Recent results from Kalman track fitting

1. Recent results from Kalman track fitting WANG Dayong Oct.26,2005

2. outline • Progress and status of KalFit • Some words for memory leakage problem • Present performance of the package

3. Progress and status of KalFit • Software re-arrange and evolvement • Improved event model for TDS and DST • Calculation of path length total&in each layer • Performance study and detailed debugging • Memory leakage • Running with MdcTrkRecon • Running together with TrkReco

4. int m_id; float m_mass; // mass assumption during the fit float m_length; //track path length from closest point to coord. origin to the last hit float m_tof; // time of flight correspond to length; int m_nhits; // number of total hits contained int m_stat; // status flag float m_chisq[2]; // chi square of fit 0:filter 1:smoother float m_ndf[2]; // degree of freedom for the fit 0:filter 1:smoother float m_pathl[43]; // path length in each layer HepVector m_zhelix; // 5 track parameters at zero point HepPoint3D m_zpivot; // position of pivot at zero point HepSymMatrix m_zerror; // error matrix at zero point HepVector m_fhelix; // 5 track parameters at 1st hit HepPoint3D m_fpivot; // position of pivot at 1st hit HepSymMatrix m_ferror; // error matrix at 1st hit HepVector m_lhelix; // 5 track parameters at last hit HepPoint3D m_lpivot; // position of pivot at last hit HepSymMatrix m_lerror; // error matrix at last hit Old reconstruction event model

5. int m_id; float m_mass[5]; // mass assumption during the fit float m_length[5]; //track path length from closest point to coord. origin to the last hit float m_tof[5]; // time of flight correspond to length; int m_nhits[5]; // number of total hits contained int m_stat[5]; // status flag float m_chisq[2][5]; // chi square of fit 0:filter 1:smoother float m_ndf[2][5]; // degree of freedom for the fit 0:filter 1:smoother HepVector m_zhelix; // 5 track parameters at zero point for pion HepSymMatrix m_zerror; // error matrix at zero point for pion HepVector m_zhelix_e; // 5 track parameters at zero point for electron HepSymMatrix m_zerror_e; // error matrix at zero point for electron HepVector m_zhelix_mu; // 5 track parameters at zero point for muon HepSymMatrix m_zerror_mu; // error matrix at zero point for muon HepVector m_zhelix_k; // 5 track parameters at zero point for kaon HepSymMatrix m_zerror_k; // error matrix at zero point for kaon HepVector m_zhelix_p; // 5 track parameters at zero point for proton HepSymMatrix m_zerror_p; // error matrix at zero point for proton HepVector m_lhelix; // 5 track parameters at last hit for pion HepSymMatrix m_lerror; // error matrix at last hit for pion HepVector m_lhelix_e; // 5 track parameters at last hit for pion HepSymMatrix m_lerror_e; // error matrix at last hit for pion HepVector m_lhelix_mu; // 5 track parameters at last hit for pion HepSymMatrix m_lerror_mu; // error matrix at last hit for pion HepVector m_lhelix_k; // 5 track parameters at last hit for pion HepSymMatrix m_lerror_k; // error matrix at last hit for pion HepVector m_lhelix_p; // 5 track parameters at last hit for pion HepSymMatrix m_lerror_p; // error matrix at last hit for pion New event model for reconstruction and DST

6. memory leakage problem • Problem description • Sources of memory leakage I encountered a. unmathed “new” or “alloc” b. condition skipped c. bad-writtern deconstructors d. problems in event model design • Problems unresolved

7. MdcRecEvent Model: a problem • If MdcRecHitCol not registered, Gaudi could not take care of everything • If we delete objects,conficts; if not obvious memory leak;

8. Possible solutions • Bi-directional relation preserved ,using vector(or SmartRef) of object itself • Only trk->vechits relation maintained, reverse relation with other means( e.g, id), MdcRecHitCol is independent • MdcRecHit modeled as a common class and no longer using MdcRecHitCol

9. Present performance • Run smoothly with both Pattern Recognition algorithms • Similar helix results can be obtained compared to 3-D Helix fitting • Material parameters compared to MC: difffences exist, but not substantial • Material effects correction can make the helix parameters better • No obvious efficiency loss

10. 1.0GeV/c Muon, theta=90degreeTrkReco+KalFit Entries 998 Mean 1.00 Sigma 0.005759

11. 1.0GeV/c Muon, theta=90degreeTrkReco Entries 998 Mean 0.9999 Sigma 0.005743

12. 1.0GeV/c Muon, theta=90degreeMdcTrkRecon+KalFit Entries 1000 Mean 1.00 Sigma 0.005459

13. 1.0GeV/c Muon, theta=90degreeMdcTrkRecon Entries 1000 Mean 0.9999 Sigma 0.005566

14. Some estimations • For He/C3H8(60/40), I=43.9eV, from Bethe-Bloch formula, for M.I.P, energy loss is about 1.1keV/cm, for Mdc, L~80cm, dE~90KeV(for small p,L is longer,dE is larger). For Al barrel in MC,d~1mm, 1.615MeV/g/cm2*2.7g/cm3~4.36MeV/cm,so in Al, dE~0.44MeV. For 400MeV/c proton, Beta*gamma~0.43,energy loss ~4 MIP, so totally MIP~0.8MeV. • From E-p relation, dP=dE/beta~2MeV • For helix fitting, its parameter is something of an average, so the differences of momentum should be half of dP,i.e,~1MeV • From the figures below, this argument roughly holds.

15. 0.4GeV/c Proton, theta=90degreeTrkReco+KalFit Entries 993 Mean 0.03915 Sigma 0.002117

16. 0.4GeV/c Proton, theta=90degreeTrkReco Entries 993 Mean 0.03903 Sigma 0.002144

17. 0.4GeV/c Proton, theta=90degreeMdcTrkRecon+KalFit Entries 988 Mean 0.3910 Sigma 0.002092

18. 0.4GeV/c Proton, theta=90degreeMdcTrkRecon Entries 988 Mean 0.3901 Sigma 0.002079

19. 0.4GeV/c Proton, theta=90degree Mean 0.006134 Sigma 0.05802 Mean 0.02898 Sigma 0.05809

20. 0.4GeV/c Proton, theta=90degree Mean 2.555 Sigma 0.01729 Mean 2.563 Sigma 0. 01754 Mean 0.05201 Sigma 0.2271 Mean 0.0520 Sigma 0. 2270

21. 0.4GeV/c Proton, theta=90degree Mean -0.002364 Sigma 0.01272 Mean -0.002364 Sigma 0.01272

22. Further check • More events with broader range of momentum and particle type • Using physics event to check • Using kinematic fitting to check error matrix thus given • Algorithm checking : stereo wires • Users are welcome to try it to help improvements