double parton scattering at the LHC in the W W channel marc dnser - - PowerPoint PPT Presentation
double parton scattering at the LHC in the W W channel marc dnser (CERN) 20th of march 2019 outline 1) introduction to double parton interactions -> DPS vs. SPS -> factorization -> sigma effective -> problems of
2
1) introduction to double parton interactions
2) non-factorization in DPS
3) two examples of DPS analyses
4) outlook to the future
3
most analyses at the LHC focus on single parton-parton interactions (SPS)
most theoretical effort focuses on SPS as well
conversely, double parton scattering (DPS) is not very ‘popular’
a small group of theorists there are good reasons to concentrate on SPS
SPS DPS
4
as mentioned, we are usually interested in SPS processes
we can describe the cross section of an SPS process (example: higgs)
pdf term partonic cross section
5
even naively, once the parton model is introduced, DPS must exist
we can write the cross section of any DPS process similar to before
P.V. Landshoff, J.C. Polkinghorne,
pdf terms partonic cross sections distance between partons
6
this integral is clearly a bit more complicated than before
but none of the other things are quite the same
not the single pdfs from before!
how to deal with this complication?
correlations between partons
pdf terms partonic cross sections distance between partons
7
we can assume that the two parton-parton interactions are factorizable
decompose in longitudinal versus transverse components
can also assume longitudinal factorization
pdf terms partonic cross sections distance between partons
8
if those factorizations are assumed, the cross sections simplifies
write down the transverse component as a cross section
the rest are now exactly the SPS cross sections for processes A and B
really simple to calculate cross-sections on the back of an envelope
9
derived from the transverse extend of the partons in the proton
in the factorization approach sigma effective is a constant
quite a number of experimental measurements
in any case: ≃ 10-20 mb
10
can make a quick estimate of some interesting cross sections
compare: σHiggs = 50 pb, σWZ->3l = 5 pb
σSPS13 TeV 832 pb 61 nb 6 nb 170 nb 5.4 µb 430 pb tt W->lν Z->ll J/ψ 2jets 2γ tt << 2.56 fb 0.23 fb 7 fb 2.2 pb << W->lν
17 fb 523 fb 166 pb 1.3 fb Z->ll
50 fb 15 pb << J/ψ
460 pb 3.7 fb 2jets
1.1 pb 2γ
11
clearly the factorization assumption must break down
if both x1 and x2 are large, energy conservation can be violated
difficult to test is transverse factorization
more correlations to consider:
12
there are theoretical calculations that do not assume factorization
summarizing here the works of many theorists:
Double Parton Distributions Incorporating Perturbative QCD Evolution and Momentum and Quark Number Sum Rules
Parton correlations in same-sign W pair production via double parton scattering at the LHC
Multiple parton interactions at the LHC, WorldScientific, 2019
these papers introduce complex theoretical calculations
13
any of the solutions presented imply correlations
some of these correlations have experimental implications
longitudinal effects affect especially the rapidity distributions
any probe must satisfy a few criteria
xb = e−ηµ MW √s ⎡ ⎣MW 2pT ∓ ⎛ ⎝ MW 2pT 2 − 1 ⎞ ⎠ ⎤ ⎦ xa = eηµ MW √s ⎡ ⎣MW 2pT ± ⎛ ⎝ MW 2pT 2 − 1 ⎞ ⎠ ⎤ ⎦
14
cross section for DPS WW -> lνlν: ~ 95 fb
(this number is inclusive in flavors and charge etc.) does this process fulfill the requirements?
correlations are not the only consequence
15
non-factorized calculations lead to a number of observable effects
gaunt&stirling define an asymmetry that maximizes sensitivity
looks more complicate than it is
16
is a measure of how a W at large rapidity affects the probability
black dots are with sophisticated dPDFs
17
Ceccopieri et al predict more observables related to correlations
another effect again in the rapidities
subtle effect of ~10% in the cross section
18
just to understand what is implemented in current MC
i will be talking about pythia, because this is what i know best
things that are taken into account:
in other words: energy conservation
i.e. color conservation missing:
pythia and herwig the only generators that allow specific second hard scatter!
19
why do it at all?
for searches/Higgs/etc there are many ways of measuring DPS at the LHC
it very much depends on the goal
important point: we need a hadron collider for this!
15 orders of magnitude in cross section
20
besides full-blown DPS, there is also the “underlying event”
very important for MC tuning
derived from underlying event information
in general: DPS is “high enough” in scale to be treated perturbatively
hard object recoil most sensitive to UE
21
a few things necessary
CMS detector at the LHC
for leptons and γs
ATLAS detector
for leptons and γs
MET resolution LHCb works too!
22
a fairly old analysis out of ATLAS: 36 pb-1 of 7 TeV data
the ‘good’ object is the leptonic W: isolated lepton, excellently measured
going back to the simplified factorization approach: we can extract the fraction fDPS of DPS events over the total events and versus
23
definition of fDPS quite straightforward need to construct 2 templates: one for W+2jSPS (A), one for W+2jDPS (B)
the variable is the momentum of the normalized di-jet system trivially we want to fit: simulation
24
throw the model at the data, or the data at the model fairly large statistical uncertainties with that little data
can interpret fDPS in terms of a measurement of σeff
25
newest DPS analysis from the LHC with 77 fb-1 at 13 TeV
pro: SPS process is highly suppressed!
con: pretty low cross section, very crowded phase space
versus
W± q0(p2) q(p1) ⌫ `± W± q0(p2) q(p1) ⌫ `±
q q ⌫ q q ⌫ W ± `± q0 q0 W ± `± q q ⌫ ⌫ q0 W ⌥ W ± Z `± W ± W ⌥ `± q0
26
this analysis does not have a single, accurate estimation of the total cross section
two options to get an estimate of the inclusive cross section: 1) calculate the DPS WW cross section via the pocket formula
2) ask generators what the cross section is
these numbers are very different, but reflect the uncertainty though
27
this process was never measured before at a hadron collider
goals different w/r/t W+2jets:
phase space rather crowded, no strong handle to suppress backgrounds
28
backgrounds are plentiful in this region of phase-space
two most important backgrounds:
if the right Z-lepton is lost, it’s very similar
estimated with standard fakerate (tight-to-loose) method
very small contribution
29
train two BDTs in signal versus WZ and signal versus fakes
we train on 11 kinematic input variables
dphi(l1,l2), dphi(l2,met), dphi(l1l2,l2)
) (rad.)
±
µ
±
µ +
±
µ
±
(e
l2 MET
φ Δ 0.5 1 1.5 2 2.5 3 Events 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
DPS WZ non-prompt
(13 TeV)
41.4 fb
CMS Preliminary
)
±
µ
±
µ +
±
µ
±
(e
2
η *
1
η 6 − 4 − 2 − 2 4 6 Events 0.00 0.05 0.10 0.15 0.20 0.25
DPS WZ non-prompt
(13 TeV)
41.4 fb
CMS Preliminary
) (GeV)
±
µ
±
µ +
±
µ
±
ll (e
T
m 50 100 150 200 250 Events 0.00 0.05 0.10 0.15 0.20 0.25
DPS WZ non-prompt
(13 TeV)
41.4 fb
CMS Preliminary
) (rad.)
±
µ
±
µ +
±
µ
±
(l1l2 l2) (e φ Δ 0.5 1 1.5 2 2.5 3 Events 0.00 0.02 0.04 0.06 0.08 0.10 0.12
DPS WZ non-prompt
(13 TeV)
41.4 fb
CMS Preliminary
30
want to be able to fit a 1D distribution out of these two BDTs
combine the two BDT classifier into one discriminant variable with some underlying principles
large signal, low background large WZ & low fakes large fakes & low WZ
profit further from two facts:
perform a binned ML fit in four flavor and charge channels
2 4 6 8 10 12 14
)
+
µ
+
µ Bin number (
0.5 1 1.5 2
Data/bkg.
total background uncertainty DPS WW
20 40 60 80 100 120 140 160 180 200
Events
ZZ DPS WW Nonprompt WZ Rare * γ W Data
Preliminary CMS (13 TeV)
1 −77 fb 2 4 6 8 10 12 14
)
−
µ
−
µ Bin number (
0.5 1 1.5 2
Data/bkg.
total background uncertainty DPS WW
20 40 60 80 100 120
Events
ZZ DPS WW Nonprompt WZ Rare * γ W Data
Preliminary CMS (13 TeV)
1 −77 fb 2 4 6 8 10 12 14
)
+
µ
+
Bin number (e
0.5 1 1.5 2
Data/bkg.
total background uncertainty DPS WW
50 100 150 200 250 300 350
Events
ZZ DPS WW Nonprompt WZ Rare * γ W Data Charge misid. γ W/Z
Preliminary CMS (13 TeV)
1 −77 fb
31
showing postfit plots of the final 1D classifier
found a total of 4921 events in data
decreasing sensitivity
µ+µ+
µ-µ- e-µ- e+µ+
2 4 6 8 10 12 14
)
−
µ
−
Bin number (e
0.5 1 1.5 2
Data/bkg.
total background uncertainty DPS WW
50 100 150 200 250
Events
ZZ DPS WW Nonprompt WZ Rare * γ W Data Charge misid. γ W/Z
Preliminary CMS (13 TeV)
1 −77 fb
32
sensitivity large enough to claim first evidence
also extract
1 2 3 4 5 6
(pb)
DPS WW
σ Inclusive
±
µ
±
+e
±
µ
±
µ 0.28) pb ± 0.28 , ± 0.40 ( ± 1.41
+
µ
+
+e
+
µ
+
µ 0.32) pb ± 0.33 , ± 0.46 ( ± 1.36
−
µ
−
+e
−
µ
−
µ 0.51) pb ± 0.54 , ± 0.74 ( ± 1.96
Preliminary CMS (13 TeV)
77 fb total stat syst Observed stat syst Predictions: PYTHIA 8 (CP5) Factorization approach
(13 TeV)
±W
±W CMS
SMP-18-015 (2019)4l (13 TeV) ATLAS
CERN-EP-2018-274 (2018)(13 TeV)
±W
±W CMS
PAS FSQ-16-009 (2017)DPS (8 TeV)
±W
±W CMS
JHEP 02 (2018) 032(8 TeV) ψ Z+J/ ATLAS
EPJC 75 (2015) 229W+2jets (7 TeV) CMS
JHEP 03 (2014) 032W+2jets (7 TeV) ATLAS
New J. P. 15 (2013) 033038+2jets (1.96 TeV) γ 2 D0
PRD 93 (2016) 052008+b/c+2jets (1.96 TeV) γ D0
PRD 89 (2014) 072006+3jets (1.96 TeV) γ D0
PRD 89 (2014) 072006+3jets (1.8 TeV) γ CDF
PRL 79 (1997) 584(mb)
eff.
σ 5 10 15 20 25 30 35 extractions (vector boson final states)
eff
σ
33
the LHC is only at the beginning of data-taking
focus here on DPS W±W± process
reminder: if parton correlated σeff will vary with η1η2
2
η ⋅
1
η 8 − 6 − 4 − 2 − 2 4 6 8 (mb)
eff
σ 10 15 20 25 30 35 40
@14 TeV (stat. err.)
3 ab
eff
σ projected projected (stat. err.) | < 2.4 η
µ @ 13 TeV
36 fb
eff
σ meas. (theory)
eff
σ constant (theory)
eff
σ variable
CMS Phase-2 Simulation
14 TeV, 200 PU
< 0
2η ⋅
1η eff
σ /
> 0
2η ⋅
1η eff
σ
0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15
CMS Phase-2 Simulation
14 TeV, 200 PU
ratio (theory)
eff
σ constant ratio (theory)
eff
σ variable (stat err. and stat+syst err.) ratio HL-LHC (projected)
eff
σ (stat+syst err.) < 2.4
µ
| η ratio coverage |
eff
σ
34
more questions to answer down the road: how well can we measure aη
with less than 3000 fb-1? does σeff depend on the production mode?
and LHCb see σeff < 10 mb how high can we push the mass scale?
produce better MC, including correlations