Diffractive dijets update - University of Birmingham

Update on Diffractive Dijets
Hardeep Bansil
University of Birmingham
22/07/2013
Contents
Calculating ξ with Transerve Energy Flow method (and other cuts)
Unfolding “closure tests”
Miscellaneous points
2
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)
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 fiducial η region [-4.9, 4.9]
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+
What effect does changing from soft diffractive to Et flow method have?
What effect does going to low pT dijets have?
Can we use the Et flow p cuts at truth level (as we would not normally for xi)
Should we stick to [-4.8, 4.8] or do what CMS do [-4.8, ∞], [-∞, 4.8]
3
MC Scaling factors in Et flow paper
Et flow paper uses scaling factors that are applied to MC EM calibrated clusters to
match them to truth
Based on “Validation of the calorimeter energy response with π0→γγ candidates”
(/ATL-COM-CAL-2011-003)
Values α are applied to cluster values
in different η bins – scaling factors
predominantly negative other than in
forward regions
Determined using 3 different Pythia 6/8
ND+SD+DD tunes
and
1 Herwig++ ND tune
4
ξ from proton v truth ξ±
Pythia 8 SD Gap Weighted – E+pz before and after gap, dijet cuts, Et flow, no p cuts at
truth
No additional
cuts
After Gap &
Dijet cuts
After cuts, still have to deal with very small subset of events where actual truth ξ quite
large but E±pz sees something smaller (gap mismatch events c.f. Herwig++)
5
E±pz Truth ξ v E±pz Recon ξ
Pythia 8 SD Gap Weighted – Before & after gap, jet cuts
In general correlation in log10 between Truth ξ and Recon ξ± good
but affected partly by reconstructed gap not being on correct side
and slightly by truth gap not being on correct side
Truth ξ v Reconstructed ξ using E+Pz
method Before Cuts




 (E
C
 (E
C
 p zi )
i
s
 p zi )
i
s
Truth ξ v Reconstructed ξ using E+Pz
method After Cuts
Recon values are smaller in general as you cannot see all of the truth particles
in the calorimeter – gap cuts remove many large ξ values, jet cuts remove the
very small ξ values but still get events at recon level going to very low ξ
6
E±pz Truth ξ v E±pz Recon ξ
Pythia 8 SD Gap Weighted – After gap, Ak6 jet cuts




 (E
C
 (E
C
 p zi )
i
s
 p zi )
i
s
Will have to apply correction factor at recon level to shift distribution back to
truth as differences currently too extreme
And potentially cut off after certain ξ range like CMS (0.0003 – 0.002, 0.002 - 0
0.0045, 0.0045 – 0.01)
0.0003 ≈ -3.5 on log scale, 0.002 = -2.70, 0.0045 = -2.35
7
E±pz Truth ξ v E±pz Recon ξ
Pythia 8 SD Gap Weighted – Before & after gap, jet cuts
In general correlation in log10 between Truth ξ and Recon ξ± good
but affected partly by reconstructed gap not being on correct side
and slightly by truth gap not being on correct side
Truth ξ v Reconstructed ξ using E+Pz
method Before Cuts




 (E
C
 (E
C
 p zi )
i
s
 p zi )
i
s
Truth ξ v Reconstructed ξ using E+Pz
method After Cuts
Recon values are smaller in general as you cannot see all of the truth particles
in the calorimeter – gap cuts remove many large ξ values, jet cuts remove the
very small ξ values but still get events at recon level going to very low ξ
8
E±pz Truth ξ v E±pz Recon ξ


 (E
C
i
 p zi )
Pythia 8 SD Gap Weighted – After gap, jet cuts
s
( E i  p zi )
Et flow method shifts events to lower ξ values, both at recon


 C
and truth levels – attributed to scaling factors that
s
are applied to cluster kinematics and slightly tighter fiducial requirement
Rapidity gaps
Et Flow
More events at recon level than truth level due to more dijets passing jet level
cuts
9
Fractional shifts
Pythia 8 SD (Gap Weighted) : (Recon - Truth)/Truth
Ak6 20, 20 cuts and gap requirement met
Rapidity gap method (fiducial, no pt cuts applied
at truth or recon level)
Mean value = -0.5049
C = 2.02
Rapidity gap method (not fiducial, no pt cuts applied
at truth or recon level)
Mean value = -0.4838
C = 1.94
E+-pz should not make that much difference once
very forward, but does shift mean closer to 0
10
Fractional shifts
Pythia 8 SD+DD+ND (Gap Weighted) : (Recon - Truth)/Truth
Ak6 20, 20 cuts and gap requirement met
Rapidity gap method (fiducial, no pt cuts applied
at truth or recon level)
Mean value = -0.4686
C = 1.88
The shift for SD+DD+ND is smaller than for SD only
as ND has typically smaller shifts
ND, however, more likely to get gap / xi matching
incorrectly compared to SD or DD
Rapidity gap method (not fiducial, no pt cuts applied
at truth or recon level)
Mean value = -0.4516
C = 1.82
11
Fractional shifts
Pythia 8 SD (Gap Weighted) : (Recon - Truth)/Truth
Ak6 20, 20 cuts and gap requirement met
Et flow method (fiducial, no p cuts applied at truth
or recon level)
Mean value = -0.5263
C = 2.11
More events with where truth much greater than
recon xi
Et flow method (not fiducial, no p cuts applied at
truth or recon level)
Mean value = -0.5271
C = 2.11
Fiducial cuts have opposite effect here to rapidity
gaps method bit only a small difference
12
Fractional shifts
Pythia 8 SD (Gap Weighted) : (Recon - Truth)/Truth
Ak6 20, 20 cuts and gap requirement met
Et flow method (fiducial, with p cuts applied)
Mean value = -0.4859
C = 1.95
Et flow method (not fiducial, with p cuts applied )
Mean value = -0.4883
C = 1.95
Correction factor with p cuts becomes smaller
because truth values become smaller
13
Unfolding Gap Size
“Closure tests” using different models – Previously saw that Pythia 8 unfolding Herwig++
and vice versa did not work that well
Other tests, try unfolding same model with different weighting
Very good until statistics get smaller at lower gap sizes
Pythia 8 (Dijet Filt.) Hadron Level
Pythia 8 (Dijet Filt.) Reconstructed
Pythia 8 (Gap Filt.) Reconstructed
Pythia 8 (Gap Filt.) Unfolded
Pythia 8 (Gap Filt.) Actual Hadron Level
Pythia 8 (Dijet Filt.) Recon / Hadron Level
Pythia 8 (Gap Filt.) Recon / Unfolded
Pythia 8 (Gap Filt.) Hadron Level / Unfolded
Pythia 8 Gap Filtered unfolded with
Dijet Filtered
Pythia 8 (Gap Filt.) Hadron Level
Pythia 8 (Gap Filt.) Reconstructed
Pythia 8 (Dijet Filt.) Reconstructed
Pythia 8 (Dijet Filt.) Unfolded
Pythia 8 (Dijet Filt.) Actual Hadron Level
Pythia 8 (Gap Filt.) Recon / Hadron Level
Pythia 8 (Dijet Filt.) Recon / Unfolded
Pythia 8 (Dijet Filt.) Hadron Level / Unfolded
Pythia 8 Dijet Filtered unfolded with
Gap Filtered
14
Unfolding Gap Size
Same as before but increase to 3 iterations for unfolding
Where statistics are abundant, still works well
Pythia 8 (Dijet Filt.) Hadron Level
Pythia 8 (Dijet Filt.) Reconstructed
Pythia 8 (Gap Filt.) Reconstructed
Pythia 8 (Gap Filt.) Unfolded
Pythia 8 (Gap Filt.) Actual Hadron Level
Pythia 8 (Dijet Filt.) Recon / Hadron Level
Pythia 8 (Gap Filt.) Recon / Unfolded
Pythia 8 (Gap Filt.) Hadron Level / Unfolded
Pythia 8 Gap Filtered unfolded with
Dijet Filtered
Pythia 8 (Gap Filt.) Hadron Level
Pythia 8 (Gap Filt.) Reconstructed
Pythia 8 (Dijet Filt.) Reconstructed
Pythia 8 (Dijet Filt.) Unfolded
Pythia 8 (Dijet Filt.) Actual Hadron Level
Pythia 8 (Gap Filt.) Recon / Hadron Level
Pythia 8 (Dijet Filt.) Recon / Unfolded
Pythia 8 (Dijet Filt.) Hadron Level / Unfolded
Pythia 8 Dijet Filtered unfolded with
Gap Filtered
15
Unfolding Gap Size
Try same gap weighting method but comparing Pythia 8 ND to SD+DD and vice versa
Ratio of actual hadron level to unfolded not centred at 1 due to differences of behaviour at
low gaps but does stay very stable up to very large gap sizes when ND statistics are slightly
lower
- Plan to look at removing first couple of bins and see how well it is controlled
Pythia 8 SD+DD Hadron Level
Pythia 8 SD+DD Reconstructed
Pythia 8 ND Reconstructed
Pythia 8 ND Unfolded
Pythia 8 ND Actual Hadron Level
Pythia 8 SD+DD Recon / Hadron Level
Pythia 8 ND Recon / Unfolded
Pythia 8 ND Hadron Level / Unfolded
Pythia 8 SD+DD Gap Filtered unfolded
with Pythia 8 ND Gap Filtered
Pythia 8 (Gap Filt.) Hadron Level
Pythia 8 (Gap Filt.) Reconstructed
Pythia 8 (Dijet Filt.) Reconstructed
Pythia 8 (Dijet Filt.) Unfolded
Pythia 8 (Dijet Filt.) Actual Hadron Level
Pythia 8 (Gap Filt.) Recon / Hadron Level
Pythia 8 SD+DD Recon / Unfolded
Pythia 8 (Dijet Filt.) Hadron Level / Unfolded
Pythia 8 ND Gap Filtered unfolded with
Pythia 8 SD+DD Gap Filtered
16
Potential systematics
1 vertex cut – loosen cut or at least allow 0 to study effect
Gap start side asymmetry
Jet cleaning efficiency uncertainty
Existing
Unfolding
Trigger
Jet Energy Scale
Jet Angular Res.
Jet Energy Res.
Jet Recon. Eff.
Cluster Energy Scale.
Cluster threshold unc.
Tracking
Luminosity
Implemented Cluster Energy Scale systematic for Et flow method, speaking to
Emily/Robindra about finer details
Detector material uncertainty? Take from Et flow paper / soft diffractive paper?
17
Next steps
Agree which exact cuts to use for E±pz calculation
Complete ξ calculation by determining C parameter
Get unfolding working
Combine different samples together correctly
Try for more distributions
Investigate if other corrections are necessary e.g. Jet cleaning
Start running systematics again
Investigate noisy cells using Et flow method
Scaling background events that have passed cuts
18