TimeExtractionUtils Class Reference

Helper class to perform all kind of track extrapolations using the methods from arXiv:0810.2241. More...

#include <TimeExtractionUtils.h>

Static Public Member Functions
static std::pair< double, double >	getChi2WithFit (const std::vector< RecoTrack * > &recoTracks, bool setDirtyFlag)
	Fit the tracks and extract the reduced chi2.
static std::pair< double, double >	getExtractedTimeAndUncertaintyWithFit (const std::vector< RecoTrack * > &recoTracks, bool setDirtyFlag)
	Fit the tracks and extract the chi2 derivatives.
static void	addEventT0WithQuality (const std::vector< RecoTrack * > &recoTracks, StoreObjPtr< EventT0 > &eventT0, std::vector< EventT0::EventT0Component > &eventT0WithQualityIndex)
	Append an event-t0 value with quality information.
static double	extractReducedChi2 (const RecoTrack &recoTrack)
	Small helper function to extract the reduced chi^2 (chi^2/ndf). Returns NaN if ndf is 0.
static std::pair< double, double >	getChi2Derivatives (const RecoTrack &recoTrack)
	Extract the derivatives d chi^2 / d alpha and d^2 chi^2 / (d alpha)^2 from the reco track by calculating the full residual covariance matrix, where alpha is the global event time. More...

Static Private Member Functions
static std::vector< int >	getMeasurementDimensions (const RecoTrack &recoTrack)
	Get a list of dimensions for each measurement in the reco track. More...
static bool	buildFullCovarianceMatrix (const RecoTrack &recoTrack, TMatrixDSym &fullCovariance)
	Build the full covariance matrix of the given reco track and fills it into the given TMatrixDSym. More...
static bool	buildFullResidualCovarianceMatrix (const RecoTrack &recoTrack, const std::vector< int > &vDimMeas, const TMatrixDSym &fullCovariance, TMatrixDSym &fullResidualCovariance, TMatrixDSym &inverseFullMeasurementCovariance)
	Build the full covariance matrix of the residuals of the measurements out of the full covariance matrix of the track states. More...
static void	buildResidualsAndTimeDerivative (const RecoTrack &recoTrack, const std::vector< int > &vDimMeas, TVectorD &residuals, TVectorD &residualTimeDerivative)
	Calculate the weighted residuals (weighted with the DAF weight) of each hit and the time derivative of the residuals by calling the timeDerivativesMeasurementsOnPlane method of the CDCRecoHits. More...

Detailed Description

Helper class to perform all kind of track extrapolations using the methods from arXiv:0810.2241.

The event time is calculated as one would calculate an alignment parameter. We follow the prescription in arXiv:0810.2241 [physics.ins-det] to build the full covariance matrix for the Kalman-fitted track (DAF counts as Kalman). We then evaluate the change in event time that minimizes the sum of chi^2s of the tracks following the same procedure that is decribed in loc.cit.

Unlike the case of alignment, we only have one free parameter (time), and therefore there are no large matrices to invert. The necessary timeshift is calculated from the estimated derivatives of chi2^2 in the linear approximation as dchi^2 /d^2chi^2 delta t = - -----— / -----— . dt / dt^2 Here division replaces a matrix inverse (this is loc.cit. Eq. (8)).

Time-dependence of the fit result comes from the CDC hits, where a later time of particle passage corresponds to a greater drift circle.

Definition at line 48 of file TimeExtractionUtils.h.

Member Function Documentation

buildFullCovarianceMatrix()

bool buildFullCovarianceMatrix ( const RecoTrack &  recoTrack, TMatrixDSym &  fullCovariance  )

staticprivate

Build the full covariance matrix of the given reco track and fills it into the given TMatrixDSym.

Needed for the derivatives.

The full covariance contains the covariances of the different states of the reco track at different hits, this means is connects the calculates state at one measurement to one at another measurement. This is why it has the dimension number of hits * dimensionality of the track state. The diagonal blocks are filled with the covariance matrices at a single measurement whereas the off diagonal elements connect different measurements.

The off-diagnal elements are calculated, by either using one prediction step forward (because the two hits are neighbors), by using the fact that if thei are not neighbors they can be connected by a finite number of forward predictions which breaks this down to the first case and some multiplications again or by using symmetry aspects (e.g. no background prediction is needed).

The notation follows 0810.2241.

Returns false if something strange happened, otherwise true.

Definition at line 272 of file TimeExtractionUtils.cc.

{
  const auto* kfs = dynamic_cast<const genfit::KalmanFitStatus*>(recoTrack.getTrackFitStatus());
 
  if (!kfs) {
    B2ERROR("Track not fitted with a Kalman fitter.");
    return false;
  }
 
  if (!kfs->isFitConverged()) {
    B2ERROR("Track fit didn't converge.");
    return false;
  }
 
  if (!kfs->isFittedWithReferenceTrack()) {
    B2ERROR("No reference track.");
    return false;
  }
 
  const auto& hitPoints = recoTrack.getHitPointsWithMeasurement();
  const unsigned int nPoints = hitPoints.size();
  const genfit::AbsTrackRep* rep = recoTrack.getCardinalRepresentation();
  const int nDim = rep->getDim();
 
  fullCovariance.ResizeTo(nPoints * nDim, nPoints * nDim);
  std::vector<TMatrixD> vFitterGain;
  vFitterGain.reserve(nPoints);
  for (unsigned int i = 0; i < nPoints; ++i) {
    const genfit::TrackPoint* tp = hitPoints[i];
    const genfit::KalmanFitterInfo* fi = tp->getKalmanFitterInfo();
    if (!fi) {
      B2DEBUG(25, "Missing KalmanFitterInfo - skipping track");
      return false;
    }
 
    // Diagonal part of the full covariance matrix are the covariances
    // of the smoothed states.
    const genfit::MeasuredStateOnPlane& mop = fi->getFittedState();
    setSubOnDiagonal(fullCovariance, i * nDim, mop.getCov());
 
    // Build the corresponding smoother gain matrix.
    if (i + 1 < nPoints) {
      const genfit::TrackPoint* tpNext = hitPoints[i + 1];
      const genfit::KalmanFitterInfo* fiNext = tpNext->getKalmanFitterInfo();
      if (!fiNext) {
        B2DEBUG(25, "Missing next KalmanFitterInfo - skipping track");
        return false;
      }
 
      // update at i
      const genfit::MeasuredStateOnPlane* update = fi->getForwardUpdate();
      // transport to i+1 prediction at i+1
      const genfit::ReferenceStateOnPlane* rsop = fiNext->getReferenceState();
      if (rsop) {
        const genfit::MeasuredStateOnPlane* pred = fiNext->getForwardPrediction();
 
        // Evaluate (C^i_i+1)^-1 F_i
        TDecompChol decomp(pred->getCov());
        TMatrixD F = rsop->getForwardTransportMatrix();
        decomp.MultiSolve(F);
 
        // Calculate gain matrix as
        //   C_i F_i^T (C^i_i+1)^-1 = C_i ((C^i_i+1)^-1 F_i)^T
        // in the notation of 0810.2241
        vFitterGain.emplace_back(TMatrixD(update->getCov(),
                                          TMatrixD::kMultTranspose, F));
      } else {
        B2DEBUG(150, "No reference state, substituting empty fitter gain matrix.");
        vFitterGain.emplace_back(TMatrixD(5, 5));
      }
    }
 
    // Build the off-diagonal elements.
    TMatrixD offDiag = mop.getCov();
    for (int j = i - 1; j >= 0; --j) {
      offDiag = TMatrixD(vFitterGain[j], TMatrixD::kMult, offDiag);
      setSubOffDiagonal(fullCovariance, j * nDim, i * nDim, offDiag);
    }
  }
 
  return true;
}

buildFullResidualCovarianceMatrix()

bool buildFullResidualCovarianceMatrix ( const RecoTrack &  recoTrack, const std::vector< int > &  vDimMeas, const TMatrixDSym &  fullCovariance, TMatrixDSym &  fullResidualCovariance, TMatrixDSym &  inverseFullMeasurementCovariance  )

staticprivate

Build the full covariance matrix of the residuals of the measurements out of the full covariance matrix of the track states.

The returned matrix has the dimensionality of the number of measurements * their dimensionality and relates the residuals of each measurements to each other residuals of an (other) measurement.

This can easily be done using the full covariance matrix of the states, which is calculated using the buildFullCovarianceMatrix method. The ful covariance matrix of the residuals and their inverse is returned.

The fullResidualCovariance is eq. (17) in 0810.2241.

Definition at line 356 of file TimeExtractionUtils.cc.

buildResidualsAndTimeDerivative()

void buildResidualsAndTimeDerivative ( const RecoTrack &  recoTrack, const std::vector< int > &  vDimMeas, TVectorD &  residuals, TVectorD &  residualTimeDerivative  )

staticprivate

Calculate the weighted residuals (weighted with the DAF weight) of each hit and the time derivative of the residuals by calling the timeDerivativesMeasurementsOnPlane method of the CDCRecoHits.

The residuals as well as the residualTimeDerivative are returned by reference to a vector as large as there are measurements in the reco track (exactly: number of measurements * dimensionality of each measurement).

Definition at line 425 of file TimeExtractionUtils.cc.

getChi2Derivatives()

std::pair< double, double > getChi2Derivatives ( const RecoTrack &  recoTrack )

static

Extract the derivatives d chi^2 / d alpha and d^2 chi^2 / (d alpha)^2 from the reco track by calculating the full residual covariance matrix, where alpha is the global event time.

The two derivatives are returned as a pair.

Definition at line 199 of file TimeExtractionUtils.cc.

getMeasurementDimensions()

std::vector< int > getMeasurementDimensions ( const RecoTrack &  recoTrack )

staticprivate

Get a list of dimensions for each measurement in the reco track.

Needed for the derivatives. E.g. the residuals vector is as large as the sum of the returned vector here.

Parameters

recoTrack

The reco track from which the measurements should be taken.

Returns

An std::vector with the number of dimensions for each measurement. Has as many entries as there are measurements in the reco track.

Definition at line 257 of file TimeExtractionUtils.cc.

---

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

• tracking/eventTimeExtraction/utilities/include/TimeExtractionUtils.h
• tracking/eventTimeExtraction/utilities/src/TimeExtractionUtils.cc