Software Workshop 05 May Report

1
Software Workshop05 May Report

• Grant Gorfine
• Marcus Elsing
• Sergio Gonzalez
• Roland Haertel
• Alexandre Rozanov

2
Grant GorfineGeometry Target precision

• AIM
• 1 level in terms of radiation length at eta lt
  2.5.
• Motivation
• Desire to have good E/p calibration. 2 is
  needed for this.
• Calibration mainly from z ? ee which has more
  relaxed requirements but E/p calibration is an
  important cross check.
• Doesnt mean 1 for every component but 1 on
  total in regions of a few centimeters.
• While this is our aim we will be quite
  satisfied if we reach 2. Even at 5 it is a
  challenge and many feel it will be hard to reach
  that.
• This requirement is more important the closer we
  are to the interaction point.

3
Geometry People

• Pixel
• Collection of Info Alexandre Rozenov
• Implementation Grant Gorfine, Jean-Baptiste
  Devivie
• SCT
• Collection of Info Stephen Haywood
• Implementation Grant Gorfine, Daisuke Naito
• TRT
• Collection of Info Fred Luehring
• Implementation Thomas Kittleman
• Service Material
• Collection of Info Vadim Kostyukhin
• Implementation Vadim Kostyukhin and shared by
  sub systems.

4
Cosmics - Geometry Configurations

• Cosmic testing (if time) in SR1 surface building
  before going to pit.

5
Alignment

• Alignment infrastructure.
• Alignment nodes at different levels in hierarchy
• Level 1 Whole pixel, SCT Barrel, SCT Endcap, TRT
  Barrel, TRT Endcap
• Level 2. Whole layers and disks in Si. Modules in
  TRT
• Level 3. Individual Detector Elements in Si.
• Alignment shifts are accessed from the CondDB
  using IoV callback mechanism.
• Transforms are then applied to the appropriate
  alignment nodes in GeoModel description. Cached
  values in readout geometry are invalidated.
• Infrastructure successfully used in CTB and by
  alignment community.

6
Alignment

• Data sets being defined by Calib/Align group.
  Proposed sets.
• 1 set with know random misalignments.
• 1 set with systematic displacement of high level
  structures (such as the 2mm shift between whole
  pixel wrt to SCTTRT)
• 1 set with systematic shifts such as sagitta
  distortion of layers.
• 1 blind set.
• For these sets we would need samples with high
  energy tracks (pT gt 2 GeV).
• muons in particular such as a trigger sample.
• other particles OK as well (except electrons
  because of bremsstrahlung)

7
Alignment Solenoid axis

• Need to simulate misalignment of solenoid field
  wrt to ID.
• Need to converge on coordinate system for
  Simulation.
• Do we move ID or the field map or both?
• If we move the field then ID tracking needs to
  shift frame to solenoid frame - This requires
  some work still.
• Implications for combined reconstruction
• We should sort this out for DC3!!

8
Digitization - Calibration/Conditions

• The conditions type data used in reconstruction
  is generally a smaller sub set of what is need in
  digitization to reproduce run conditions.
• Those needed by reconstruction.
• Silicon
• Those effecting Lorentz angle (Temp, Bias and
  Depl Voltage - still to be put in as conditions)
• Noisy/dead channels (and perhaps partially
  efficient ones)
• Used in CTB with Cond DB.
• TRT
• Tzero, R-t relationship
• Used CondDB in CTB for real data.
• Dead channels.
• Also in CTB for real data.
• Others needed to reproduce run conditions
• Long list see web pages.

9
Reconstruction - High Level Design

• Updated view on Inner Detector Reconstruction
• New Finding (RTF) replaces old packages
• New Reconstruction Tools
• Foreseen to have 2nd Stage (e.g., Pixel
  re-clustering) and V0/Brem/e/m code

10
RTF Tracking - Considerations

• Past packages aim at resolving the EVENT !
• This implies they give the ONLY and FINAL answer
• Not an open system !
• New RTF Tracking uses different approach
• Multiple Track Finding Algorithms
• Aim at finding ALL relevant candidates (depends
  on the task specified)
• Aim at reasonable rate of fakes
• Ambiguity Solution Algorithm to resolve event
• Optimal solution w.r.t. list of criteria (b-layer
  hits, NDF, chi2, )
• Like a chess problem
• We will have STAGES
• SiSPSeededTrackFinding -gt Ambi -gt TRTExtension -gt
  2nd Stage
• Specific Finding code for Brem, V0, integrated
  later in chain
• This is not new - both IPATREC and XKALMAN had
  similar approaches
• But these were used in the box, not visible

11
Status of RTF Tracking

• SiSPSeededTrackFinding
• Most Tools needed are in nightlies
• Igor finishes prototype implementation of
  Algorithm
• Ambiguity Processing
• Prototype (so called N2 method) is in the
  Nightlies (Ed)
• Needs caching and should become recursive for
  timing optimization
• 2nd method xKalman based - later
• 3rd method so called Full method - later
• TRT_Extension
• Next for Igor to do after SiSPSeededTrackFinding
• Needs TRT_DetElementsRoadMaker and a few other
  things
• Aim at Inner Detector week for first
  implementation and tests
• Material integration another major step coming in
  the next weeks
• InDet prototype under testing (Andi)
• Still less complex than (Combined) Muons -
  torture meeting next week

12
Status for Release 11.0.0

• Release 11.0.0 prototype strategies equivalent
  to default xkal/ipat
• Aim to complete Event Filter integration for
  Prototype
• Performance and Timing will not be compatible for
  first prototype
• Work between Release 11.0.0 and 12.0.0 to tune
  those
• Proposal Use IPatRec/xKalman for Physics
  production based on 11.0.0
• Reconstruction for Commissioning uses new
  Tracking Tools and EDM
• No legacy packages, InDetCTBTracking is 100 new
  code
• See talk of Thijs
• Towards Release 12.0.0
• Change baseline in 11.x.0 when physics
  performance of new prototype equivalent to
  default IPatRec/xKalman
• Then work on timing performance for release
  12.0.0
• In Parallel new features to complete Inner
  Detector reconstruction
• Migration of IPatRec is becoming a major concern
• Muon support is eating up Alans time
• Strategy to move MuID/Moore and then IPatRec -
  but risk to time out, solution ?
• Open system - possible to reintroduce IPatRec
  pattern modules for Release 13.0.0

13
Alignment issues

• Peter Hansen and others observed a matching
  problem in the TRT from silicon standalone
  extrapolated tracks, depending on the y-value of
  the track intercept in the TRT detector
• Possible problem in one Si sector ???

• Reconstructed momentum
• black pixel SCT
• blue pixel SCT for tracks with d0gt0
• red pixel SCT for tracks with d0lt0
• green SCT only

Nabil Ghodbane
14
More checks

• Peter Hansen also reported an imrovement for the
  rms of the residual
• (hit track ) of the extrapolated full silicon
  track to the TRT from 200 mm
• to 165 mm.
• The reconstructed momentum using individual
  detectors was also better, and now there are no
  clear differences acoording to the sign of the
  track d0 parameter

• Reconstructed momentum
• black pixel SCT
• blue pixel SCT for tracks with d0gt0
• red pixel SCT for tracks with d0lt0
• green SCT only

15
RobustAlign Roland Haertel
16
ESD by XKalman
17
Aligned one Aligned, rms should be 1
18
Outline

• b-tagging from AOD
• Methods 2D, 3D, SV1
• Datasets WH(400), tt, ttH/ttjj
• Rome AOD b-tagging vs DC1

19
Data sample of ttH/ttjj/tt events
ttH(4854)/ttjj(4856) events, mH120 GeV, Rome
initial 2 pixel layers layout, b-layer 400 µm, no
pileup, no inefficiencies

• b- from ttH, u- from ttjj
• full simulation jets ?R lt 0.4
• reconstructed primary vertex
• b,c labelling with ?Rjbc lt 0.3 , no
  additional jet cleaning
• release 10.0.1 few patches
• generic b-calibration (DC1 datasets
  ttHttjjttbbWH), all b and udsg ,?Rjbc lt 0.8,
  ATHENA 9.3.0, with inefficiencies, low
  luminosity
• iPatrec track finding
• missing shared/ambigious hits
• missing suppression secondary/V0 tracks
• missing BadTrack algorithm

20
WH (400 GeV) dataset
b-quarks from WH(400)-gtbb (4852), light quarks
from WH(400)-gtuu(4852)
21
tt dataset
b-quarks from tt(4100), light quarks also from
tt(4100)
22
ttH/ttjj datasets
b-quarks from ttH(4854), light quarks from
ttjj(4856)
23
Compairison of Rome AOD vs DC1
ttH(4854)/ttjj(4856) events, mH120 GeV, Rome
layout, initial 2 pixel layers layout, b-layer
400 µm, no pileup, no inefficiencies

• correction for low pile-up and inefficiency
  0.970.920.89
• DC1 ATRECON-Xkalman-7.5
• DC1 Primary vertex from VKalVrt
• DC1 ATLFAST jets ?Rjbc gt 0.3 cleaning
• DC1 G3 Rome G4
• DC1 Specific calibration, selection of
  secondary/V0 tracks, BadTrack method,
  shared/ambigious hits

24
Rome vs DC1 for ttH/ttjj -2D
2D method, b-quarks from ttH(4854), light
quarks from ttjj(4856)
25
Rome vs DC1 for ttH/ttjj 3D
3D method, b-quarks from ttH(4854), light
quarks from ttjj(4856)
26
Rome vs DC1 for ttH/ttjj SV1
SV1 method, b-quarks from ttH(4854), light
quarks from ttjj(4856)
27
Conclusions

• b-tagging with AOD works
• 4 datasets show reasonable performance
• specific difference should still be traced and
  understood
• despite many (more than 10) major differences
  in software between DC1 and Rome AOD b-tagging,
  the final b-tagging results are very close.
  Probably the law of big numbers works
• many improvements still to be implemented after
  Rome