-- ChengguangZhu - 25 Nov 2007

Useful links : AtlfastProcedure StoregateKeysForAODRelease12

Introductions

To tune MC with the first data of ATLAS, atlfast is a better selection, which needs short time and less resources for a MC loop. So the performance of atlfast data and full simulation data are compared here.

A comparison with only events passed selection and quality cut, with AcerMC data, can be found at ZhuSelEvnetsCom

Used data ZhuGeneratorWithAcerMC

MCatNLO Data sample: trig1_misal1_mc12.005200.T1_McAtNlo_Jimmy.recon.AOD.v12000601_tid005997: totally 344k weighted events, 182.8k semileptonic e/mu events events are used.

Full simulation with 1mm bug, fixed at AOD level

Atlfast data: The bjet tag in atlfast emulates (code from J.B. Devivie) the b tag of full simulation. In the emulation, weight are set to jets in a specific (pt, eta) window randomly according to weight distribution of full-sim jets in such a window. %h% important: weight is caculated according to jet's specific characteristics, tagged jets should have same characteristics of course, meaning that the randomly set weight may bring bias in measurements.

Both data read out from the same data sample, the jet energy is calibrated to quarks.

Selection efficiencies

Qualities

• Electrons: in full sim, electron->parameter(ElectronParameters::etcone20) same as author()= 1 or 3.
  

Qalities	atlfast	full sim (e/mu ID)	Comments
b tag efficiency for truth b jets (42% truth b jets)	62.7%	60.8%	emulation weight is slightly higher
purity for b tagged jets	93.1%	93.4%	atlfast purity even lower, btag increase, c/l jet rejection worse
c jet rejection (10% truth c jets )	6.6	6.9
tau jet rejection (truth tau jets,not in signal )	16.9	20.5
l jet rejection( 48% truth l jets)	118	131
electron reconstruction efficiency true pt>20GeV abs(eta)<2.5	89.9%	69.6%	artificially set reco effi to 89.5%?
purity of selected electrons	95.4%	98.3%	aflfast electron is not isolated, I should use isolation with comparing to jets, jets e/p matched to a ele should be regarded as electron? 68.2%/70.6%=96.6%(purity)
muon reconstruction efficiency	84.4%	91.7%	artificially set reco effi to 84.4%?
purity of selected muons	99.4%	98.7%	atlfast muon is isolated in AtlfastMuonCollection, unisolated muon is stored in AtlfastNonIsoMuonCollection

Plots (red: atlfast, blue: full simulation)

* comp_fast_full.cxx: code for producing the plots

program flowing

for AcernMC red for atlfast and blue for full sim.: 0 -->1 lepton-->1--> >=4jets --> 2--> >=2bjets and >=2ljets --> 3--> etmiss>20G --> 4-->wln-->5-->wjj and jjb-->7

electron

• pt, eta and phi distribution of selected elections and resolutions:
  

electron reco efficiency vs eta phi pt

• e reconstruction efficiency along eta, pt and phi for truth electron with pt>20GeV |eta|<2.5:
  

electron reco efficiency vs energy

plot 1: reconstructed coslep with e_pt_cut>25GeV (red) and >60GeV (black)	coslep<0 is affected(suppressed) by cut on low pt electron
plot 2: reconstructed electron energy (p) distribution, for atlfast (red) and full sim (blue)	full sim reconstructed less high energy electron
plot 3: electron reconstruction efficiency v.s. energy(p)	as plot 2

Conclusion: full sim reconstructed less high energy electrons ---> suppress the reconstruction of coslep (>0) as see (blue in plot 1 and 2 at below)

• 

electron selection

• p1:Before selection: aflfast: 17% with no ele at all (88.5%*89.5%+noisolated electron); p2: after selection; p3; after selection, full sim with isEM ID, matched distribution.
  

muon

• pt, eta and phi distribution of selected muons and resolutions:
  

muon reconstruction efficiency

• muon reconstruction efficiency along eta, pt and phi for truth muonon with pt>20GeV |eta|<2.5:
  

jets

• pt, eta and phi distribution of selected jets and multiplicity of l jets, bjets :
  

• AcerMC atlfast jet resolution.
  • There was a bug in reading out truth information previously--->the truth b quark readout is not directly from top, being a b after radiating, and the calibration with full sim parameter looks making jet reso shift rightwards. Look at the 7th sub-plot in the plot above, the red one for atlfast is in fact before calibration.
  • but it doesn't affect analysis:
  • Now it's fixed. The pre-calibration with full sim parameter looks working well--->shift being corrected and the shape of the distribution is more reasonable (left tail):
    

jet weight

• full sim weight and atlfast emulation weight:
  
• do jet pre-calibration should re-valuate etmiss in the same time? No, precalibration and the etmiss artificially shift correction is in the same level, which is called calibration and correlate all objects each other. but in study of the uncertainty of the jet miscalibration, we think about the errors in jet calibration, then other correlated objects, e.g. etmiss should be recalibrated again according to their relation.

btag

• c/l jet rejection v.s. btag effi:
  

etmiss

• etmiss 1. reco pt 2. reco phi 3. pt resolution 4. phi resolution:
  
• artificial etmiss shift 1.042 in the 12.0.6 should be removed? No in CSC 1206 data, the etmiss resolution is -5.5% shifted, in fact need to multiply a larger value for calibrate it.

reconstructed mass of w and t

• invariant mass

• 	Whad atlfast	Whad fullsim	thad atlfast	thad fullsim	tlep atlfast	tlep fullsim
  Mass	80.7±0.12	83.6±0.18	174.7±0.21	178.6±0.26	173.1±0.18	173.5±0.19
  Sigma	7.6±0.16	9.8±0.31	12.6±0.34	16.5±0.41	11.2±0.28	10.9±0.31

  :
  

cospsi

• resolution of cospsi, psi is the angle between the direction of lepton in w rest frame and w direction in t rest frame :
  
  • use only pure signal (without even cmbk) for full sim:
    

All angles and angle resolution:

• first raw, resoluation of reconstructed , , coslep, coshad_lej, coslep*coshad_lej:
  
• second raw: distribution of them.

Angle distributions with quality cut

• Distributions are reconstructed angles with cuts: qualitycut*mttbar*ept25. Left-up corner: entries of fast data, right-up entries of full simulaiton
• for coslep>0, full simulation has lower reconstruction efficiency, see reason
• for coshad_lej, the reconstruction efficiency for full and fast simulation match better. Because accordingly coshad_lej<0 is slightly suppressed due to correlation with coslep, see left tail of plot 3

coslep (coshad_lej>0) 60-75% more events than right one for fast

and

coslep (coshad_lej<0)

| coshad_lej (coslep>0) similar events with the right one | and | coshad_lej (coslep<0) |

ttbar mass and resolution:

t and ttbar pt resolution

• ttbar pt is closer to had top in full than in fast (Used semilep events and quality cut):
  

semiemu and semitau events

• Because 3 neutrinos in the semitau events fly in different direction, the measured pt_miss is the sum pt of the 3 neutrinos, but take the pt_miss as their sum energy, it will be low estimated. This affect the reconstruction of w and further t
• The following are plots of lep_w pt resolution, lep_w eta resolution, lep_w phi resolution and lep_t eta resolution, lep_t mass resolution and lep_t mass distribution. blue for semiemu events and red for the semitau events
• It means that although the tau decayed leptons fly almost in the direction of tau, but when we want its direction in the w rest frame or t rest frame, the reconstructed one is much worse. :
  
• resolution of coslep, cospsi, cosphilep_lej and coslep*coshad_lej, blue for semiemu events and red for semitau events.:
  

Assigned to	Due date	Description	State	Closed	Notify
ChengguangZhu	2007-12-20	plots of jet reconstruction efficiency and purity and fake jets	openclosed			edit
ChengguangZhu	2007-12-09	atlfast electron muon reconstruction and effieicnys	openclosed			edit