Transcript Relative Tuning of the Pythia Underlying Event for Recent PDFs

Relative Tuning of the Pythia
Underlying Event for Recent PDFs
OUTLINE
I.
II.
III.
IV.
V.
S. Muanza
Introduction and Methodology
Tools utilized
Comparison Method
Current Results
Prospects
Simulation Meeting
16 September 2005
I. Introduction
• A quite detailed study of the underlying event has been performed by Rick Field a
theorist working in the CDF collaboration
(http://www.phys.ufl.edu/~rfield/cdf/rdf_talks.html)
• This study has been sustained for more than 5 years
• Working definition of the Underlying Event:
• All but the hard scattering process
• ie: beam-beam remnants (spectator partons), plus possible ISR gluon radiations ,
plus the possible Multiple Parton Interactions (MPI)
• Systematic comparisons of CDF Run I data (min.bias and soft jets) to different MC
models have been performed and finally led to a tuning of Pythia underlying event
model:
S. Muanza
Simulation Meeting
16 September 2005
I. Introduction
First Step
Second Steps
Pythia
Version
6.115
6.206
PDF
CTEQ4L
CTEQ5L
Tuning Name
“““Tune 0”””
Tune A, B, C, D
MSTP Values
MSTP(81)=1 (MPI on)
MSTP(81)=1
MSTP(82)=4
MSTP(82)=4 (dble gauss. had. matter dens.)
PARP Values
PARP(82)=2.4 (MPI pT cut-off)
PARP(67)=4.0;PARP(82)=2.0
PARP(83)=0.5;PARP(84)=0.4
PARP(85)=0.9;PARP(86)=0.95
PARP(89)=1800.0;PARP(90)=0.25
Usage at D0
mcp10-mcp14
cardfiles/np/v00-02-01 to v00-04-58
mcp14
cardfiles/np/v00-04-59 to v00-08-53
cardfiles/dzero/v00-05-01 to v00-08-53
S. Muanza
Simulation Meeting
16 September 2005
I. Introduction
• This tuning is PDF dependent
(http://cepa.fnal.gov/patriot/mc4run2/MCTuning/run2mc/R_Field.pdf)
• This tuning fits CDF Run IIA min.bias+soft jets data
• Provided decent choice for the renormalization scales, this tuning also fits the
UE for the bbbar, di-photon, Z+jets processes
(http://www.phys.ufl.edu/~rfield/cdf/RickField_Workshop_6-11-04.pdf)
S. Muanza
Simulation Meeting
16 September 2005
I. Methodolgy
• Since the UE tuning is PDF dependent it should in principle be redone whenever
changing from the “reference PDF” (CTEQ5L for Tune A)
• However this is obviously cumbersome since it requires correcting either the data or
the detailed MC and re-doing the full tuning procedure each time
• I propose instead to start from a reference (CTEQ5L for Tune A) that was properly
tuned to data and just to reproduce its UE properties
• This only requires generator level or fast simulation scan over the UE parameters:
whatever set of parameters that reproduces the reference UE constitutes the relative UE
tuning for a given PDF
• I assume the p/pbar hadronic matter is described by a double gaussian (MSTP(82)=4
as in Tune A), so I’m left w/ scanning “only” over 7 PARP parameters (67,82-86,90)
since PARP(89)=1800.0 keeps its fixed value (all the evolutions to another CoM
energy are internally treated within Pythia)
S. Muanza
Simulation Meeting
16 September 2005
I. Methodolgy
Scan over the UE Parameters
UE Parameter Min
Max
Scan Step
Default
PARP(67)
1.0
4.0
1.0
1.0
PARP(82)
1.8
2.1
0.1
2.0
PARP(83)
0.4
0.6
0.1
0.5
PARP(84)
0.3
0.5
0.1
0.2
PARP(85)
0.33
1.00
~0.33
0.33
PARP(86)
0.33
1.00
~0.33
0.66
PARP(90)
0.20
0.30
0.05
0.16
This scan contains 3888 different PARP configurations
S. Muanza
Simulation Meeting
16 September 2005
II. Tools Utilized
• Generator:
Pythia v6.320
• PDF Library:
LHAPDF v4.0
• Fast detector simulation:
ATLFAST v2.60 (Atlas Collaboration), including smeared tracks and jets
• Events production:
• Process:
Pythia minbias  MSEL=2  MSUB(91-95)
 elastic scattering+ diffraction + low pT QCD, w/ pT* > 0 GeV
Note: the soft jets part is not yet produced ( MSEL=1, w/ pT* > 5 GeV)
• Statistics:
25k / sample (ie per PDF/ & per PARP combination)
• PDF:
ref. sample:
• CTEQ5L (LO fit & LO aS)
compar. sample:
• CTEQ6LL, ALEKHIN02LO, MRST01LO (LO fit & LO aS)
• CTEQ6L (LO fit & NLO aS)
S. Muanza
Simulation Meeting
16 September 2005
II. Tools Utilized
• Events selection:
• Similar to R. Field's:
events w/ 1 or 2 jets, pT(jets)> 0 GeV, |eta(jets)|<2.0
• The transverse plane is divided into 4 regions:
• towards: |Df(ojbect,leading jet)|<60°
• away: | Df(ojbect, leading jet)|>120° (only for 2 jet events)
• transverse regions: 60°<|Df(ojbect,leading jet)|<120°
• Look at tracks w/ pT(tracks)>0.5 GeV and |eta(tracks)|<1.0 in the transverse
regions
• Construct 2-D histos:
• Ntracks/Dh/Df/(1 GeV) vs leading jet pT
• SpT/ Dh/Df/(1 GeV)
vs leading jet pT (scalar pT sum)
• Differences wrt R. Field:
I used "calorimeter jets" instead of “track jets”
=> pT(jets)>6 Gev instead of 0 GeV
• Note:
the overall efficiency is rather low (~12%) and since I did not write any
filter for the produced events, the comparisons are only based on a KS
test of two 2-D histos w/ ~3 k entries!!!
S. Muanza
Simulation Meeting
16 September 2005
Charged Particle Df
Correlations
Charged Jet #1
Direction
“Toward-Side” Jet
Df
“Toward”
“Transverse”
“Transverse”
Charged Jet #1
Direction
Df
2p
Away Region
Transverse
Region
“Toward”
“Transverse”
“Transverse”
f
Leading
Jet
Toward Region
“Away”
Transverse
Region
“Away”
Away Region
0
“Away-Side” Jet
-1
h
+1
• Look at charged particle correlations in the azimuthal angle Df relative to the
leading charged particle jet.
• Define |Df| < 60o as “Toward”, 60o < |Df| < 120o as “Transverse”, and
|Df| > 120o as “Away”.
• All three regions have the same size in h-f space, DhxDf = 2x120o = 4p/3.
S. Muanza
Simulation Meeting
16 September 2005
Tuned PYTHIA 6.206 vs HERWIG 6.4
“TransMAX/MIN” vs PT(chgjet#1)
Charged Jet #1
Direction
<Nchg>
“Toward”
“TransMAX”
“TransMIN”
“Away”
<PTsum>
"Transverse" <Nchg> in 1 GeV/c bin
Df
"Max/Min Transverse" Nchg
3.0
Tuned PYTHIA 6.206
PARP(67)=1
CDF Preliminary
2.5
data uncorrected
theory corrected
Tuned PYTHIA 6.206
PARP(67)=4
"Max Transverse"
2.0
1.5
CTEQ5L
HERWIG 6.4
1.0
"Min Transverse"
0.5
1.8 TeV |h|<1.0 PT>0.5 GeV
0.0
0
• Plots shows data on the “transMAX/MIN”
S. Muanza
10
15
20
25
30
35
40
45
50
PT(charged jet#1) (GeV/c)
"Max/Min Transverse" PTsum
3.5
<PTsum> (GeV/c) in 1 GeV/c bin
•
<Nchg> and “transMAX/MIN” <PTsum> vs
PT(chgjet#1). The solid (open) points are
the Min-Bias (JET20) data.
The data are compared with the QCD
Monte-Carlo predictions of HERWIG 6.4
(CTEQ5L, PT(hard) > 3 GeV/c) and two
tuned versions of PYTHIA 6.206
(PT(hard) > 0, CTEQ5L, PARP(67)=1 and
PARP(67)=4).
5
Tuned PYTHIA 6.206
PARP(67)=1
CDF Preliminary
3.0
data uncorrected
theory corrected
Tuned PYTHIA 6.206
PARP(67)=4
"Max Transverse"
2.5
2.0
1.5
HERWIG 6.4
CTEQ5L
1.0
"Min Transverse"
0.5
1.8 TeV |h|<1.0 PT>0.5 GeV
0.0
0
Simulation Meeting
5
10
15
20
25
30
35
40
45
50
PT(charged jet#1) (GeV/c)
16 September 2005
III. Comparison Method
• Histo Comparisons: for each PARP configuration and for each PDF, the two
2-D histos are compared using a 2-D Kolmogorov-Smirnov test to those of the ref.
sample (just the shapes enter the comparison, not the normalizations)
• Global probability: the probability assigned to each comparison sample is simply
the product [1] of the individual probability of comparing on one hand the
charged tracks density and on the other hand the pTsum density
PKS ((var 1) &(var 2))  PKS (var 1)  PKS (var 2)
Valid if & only if var1 and var2 are not correlated!!!
PKS ((var 1) &(var 2))  PKS (var 1 | var 2)  PKS (var 2)
Have to calculate a conditional probability if var1 and var2 are correlated!!!
• Tools: all the histos and comparison methods are taken from ROOT v4.04.02b
S. Muanza
Simulation Meeting
16 September 2005
IV. Current Results
• PDF: ALEKHIN02LO
• LO fit and LO aS
• 3881/3888 configs
S. Muanza
• Max(PKS)=0.967 008 (8 max configs)
• Min(PKS)=2.058x10-10 (4 min configs)
UE Parameter Best
Worst
CTEQ5L
Tune A
PARP(67)
FLAT
FLAT
4.0
PARP(82)
2.0
1.8
2.0
PARP(83)
0.6
0.4
0.5
PARP(84)
0.3
0.4
0.4
PARP(85)
0.66
0.33
0.9
PARP(86)
0.33-0.66
1.0
0.95
PARP(90)
0.30
0.30
0.25
Simulation Meeting
16 September 2005
IV. Current Results
• PDF: MRST01LO
• LO fit and LO aS
• 3820/3888 configs
S. Muanza
• Max(PKS)=0.956524 (12 max configs)
• Min(PKS)=4.433x10-11 (4 min configs)
UE Parameter Best
Worst
CTEQ5L
Tune A
PARP(67)
FLAT
FLAT
4.0
PARP(82)
1.80
1.90
2.0
PARP(83)
0.6
0.4
0.5
PARP(84)
0.3
0.5
0.4
PARP(85)
1.0
0.33
0.9
PARP(86)
FLAT
1.0
0.95
PARP(90)
0.20
0.20
0.25
Simulation Meeting
16 September 2005
IV. Current Results
• PDF: CTEQ6L
• LO fit and NLO aS
• 3867/3888 configs
S. Muanza
• Max(PKS)=0.954313 (12 max configs)
• Min(PKS)=2.924x10-10 (4 min configs)
UE Parameter Best
Worst
CTEQ5L
Tune A
PARP(67)
FLAT
FLAT
4.0
PARP(82)
2.10
1.80
2.0
PARP(83)
0.6
0.4
0.5
PARP(84)
0.5
0.4
0.4
PARP(85)
1.0
0.33
0.9
PARP(86)
FLAT
1.0
0.95
PARP(90)
0.25
0.30
0.25
Simulation Meeting
16 September 2005
IV. Current Results
• PDF: CTEQ6LL
• aka CTEQ6L1
• LO fit and LO aS
• 3886/3888 configs
S. Muanza
• Max(PKS)=0.977060 (12 max configs)
• Min(PKS)=1.815x10-11 (4 min configs)
UE Parameter Best
Worst
CTEQ5L
Tune A
PARP(67)
FLAT
FLAT
4.0
PARP(82)
2.00
2.00
2.0
PARP(83)
0.4
0.4
0.5
PARP(84)
0.5
0.4
0.4
PARP(85)
1.0
0.33
0.9
PARP(86)
FLAT
1.0
0.95
PARP(90)
0.20
0.30
0.25
Simulation Meeting
16 September 2005
IV. Current Results
Example w/ Alekhin 2002 LO PDF
ref
best
worst
« same »
HT (GeV)
S. Muanza
Simulation Meeting
16 September 2005
IV. Current Results
ref
best
worst
« same »
mET (GeV)
S. Muanza
Simulation Meeting
16 September 2005
IV. Current Results
ref
best
worst
« same »
N(jets)
S. Muanza
Simulation Meeting
16 September 2005
IV. Current Results
ref
best
worst
« same »
Total N(tracks)
S. Muanza
Simulation Meeting
16 September 2005
IV. Current Results
• After fixing the correlation issue:
Var2Ntracks/Dh/Df/(1 GeV)
Var1SpT/ Dh/Df/(1 GeV)
Attaching file plots.root as _file0...
root [1] h2_hist1_mix0->GetCorrelationFactor(1,2)
(const Stat_t)1.22289661104907951e-01
root [2] h2_hist1_mix1->GetCorrelationFactor(1,2)
(const Stat_t)8.79092032677424529e-01
root [3] h2_hist2_mix0->GetCorrelationFactor(1,2)
(const Stat_t)1.18224432124161408e-01
root [4] h2_hist2_mix1->GetCorrelationFactor(1,2)
(const Stat_t)8.23833825978842360e-01
Var1(SpT/Ntracks)/ Dh/Df/(1 GeV)
• The correlation coefficient drops from 80% downto 12%
• This makes the marginal probabilities product an acceptable approximation
S. Muanza
Simulation Meeting
16 September 2005
VI. Conclusions & Prospects
•Conclusions:
• There are flat directions (as expected in multivariate analyses, especially w/
coarse scans and limited statistics). In this case I propose to pick the PARP value
which is the closest to the reference one (CTEQ5L+Tune A)
• As expected the shape of the so-called “best” configuration (green histos) is the
closest to that of the reference (black histos). This demonstate that there is a
measurable difference between different UE settings for a given PDF and that the
UE is PDF-dependent.
Prospects:
• Produce the low pT QCD samples
• Add them to the 2-D histos for the comparisons
• Couple of additional cross checks
• Increase the statistics
S. Muanza
Simulation Meeting
16 September 2005
VI. Prospects
•
• Produce the low pT QCD samples
• Add them to the 2-D histos for the comparisons
• Couple of additional cross checks
• Increase the statistics
S. Muanza
Simulation Meeting
16 September 2005
Back Up
S. Muanza
Simulation Meeting
16 September 2005
Pythia UE Parameters Definition
UE Parameter
Definition
MSTP(81)
MPI on/off
MSTP(82)
3 / 4: resp. single or double gaussian hadronic matter
distribution in the p / pbar
PARP(67)
ISR Max Scale Factor
PARP(82)
MPI pT cut-off
PARP(83)
Warm-Core: parp(83)% of matter in radius parp(84)
PARP(84)
Warm-Core:
PARP(85)
prob. that an additional interaction in the MPI formalism gives
two gluons, with colour connections to NN in momentum space
PARP(86)
prob. that an additional interaction in the MPI formalism gives
two gluons, either as described in PARP(85) or as a closed gluon
loop. Remaining fraction is supposed to consist of qqbar pairs.
PARP(89)
ref. energy scale
PARP(90)
energy rescaling term for PARP(81-82)~ECM^PARP(90)
S. Muanza
Simulation Meeting
”
16 September 2005
VI. Final Checks on Shapes
S. Muanza
Simulation Meeting
16 September 2005