Muon ID work UCSB Claudio Jan 25, 07

1
Muon ID work _at_ UCSBClaudio Jan 25, 07

• At the 1st meeting we gave a brief summary of
  what we have been working on
• Time on the agenda today ? give a few more
  details
• This is talk that we gave at 11-28-06 ?-PRS
• http//indico.cern.ch/conferenceDisplay.py?confId
  9118
• With additions for background (in purple)

2
Muons at CMS

• (Up to) 4 muon stations _at_ CMS
• Barrel Drift Tubes (DT), Europe
• EndCaps Cathode Strip Chambers (CSC), USA
• In return B-field
• Each station measures a segment (X, Y, X', Y')
• X Y local coordinates
• Outermost station only X view

3
Traditional ? reco _at_ CMS

• Reconstruct track in muon detector only
• StandAlone muon
• Put together with silicon track into single fit
• Global muon

• PT improvement for global muon vs Si-track only
  above 250 GeV
• if dE/dX multiple scattering treated correctly
  (remember that TeV muon starts to shower!)
• if B field is OK
• if muon system aligned to O(100 ?)

4
High PT ?

• Slide from Valuev, Z'?? ? PTDR analysis

• Three different methods for Pt reco
• Global Muon Reconstructor (GMR) standard ORCA.
  Fit to all tracker and muon hits on track.
  Vulnerable to muon bremsstrahlung.
• Truncated Muon Reconstructor (TMR), UCLA (J.
  Mumfords talk at PRS/µ meeting on Sept. 16th,
  2003). Optimized combination of
  tracker-plus-first-muon-station and tracker-only
  fits.
• Picky Muon Reconstructor (PMR), Warsaw (P.
  Traczyks talk at PRS/µ meeting on March 15th,
  2005). Fit to tracker plus all muon chambers not
  appearing to contain electromagnetic showers.

• Best possible job at high PT is not simple

5
How we got involved

• Obviously need code to propagate track (and
  covariance) into muon detectors
• knowing about MS, dE/dX, geometry, B-field
• ORCA code based on GEANT3 ? not for CMSSW
• Were asked to step in and do quick-and-dirty job
  for CMSSW while waiting for more better code
• Long story short only code that works and is
  being used for now and foreseeable future is
  SteppingHelixPropagator (Slava)
• Slava committed to maintaining and improving it

6
Alternative approach to muons (1)

• Take Si-track, swim it to muon chambers, see if
  there are matching segments
• also calorimeter energy deposition
• No loss of PT resolution up to 250 GeV
• and above things are not so simple....
• Less requirements on ? system
• alignment, material, geometry, B-field,
  fitting...
• Get more information on ? quality
• Useful for BG rejection
• Global ? object only contains ?2 of global fit
  !!!!

7
Alternative approach to muons (2)

• Complementarity with global ? approach
• Play one against the other to find problems
• Good monitoring tool
• eg, quick plots of track-segment matches, etc
• Possibility for alternative HLT approach
• comments at the end

8
Alternative approach to muons (3)

• Not a new idea
• was in ORCA, PTDR
• CMS-speak Muon-ID as opposed to
  Muon-Reconstruction (?)
• low PT muons for b-tagging
• I think that, potentially, can be more than a
  niche-algorithm for b-tagging
• particularly at startup
• Opportunity to do it "our way"

9
Update on Sitrack--?segment match algorithm

• Claudio Campagnari, Dmytro Kovalskyi, Slava
  Krutelyov, Jake Ribnik
• UC Santa Barbara

10
Reminder

• ? ID starting from Si-track propagated to
  calorimeter and ? detector
• Calculate energy deposition in calorimeter
• Calculate distance between extrapolated track and
  reconstructed segments
• Based on
• SteppingHelixPropagator (Slava)
• TrackAssociator (Dmytro)
• Last ?PRS presentation Jake, August 22 2006
• Demonstrated that basic code works
• Track-segment pulls 1
• Calorimeter energies make sense

11
Some Plots from Jake's talk (1)
Single ? gun Monte Carlo events PT 5? 40
GeV Plot of energy in crossed ECAL and HCAL cells
12
Some Plots from Jake's talk (2)
typical pulls
CSC, X station 2
DT, X station 3
DT, Y station 1
CSC, Y station 1
13
What is new

• Started to work on algorithm (producer) to
  make information persistent for a subset of
  Si-tracks

14
Stuff in red to be determined. Much of it
configurable at run time
15
What is arbitration?

• Can have gt 1 Si-track reasonably well matched to
  a segment
• Can happen esp. in jets
• Arbitration
• decide which track the segment really belongs to
• essentially pattern recognition problem
• Leave it to the user or do it in producer?
• our sense is that we should do it in producer
• but code should be written such that this feature
  can be easily turned off

16
Producer

• Built around TrackAssociator
• which works!
• Simple matter to wrap some code around it to do
  what we want
• but still needs to be written!
• Goals of this study
• give a 1st conservative guess for the
  requirements (in red) on page 4
• Take a look at arbitration, with conservative
  requirements
• is it a serious issue or a 2nd order effect?

17
Data Sets

• ? gun events
• CMSSW 0_9_2
• 5 lt Pt lt 50 GeV/c
• CSA06 tt events
• Simulated with CMSSW 0_8_2
• Reconstructed with CMSSW 0_9_2

18
1st conservative guess at requirements
Loop over all Si tracks P or Pt gt min
PT gt 2 GeV/c
quite low.. maximizes confusion
Propagate to calo ? detector Assign segment to
track if matching is good enough
?Xlocal lt 5? OR ?Xlocal lt 5 cm No requirement
on Ylocal
Discard track unless segments from enough
stations are there
If tracks share segment perform arbitration (or
perhaps not)
Nstations 2
Make information from surviving tracks persistent
19
Nstations 2
3
4
2
0
1
Number of stations on single ? gun events
20
Arbitration Issues

• tt MC has isolated ? from W decays and
  non-isolated ? from b and c decays
• A good testing ground for these issues
• Before arbitration we find (PTgt2 GeV/c)
• 15 of ? from W share segments with other tracks
• 33 of ? from other sources share segments with
  other tracks
• For PT gt 5 GeV/c, fractions are 5 and 20

21
Two simple arbitration algorithms

• Pick segments with best Dx
• Pick segments with best
• Results (PT gt 2 GeV/c and PTgt 5 GeV/c)

22
Conclusions about arbitration

• It does not seem to be a major issue
• We'll implement these two algorithms in the
  producer
• Can be re-visited at a later stage

23
What is missing

• For ultimate ? ID, a candidate ? that goes
  through the middle of a (working) station and
  leaves no segment, should be "penalized"
• Current code cannot distinguish between cracks in
  coverage and true misses
• Fuzzy distinction because of multiple scattering
• TrackAssociator is built around concept of
  "calculate distance between track and segment"
• At the moment provides no information for
  stations where there are no segments
• Need to add functionality

24
Future Plans

• Implement producer as outlined in this talk
• Paying some attention to execution time
• Add (fuzzy) information on whether segment in
  station was expected or not
• Necessitates some rework of TrackAssociator
• in progress (Dmytro)
• Anticipate release of producer in January

25
What is new since then

• Dmytro reworked TrackAssociator to be able to
  answer the question of missing segments
• Does track extrapolate into or out of chamber
  acceptance?
• What is the distance from the edge, in cm as well
  as number-of-MS-sigmas
• Now debugging phase
• Have plan on how to store info in "muonID" object
• This kind of info will be useful for global muon
  too!
• Currently, global muon is fit to 2 or more
  stations
• If station 3 and 4 are missing, is it OK, or is
  it because the "muon" is really a pion that
  stopped?

26
Plan

• Hope to put producer in 1_3_0
• Interest from people to start using it
• eg, to define b?? tagging algorithm

27
A word about HLT (1)

• Default plan
• Level 2 Standalone muon reconstruction
• HLT Global muon reconstruction
• either by "standard" Si-tracking pattern
  recognition done in "region of interest"
• or by "custom" pattern recognition seeded by L2
  muon (Richman, Vlimant, UCSB)
• Can imagine alternative
• Skip L2, go directly to HLT, with "custom"
  pattern recognition seeded by L1, only
  reconstruct Si-track, then use "?-ID" for final
  decision
• Or some combination

28
A word about HLT (2)

• Optimal strategy depends on details of timing and
  what standalone (L2) reconstruction really
  rejects
• Hard to tell from MC
• Not pursuing these ideas at the moment
• Richman Vlimant will try reconstruction from L1
  seed