Presentation is loading. Please wait.

Presentation is loading. Please wait.

José E. García b -tagging performance with xKalman and iPatRec With the help of: S. Correard, I. Gavrilenko, S. González, A. Poppleton, E. Ros, A. Rozanov,

Similar presentations


Presentation on theme: "José E. García b -tagging performance with xKalman and iPatRec With the help of: S. Correard, I. Gavrilenko, S. González, A. Poppleton, E. Ros, A. Rozanov,"— Presentation transcript:

1 José E. García b -tagging performance with xKalman and iPatRec With the help of: S. Correard, I. Gavrilenko, S. González, A. Poppleton, E. Ros, A. Rozanov, J. Sánchez and M. Vos

2 Tools S/W Workshop – December 03 Jose E. Garcia b-tagging … ATLAS kit release 7.0.0 xKalmanppAtrecon-00-00-89  Reconstruction with default values xKalmanppAthena-00-00-87  Selectivity = 2 & xKalman field iPat-02-09-00  Reconstruction with default values o Saved truth information for all particles DC1 samples WH(120), WH(400) and WH(400) + pileup 50 K events: H  uu 20 K events: H  bb Reconstruction @ farm

3 Method S/W Workshop – December 03 Jose E. Garcia b-tagging … ATLAS Communication, M. Vos et al. Standard ATLAS 2D b-tagging and the 3D (2D+Z) algorithm ( ATLAS Communication, M. Vos et al. ) Significance = impact parameter divided by it error Track weight calculated depending on significance S w 2D = log [ f b (S)/f u (S)] w 3D = log [ f b (S d,S z )/f u (S d,S z )] Primary vertex reconstruction with a  30  m error assumed Same events for likelihood and analysis ATLFAST jets used in the analysis Jets coinciding with the direction of the parton from the Higgs decay (  R jp < 0.2)

4 Cuts S/W Workshop – December 03 Jose E. Garcia b-tagging … |d 0 | < 1 mm Number of precision hits  9 Number of pixel hits  2 At least one hit in the b-layer.  2 of the track:  2 < 3 Longitudinal impact parameter: z 0 < 1.5 mm No shared hits in the b-layer Maximum two shared hits in Pixels + SCT p T > 1 GeV, |  | < 2.5 No ambiguity in the first wafer ID cuts: High quality cuts Standard b -tag cuts Very high quality cuts

5 Resolution S/W Workshop – December 03 Jose E. Garcia b-tagging … No difference in the p T resolution between the three algorithms xKalmanAthena has less resolution in the impact parameter. iPatRec and xKalmanAtrecon have similar performance

6 Significances S/W Workshop – December 03 Jose E. Garcia b-tagging … Signed track significance for u and b -jets from H(120) decay Distribution for u-jets is independent of the algorithm xKalmanAtrecon and iPatRec distributions have the same shape. Small difference for the high significance region in the xKalmanAthena algorithm. u b

7 Weights S/W Workshop – December 03 Jose E. Garcia b-tagging results Jet weights distribution for u and b -jets from H(120) decay ATLFAST used for jet reconstruction Exactly the same jets used for the two Athena algorithms. No much difference between jets in xKalmanAtrecon and Athena. Jet weight calculated as the sum of all track weights. b u

8 b-tagging S/W Workshop – December 03 Jose E. Garcia b-tagging … u -jet rejection as function of b-jet tagging efficiency for H(120) sample xKalmanAtrecon performance has not changed since release 5.3.0 xKalmanAthena and iPatRec improve performance with increasing releases (see previous talks from Marseille group) Performance of xKalmanAtrecon is still better than both Athena algorithms. R u for  b = 50% is bigger than 100.

9 Rejections S/W Workshop – December 03 Jose E. Garcia b-tagging … 2D algorithm3D algorithm R(  b ) AtreconxKalmaniPatRecAtreconxKalmaniPatRec R (  b = 50 %) 180 ± 8133 ± 5150 ± 6397 ± 24292 ± 17372 ± 22 R (  b = 60 %) 57 ± 243 ± 147 ± 1117 ± 483 ± 296 ± 4 2D algorithm3D algorithm R(  b ) AtreconxKalmaniPatRecAtreconxKalmaniPatRec R (  b = 50 %) 139 ± 693 ± 395 ± 3207 ± 8139 ± 6145 ± 5 R (  b = 60 %) 55 ± 138 ± 140 ± 178 ± 252 ± 161 ± 1 WH(400) sample WH(120) sample

10 Pt dependency S/W Workshop – December 03 Jose E. Garcia b-tagging … Rejection of u -jets for  b = 60 % versus transverse momentum of jets Rejection for u -jets with combined WH(120) and WH(400) data samples Athena algorithms have around 25% less rejection than Atrecon Slightly better performance of iPatRec comparing the two Athena algorithms

11 Eta dependency S/W Workshop – December 03 Jose E. Garcia b-tagging … Rejection of u -jets for  b = 60 % versus |  | Combined samples WH(120) and WH(400) For |  | > 1.5 results from the three algorithms are almost equal Better performance of xKalmanAtrecon at small |  |

12 WH(400) + pileup S/W Workshop – December 03 Jose E. Garcia b-tagging … High luminosity: 23 minimum bias events High luminosity: 23 minimum bias events After cuts up to 30% of tracks in the jet come from pileup After cuts up to 30% of tracks in the jet come from pileup Pileup tracks primary vertices are generally displaced by large distances (~ 5.6 cm) Pileup tracks primary vertices are generally displaced by large distances (~ 5.6 cm) Z cut becomes more important Z cut becomes more important More information on the xKalmanAtrecon: The b-tagging performance of the complete ATLAS DC1 layout using WH events, M.Vos et al., ATLAS communication More information on the xKalmanAtrecon: The b-tagging performance of the complete ATLAS DC1 layout using WH events, M.Vos et al., ATLAS communication

13 iPatRec - PRELIMINARY WH(400) + pileup (II) S/W Workshop – December 03 Jose E. Garcia b-tagging … RuRu 2D algorithm3D algorithm R(  b ) with pileupwithout pileupratiowith pileupwithout pileupratio R (  b = 50 %) 130 ± 5139 ± 60.93189 ± 8207 ± 80.91 R (  b = 60 %) 49 ± 155 ± 10.8971 ± 278 ± 20.91 RuRu 2D algorithm3D algorithm R(  b ) with pileupwithout pileupratiowith pileupwithout pileupratio R (  b = 50 %) 92 ± 395 ± 30.96129 ± 4145 ± 50.89 R (  b = 60 %) 40 ± 1 0.9957 ± 161 ± 10.92 xKalmanAtrecon WH(400) + pileup

14 Conclusions S/W Workshop – December 03 Jose E. Garcia b-tagging … xKalmanAtrecon still has better performance than the Athena algorithms Needed a check with the error parameterization “a la Atrecon” for SiClusters Needed a check with the error parameterization “a la Atrecon” for SiClusters 3D method improves considerably rejection 3D method improves considerably rejection Pileup gives a deterioration around 10% in the rejection Pileup gives a deterioration around 10% in the rejection iPatRec has less deterioration than xKalmanAtrecon for the 2D method, being a bit worse for the 3D xKalmanAthena results maybe more sensitive than iPatRec to error in SiClusters WAIT for release 7.4.0


Download ppt "José E. García b -tagging performance with xKalman and iPatRec With the help of: S. Correard, I. Gavrilenko, S. González, A. Poppleton, E. Ros, A. Rozanov,"

Similar presentations


Ads by Google