1. Physics Performance of the CMS Reconstruction Software with the First LHC Collisions
David Lange
Lawrence Livermore National Laboratory
Representing the CMS collaboration
CHEP 2010
David Lange, LLNL

2. d
CHEP 2010
David Lange, LLNL

3. Reconstruction process turns detector hits into analysis level objects
Local Detector Reconstruction
Physics object reconstruction
High-level reconstruction
Tracker hits
HCAL clusters
ECAL clusters
Muon segments
Tracks
Photons
Muons
Electrons
Jets
Missing energy
b tagging
Particle Flow
Tau
Vertexing
We have the ability to redo high-level reconstruction components at the analysis stage. This provides flexibility to include latest developments
CHEP 2010
David Lange, LLNL

4. Results of prompt reconstruction is already of sufficient quality for analysis
High-level trigger Real-time local and regional reconstruction
Express reconstruction
Availability: ~60 minutes
Usage: Calibrations, data quality
Prompt reconstruction
Availability: 48 hours
Re-reconstruction
Usage: Make coherent data sets (e.g., consistent software release, calibrations)
Monte Carlo
(Re)Reconstruct Geant4 samples to match reconstruction used in data
Already achieved good operational stability. Working to incorporate lessons learned from initial data analysis into official reconstruction algorithms
CHEP 2010
David Lange, LLNL

5. CMS production scheme has led to reliable offline operations
Code development cycle:
Nightly and ~weekly validations
6-8 week dev. cycle
Production using small subset of data/MC
Conditions updates
Production release
Pre- Production
Validation group signoff?
Code bug fixes No
Validation procedure is transitioning to be data driven in both the code development and pre-production stage
At this stage, the validation sign-off is largely performed by hand
Yes
Production start
CHEP 2010
David Lange, LLNL

6. Lightning tour of CMS reconstruction methods and results
Validation procedures enabled us to derive most ICHEP results from prompt reco
Results are primarily from ICHEP2010
CHEP 2010
David Lange, LLNL

7. CMS tracking in a nutshell
CHEP 2010
David Lange, LLNL

8. CMS tracking in a nutshell
CHEP 2010
David Lange, LLNL

9. CMS tracking in a nutshell
CHEP 2010
David Lange, LLNL

10. We achieved excellent tracking performance
Energy scale: KS mass to 0.3 MeV over kinematic range
Material model: ~10% agreement with MC model)
Track parameters / impact parameter agree with MC
99+% tracking efficiency measured on J/ymm data
CHEP 2010
David Lange, LLNL

11. Muon reconstruction
Tight muon selection requires combined fit of tracker and muon system hits and hits in at least 2 muon stations
Soft muon selects muons down to 0.5 GeV
Tracker matched to 1+ CSC or DT segs.
CHEP 2010
David Lange, LLNL

12. Muon ID performance: 90+% efficiency above 7 GeV
Tag/Probe efficiency measurement
Mis-ID measurement
CHEP 2010
David Lange, LLNL

13. Di-muons in CMS
CHEP 2010
David Lange, LLNL

14. Photon candidates built from superclusters.
Purity increases with pT.
Observe p0 and h resonances in agreement with MC expectations. Correct energy scale (O(1%))
CHEP 2010
David Lange, LLNL

15. Electron Identification
Utilize two complementary approaches
Start from ECAL superclusters, then look for hits in tracker
Start from tracks, then look for ECAL hits
Brem. energy-loss model to fit electron track (GSF)
Preselect candidates based on matching criteria. Apply further selections at analysis level based on efficiency and fake requirements (fake rate measured to be depending on working point and pT)
CHEP 2010
David Lange, LLNL

16. Particle Flow reconstruction in CMS
Identify individual particles (tracks and clusters).
Use to build jets, electrons, identify taus from their decay products, tag b jets, etc…
Example application: J/yee
CHEP 2010
David Lange, LLNL

17. Jet reconstruction Single particle response Jet pT resolution
Data vs MC agreement at about 10% level or better already
PF approach improves resolution significantly over calorimetric method
CHEP 2010
David Lange, LLNL

18. Missing ET : Calorimeter noise cleaning algorithms bring MET performance to MC expectations
Data/MC agreement in MET distribution is seen over several orders of magnitude.
Particle-flow based MET techniques show improved performance over calorimetry based techniques
Both event and cluster level calorimeter cleaning algorithms developed and applied for ICHEP results
CHEP 2010
David Lange, LLNL

19. B-tagging discriminator algorithms: Commissioned from the very beginning
Jet Probability
Algorithm
tags jets according to the probability of all the tracks in the jet to originate from the primary vertex, given their IP significances
Track Counting Algorithm
tags jets containing N tracks with Impact Parameter (IP) significance exceeding S
High Purity configuration: N=3
SSV Algorithm
tags jets according to the 3D flight distance significance of the reconstructed secondary vertex
High Purity configuration:
Vertices with 3 or more tracks
CHEP 2010
David Lange, LLNL

20. CMS is in the midst of a successful first run
Thanks to careful preparation prior to the first data: the reconstructed data shows an exceptional agreement with expectations for all the high-level physics objects
The rapid increase in luminosity will bring further improved calibrations and algorithms
Zmm
CHEP 2010
David Lange, LLNL