLepKinSolver::n1 [private]

Definition at line 80 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::n2 [private]

Definition at line 81 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::n3 [private]

Definition at line 82 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::pom [private]

Definition at line 69 of file TtFullLepKinSolver.h.

Referenced by FindCoeff(), and TopRec().

double TtFullLepKinSolver::pxmiss_ [private]
double TtFullLepKinSolver::pymiss_ [private]

Definition at line 59 of file TtFullLepKinSolver.h.

Referenced by addKinSolInfo(), getNuSolution(), and TtFullLepKinSolver().

Definition at line 60 of file TtFullLepKinSolver.h.

Referenced by addKinSolInfo(), getNuSolution(), and TtFullLepKinSolver().

Definition at line 61 of file TtFullLepKinSolver.h.

Referenced by addKinSolInfo(), getNuSolution(), and TtFullLepKinSolver().

Definition at line 89 of file TtFullLepKinSolver.h.

Referenced by addKinSolInfo(), and useWeightFromMC().