Wiki SiStripDeDxStudies

How to use dE/dx for hadron-id in your analysis

Analysis Note on CVS

At FYNU:

export CVSROOT=:pserver:[login]@server06.fynu.ucl.ac.be:/home/cvsfynu/cvshe
cvs login
cvs co Documents/note-HSCP
cd Documents/note-HSCP/
latex main.tex

Talks:

• HSCP meeting, 2/2/2009

Code is in cd /afs/cern.ch/cms/CAF/CMSALCA/ALCA_TRACKERCALIB/SiStrip/pIDAnalysis/CMSSW_3_2_5/src/Andrea/DeDxVsTime/

Talks:

• Tracker DPG general meeting, 17/9/2009
• Calibration meeting, 4/9/2009

Code:

• Instructions for Andrea V.'s tools for time correlations
• DiscriminationPower.cc
• Cosmics_dbxfromevaux_cfg.py
• retriggering.C

Talks:

• Calibration meeting, 3/4/2009

Index:

• Analysis
• Standard plots
• Likelihood
• Weighted mean
• Neural network analysis
• Cosmics at MTCC
• Cosmics at TIF
• Cosmic showers at TIF
• Attenuation of the signal as a function of position and β
• Applications to Minimum Bias, jets, Heavy Stable Charged Particles
• Calibration with V0's
• Systematics
• Capacitive coupling and cluster thresholds
• Talks and publications
• References

1st step: generation of tracks with a particle gun

Particle gun (flat in energy or pt) + detector simulation + digitization (only pixels+strips) + track reco (instructions here).
Script to send jobs to lxbatch: gensimudigi.sh. Create configuration files with: makecfg.sh.
I store my files on castor, directory /castor/cern.ch/user/g/giamman/tracks. Particles types: e^-, μ^-, π⁺, K⁺, p.

2nd step: reading tracks, retrieving rechits, calculating dE/dx

Content of my RecoTracker/DeDx directory (CMSSW_1_2_0):

• BuildFile
• src/DeDx.cc
• interface/DeDx.h
• test/BuildFile
• test/AnalyzeTracks.h
• test/AnalyzeTracks.cc
• test/SealModules.cc
• test/AnalyzeTracks.cfg

How to use the DeDx class in your EDAnalyzer:

• In your constructor: DeDx* myDeDx = new DeDx(conf);
  (where conf is an edm::ParameterSet const&)
• Once per event (e.g. in the analyze() routine): myDeDx->init(event,eventsetup);
• For each track: myDeDx->compute(*track);
  After this call, it is possible to call the following functions for each track:
• truncated arithmetic mean: double mean = myDeDx->trunc_mean(trunc);
  (trunc: fraction of measurements to throw away; use 0 for the usual mean)
• median: double median = myDeDx->median()
• harmonic mean of a given power of dE/dx: double harm = myDeDx->harmonic_mean(power);
  (power=1/2 was used in H1, but I found a better separation with power=2, see plots)
• list of all dE/dx measurements: vector dedxMeas = myDeDx->allMeasurements();
• In your endJob() routine: TH1F* myHistogram = (myDeDx->getHistograms(i)).at(j);
  gives an histogram filled hit by hit
  i = 0: all detectors, 1: pixel barrel, 2: pixel endcap, 3: TIB, 4: TID, 5: TOB, 6: TEC
  j = 0: energy deposit, 1: module thickness, 2: cos θ (angle between the track and the normal to the module), 3: dE/dx, 4: number of hits
• In your destructor: delete myDeDx;

Script to send jobs to lxbatch: dedx.sh. (Full chain, from particle gun to dE/dx: fullchain.sh.)
So far there's no way to access directly the track measurements from the RecoTracks, so I'm re-fitting the track (thanks to Chiara); in 1_1_0 this should not be necessary.

3st step: macros for high level analysis

plotdedx.C
Script for full analysis for all particle types and several estimators: plotdedx.sh

nndedx.C (for neural network analysis)

Main plots and numerical results:

[legend: muons, pions, kaons, protons, electrons]

Number of hits per track (only SST, muons 3<p_T<5 GeV):

Estimator: generalized harmonic mean of dE/dx (only strips, no pixel; at least 3 hits)

profile histogram and separation vs P
In the [900 MeV, 1100 MeV] momentum range: linear superimposed plot, individual log plots with gaussian fit
Main quality criteria [900 MeV, 1100 MeV]:
muon chi2/dof: 3.8
muon sigma: 6.7 %
pion/muon separation: 0 sigmas
kaon/pion separation: 1.3 sigmas
proton/kaon separation: 3.7 sigmas
proton/pion separation: 4.8 sigmas

Likelihood hit-by-hit:
Macro used to create table of integrals from -inf to x, for a standard Landau
Macro used for the following plots

• Momentum scan (dE/dx hit-by-hit):
• 0 < dx < infinity
• muons
• pions
• kaons
• protons
• 0 < dx < 350 um
• muons
• pions
• kaons
• protons
• 350 um < dx < 509 um
• muons
• pions
• kaons
• protons
• 509 um < dx < 560 um
• muons
• pions
• kaons
• protons
• 560 um < dx < infinity
• muons
• pions
• kaons
• protons
• Peak and width/peak vs momentum:
• 0 < dx < infinity
• muons
• pions
• kaons
• protons
• 0 < dx < 350 um
• muons
• pions
• kaons
• protons
• 350 um < dx < 509 um
• muons
• pions
• kaons
• protons
• 509 um < dx < 560 um
• muons
• pions
• kaons
• protons
• 560 um < dx < infinity
• muons
• pions
• kaons
• protons

Likelihood at track level:
Macro used for the following plots

• harm2:
• muons, scan
• pions, scan
• kaons, scan
• protons, scan
• muons, curve
• pions, curve
• kaons, curve
• protons, curve
• harm4:
• muons, scan
• pions, scan
• kaons, scan
• protons, scan
• muons, curve
• pions, curve
• kaons, curve
• protons, curve
• harm6:
• muons, scan
• pions, scan
• kaons, scan
• protons, scan
• muons, curve
• pions, curve
• kaons, curve
• protons, curve
• mean04:
• muons, scan
• pions, scan
• kaons, scan
• protons, scan
• muons, curve
• pions, curve
• kaons, curve
• protons, curve

Weighted mean:
Macro for estimating the weights
Plots of dE and log(dE) vs dx, and spreads:

• 0.8<p_T<0.9 GeV:
• muons
• pions
• kaons
• protons
• 0.9<p_T<1.0 GeV:
• muons
• pions
• kaons
• protons
• 1.0<p_T<1.5 GeV:
• muons
• pions
• kaons
• protons
• 1.5<p_T<2.0 GeV:
• muons
• pions
• kaons
• protons
• 2.0<p_T<3.0 GeV:
• muons
• pions
• kaons
• protons
• 3.0<p_T<5.0 GeV:
• muons
• pions
• kaons
• protons

Neural network analysis:

My best networks so far:

• protons vs pions (inputs: n.hits, P, η, dP/dr, Pin-Pout, dE/dx), SteepestDescent, 4+0 hidden neurons, 2800 cycles
  S/B for validation sample: P<1.5 GeV, |η|<0.8, P<1.5 GeV, |η|>0.8, P>1.5 GeV
  All particle types: Profile histogram and separation vs P
  All particle types: In the [900 MeV, 1100 MeV] momentum range
• electrons vs pions (inputs: n.hits, P, η, dP/dr, Pin-Pout, dE/dx), SteepestDescent, 4+0 hidden neurons, 3000 cycles
  S/B for validation sample: P<1.5 GeV, |η|<0.8, P<1.5 GeV, |η|>0.8, P>1.5 GeV
  All particle types: Profile histogram and separation vs P
  All particle types: In the [900 MeV, 1100 MeV] momentum range
  All particle types: In the [5 GeV, 10 GeV] momentum range
• Scatter plot of the two nets:
• In the [900 MeV, 1100 MeV] momentum range
• In the [3 GeV, 10 GeV] momentum range
• In the [5 GeV, 10 GeV] momentum range

Cosmics at MTCC:
Reconstructed cosmic tracks from MTCC are in /castor/cern.ch/cms/testbeam/tkmtcc/P5_data/tracker_reprocessing/pass3_with_alignment/[run number]_reco_full.root
Massimiliano's simulated samples (with MTCC setup) are in lxmcsf1:/data/chiorbo/rootuple/CMSSW_1_1_1_SimHits/.
High-level (re-)reconstruction (from hits onward): reprocess_from_simhits.sh.

My configuration file: AnalyzeTracks_MTCC.cfg
It needs the following cfi's:
mtcc_data_pass3.cfi (data @ 3.8 T)
mtcc_simu.cfi (simulation @ 4.0 T)

Analysis root macro: plotmtcc.C

Intercalibration extracted from MTCC data: InterCalibration.h, InterCalibration.cc
It uses the peak position as extracted from APV pairs with at least 60 hits; otherwise, it uses the tickmark method.

• cos(angle):
  • data/MC comparison (Geant 4.8.1)
  • as above, for 0<eta<0.5
• dE/dx per hit:
  • data/MC comparison (Geant 4.8.1) (c.c.=7%)
    • same as above, for ADC counts * cos(angle)
  • data/MC comparison (Geant 4.7.1) (c.c.=7%) --- MC IN DECO MODE!!!
    • data/MC comparison (Geant 4.7.1) (peak mode)
    • as above, for ADC counts * cos(angle)
  • comparison in MC (Geant 4.7.1) between c.c.=6% and c.c.=12%
    No effect.
  • comparison in MC (g4.8.1) between reconstructions with different B fields
  • comparison in MC between g4.7.1 and g4.8.1 --- DECO MODE
    • as above, in peak mode
  • comparison in MC (Geant 4.8.1) between peak and deco
  • comparison in data between uncorrected and corrected
    The shift between TIB and TOB is solved by construction.
    χ²: uncorrected 0.04(TIB)0.08(TOB), tickmark 0.03(TIB)0.08(TOB), particle 0.03(TIB)0.04(TOB)
    TOB-TIB peak shift: uncorrected +0.116 MeV/cm, tickmark +0.055 MeV/cm, particle -0.071 MeV/cm
• number of hits per layer: B=0 T, B=3.8 T
• dE/dx per track (estimator: harmonic mean of (dE/dx)^2):
• B = 0.0 T: scatter plot vs P, linear plot, log plot with gaussian fit, profile histogram vs P
  Main quality criteria:
  cosmic chi2: 24
  cosmic sigma: 20.8 %
• B = 3.8 T: scatter plot vs P, linear plot, log plot with gaussian fit, profile histogram vs P
  Main quality criteria:
  cosmic chi2: 53
  cosmic sigma: 16.1 %

Cosmics at TIF:

I'm using the standard program, with AnalyzeTracks_TIF.cfg. I redo the track-finding, instead of calling TrackRefitter, because it orders the hits by distance from (0,0,0), which is not appropriate for cosmics.
TIB+TOB runs: tif.cfi
plottif.C
Comparison TIF/MTCC

Study of multi-track events at TIF:

• Talk by David Stuart
• Patrizia's plots
• 18/6/2007, CMSTK INFN meeting, Multitrack events [cosmic showers] at TIF (T.Dorigo)

Attenuation of the signal as a function of position and β (thanks to A. Rizzi):

niceplots.C (for plots of signal attenuation as a function of position and β)

• Pulse shape (and how it is implemented in CMSSW: look at SiLinearChargeDivider::DeconvolutionShape)
• Tracker layout, with distances in cm
• Attenuation as function of (r,β)
• Attenuation as function of (x,y) for β=0.40
  M=1 GeV: beta = 0.4 means P=0.436436
  M=100 GeV: beta = 0.4 means P=43.6436
• Attenuation as function of (x,y) for β=0.45
  M=1 GeV: beta = 0.45 means P=0.503903
  M=100 GeV: beta = 0.45 means P=50.3903
• Attenuation as function of (x,y) for β=0.50
  M=1 GeV: beta = 0.5 means P=0.57735
  M=100 GeV: beta = 0.5 means P=57.735
• Attenuation as function of (x,y) for β=0.55
  M=1 GeV: beta = 0.55 means P=0.658553
  M=100 GeV: beta = 0.55 means P=65.8553
• Attenuation as function of (x,y) for β=0.60
  M=1 GeV: beta = 0.6 means P=0.75
  M=100 GeV: beta = 0.6 means P=75
• Attenuation as function of (x,y) for β=0.65
  M=1 GeV: beta = 0.65 means P=0.855337
  M=100 GeV: beta = 0.65 means P=85.5337
• Attenuation as function of (x,y) for β=0.70
  M=1 GeV: beta = 0.7 means P=0.980196
  M=100 GeV: beta = 0.7 means P=98.0196
• Attenuation as function of (x,y) for β=0.75
  M=1 GeV: beta = 0.75 means P=1.13389
  M=100 GeV: beta = 0.75 means P=113.389
• Attenuation as function of (x,y) for β=0.80
  M=1 GeV: beta = 0.8 means P=1.33333
  M=100 GeV: beta = 0.8 means P=133.333

Applications of the method:

bands.C

Minimum Bias:

With bilateral probability (minbias): like_pi, like_k, like_proton, LR proton/pion, LR kaon/pion, LR proton/kaon

MC truth:

Jets:

Heavy Stable Charged Particles:

More infos on the wiki of the SUSY working group on HSCPs

Calibration with V0's:
Principle:
Identify K_S⁰ and Λ.
68% of K_S⁰ decays are into π⁺π^-, 64% of Λ decays are into pπ^-. Thus, the first selection gives a very pure sample of π's, useful for absolute dE/dx calibration, while the second gives a π and a proton, useful to test the separation between the two bands.
Identification of &Lambda's can be done exploiting the V-A structure of the decay: as explained here, the proton predominantly is found going off in the direction of the spin of the Lambda.
Note on V0's in CMS by Ferenc
Podolanski-Armenteros variables are described at page 8. They are qT and α, where qT is p1*p2*sen θ divided by the mother's p, and α is (p1²-p2²)/p².
K_S⁰ is symmetric in α and with sizeable qT, photons are symmetric with negligible qT, Λ and anti-Λ are asymmetric and with intermediate qT.
(In order to have Λ's on the right and anti-Λ's on the left, choose particle 1 to be the positive track and particle 2 to be the negative one.)

K_S⁰:

Λ:

MC truth:

Systematics:

The average gain increase with temperature is -0.0064 ^oC^-1. I assume a 3 degrees uncertainty, corresponding to 2% uncertainty on gain.
The timing uncertainty is 1 ns. Each 1 ns delay correspond to a decrease of 4% on signal.

Capacitive coupling and cluster thresholds:

Particle gun, 0.9<pt<1.1 GeV

• Eta = 0:
• C.c.=6%
• C.c.=12%
• Eta = 1:
• C.c.=6%
• C.c.=12%

Publications:

• CMS NOTE-2008/005

Talks:

• 11/3/2008, Joint Tracker - Physics Workshop
• 4/9/2007, Tracker DPG
• 19/6/2007, tracking meeting, Producer for dE/dx
• 14/6/2007, Strip Tracker simulation tuning meeting: Producer for dE/dx (A.Rizzi), Signal properties at TIF (M. De Mattia), dE/dx studies at MTCC and TIF (P.Azzi)
• 11/4/2007, "lunch seminar" in Louvain-la-neuve, dE/dx in the search for exotic charged stable particles
• 29/3/2007, Strip Tracker simulation tuning meeting, data/MC comparison in MTCC
• 2/2/2007, Strip Tracker simulation tuning meeting, data/MC comparison in MTCC
• 20/10/2006, internal seminar in Louvain-la-neuve
• 28/9/2006, energy flow meeting
• 29/8/2006, b-tau meeting (minutes)

References:

• Coordination page for Paper on Detector Performance at TIF
• A method to improve tracking and particle identification in TPCs and silicon detectors.
  H. Bichsel (Washington U., Seattle) . 2006. 44pp.
  Published in Nucl.Instrum.Meth.A562:154-197,2006.
• The tracking, calorimeter and muon detectors of the H1 experiment at HERA
• Presentation on low-Pt tracking and dE/dx by the Budapest group (see "Charged Particle Tracking")
• Notes in preparation by the Budapest group
• CMS IN 2001/052
• CMS NOTE-2006/145 - Predicting the Gain Spread of the CMS Tracker Analog Readout Optical Links
• Andrea R.'s talk on R-hadron searches (with TOF and dE/dx): [1] [2]
• Analysis note on R-hadrons
• CMS note on misalignment
• Brian's talks on V0 reconstruction in CMSSW
• Deconvolution of the inter-strip couplings: /afs/cern.ch/user/t/tomalini/tracker/ORCA_8_13_1/src/TrackerReco/HitResolution/test/max/capcorrection.ps
• See Sec.3.1 of this note for the energy deposit simulation: /afs/cern.ch/user/t/tomalini/text/resolution/res.pdf
• Stopping-Power and Range Tables
• Brian Drell's research log on V0s

CMSSW main page
CMSSW code browser - RecTrack.h
CMSSW doxygen
CMSSW tutorial
LXR search
CMSSW @ LLN
DeDxTools in ORCA
MTCC track reconstruction and filter
Table of MTCC tracker runs
Energy Flow tutorial (slides)
QCD cross sections in pthat bins
How to run on TIF data
Tracker publications