Transcript 20140127_SoftQcd_Diff - Elementary Particle Physics

Update on Diffractive Dijets
Analysis
Hardeep Bansil
University of Birmingham
Soft QCD / Diffraction WG Meeting
27/01/2014
Contents
• Introduction
• Updates
– Unfolding systematic
– 2D unfolding
– Reweighting Pythia8 SD/DD
– Efficiencies in data
– Note progress
• Further plans
2
Diffractive dijets
• Single diffractive events (pppX)
•
Rapidity gap from colourless exchange with vacuum quantum numbers
“pomeron”
• Search for hard diffraction with a hard scale set by 2 jets
•
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 diffractive to inclusive dijet events
• Gap Survival Probability – the chance of the gap between the
intact proton and diffractive system staying intact due to
scattering
•
•
Rescatter with p ?
Tevatron have Gap Survival Probability of 0.1 relative to H1 predictions
Khoze, Martin and Ryskin predict LHC to have GSP of ~ 0.03
ξ
3
Analysis
• 2010 Period B data with GRL (L1Calo and MinBias streams) - ∫L dt = 6.753 nb-1
–
Low pile-up (Peak <μ> for selected runs < 0.15)
–
Trigger using mixture of L1_MBTS_1 (prescaled) and L1_J5 (unprescaled)
–
Vertex requirement - 1 PV (5+ associated tracks), no additional vertices (2+ tracks)
• PYTHIA8 samples of ND, SD and DD events with ATLAS UE Tune AU2-CT10
–
Samples produced with jets in different pT ranges, no pile-up
–
SD+DD use Schuler-Sjöstrand for IP flux (ε = 0), unconventional
• At least 2 anti-kT jets (R=0.4 or R=0.6) with pT > 20 GeV, |η| < 4.4
–
Require medium quality jet cleaning cuts in data
• Calculate forward gap (ΔηF) and xi (ξ±) in range |η|<4.8


 (E
C
i
 p zi )
s
• Using “hybrid” p and pT cuts for clusters / stable truth particles
4
(Debugging) Unfolding systematic
• Data-driven method determined from difference between unfolded reweighted reco
MC and reweighted true MC (reweighting MC based on reco MC/data comparison)
• Typically up to 15%, increasing to 45% difference where statistics are limited so
splitting up process into individual steps
– Strongly affects distributions with gap > 3.0 cut (which we are interested in)
•
Reco data distribution
•
Reco MC distribution
•
Ratio of reco data / reco MC
•
Fitted function to this distribution
•
True MC distribution
•
True MC distribution weighted by weighting
function (True-weighted MC)
•
True-weighted MC after being folded through
a statistically independent matrix (Reco MC*)
•
Ratio of reco data / reco MC*
•
Unfolded reco data
•
Unfolded reco MC*
•
Ratio of unfolded reco data / unfolded reco MC*
5
(Debugging) Unfolding systematic
• Data-driven method determined from difference between unfolded reweighted reco
MC and reweighted true MC (reweighting MC based on reco MC/data comparison)
• Much worse in ξ± than gap size at larger xi values – due to corrections applied?
6
(Debugging) Unfolding systematic
•
Reco data distribution, Reco MC distribution
•
Ratio of reco data / reco MC, fitted function to ratio
•
Should these be scaled based on integral or not?
•
Tried several different polynomials as tests
•
Ended up with 5th order polynomial
•
No unique solution for different variables so could
be optimised
•
Not perfect description either but is this reasonable?
7
(Debugging) Unfolding systematic
•
True MC distribution
•
True MC distribution weighted by weighting function (True-weighted MC)
• Provided scaling not necessary in Step 1 then should be OK
8
(Debugging) Unfolding systematic
•
True-weighted MC after being folded through a
statistically independent matrix (Reco MC*)
•
Not sure that smearing on original response matrix being
applied properly (will ask offline)
Response matrix: Representing reconstructed
physical quantity vs. true physical quantity
passing cuts
Folding matrix: matrix giving probability
for a value of the true physical quantity
to be reconstructed at another value
9
(Debugging) Unfolding systematic
•
Ratio of reco data / reco MC*
•
Unfolded reco data, Unfolded reco MC* +
ratio
•
Average of ratio used for fit around 5, explaining
increased difference between reco MC* and data
•
Data and reco MC* unfolded using the same
original response matrix
•
Ratio of unfolded data and reco MC* roughly the
same as LHS plot
10
2D Unfolding
•
Correction process uses RooUnfold software to unfold individual distributions
separately but there are multiple exponentially falling distributions in this analysis:
∆ηF and jet pT
•
•
Both produce a net migration to larger rapidity gaps
Improve by using simultaneous unfolding procedure for both variables
•
Removed 5% pT reduction previously being applied to reconstructed jets
• ∆ηF against leading jet pT (R=0.4)
Raw data
Unfolded data
11
2D Unfolding
•
∆ηF against leading jet pT (R=0.4) – Supporting plots from MC
•
Fill (passes cuts at both truth and recon levels) – enough stats for measurement
Recon MC
Fill
•
Fakes (pass at recon, fails at truth) and Miss (fails at recon, pass at truth)
Recon MC
Fake
•
Truth MC
Fill
Truth MC
Miss
Fakes have very large spread across gaps and pt
12
2D Unfolding
•
Comparing to 1D unfolding measurement
•
Ratio of unfolded to raw data decreases in every bin but overall change not that large
•
Raw and unfolded data agree in first bin (may also be consequence of putting fakes and misses
into response)
2D Unfolding
•
1D Unfolding
Big drop in MC and data for bin at 6-6.5 in data
•
Phase space limit should be at gaps greater than 7? Other effect? Drop bin?
13
2D Unfolding
•
Currently have the capability to unfold (for anti-kt R=0.4 and R=0.6)
•
∆ηF against leading jet pT
•
ξ± against leading jet pT
•
Still need to add this to cross section measurement
•
Should we do 2D unfolding for other variables e.g. leading jet eta?
•
How will unfolding systematic change if going to 2D?
14
Reweighting Pythia8 SD/DD
•
Analysis uses one set of Monte Carlo samples generated using PYTHIA8
•
PYTHIA8 diffractive samples use
unconventional model (Schuler-Sjöstrand)
in generation
•
Reweight SD based on ξ distribution
obtained from inclusive diffractive
PYTHIA8 samples
15
Reweighting Pythia8 SD/DD
•
Reweight SD based on ξ distribution obtained from inclusive diffractive PYTHIA8 samples
•
Ratio fitted with 3rd order polynomial for B-S model (although did not work for D-L so will revisit)
•
DD trickier as cannot measure ξ directly
•
Can try determining MX from largest gap between stable truth particles but not
guaranteed to be perfect
•
Any other quantities that can be used to test with?
16
Trigger efficiencies in data
• Biggest difference between Bham/Prague down to trigger approach for data
• Gives different numbers of events in MinBias/L1Calo streams
• Matching of jets to Jet RoIs same but studied for different η ranges
J5 efficiency
MinBias Data
Excl. EM Transition
• Bham studies up to |η|<2.9, looks at
EM barrel-transition (1.3<|η|<1.6) separately
–
Separate fits for anti-kt R=0.4/0.6 jets (good/bad thing?)
• Prague follows method 2010 dijet analysis
•
–
|η|<0.3,0.3<|η|<0.8,0.8<|η|<1.2, 1.2<|η|<2.1,2.1<|η|<3.2
–
Better accounts for the η dependence that exists at low pT
–
Better as a reference to trigger group to defend this
–
Currently no difference in
J5 efficiency
Pythia8 ND
(Zoomed in)
Looking at 2010 dijet analysis as well to confirm
Prague results (results soon)
17
Note progress
• Thesis converted into ATLAS format
• Currently being edited (Paul, myself)
• Need to introduce Prague studies as well
– Need immediate agreement on analysis (trigger) selection, consistency checks
etc. at same time otherwise will be rewriting more sections
– Finalise this week?
• Will make it available on a CERN SVN repository accessible to
everyone working on analysis
– How to set this up?
• When is realistic time to have first combined draft ready  EB etc?
– Realistically 4 months left for me (similar timescale for Vlasta?)
18
Further plans
• Understand why unfolding systematic large in places and add to csx measurement
• Updated efficiency method in next few days
• Hopefully finalise Bham/Prague analysis definitions this week
• Sort out reweighting in Pythia8 SD/DD – show effect on distributions
• Some systematics used in 2010 dijet analysis derived from PYTHIA6 – should
determine from PYTHIA8 where possible?
• POMWIG
– Currently generated 800,000 Pomwig events at truth level (EVNT)
– 100,000 for each jet pt range, intact proton direction available in job options
– Need to convert again to NTUP_TRUTH as missing anti-kt R=0.6 jet collection
– Anyone know what settings need to be added to Athena transform for this?
• Other MCs to generate for want model independent conclusions?
19