- fCov[0] = kScalePulls*cov[0]; // variance of y0
- fCov[1] = kScalePulls*cov[2]; // covariance of y0, dydx
- fCov[2] = kScalePulls*cov[1]; // variance of dydx
+
+ //
+ // MI comment modification (03.05.2014)
+ // In currently used code variance is used to estimate the tracklet error
+ // see AliTRDtrackerV1::AliTRDLeastSquare::Eval()
+ // 2 problems:
+ // a.) Assumption about the intdpendent error not valid beacause of time response function
+ // b.) Constuction of cluster errors used in cl->GetSigmaY2() and AliTRDtrackerV1::AliTRDLeastSquare::Eval
+ // to be parameterized
+ // c.) For crossing pad-rows the linear fit AliTRDtrackerV1::AliTRDLeastSquare::Eval do not take into account step
+ // Therefore y possition is biased, bias is y dependent
+ // Effective correction factor used to rescale error estimates
+ // In future additional angular dependent error should be investigated ( as in TPC and in previous versions of the TRD tracking)
+ //
+ const Double_t kCorrelError=1.5;
+
+ fCov[0] = kCorrelError*kScalePulls*cov[0]; // variance of y0
+ fCov[1] = kCorrelError*kScalePulls*cov[2]; // covariance of y0, dydx
+ fCov[2] = kCorrelError*kScalePulls*cov[1]; // variance of dydx