L1Calo performance with electrons and photons in 2009 data

Update on Diffractive Dijets
Hardeep Bansil
University of Birmingham
Diffraction Analysis Meeting
17/06/2013
Contents
Analysis
D3PD trigger variables
ξ resolutions
Efficiencies
Unfolding
Gap asymmetry
Next steps
2
Diffractive dijets
•
Single diffractive events (pppX)
•
•
Search for hard diffraction with a hard scale set by 2 jets
•
•
Rapidity gap from colourless exchange with vacuum quantum numbers
“pomeron”
Described by diffractive PDFs + pQCD cross-sections
Previous measurements of hard diffractive processes at HERA
and Tevatron
•
•
At Tevatron, ratio of yields of single diffractive to inclusive dijets ≈ 1%
Likely to be smaller than this at LHC
•
Measure the ratio of the single diffractive to inclusive dijet
events
•
Gap Survival Probability – the chance of the gap between the
intact proton and diffractive system being lost due to scattering
•
•
•
Tevatron have Gap Survival Probability of 0.1 relative to H1 predictions
Khoze, Martin and Ryskin predict LHC to have GSP of ~ 0.03
Rescatter with p?
Understand the structure of the diffractive exchange by
comparison with predictions from electron-proton data and be
able to get a measure of FDjj
ξ
3
Analysis
• Using data10 L1Calo stream period A and B, MinBias stream
period B (GRL applied)
• Herwig++ (SD, ND) and Pythia 8 (SD, ND, DD) samples
•
•
•
Separate samples – one unfiltered in gap size, the other filtered to get larger gap
sizes
No rapidity gap destruction built in
Look at DD samples to put limit on SD + DD (MY < 7 GeV)
• Event selection: GRL, PV0 with ntracks>4, No PU vtx (having
ntracks>1)
• Anti-kt jets with R=0.6 or R=0.4:
•
•
Require >= 2 jets in event passing medium jet cleaning cuts
ET Jet1,2 |η| < 4.4, ET Jet1 > 30 GeV, ET Jet2 > 20 GeV
• Ask for a forward gap: |ηstart| = 4.9, ΔηF ≥ 3.0
•
Biggest region in η from edge of detector absent of particles
•
Defined wrt tracks and noise suppressed EM clusters with pT>200 MeV
•
Stable truth particles with pT>200 MeV
• Trigger: J5 in central region / MBTS 1 in forward region
4
D3PD Trigger Variables
• L1 CTP keeps track of whether trigger item passed at three different levels:
•
•
•
Trigger before prescale (TBP)
Trigger after prescale (TAP)
Trigger after veto (TAV) – accounting for simple/complex deadtime
• Previously using the D3PD variables to access results for L1_J5 and
L1_MBTS_1 but these correspond to the TBP result – switch to
TriggerDecisionTool for all analyses
• Little effect overall, still
reaches gap sizes up to 7
in data (see control csx plot)
• Majority of events triggered
by L1_J5
5
Reconstructing ξ
Proper way to calculate ξ can be done in SD MC by looking at proton / pomeron
Can calculate MX from invariant mass then convert to ξ (= MX2 /s)
MX
Largest rapidity gap between truth particles
MY
Truth level: all final state particles excluding intact proton from diffractive exchange
(if there is one in event)
Reconstructed level: all caloClusters
~
Calculate ξ using E±pz method using particles in η region [-4.9, 4.9] in order
to have consistent definition of observable between reconstruction/hadron level


 (E
C
i
 p zi )
s


 (E
C
i
 p zi )
s
(C=1 for truth, determine for data)
Now base choice on position of where forward gap starts using gap algorithm
Gap start at -4.9 uses ξ- and gap start at +4.9 uses ξ+
6
ξ Resolution v ξ
Pythia 8 SD & DD – Plot ξ resolution (fractional) against reconstructed ξ± to
observe potential ξ dependency on resolution after gap and jet cuts are applied
Pythia 8 SD Gap Weighted sample
Truth
larger
Truth
larger
Recon
larger
Recon
larger
Pythia 8 DD Gap Weighted sample
Both axes are on log scale
In SD, get a tail at recon ξ± around log10(ξ±) = -2 from “gap mismatch” events
No strong dependency as a function of recon ξ± but clear shift to larger truth
values
Looking to fit across the range -3.5 to -1.5 to determine variation, also try linear
7
ξ Resolution v ξ
Pythia 8 SD & DD – Plot ξ resolution (fractional) against reconstructed ξ± to
observe potential ξ dependency on resolution after gap and jet cuts are applied
Pythia 8 SD Gap Weighted sample
Truth
larger
Truth
larger
Recon
larger
Recon
larger
Pythia 8 DD Gap Weighted sample
Both axes are on log scale
No strong dependency as a function of recon ξ± but clear shift to larger truth
values
Small difference between these plots and previous slide show that after cuts,
very little difference between truth ξ± and actual ξ values
Looking to fit across the range -3.5 to -1.5 to determine variation, also try linear
8
Resolutions


 (E
C
 (E
C
i
 p zi )
s
i
 p zi )

Gap Weighted samples – after gap, jet cuts
s
Now able to reconsider the correction to reconstruction level necessary to match back onto
truth level, accounting for particles not seen in calorimeter
Using linear ξ values for resolutions, fit with truncated gaussian

If comparing ξ± resolution, both Pythia 8 SD & DD have similar μ = -0.46, σ = 0.11
Would imply correction factor C = 1.85 ± X
(cf. CMS factor = 1.45 ± 0.04 with PF pt > 0.2 GeV central, E > 4 GeV forward)
Pythia 8 SD Resolution ξ E+Pz
method after Cuts
Pythia 8 DD v Resolution ξ E+Pz
method after Cuts
9
Resolutions


 (E
C
 (E
C
i
 p zi )
s
i
 p zi )

Gap Weighted samples – after gap, jet cuts
s
Now able to reconsider the correction to reconstruction level necessary to match back onto
truth level, accounting for particles not seen in calorimeter
Using linear ξ values for resolutions, fit with truncated gaussian

If comparing ξ± resolution, both Pythia 8 SD & DD have similar μ = -0.46, σ = 0.11
Would imply correction factor C = 1.85 ± X (cf. CMS factor = 1.45 ± 0.04)
Will try applying correction factor to recon level in data & MC to test
Pythia 8 SD Resolution ξ E+Pz
method after Cuts
Pythia 8 DD v Resolution ξ E+Pz
method after Cuts
10
Efficiencies
Moving method of calculation to be more like SM2010 analysis in that efficiencies are
calculated for J5 in EM crack region (1.3<|n|<1.6) separately to those in remaining central
area (|n|<2.9) due to detector geometry + L1Calo configuration at time
Smaller statistics now that triggering has been fixed
Use a fit of form a1*Erf(a2(x-a3)) where a1,2,3 are fit parameters above 20 GeV
EM Crack only
anti-kT R=0.4
anti-kT R=0.6
Fit slightly overestimates rise to plateau for Ak4 jets, worse for Ak6 jets – use for now,
improve fit at later date
Will also look at efficiency v gap size  gap size dependency for J5 could be potential
systematic
11
Efficiencies
Moving method of calculation to be more like SM2010 analysis in that efficiencies are
calculated for J5 in EM crack region (1.3<|η|<1.6) separately to those in remaining central
area (|η|<2.9) due to detector geometry + L1Calo configuration at time
Also look at efficiency v η and gap size above 30 GeV  gap size dependency for J5 could
be potential systematic
MinBias data (All Jets)
Excluding EM Transition
Region
anti-kT R=0.4 Jets
anti-kT R=0.6 Jets
MinBias data (All Jets)
Excluding EM Transition
Region
anti-kT R=0.4 Jets
anti-kT R=0.6 Jets
Not on plateau at 30 GeV so would not expect 100% efficiency, can see difference in η
clearly (L1Calo jet triggers not optimised yet so acceptable)
Only get up to gap size of 5 as determining efficiencies with MinBias stream and MBTS_1
so requirement of jets and MBTS prescales do not give enough events
Hard to determine full dependency at the moment unless try tag and probe with L1Calo
12
Efficiencies
Moving method of calculation to be more like SM2010 analysis in that efficiencies are
calculated for J5 in EM crack region (1.3<|η|<1.6) separately to those in remaining central
area (|η|<2.9) due to detector geometry + L1Calo configuration at time
Also look at efficiency v η and gap size above 30 GeV  gap size dependency for J5 could
be potential systematic
Pythia 8 ND (All Jets)
Excluding EM Transition
Region
anti-kT R=0.4 Jets
anti-kT R=0.6 Jets
Pythia 8 ND (All Jets)
Excluding EM Transition
Region
anti-kT R=0.4 Jets
anti-kT R=0.6 Jets
Not on plateau at 30 GeV so would not expect 100% efficiency, can see difference in η
clearly (L1Calo jet triggers calibrated in MC)
No particular dependency in MC for gap size where there are significant statistics
13
Unfolding
• First attempt at unfolding data with MC for gap size distributions
• 1D - adding in events that passed cuts at either truth & reconstruction level
• Includes 'missed' dijets (truth dijet was not reconstructed)
• Includes 'fake‘ dijets (dijet was reconstructed which did not pass truth
requirements)
• 2D – adding only events where both truth & reconstruction requirements are met
• Scaling ND
• No mention of scaling ND by
1.0/1.5 in SM analysis (from looking
through supporting note)
•Need to understand why the difference
between recon & truth gap sizes is so large
•Especially when distributions like dijet
eta and pt have very small differences
•Difference comes particularly large after
gaps of 2, where recon gap jumps from
FCAL to HEC
MC Hadron Level
MC Reconstructed
Data to Unfold
Data Unfolded
•Need to check how many events pass
recon and truth jet cuts
Pythia 8 SD+DD+ND after Cuts
14
Unfolding (Closure Tests)
Closure tests for Herwig++ combined samples and Pythia 8 combined samples, just to
check that things make sense (unfold HW with HW, P8 with P8)
MC reconstructed sits on MC to Unfold
MC Unfolded should give same results as MC hadron level
Some errors are slightly different but tests work for both sample types
Code for unfolding works as expected
MC Hadron Level
MC Reconstructed
MC to Unfold
MC Unfolded
Herwig++ SD+ND after Cuts
MC Hadron Level
MC Reconstructed
MC to Unfold
MC Unfolded
Pythia 8 SD+DD+ND after Cuts
15
Unfolding
• Need to understand why the difference between recon & truth gap sizes is so large
• May be an issue with differences in the first bin between recon & truth gap size that
enhance the effect of smearing the gap size distribution
•
•
FCAL readout bins are 0.2 x 0.2 in eta, phi – cluster needs significant deposit so hard for cluster
barycentre to be so close to the edge of the detector  leads to more events with slightly larger
gaps.
With larger rebinning, get first bin to agree better in both MC and data (better acceptance factor)
MC Hadron Level
MC Reconstructed
Data to Unfold
Data Unfolded
Pythia 8 SD+DD+ND after Cuts
MC Hadron Level
MC Reconstructed
Data to Unfold
Data Unfolded
Pythia 8 SD+DD+ND after Cuts
16
Unfolding
• Need to understand why the difference between recon & truth gap sizes is so large
• Take MC samples scaled by Ldata/LMC sample and look at ratio of Recon/Truth (bin by bin
acceptance corrections)
• Looks like DD contribution has significant difference between recon & truth although all
samples have big factor between 2-6 in gap size
• Only a few bins where recon/truth < 1 so must be more recon jets passing cuts than truth
jets
Combined HW, P8 Gap Filtered after jet Cuts
Separate HW, P8 Gap Filtered after jet Cuts
17
Unfolding (Scaling)
Looking at different scales for the different components of ND:SD:DD in Pythia 8
Default is on left - 1.00:1.00:1.00
Pythia 8 ND scaled down by 1.5 on right - 0.67:1.00:1.00
Scaling down by 1.5 makes ratio to unfolded data smaller
Assuming can normalise ND as mainly from first bin, determine ideal factor
MC Hadron Level
MC Reconstructed
Data to Unfold
Data Unfolded
Pythia 8 SD+ND+DD no additional
scaling
MC Hadron Level
MC Reconstructed
Data to Unfold
Data Unfolded
Pythia 8 SD+ND+DD with ND scaled
down by 1.5
18
Unfolding (Scaling)
Looking at different scales for the different components of ND:SD:DD in Pythia 8
Default is on left - 1.00:1.00:1.00
Middle has SD up by 15%, DD down by 15% - 1.00:1.15:0.85
Right has SD up by 15%, DD down by 15% - 1.00:1.15:0.85
Scaling by SD and DD have effect of making difference between truth and recon mc larger
so unfolding correction larger. May need to use more sensible fractions to make differences
more noticeable
Pythia 8 SD+ND+DD default
Pythia 8 SD+ND+DD SD up 15%, DD
down 15%
Pythia 8 SD+ND+DD SD down 15%, DD
up 15%
19
Gap Start Asymmetry
Look at the distributions of gap size and other variables as a function of the side that the
gap is on and look for asymmetry
Plots here have dijet cuts applied
Unlike Vlasta, do not see any bins in ratio plot where value is ratio is significantly away
from 1 (apart from where we have low statistics), even after finer binning
20
Gap Start Asymmetry
Look at the distributions of gap size and other variables as a function of the side that the
gap is on and look for asymmetry
Dijet and also gap cuts applied for pT and η plots
Data
Pythia 8 SD gap filtered
Pythia 8 ND gap filtered
With gap cuts applied, can now start seeing bins with limited statistics where ratio out by a
couple of standard deviations from 1
21
Gap Start Asymmetry
Look at the distributions of gap size and other variables as a function of the side that the
gap is on and look for asymmetry
Dijet and also gap cuts applied for pT and η plots
Data
Pythia 8 SD gap filtered
Pythia 8 ND gap filtered
With gap cuts applied, can now start seeing bins with limited statistics where ratio out by a
couple of standard deviations from 1
22
Next steps
Understand differences between recon v truth in gap size distributions in MC
and effect it has on unfolding
Investigate statistics in DD events
Complete ξ calculation by applying C factor to data
Quantify the amount of migration over the pt jet threshold on gap spectrum
using matched jets
Test scaling of contributions of ND:SD:DD
Look at other corrections & other systematics
Scaling background events that have passed cuts
23
Backup Slides
24
Unfolding
• Need to understand why the difference between recon & truth gap sizes is so large
Possibility:
•At MC, no trigger corrections applied and both recon and truth are filled with same weight
Smearing matrix shows truth events get pushed to larger gap sizes meaning that larger
weights being applied at larger recon values, creating big difference
•Should see same result with gap filtered and non-filtered samples (slightly different gap
spectra due to weighting but
MC Hadron Level
MC Reconstructed
Data to Unfold
Data Unfolded
Pythia 8 SD+DD+ND Gap Filtered after Cuts
MC Hadron Level
MC Reconstructed
Data to Unfold
Data Unfolded
Pythia 8 SD+DD+ND Not Filtered after Cuts
25
Unfolding
• Need to understand why the difference between recon & truth gap sizes is so large
• Take MC samples scaled by Ldata/LMC sample and look at ratio of Recon/Truth (bin by bin
acceptance corrections)
• Looks like DD contribution has significant difference between recon & truth although all
samples have big factor between 2-6 in gap size
• Only a few bins where recon/truth < 1 so must be more recon jets passing cuts than truth
jets
•Should not expect to see a bin correction of 800 in DD – not enough statistics at large gap
sizes here
Combined HW, P8 Not Filtered after jet Cuts
Separate HW, P8 Not Filtered after jet Cuts
26