genf::GFKalman Class Reference

#include <GFKalman.h>

Public Member Functions
	GFKalman ()
	~GFKalman ()
void	operator() (GFTrack *track)
	Operator for use with STL. More...
void	operator() (std::pair< int, GFTrack * > tr)
	Operator for use with STL. More...
void	setLazy (Int_t)
	Switch lazy error handling. More...
void	setNumIterations (Int_t i)
	Set number of iterations for Kalman Filter. More...
void	processTrack (GFTrack *)
	Performs fit on a GFTrack. More...
void	fittingPass (GFTrack *, int dir)
	Performs fit on a GFTrack beginning with the current hit. More...
double	getChi2Hit (GFAbsRecoHit *, GFAbsTrackRep *)
	Calculates chi2 of a given hit with respect to a given track representation. More...
void	setInitialDirection (int d)
	Sets the inital direction of the track fit (1 for inner to outer, or -1 for outer to inner). The standard is 1 and is set in the ctor. More...
void	setBlowUpFactor (double f)
	Set the blowup factor (see blowUpCovs() ) More...
void	setMomLow (Double_t f)
void	setMomHigh (Double_t f)
void	setMaxUpdate (Double_t f)
void	setErrorScaleSTh (Double_t f)
void	setErrorScaleMTh (Double_t f)

Private Member Functions
void	processHit (GFTrack *, int, int, int)
	One Kalman step. More...
void	switchDirection (GFTrack *trk)
	Used to switch between forward and backward filtering. More...
TMatrixT< Double_t >	calcGain (const TMatrixT< Double_t > &cov, const TMatrixT< Double_t > &HitCov, const TMatrixT< Double_t > &H)
	Calculate Kalman Gain. More...
TMatrixT< Double_t >	calcCov7x7 (const TMatrixT< Double_t > &cov, const genf::GFDetPlane &plane)
double	chi2Increment (const TMatrixT< Double_t > &r, const TMatrixT< Double_t > &H, const TMatrixT< Double_t > &cov, const TMatrixT< Double_t > &V)
	this returns the reduced chi2 increment for a hit More...
void	blowUpCovs (GFTrack *trk)
	this is needed to blow up the covariance matrix before a fitting pass drops off-diagonal elements and blows up diagonal by blowUpFactor More...
void	blowUpCovsDiag (GFTrack *trk)

Private Attributes
int	fInitialDirection
Int_t	fNumIt
double	fBlowUpFactor
Double_t	fMomLow
Double_t	fMomHigh
Double_t	fMaxUpdate
Double_t	fErrScaleSTh
Double_t	fErrScaleMTh
bool	fGENfPRINT

Detailed Description

Definition at line 53 of file GFKalman.h.

Constructor & Destructor Documentation

genf::GFKalman::GFKalman

(

)

Definition at line 39 of file GFKalman.cxx.

                      :fInitialDirection(1),fNumIt(3),fBlowUpFactor(50.),fMomLow(-100.0),fMomHigh(100.0),fMaxUpdate(1.0),fErrScaleSTh(1.0),fErrScaleMTh(1.0), fGENfPRINT(false)
{
  art::ServiceHandle<art::TFileService const> tfs;
}

genf::GFKalman::~GFKalman

(

)

Definition at line 44 of file GFKalman.cxx.

{;}

Member Function Documentation

void genf::GFKalman::blowUpCovs ( GFTrack *  trk )

private

this is needed to blow up the covariance matrix before a fitting pass drops off-diagonal elements and blows up diagonal by blowUpFactor

Definition at line 154 of file GFKalman.cxx.

                                         {
  int nreps=trk->getNumReps();
  for(int irep=0; irep<nreps; ++irep){
    GFAbsTrackRep* arep=trk->getTrackRep(irep);
    //dont do it for already compromsied reps, since they wont be fitted anyway
    if(arep->getStatusFlag()==0) {
      TMatrixT<Double_t> cov = arep->getCov();
      for(int i=0;i<cov.GetNrows();++i){
        for(int j=0;j<cov.GetNcols();++j){
          //if(i!=j){//off diagonal
          //  cov[i][j]=0.;
          //}
          //else{//diagonal
          cov[i][j] = cov[i][j] * fBlowUpFactor;
          //}
        }
      }
      arep->setCov(cov);
    }
  }
}

void genf::GFKalman::blowUpCovsDiag ( GFTrack *  trk )

private

Definition at line 132 of file GFKalman.cxx.

                                             {
  int nreps=trk->getNumReps();
  for(int irep=0; irep<nreps; ++irep){
    GFAbsTrackRep* arep=trk->getTrackRep(irep);
    //dont do it for already compromsied reps, since they wont be fitted anyway
    if(arep->getStatusFlag()==0) {
      TMatrixT<Double_t> cov = arep->getCov();
      for(int i=0;i<cov.GetNrows();++i){
        for(int j=0;j<cov.GetNcols();++j){
          if(i!=j){//off diagonal
            cov[i][j]=0.;
          }
          else{//diagonal
            cov[i][j] = cov[i][j] * fBlowUpFactor;
            //      cov[0][0] = 0.1;
          }
        }
      }
      arep->setCov(cov);
    }
  }
}

TMatrixT< Double_t > genf::GFKalman::calcCov7x7 ( const TMatrixT< Double_t > &  cov, const genf::GFDetPlane &  plane  )

private

Definition at line 671 of file GFKalman.cxx.

{
  // This ends up, confusingly, as: 7 columns, 5 rows!
  TMatrixT<Double_t> jac(7,5); // X,Y,Z,UX,UY,UZ,Theta in detector coords

  TVector3 u=plane.getU();
  TVector3 v=plane.getV();
  TVector3 w=u.Cross(v);

  // Below line has been done, with code change in GFKalman that has updated
  // the plane orientation by now.
  //  TVector3 pTilde = 1.0 * (w + state[1][0] * u + state[2][0] * v);
  TVector3 pTilde = w;
  double pTildeMag = pTilde.Mag();

  jac.Zero();
  jac[6][0] = 1.; //  Should be C as in GFSpacepointHitPolicy. 16-Feb-2013.

  jac[0][3] = u[0];
  jac[1][3] = u[1];
  jac[2][3] = u[2];

  jac[0][4] = v[0];
  jac[1][4] = v[1];
  jac[2][4] = v[2];

  // cnp'd from RKTrackRep.cxx, line ~496
  // da{x,y,z}/du'
  jac[3][1] = 1.0/pTildeMag*(u.X()-pTilde.X()/(pTildeMag*pTildeMag)*u*pTilde);
  jac[4][1] = 1.0/pTildeMag*(u.Y()-pTilde.Y()/(pTildeMag*pTildeMag)*u*pTilde);
  jac[5][1] = 1.0/pTildeMag*(u.Z()-pTilde.Z()/(pTildeMag*pTildeMag)*u*pTilde);
  // da{x,y,z}/dv'
  jac[3][2] = 1.0/pTildeMag*(v.X()-pTilde.X()/(pTildeMag*pTildeMag)*v*pTilde);
  jac[4][2] = 1.0/pTildeMag*(v.Y()-pTilde.Y()/(pTildeMag*pTildeMag)*v*pTilde);
  jac[5][2] = 1.0/pTildeMag*(v.Z()-pTilde.Z()/(pTildeMag*pTildeMag)*v*pTilde);

  // y = A.x => x = A^T.A.A^T.y
  // Thus, y's Jacobians Jac become for x (Jac^T.Jac)^(-1) Jac^T

  TMatrixT<Double_t> jac_t(TMatrixT<Double_t>::kTransposed,jac);
  TMatrixT<Double_t> jjInv(jac_t * jac);
  //jInv.Zero();

  double det(0.0);
  try {
    jjInv.Invert(&det); // this is all 1s on the diagonal, perhaps
                        // to no one's surprise.
  }
  catch (cet::exception &)
    {
      throw GFException("GFKalman: Jac.T*Jac is not invertible. But keep plowing on ... ",__LINE__,__FILE__).setFatal();
    }
  if(TMath::IsNaN(det)) {
    throw GFException("GFKalman: det of Jac.T*Jac is nan",__LINE__,__FILE__).setFatal();
  }

  TMatrixT<Double_t> j5x7 = jjInv*jac_t;
  TMatrixT<Double_t> c7x7 = j5x7.T() * (cov * j5x7);
  return c7x7;
}

TMatrixT< Double_t > genf::GFKalman::calcGain ( const TMatrixT< Double_t > &  cov, const TMatrixT< Double_t > &  HitCov, const TMatrixT< Double_t > &  H  )

private

Calculate Kalman Gain.

Definition at line 733 of file GFKalman.cxx.

                                                     {

  // calculate covsum (V + HCH^T)

  // Comment next 3 out for normal running.
  //cov.Print();
  //H.Print();
  //HitCov.Print();
  TMatrixT<Double_t> covsum1(cov,TMatrixT<Double_t>::kMultTranspose,H);
  //  covsum1.Print();
  TMatrixT<Double_t> covsum(H,TMatrixT<Double_t>::kMult,covsum1);

  //covsum.Print();

  covsum+=HitCov;
  //covsum = (covsum,TMatrixT<Double_t>::kPlus,HitCov);
  // covsum.Print();

  // invert
  double det=0;
  covsum.SetTol(1.0e-23); // to avoid inversion problem, EC, 8-Aug-2011.
  covsum.Invert(&det);
  //    std::cout << "GFKalman:: calGain(), det is  "<< det << std::endl;
  if(TMath::IsNaN(det)) {
    throw GFException("Kalman Gain: det of covsum is nan",__LINE__,__FILE__).setFatal();
  }

  if(det==0){
        GFException exc("cannot invert covsum in Kalman Gain - det=0",
                                                __LINE__,__FILE__);
        exc.setFatal();
        std::vector< TMatrixT<Double_t> > matrices;
        matrices.push_back(cov);
        matrices.push_back(HitCov);
        matrices.push_back(covsum1);
        matrices.push_back(covsum);
        exc.setMatrices("cov, HitCov, covsum1 and covsum",matrices);
        throw exc;

  }

  // gain is CH^T/(V + HCH^T)
  // calculate gain
  TMatrixT<Double_t> gain1(H,TMatrixT<Double_t>::kTransposeMult,covsum);
  TMatrixT<Double_t> gain(cov,TMatrixT<Double_t>::kMult,gain1);

  return gain;
}

double genf::GFKalman::chi2Increment ( const TMatrixT< Double_t > &  r, const TMatrixT< Double_t > &  H, const TMatrixT< Double_t > &  cov, const TMatrixT< Double_t > &  V  )

private

this returns the reduced chi2 increment for a hit

Definition at line 222 of file GFKalman.cxx.

                                                                                       {

  // residuals covariances:R=(V - HCH^T)
  TMatrixT<Double_t> R(V);
  TMatrixT<Double_t> covsum1(cov,TMatrixT<Double_t>::kMultTranspose,H);
  TMatrixT<Double_t> covsum(H,TMatrixT<Double_t>::kMult,covsum1);

  R-=covsum;

  // chisq= r^TR^(-1)r
  double det=0.;
  TMatrixT<Double_t> Rsave(R);
  R.SetTol(1.0e-30); // to avoid inversion problem, EC, 8-Aug-2011. Was23, 9-Jan-2012.

  try
    {
      R.Invert(&det);
    }
  catch (cet::exception &)
    {
      GFException e("Kalman Chi2Increment: R is not even invertible. But keep plowing on ... ",__LINE__,__FILE__);
      //e.setFatal();
      //throw e;
    }
  if(TMath::IsNaN(det)) {
    throw GFException("Kalman Chi2Increment: det of covsum is nan",__LINE__,__FILE__).setFatal();
  }
  TMatrixT<Double_t> residTranspose(r);
  residTranspose.T();
  TMatrixT<Double_t> chisq=residTranspose*(R*r);
  assert(chisq.GetNoElements()==1);

  if(TMath::IsNaN(chisq[0][0])){
        GFException exc("chi2 is nan",__LINE__,__FILE__);
        exc.setFatal();
        std::vector<double> numbers;
        numbers.push_back(det);
        exc.setNumbers("det",numbers);
        std::vector< TMatrixT<Double_t> > matrices;
        matrices.push_back(r);
        matrices.push_back(V);
        matrices.push_back(Rsave);
        matrices.push_back(R);
        matrices.push_back(cov);
        exc.setMatrices("r, V, Rsave, R, cov",matrices);
    throw exc;
  }

  return chisq[0][0];
}

void genf::GFKalman::fittingPass	(	GFTrack *	trk,
		int	dir
	)

Performs fit on a GFTrack beginning with the current hit.

Definition at line 177 of file GFKalman.cxx.

                                                    {
  //loop over hits

  unsigned int nhits=trk->getNumHits();
  unsigned int starthit=trk->getNextHitToFit();

  int nreps=trk->getNumReps();
  int ihit=(int)starthit;

  //clear chi2 sum and ndf sum in track reps
  for(int i=0;i<nreps;++i){
    GFAbsTrackRep* arep=trk->getTrackRep(i);
    arep->setChiSqu(0.);
    arep->setNDF(0);
  }

  //clear failedHits and outliers
  trk->getBK()->clearFailedHits();

  while((ihit<(int)nhits && direction==1) || (ihit>-1 && direction==-1)){
    //    GFAbsRecoHit* ahit=trk->getHit(ihit);
    // loop over reps
    for(int irep=0; irep<nreps; ++irep){
          GFAbsTrackRep* arep=trk->getTrackRep(irep);
          if(arep->getStatusFlag()==0) {
            try {
              processHit(trk,ihit,irep,direction);
            }
            catch(GFException& e) {
              trk->addFailedHit(irep,ihit);
              std::cerr << e.what();
              e.info();
              if(e.isFatal()) {
                arep->setStatusFlag(1);
                continue; // go to next rep immediately
              }
            }
          }
    }// end loop over reps
    ihit+=direction;
  }// end loop over hits
  trk->setNextHitToFit(ihit-2*direction);
  //trk->printGFBookkeeping();
}

double genf::GFKalman::getChi2Hit	(	GFAbsRecoHit *	hit,
		GFAbsTrackRep *	rep
	)

Calculates chi2 of a given hit with respect to a given track representation.

Definition at line 276 of file GFKalman.cxx.

{
  // get prototypes for matrices
  int repDim=rep->getDim();
  TMatrixT<Double_t> state(repDim,1);
  TMatrixT<Double_t> cov(repDim,repDim);;
  GFDetPlane pl=hit->getDetPlane(rep);
  rep->extrapolate(pl,state,cov);


  TMatrixT<Double_t> H = hit->getHMatrix(rep);

  // get hit covariances
  TMatrixT<Double_t> V=hit->getHitCov(pl);
  V[0][0] *= fErrScaleMTh;
  TMatrixT<Double_t> r=hit->residualVector(rep,state,pl);
  assert(r.GetNrows()>0);

  r[0][0] = fabs(r[0][0]);
  //this is where chi2 is calculated
  double chi2 = chi2Increment(r,H,cov,V);

  return chi2/r.GetNrows();
}

void genf::GFKalman::operator() ( GFTrack *  track )

inline

Operator for use with STL.

This operator allows to use the std::foreach algorithm with an STL container o GFTrack* objects.

Definition at line 68 of file GFKalman.h.

{processTrack(track);}

void genf::GFKalman::operator() ( std::pair< int, GFTrack * >  tr )

inline

Operator for use with STL.

This operator allows to use the std::foreach algorithm with an STL container o GFTrack* objects.

Definition at line 75 of file GFKalman.h.

{processTrack(tr.second);}

void genf::GFKalman::processHit ( GFTrack *  tr, int  ihit, int  irep, int  direction  )

private

One Kalman step.

Performs

• Extrapolation to detector plane of the hit
• Calculation of residual and Kalman Gain
• Update of track representation state and chi2

Definition at line 303 of file GFKalman.cxx.

                                                                     {
  GFAbsRecoHit* hit = tr->getHit(ihit);
  GFAbsTrackRep* rep = tr->getTrackRep(irep);

  // get prototypes for matrices
  int repDim = rep->getDim();
  TMatrixT<Double_t> state(repDim,1);
  TMatrixT<Double_t> cov(repDim,repDim);
  static TMatrixT<Double_t> covFilt(cov);
  const double pi2(10.0);
  GFDetPlane pl, plPrev;
  unsigned int nhits=tr->getNumHits();
  int phit=ihit;
  static TMatrixT<Double_t> oldState(5,1);
  static std::vector<TVector3> pointsPrev;

  const double eps(1.0e-6);
  if (direction>0 && ihit>0)
    {
      phit = ihit - 1;
    }
  if (direction<0 && ihit<((int)nhits-1))
    {
      phit = ihit + 1;
    }
  GFAbsRecoHit* hitPrev = tr->getHit(phit);

  /* do an extrapolation, if the trackrep irep is not given
   * at this ihit position. This will usually be the case, but
   * not if the fit turns around
   */
  //  std::cout << "GFKalman::ProcessHit(): ihit is " << ihit << std::endl;
  //  std::cout << "GFKalman::ProcessHit(): direction is "  << direction << std::endl;
  //rep->Print();


  Double_t thetaPlanes(0.0);
  if(ihit!=tr->getRepAtHit(irep)){
    //std::cout << "not same" << std::endl;
    // get the (virtual) detector plane. This call itself calls
    // extrapolateToPoint(), which calls Extrap() with just 2 args and
    // default 3rd arg, which propagates track through
    // material to find the next plane. But it in fact seems
    // to just return this pl and plPrev, as desired. Something in Extrap()
    // kicks it out... even though I can see it walking through material ...
    // A mystery.

    pl=hit->getDetPlane(rep);
    plPrev=hitPrev->getDetPlane(rep);

      // std::cout << "GFKalman::ProcessHit(): hit is ... " << ihit << std::endl;
      //hit->Print();
      //std::cout << "GFKalman::ProcessHit(): plane is ... " <<  std::endl;
      //pl.Print();

    //do the extrapolation. This calls Extrap(), which wraps up RKutta and
    // GFMaterials calls. state is intialized, pl is unchanged. Latter behavior
    // I will alter below.
    try{
      rep->extrapolate(pl,state,cov);
      /*
      if ( std::isnan(cov[0][0]) )
        {
          cov = covFilt;
        }
      else
        {
          covFilt = cov;
        }
      */
    }
    catch (cet::exception &)
      {
        throw cet::exception("GFKalman.cxx: ") << " Line " << __LINE__ << ", " << __FILE__ << " ...Rethrow. \n";
      }

  }
  else{
    pl = rep->getReferencePlane();
    plPrev = hitPrev->getDetPlane(rep);
    state = rep->getState();
    cov = rep->getCov();
  }

  // Update plane. This is a code change from Genfit out of box.
  // state has accumulated the material/magnetic field changes since last
  // step. Here, we make the *plane* at this step know about that bend.
  // We will not, in the end, save this plane, cuz Genfit knows to properly
  // calculate it later.
  // Actually, I do *not* change the plane. EC, 11-May-2012.
  TVector3 u(pl.getU());
  TVector3 v(pl.getV());
  TVector3 wold(u.Cross(v));
  //Double_t sign(1.0);
  //if ((direction==-1) && ihit==((int)nhits-1)) sign = -1.0;

  TVector3 pTilde = direction * (wold + state[1][0] * u + state[2][0] * v);
  TVector3 w(pTilde.Unit());
  // Find angle/vector through which we rotate. Use it subsequently.
  TVector3 rot(wold.Cross(w));
  Double_t ang(TMath::ACos(w*wold));
  ang = TMath::Min(ang,0.95*TMath::Pi()/2.0);
  /*
    {
      u.Rotate(ang,rot);
      v.Rotate(ang,rot);

      pl.setNormal(w);
      pl.setU(u.Unit());
      pl.setV(v.Unit());
    }
  */
  if(cov[0][0]<1.E-50 || TMath::IsNaN(cov[0][0])){
    //std::cout<<"GFKalman::processHit() 0. Calling Exception."<<std::endl;
        GFException exc(COVEXC,__LINE__,__FILE__);
        if (fGENfPRINT) cov.Print();
        if (fGENfPRINT) std::cout<<"GFKalman::processHit() 1. No longer throw exception. Force cov[0][0] to 0.01."<<std::endl;
        // std::cout<<"GFKalman::processHit() 1. About to throw GFException."<<std::endl;
        cov = covFilt;
        //      throw exc;
        // std::cout<<"GFKalman::processHit() 2. Back from GFException."<<std::endl;

  }

  /*
  if(direction==1){
    tr->getBK(irep)->setMatrix("fPredSt",ihit,state);
    tr->getBK(irep)->setMatrix("fPredCov",ihit,cov);
    tr->getBK(irep)->setGFDetPlane("f",ihit,pl);
  }
  else{
    tr->getBK(irep)->setMatrix("bPredSt",ihit,state);
    tr->getBK(irep)->setMatrix("bPredCov",ihit,cov);
    tr->getBK(irep)->setGFDetPlane("b",ihit,pl);
  }
  */

  //  TMatrixT<Double_t> origcov=rep->getCov();

  int pdg = tr->getPDG();
  TParticlePDG * part = TDatabasePDG::Instance()->GetParticle(pdg);
  Double_t mass = part->Mass();

  Double_t dist = (pl.getO()-plPrev.getO()).Mag();
  Double_t mom = fabs(1.0/state[0][0]);
  Double_t beta = mom/sqrt(mass*mass+mom*mom);
  const Double_t lowerLim(0.01);
  if (std::isnan(dist) || dist<=0.0) dist=lowerLim; // don't allow 0s here.
  if (std::isnan(beta) || beta<0.01) beta=0.01;
  TMatrixT<Double_t> H=hit->getHMatrix(rep,beta,dist);


  // Can force huge error here on p and du,v/dw, and thus insensitivity to p,
  // by setting V[0][0]->inf.
  TMatrixT<Double_t> V=hit->getHitCov(pl,plPrev,state,mass);
  V[0][0] *= fErrScaleMTh;
  tr->setHitMeasuredCov(V);

  // calculate kalman gain ------------------------------
  TMatrixT<Double_t> Gain(calcGain(cov,V,H));

  //std::cout << "GFKalman:: processHits(), state is  " << std::endl;
  //rep->getState().Print();

  TMatrixT<Double_t> res=hit->residualVector(rep,state,pl,plPrev,mass);
  // calculate update -----------------------------------


  //  TVector3 pointer((pl.getO()-plPrev.getO()).Unit());

  TMatrixT<Double_t> rawcoord = hit->getRawHitCoord();
  TVector3 point(rawcoord[0][0],rawcoord[1][0],rawcoord[2][0]);
  TMatrixT<Double_t> prevrawcoord = hitPrev->getRawHitCoord();
  TVector3 pointPrev(prevrawcoord[0][0],prevrawcoord[1][0],prevrawcoord[2][0]);
  pointsPrev.push_back(pointPrev);
  TMatrixT<Double_t> Hnew(H);
  if ((ihit==((int)nhits-1)&&direction==-1) || (ihit==0&&direction==1))
    pointsPrev.clear();

  TVector3 pointer((point-pointPrev).Unit());
  static TVector3 pointerPrev(pointer);
  if (ihit==0&&direction==1   )
    {pointer[0] = 0.0;pointer[1] = 0.0;pointer[2] = 1.0;}
  if (ihit==((int)nhits-1)&&direction==-1)
    {pointer[0] = 0.0;pointer[1] = 0.0;pointer[2] = -1.0;}
  double thetaMeas = TMath::Min(fabs(pointer.Angle(pointerPrev)),0.95*TMath::Pi()/2.0);
  // Below line introduced because it's not true we predict the angle to be
  // precisely ang. If we'd taken this quantity from our transport result
  // (with measurement errors in it) we'd expect a smear like this on ang.
  // EC, 22-Feb-2012.
  // Error is taken on th^2
  ang = (Hnew*state)[0][0]; // H->Hnew

  // fErrScale is extra-fun bonus factor!
  //thetaPlanes = ang*ang; // + fErrScaleSTh*gRandom->Gaus(0.0,V[0][0]);
  thetaPlanes = gRandom->Gaus(0.0,ang*ang);
  thetaPlanes = TMath::Min(sqrt(fabs(thetaPlanes)),0.95*TMath::Pi()/2.0);

  Double_t dtheta =  thetaMeas - thetaPlanes; // was fabs(res[0][0]). EC, 26-Jan-2012
  if (((ihit==((int)nhits-1)||ihit==((int)nhits-2))&&direction==-1) || ((ihit==0||ihit==1)&&direction==1))
    {
      dtheta = 0.0; // at the bare minimum (Jan-2013). Now (Feb-2013), ....
      // Do not let these 4 points*direction influence the state or chi2.
      rep->setData(state,pl,&cov); // This is where fState,
                                   // fRefPlane and fCov are updated.
      pointerPrev = pointer;
      return;
    }

  oldState = state;

  res[0][0] = dtheta/Hnew[0][0];
  // The particle was propagated through material until its poca
  // to this current hit was found. At that point its plane is defined,
  // in which du/dw=dv/dw=0 by construction. But, there's a deviation
  // in those two quantities measured in the *previous* plane which we want.
  TVector3 uPrev(plPrev.getU());
  TVector3 vPrev(plPrev.getV());
  TVector3 wPrev(u.Cross(v));
  // We want the newest momentum vector's d{u,v}/dw defined in the prev plane.
  // That is the predicted du,v/dw. We will subtract from the actual.
  // But the u's and v's need to come from the State not the Planes, cuz I
  // haven't updated pl,plPrev per latest State. plFilt, the updated plPrev,
  // is what we want. w is from updated new plane pl, per latest State.
  // e.g. plFilt.
  static GFDetPlane plFilt;//(pl);
  if (plFilt.getO().Mag()>eps)
    {
      uPrev = plFilt.getU();
      vPrev = plFilt.getV();
      wPrev = plFilt.getNormal();
    }
  //res[1][0] = (pointer*uPrev)/(pointer*wPrev) - (w*uPrev)/(w*wPrev);
  //res[2][0] = (pointer*vPrev)/(pointer*wPrev) - (w*vPrev)/(w*wPrev);
  //  res[1][0] = 0.0;
  //  res[2][0] = 0.0;

  TMatrixT<Double_t> update=Gain*res;

  // Don't let crazy ass spacepoints which pull the fit all over the place
  // be allowed to update the state. Use __ xRMS for fMaxUpdate.
  // Use a larger limit (~0.1) when starting with too-high seed momentum.
  // Smaller, when starting with an underestimate.
  //
  // Don't allow fits above 20 GeV, or so. Else, 1/p will likely
  // migrate across zero. Similarly, don't allow tiny momentum.
  //
  tr->setHitUpdate(update);
  if (fabs(update[0][0])>fMaxUpdate )
    { // zero the gain so as to not update cov, state
      // zero the updates
      update[0][0] = 0.0;/* */
      //      update[0][0] = fMaxUpdate*update[0][0]/fabs(update[0][0]);
      // either of below leads to craziness, somehow.
      // update.Zero();
      // Gain.Zero();
    }
  state+=update; // prediction overwritten!

  // Debugging purposes
  TMatrixT<Double_t> GH(Gain*Hnew);
  if (GH[0][0] > 1.)
    {
      //      std::cout << "GFKalman:: Beginnings of a problem." << std::endl;
      Hnew[0][0] = Hnew[0][0] - eps/Gain[0][0];
    }
  TMatrixT<Double_t> GHc(GH*cov);

  cov-=Gain*(Hnew*cov);

  // Below is protection required at end of contained track when
  // momentum is tiny and cov[0][0] gets huge.

  //  if (cov[0][0]>pi2/Hnew[0][0] || cov[0][0] <= 0.0 || TMath::IsNaN(cov[0][0]))
  if (cov[0][0] <= 0.0 || TMath::IsNaN(cov[0][0]))
                                                                                                                                                                                                                                                                                             {
      cov[0][0]=pi2/Hnew[0][0]; // some big value.a
      //cov = covFilt;
    }
  else  // store away a good cov matrix
    {
      covFilt = cov;
    }


  TMatrixT<double> cov7x7(calcCov7x7(cov,pl));
  tr->setHitCov(cov);
  tr->setHitCov7x7(cov7x7);
  tr->setHitState(state);

  if (fabs(1.0/state[0][0])<fMomLow)
    state[0][0] = 1.0/fMomLow*fabs(state[0][0])/state[0][0];
  if (fabs(1.0/state[0][0])>fMomHigh)
    state[0][0] = 1.0/fMomHigh*fabs(state[0][0])/state[0][0];


  // Let's also calculate the "filtered" plane, and pointer here.
  // Use those to calculate filtered chisq and pass to setHitPlane()
  // for plane-by-plane correct "filtered" track pointing that gets
  // stuffed into the Track().
  TVector3 uf(pl.getU());
  TVector3 vf(pl.getV());
  TVector3 wf(uf.Cross(vf));
  TVector3 Of(pl.getO());
  // direction *
  //Double_t sign2(1.0);
  //if (direction==-1 && ihit==(int)nhits-1) sign2 = -1.0;

  TVector3 pf = direction*(wf + state[1][0] * uf + state[2][0] * vf);
  TVector3 pposf = Of + state[3][0] * uf + state[4][0] * vf;
  Double_t angf = TMath::Min(fabs(pf.Angle(wf)),0.95*TMath::Pi()/2.0);
  TVector3 rotf(wf.Cross(pf.Unit()));

  uf.Rotate(angf,rotf);
  vf.Rotate(angf,rotf);
  wf = uf.Cross(vf);
  plFilt.setU(uf.Unit());
  plFilt.setV(vf.Unit());
  plFilt.setO(pposf);
  plFilt.setNormal(pf.Unit());

  tr->setHitPlaneXYZ(plFilt.getO());
  tr->setHitPlaneUxUyUz(plFilt.getNormal());
  tr->setHitPlaneU(plFilt.getU());
  tr->setHitPlaneV(plFilt.getV());

  // calculate filtered chisq from filtered residuals
  TMatrixT<Double_t> r=hit->residualVector(rep,state,plFilt,plPrev,mass);
  if (direction==-1) wold.Rotate(TMath::Pi(),wold.Orthogonal());
  dtheta = thetaMeas - TMath::Min(fabs(wold.Angle(plFilt.getNormal())),0.95*TMath::Pi()/2.);
  r[0][0] = dtheta/Hnew[0][0];
  //r[1][0] = (pointer*uPrev)/(pointer*wPrev) - (wf*uPrev)/(wf*wPrev);
  //r[2][0] = (pointer*vPrev)/(pointer*wPrev) - (wf*vPrev)/(wf*wPrev);
  //r[1][0] = 0.;
  //r[2][0] = 0.;

  double chi2 = chi2Increment(r,Hnew,cov,V);
  int ndf = r.GetNrows();
  if (update[0][0]==0.0) {chi2=0.0; ndf=0;};
  rep->addChiSqu( chi2 );
  rep->addNDF( ndf );
  pointerPrev = pointer;
  tr->setHitChi2(chi2);
  /*
  if(direction==1){
    tr->getBK(irep)->setNumber("fChi2",ihit,chi2/ndf);
  }
  else{
    tr->getBK(irep)->setNumber("bChi2",ihit,chi2/ndf);
  }
  */

  // if we survive until here: update TrackRep
  //rep->setState(state);
  //rep->setCov(cov);
  //rep->setReferencePlane(pl);

  // Since I've tilted the planes, d(u,v)/dw=0 by construction.
  // But, I *haven't* tilted the planes! EC, 11-May-2012.
  //state[1][0] = 0.0;
  //state[2][0] = 0.0;
  rep->setData(state,pl,&cov); // This is where fState,
                               // fRefPlane and fCov are updated.
  tr->setRepAtHit(irep,ihit);

}

void genf::GFKalman::processTrack

(

GFTrack *

trk

)

Performs fit on a GFTrack.

The hits are processed in the order in which they are stored in the GFTrack object. Sorting of hits in space has to be done before!

Definition at line 46 of file GFKalman.cxx.

                                           {
  int direction=fInitialDirection;
  if ((direction != 1) && (direction != -1))
    throw GFException(std::string(__func__) + ": wrong direction", __LINE__, __FILE__).setFatal();
  //  trk->clearGFBookkeeping();
  trk->clearRepAtHit();
  /*
  int nreps=trk->getNumReps();
  for(int i=0; i<nreps; ++i){
    trk->getBK(i)->setNhits(trk->getNumHits());
    trk->getBK(i)->bookMatrices("fPredSt");
    trk->getBK(i)->bookMatrices("fPredCov");
    trk->getBK(i)->bookMatrices("bPredSt");
    trk->getBK(i)->bookMatrices("bPredCov");
    trk->getBK(i)->bookGFDetPlanes("f");
    trk->getBK(i)->bookGFDetPlanes("b");
    trk->getBK(i)->bookNumbers("fChi2");
    trk->getBK(i)->bookNumbers("bChi2");
  }
  */

  /*why is there a factor of two here (in the for statement)?
    Because we consider one full iteration to be one back and
    one forth fitting pass */
  for(int ipass=0; ipass<2*fNumIt; ipass++){
    //    std::cout << "GFKalman:: numreps, numhits, ipass, direction"<< trk->getNumReps()<<", "<< trk->getNumHits()<<", "<< ipass<<", " <<direction << std::endl;
    //    if(floor(ipass/2)==fNumIt && direction==1) blowUpCovsDiag(trk);
    if(ipass>0) blowUpCovs(trk); // Back to >0, not >=0 EC, 9-11-Feb-2013
    if(direction==1){
      trk->setNextHitToFit(0);
    }
    else {
      trk->setNextHitToFit(trk->getNumHits()-1);
    }


    try
      {
        fittingPass(trk,direction);
      }
    catch(GFException &)
      {
        throw cet::exception("GFKalman.cxx: ") << " Line " << __LINE__ << ", " << __FILE__ << " ...Rethrow. \n";
      }

    //save first and last plane,state&cov after the fitting pass
    if(direction==1){//forward at last hit
      int nreps=trk->getNumReps();
      for(int i=0; i<nreps; ++i){
        trk->getTrackRep(i)->
          setLastPlane( trk->getTrackRep(i)->getReferencePlane() );
        trk->getTrackRep(i)->
          setLastState( trk->getTrackRep(i)->getState() );
        trk->getTrackRep(i)->
          setLastCov( trk->getTrackRep(i)->getCov() );
      }
    }
    else{//backward at first hit
      int nreps=trk->getNumReps();
      for(int i=0; i<nreps; ++i){
        trk->getTrackRep(i)->
          setFirstPlane( trk->getTrackRep(i)->getReferencePlane() );
        trk->getTrackRep(i)->
          setFirstState( trk->getTrackRep(i)->getState() );
        trk->getTrackRep(i)->
          setFirstCov( trk->getTrackRep(i)->getCov() );
      }
    }

    //switch direction of fitting and also inside all the reps
    if(direction==1) direction=-1;
    else direction=1;
    switchDirection(trk);

  }
  return;
}

void genf::GFKalman::setBlowUpFactor ( double  f )

inline

Set the blowup factor (see blowUpCovs() )

Definition at line 114 of file GFKalman.h.

{fBlowUpFactor=f;}

void genf::GFKalman::setErrorScaleMTh ( Double_t  f )

inline

Definition at line 119 of file GFKalman.h.

{fErrScaleMTh=f;}

void genf::GFKalman::setErrorScaleSTh ( Double_t  f )

inline

Definition at line 118 of file GFKalman.h.

{fErrScaleSTh=f;}

void genf::GFKalman::setInitialDirection ( int  d )

inline

Sets the inital direction of the track fit (1 for inner to outer, or -1 for outer to inner). The standard is 1 and is set in the ctor.

Definition at line 110 of file GFKalman.h.

{fInitialDirection=d;}

void genf::GFKalman::setLazy ( Int_t  )

inline

Switch lazy error handling.

This is a historically left-over method and shall be deleted some time

Definition at line 83 of file GFKalman.h.

{std::cerr<<"Using outdates setLazy method of class GFKalman:"<<std::endl;}

void genf::GFKalman::setMaxUpdate ( Double_t  f )

inline

Definition at line 117 of file GFKalman.h.

{fMaxUpdate=f;}

void genf::GFKalman::setMomHigh ( Double_t  f )

inline

Definition at line 116 of file GFKalman.h.

{fMomHigh=f;}

void genf::GFKalman::setMomLow ( Double_t  f )

inline

Definition at line 115 of file GFKalman.h.

{fMomLow=f;}

void genf::GFKalman::setNumIterations ( Int_t  i )

inline

Set number of iterations for Kalman Filter.

One iteration is one forward pass plus one backward pass

Definition at line 89 of file GFKalman.h.

{fNumIt=i;}

void genf::GFKalman::switchDirection ( GFTrack *  trk )

private

Used to switch between forward and backward filtering.

Definition at line 125 of file GFKalman.cxx.

                                         {
  int nreps=trk->getNumReps();
  for(int i=0; i<nreps; ++i){
    trk->getTrackRep(i)->switchDirection();
  }
}

Member Data Documentation

double genf::GFKalman::fBlowUpFactor

private

Definition at line 157 of file GFKalman.h.

Double_t genf::GFKalman::fErrScaleMTh

private

Definition at line 163 of file GFKalman.h.

Double_t genf::GFKalman::fErrScaleSTh

private

Definition at line 162 of file GFKalman.h.

bool genf::GFKalman::fGENfPRINT

private

Definition at line 164 of file GFKalman.h.

int genf::GFKalman::fInitialDirection

private

Definition at line 155 of file GFKalman.h.

Double_t genf::GFKalman::fMaxUpdate

private

Definition at line 161 of file GFKalman.h.

Double_t genf::GFKalman::fMomHigh

private

Definition at line 160 of file GFKalman.h.

Double_t genf::GFKalman::fMomLow

private

Definition at line 159 of file GFKalman.h.

Int_t genf::GFKalman::fNumIt

private

Definition at line 156 of file GFKalman.h.

---

The documentation for this class was generated from the following files:

• GFKalman.h
• GFKalman.cxx