1. Study of 1D Hit Error Assignment Marco Terranova, Filippo Pisano, N. Amapane, G. Cerminara

2. Marco TerranovaSep 28, 2010 2 Introduction Correct hit position error assignment is essential for segment pattern recognition and for the STA/GLB fits. Current DB tables specify a fixed hit error,  = 500  m –Same for all chambers/SLs –A recent issue with validation of new MC IDEAL conditions has brought the attention to proper hit error assignment Reported by Javier S.: https://hypernews.cern.ch/HyperNews/CMS/get/muon-object-validation/164/2/1.html https://hypernews.cern.ch/HyperNews/CMS/get/muon-object-validation/164/2/1.html Our goal: Study the effect of 1D hit error assignment on reconstruction –Estimate uncertainty to be assigned to hits based on observed residuals –Apply it to hit reconstruction and check the results

3. Marco TerranovaSep 28, 2010 3 Sample and Event Selection Sample used: /MuOnia/Run2010A-CS_Onia-v6/RAW-RECO –Central skim of global/tracker di-muons with M  > 2 GeV –Includes RAW (necessary for re-reco) –Additional filters: good vertex, noscraping Note: New skim of golden Z + W candiates with RAW recently requested. Will be the ideal sample for this kind of studies. Consider only events with one GlobalMuon or TrackerMuon with p T >5 GeV –p T cut changes the distribution of segment phi angles! Use only 4D segments with: –|  |<25 deg –at least 7 phi hits and 4  theta hits

4. Marco TerranovaSep 28, 2010 4 Resolution and Residuals Segment residual distribution widths (  res ) are smaller than the hit resolution  because the hits themselves are used in the segment fit): E.g.: W0, MB1, theta SL L1  res = 165  m L2  res = 255  m L3  res = 256  m L4  res = 167  m  = k ∙  res Under general assumptions, for a single SL (e.g. theta): –Outer layers (1,4): k 1,4 = 1.83 –Inner layers (2,3): k 2,3 = 1.20 –Cf. Jesus Puerta’s PHD thesis

5. Marco TerranovaSep 28, 2010 5 Resolution and Residuals: Phi Extending to a 8-layer (L=1…8) segment over 2 phi SLs: K 1,8 = 1.17 K 2,7 = 1.16 K 3,6 = 1.15 K 4,5 = 1.14 Therefore, to get the resolution, we: –Fit residual distributions for each layer separately With a somewhat arbitrary tuning of the fit due to non-Gaussian tails –Average the resulting  = (  res ∙ k) for all layers in a chamber, separately for phi and theta 1 2 3 4 5 6 7 8 SL1 SL3

6. Marco TerranovaSep 28, 2010 6 We are currently overestimating the phi hit error Almost same value for all sectors of the same W/station (  average) Good symmetry between +/– wheels worse resolution in MB4 no correction for signal propagation along the wire present in MB4 (no theta) Current DB value: 500  m MB1MB2MB3MB4 –  (  m) Result: phi resolution

7. Marco TerranovaSep 28, 2010 7 Results: theta resolution Similar features, but: In some wheels of MB2, large differences between sectors –…see next slide: closer look at W-1 MB2 Current DB value: 500  m MB1 MB2 MB3 –  (  m)

8. Marco TerranovaSep 28, 2010 8 Example: 12 sectors in W-1, MB2, SL2, L1 Residuals are very different among sectors! To be investigated… –We will use sector-averaged errors in the following –If these differences are confirmed, we should assign sector-specific hit errors!

9. Marco TerranovaSep 28, 2010 9 Resolution: Data-MC (CMSSW 3.8.2) comparison In phi, resolution is worse in MC STARTUP than in data –Although the conditions deployed in 3.8.2 already improved it a lot! –In theta, STARTUP MC is similar to data (sometimes better) This is useful input for the re-tuning of the STARTUP scenario Theta Phi 500  m

10. Marco TerranovaSep 28, 2010 10 Phi pull CMSSW 3.8.2 default Theta pull Tuned errors Phi pullTheta pull Pull widths get closer to 1 Distributions not perfectly Gaussian –Resolution changes a lot within the cell, while we assign a fixed error –Estimate of pull width depends on fit range (Thanks to Javier Santaolalla for providing these plots in the validation suite) Effect on reconstruction: True MC pulls Example: W0 MB1, IDEAL MC (SingleMuPt100)

11. Marco TerranovaSep 28, 2010 11 True MC pulls: all wheels/stations MC Pulls close to 1 IDEAL (SingleMupT100), tuned errorsSTARTUP (ZMM), tuned errors Pull

12. Marco TerranovaSep 28, 2010 12 Reconstruction is not broken by the corrected errors, although they are much smaller than current ones  2 /DOF #DT hits in STA fit#DT hits in global fit GlobalStandalone GlobalStandalone – Default errors – Tuned errors Effect on Track Fit: data Global muons with |  | 10 GeV

13. Marco TerranovaSep 28, 2010 13 Same conclusion Effect on Track Fit: STARTUP MC (Zmm) Global muons with |  | 10 GeV  2 /DOF #DT hits in STA fit#DT hits in global fit GlobalStandalone GlobalStandalone – Default errors – Tuned errors

14. Marco TerranovaSep 28, 2010 14 Conclusions Uncertainties currently assigned to hits (500  m) are not optimal –First tests of tuning hit errors from residuals are encouraging –Procedure was tested on MC as well True Pulls widths get closer to 1 Also: a better picture of how STARTUP scenario should be modified to better match data We observed some effects in theta SLs of some wheels (e.g. -1) of MB2 –to be investigated –Should we adopt sector-specific uncertainties? Outlook: –We should compare with resolutions obtained with the meantimer method See Silvia’s presentation earlier today –Study how to assign different errors for the first step of hit reconstruction At the moment the same error is assigned for “step 1” (pattern recognition) and for the final hits –At some point, dependence of resolution on e.g. distance from the wire should be accounted for