Pythia 8 status report
ATLAS Process Modelling GroupSeptember 04, 2018Stefan Prestel (Lund/Fermilab)
Accurate Monte Carlo predictions
Accurate pseudodata from theory tools → better analyses of backgrounds,better analyses of signals
Colliding composite objects kick-startsmany processes:
hard scatteringradiation cascademultiparton interactionshadronization and decay
2 / 33
PYTHIA in a nutshell
Beams ee, ep, pp, γx, pA, AA, DMHard scattering Core library of internal processes, oth-
erwise from external tools. NLO+PSmatching/merging with both aMC@NLO andPOWHEG-BOX processes.
Parton shower Three alternative models: Default (w/ andw/o dipole recoil), Vincia and Dire plugins.
Multiparton interactions Regularised secondary 2 → 2 SM scatterings,interleaved with shower evolution.
Soft physics Regge-based diffraction and cross sectionmodels
Fragmentation String hadronization with Schwinger-basedor thermal transition probabilities.
PYTHIA8 convenience features: “Automatic” PS uncertaintiesPYTHIA8 news: PS developments, improved elastic & diffractive crosssections, heavy ions and photons
3 / 33
I) Hard scattering
Domain of fixed-order and precision calculations. Big community effort,including POWHEG-BOX, MADGRAPH, aMC@NLO & Pythia.
4 / 33
Matching & merging
Philosophy: Combine multi-jet calculations with each other and with subsequentshower for maximal accuracy.
Matching: Combine {n, n + 1}-partonstates with shower. Overlap handled bysubtraction. Increase precision of inclusiven-parton observables to NLO.
Merging: Combine {n, n+1, . . . , n+m}-parton states with shower. Overlaphandled by cuts & vetoes.
NLO merging: Same as merging, but withsome overlap handled by subtraction. NLOprecision of inclusive n + i-parton observ-ables for well-separated n + i jets.
5 / 33
Matching & merging: Availability through Pythia + friends
Matchingthrough POWHEG-BOX: works with Pythia shower variations.through aMC@NLO: requires global recoil for first emission.
MergingCKKW-L:default in ATLAS. Partially combined with EW corrections. Currentlynot working with shower variations.MLM: not clear if can work with shower variations.FxFx: more heavily used since more streamlined on aMC-side. not clear ifcan work with shower variations.UNLOPS: implementation in ATLAS underway. Currently not working withshower variations.
OtherDefault Pythia often includes matrix element corrections for 1st splittings.Vincia plugin contains iterated ME corrections as alternative to merging.
6 / 33
Matching & merging sample plots
…you probably have more than I do.
7 / 33
Cautionary tales…
EPJC 76 (2016) 1, 39
b
b
b
bb b b b b b
bb b b b b b b b b b b b b
bb
b
b
b
b
Datab
Default scheme
EW-improved
10−4
10−3
10−2
10−1
1
W → eν (MC) vs W → ℓν (data), dressed level
dσ
W+≥
2j/
dH
T
200 400 600 800 1000 1200 1400 1600 1800 20000.60.70.80.91.01.11.21.31.4
HT [GeV]
MC
/D
ata
Higher-orders for QCD- and EW corrections very different!Be holistic when e.g. choosing scales (Merging:allowWeakClustering = on)8 / 33
Matching/Merging Questions or Discussion?
9 / 33
II) Parton showering
Crucial part of physics modeling, as required for jet structure andevolution. NLO+PS only as good as the PS
10 / 33
Parton shower intro
a
Showers “dress” partons withradiation by iteratively generatingbranchings.
Each generated exclusive state is asolution to an evolution equation.
Summed semi-inclusively, this solu-tion recovers DGLAP evolution.
Summed fully inclusively, the inputcross section is not changed.
These boundary conditions do not determine the cascade completely→ different choices beyond simplest leading terms allowed(ordering/radiation functions/phase space mapping)
11 / 33
Shower options for Pythia
Current default⋄ Improved DGLAP evolution in p⊥⋄ ME corrections for 1st splitting.⋄ QCD, QED, EW, hidden valley⋄ Extensive tuning expertise.
a
Default with new recoil⋄ Improved DGLAP QCD evoln in p⊥.⋄ Improved (dipole) recoil forDIS/VBF-like processes.⋄ No realistic tune yet.
Vincia plugin⋄ Soft/collinear QCD evolution in1/eikonal⋄ Implements iterated LO matrix el-ement corrections.⋄ Detailed handcrafted tune.
a
Dire plugin⋄ Soft/collinear QCD evolution in1/eikonal⋄ Implements NLO corrections toevolution.⋄ Simple (LO) Professor tune.
12 / 33
Differences between parton shower optionsarXiv:1803.07977 (LH proceedings 2017)
Le
sHo
uc
he
sS
MW
G2
01
8
CSS
1/4 t ≤ µ2R ≤ 4 t
Dire
1/4 t ≤ µ2R ≤ 4 t
Pythia
1/4 t ≤ µ2R ≤ 4 t
0
0.005
0.01
0.015
0.02
0.025
dσ
/d
ph ⊥
[pb
/G
eV]
a
Theory of default and plugins differs. . . in ordering. . . in radiation functions. . . in treatment of coherence.leading to visible differences.
Shower plugins further handle onuncertainty for shower-sensitiveobservables (jet substructure…)
13 / 33
Shower variations in PYTHIAarXiv:1605.08352
PYTHIA allows UncertaintyBands:doVariations = on for automatic variations of⋄ µr in shower (fsr:muRfac=0.5 isr:muRfac=0.5…)⋄ finite pieces of splitting kernels (fsr:cNS=2.0 isr:cNS=-2.0…)⋄ PDF members in shower (e.g. isr:PDF:plus, isr:PDF:minus…)VINCIA includes vincia24.cc as illustration (slightly different syntax)DIRE includes dire03.cc as illustration (syntax slightly different) 14 / 33
Peculiarities when varying µP Sq and µP S
r arXiv:1605.08352
µP Sr/q variations will often yield regions with vanishing uncertainty band.
This is expected from “shower unitarity” (area under curve = inclusive x-section)
p⊥evol
dσ/dp ⊥
evol
µpsr =
1
2p⊥evol
µpsr = p⊥evol
µpsr = 2p⊥evol
a
PDF variations can “fill in” some regions. 15 / 33
Parton shower variations with compensation terms
arXiv:1803.07977 (LH proceedings 2017)
Le
sHo
uc
he
sS
MW
G2
01
8
CSS
1/4 t ≤ µ2R ≤ 4 t
Dire
1/4 t ≤ µ2R ≤ 4 t
Pythia
1/4 t ≤ µ2R ≤ 4 t
1 2
10−5
10−4
10−3
10−2
dσ
/d
ph ⊥
[pb
/G
eV]
1 2
0.8
0.9
1.0
1.1
1.2
CS
S
1 2
0.8
0.9
1.0
1.1
1.2
Dir
e
1 10 1 10 2
0.8
0.9
1.0
1.1
1.2
ph⊥
[GeV]
Py
thia
a
PS includes many improvementsbeyond leading-order, that mightbe nice to retain when performingvariations
→ Introduce compensating terms.
Drastic reduction realistic?
16 / 33
(Almost) complete NLO shower vs. uncertainty estimatesarXiv:1805.03757 (S. Höche, F. Dulat, SP)
Dir
eP
S
LO⊕NLO soft
κ/2 ≤ µ ≤ 2κ
LO
κ/2 ≤ µ ≤ 2κ
LO no CMW
5
10
15
20
25
30
35
dσ
/d
log
10(y
34)
[nb
]
-4 -3.5 -3 -2.5 -2 -1.5 -1
0.6
0.8
1
1.2
1.4
log10(y34)
Ra
tio
Le
sHo
uc
he
sS
MW
G2
01
8
CSS
1/4 t ≤ µ2R ≤ 4 t
Dire
1/4 t ≤ µ2R ≤ 4 t
Pythia
1/4 t ≤ µ2R ≤ 4 t
4 3 2 1
10−2
10−1
1
10 1
10 2
10 3
dσ
/d
y3
4
4 3 2 1
0.8
0.9
1.0
1.1
1.2
Dir
e
10−4 10−3 10−2 10−1
0.8
0.9
1.0
1.1
1.2
yDurham34
Py
thia
→ Reduced uncertainty in NLO calculations.→ “Compensated” LO scale variation attempts give much too small &→ distorted band.
17 / 33
Parton shower Questions or Discussion?
18 / 33
III) Tuning
A collision is more than “just” perturbative QCD. Heuristic models neededto fill the gaps. Inherent parameters need to be extracted from data →Tuning. 19 / 33
Correlations between pQCD and soft/non-perturbative modelling
Uncertainties:Short-distance cross section:µH
r , µHf , PDFH , αH
s
Parton shower:µP S
q , µP Sr , µP S
f , µP Scut, PDFP S , αP S
s
Multiple interactions:µMP I
q , PDFMP I , αMP Is …
…correlated with:µH
f with shower starting scaleµH
f , PDFH with MPIµP S
q /µHf and PDFP S/PDFH
µP Sr /µH
r and αP Ss /αH
s for NLO+PSµP S
cut with “string p⊥” & “primordial k⊥”αMP I
s and αP Ss
αMP Is and “string tension”
Tough to describe all phenomena, let alone with satisfactory uncertainty.20 / 33
Tuning and correlationsarXiv:1803.07977 (LH proceedings 2017)
b
bbb b b b b b b b b b b b b b b b b b b b b b b b b b b b
bbbbbb
b
b
bb
b
b b
bb b
bb
Datab
P8, ξ = 2, central
P8, ξ = 2, retune
10−6
10−5
10−4
10−3
10−2
10−1
Durham jet resolution 3 → 2 (ECMS = 91.2 GeV)
1/
σd
σ/dln
(y23)
1 2 3 4 5 6 7 8 9 10
0.6
0.8
1
1.2
1.4
− ln(y23)
MC/Data
b b b
b bbbb b b
bbb
b
b
b
bbbbb b b b b b b b
bbbbbbbbbb
b
b
b
b
b
b
Datab
P8, ξ = 2, central
P8, ξ = 2, retune
10−5
10−4
10−3
10−2
10−1
1
Thrust minor (ECMS = 91.2 GeV)
1/
σd
σ/dln
(Tminor)
-4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5
0.6
0.8
1
1.2
1.4
ln(Tminor)
MC/Data
Naively, would want to tune only soft/NP parameters using “specialized” observables(nch, nπ,K... scaled momentum). But NP models are very sensitive to perturbativeinput state. ⇒ Soft/NP & perturbative parameters very correlated!
Also, “perturbative” observables can have NP regions as well. Should perturbativevariations degrade the accuracy there? ⇒ One tune per variation?
Results of joint tune/scale variation seem reasonable at LEP. Run once per variation :(21 / 33
Fitting MPI at hadron colliders
Scattering composite objects adds more complexity:Underlying event, modelled by multiparton interactionsAs for fragmentation, fit MPI on “specialized” measurements (chargedmultiplicities, divided into MPI-sensitive and MPI-insensitive regions)
Transverse
Away
Toward
Leading object
22 / 33
Fitting MPI → Tuning perturbative physics?Plot from arXiv:1512.00815
Hard QCD has considerable impact on tailored MPI observables!But no generator/merging allows consistent transitionMinimum Bias → Dijets with UE → Multijets with UE. 23 / 33
Tuning Questions or Discussion?
24 / 33
IV) Semi-soft and non-perturbative effects
Non-perturbative models needed to describe the bulk of cross sections atLHC and beyond & offer exciting insights into emergent phenomena.
25 / 33
Color reconnectionIllustrative plot from arXiv:1505.01681
Many overlapping QCD interactions → collective behavior from colorrearrangement:
→ Reconnection necessary to describe data.New model of arXiv:1505.01681 emphasizes SU(3) selection rules to arrange inputcolor states for string hadronization. Feedback welcome!
26 / 33
NP processes: Total, elastic and diffractive cross-sectionsPlots from arXiv:1804.10373
Accurate modelling of complete scatteringcross section crucial to understand Min-Bias & UE, and hence jet profiles + pile-up.
Difficult to describe all LHC data w/ oldsingle Pomeron exchange model.⇒ Total + elastic σ updated to two mod-ern parametrizations w/ ≥ 4 exchanges.
Also need to dissect σ into diffractiveand non-diffractive parts to describescatterings with (partially) intact p-beams.⇒ Unification of soft & hard diffraction.
Ready for serious use + feedback fromexperimental bird’s-eye view welcome!
100 101 102
M 2X (GeV2)
0
1
2
3
4
5
d2σ
SD/dM
2 Xdt
(mb
/GeV
4) SaS
ABMST
ABMST modified
pp data
√s = 23.7 GeV, t = −0.05 GeV2
b
b
b
b
bb b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
Datab
SaS
SaS + CSCR
ABMST
ABMST mod
10−1
1
10 1
10 2
Rapidity gap size in η starting from η = ±4.9, pT > 200 MeV
dσ
/d
∆η
F[m
b]
0 1 2 3 4 5 6 7 80.50.60.70.80.9
11.11.21.31.4
∆ηF
MC
/D
ata
27 / 33
NP processes: Ultra-peripheral collisionsFigures by I. Helenius, arXiv:1510.05900, arXiv:1806.07326
In ultra-peripheral pA or AA collisions, colliding photons can also havenon-perturbative structure & illuminate nuclear PDFs.
1/N
evtdN/d
xA
PbPb,√sNN = 5.02 GeV
anti-kT, R = 0.4pleadT > 20 GeV/cmjets > 35 GeV
PythiaNNPDF2.3EPPS16resolveddirect
Ratio
toNNPDF2.3
xA
PYTHIA implementation ready to use for measurements at pp LHC, e.g.dijets in UPC. Sample usage, see main70.cc. Feedback welcome!
28 / 33
NP processes: Heavy ion cross sectionsarxiv:1806.10820
High-multiplicity (MinBias) pp collisions already suggest extreme QCDbehavior. How does that relate to the state of matter in heavy ioncollisions?
pA and AA collisions includedthrough Angantyr model:◦ Gives complete event-by-eventfinal state of nucleon-nucleonsubcollisions and of total collision◦ Includes event-by-event fluctua-tions of nucleon wavefunctions.
Pythia ready for pA and AAcollisions (see main112.cc andmain113.cc) b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
b b b b b b b b b b b b b b b bb b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b b
ATLASb
Pythia8/Angantyr (generated centrality)
Pythia8/Angantyr (∑ EPb⊥ bins from data)
-2 -1 0 1 20
20
40
60
80
100(a) Centrality-dependent η distribution, pPb,
√SNN = 5 TeV.
η
(1/
Nev)
dN
ch/
dη
29 / 33
Collective effects
Collisions at LHC are packeddensely with color.
CR mimics collective effects,but not dynamics.
Model string interactions→ Microscopic model ofcollectivity!
30 / 33
Collective effectsPlots from arXiv:1710.09725
In 2010, CMS measured long-range azimuthal multiplicity correlations.Repulsive string interactions can reproduce the “ridge” in CMS pp datawith only one parameter on top of PYTHIA.
31 / 33
IV) Plans
Perturbative side:▶ Consistently combine shower variations and merging▶ Implement matching/merging for Vincia and Dire▶ Improve shower evolution beyond leading color/order (Vincia & Dire)▶ Include Vincia & Dire as core Pythia functionality
Non-perturbative side:▶ Unified photo-production and diffractive framework▶ Extend spacetime picture of hadronization▶ Combine “ropes” with heavy-ion modelling
32 / 33
V) Summary
▶ Pythia contains sophisticated matching/merging facilities forfixed-order calculations.No real news – need to understand uncertainties & correlation withshowers and tuning. Need to consolidate UNLOPS functionality inATHENA!
▶ Renewed interested in advanced parton shower models:Vincia, Dire, improvements to default shower. Global assessment ofnew models necessary. Use for jet (sub)structure encouraged.
▶ Significant expansion in non-perturbative physics:Better diffraction, new photo-production processes, new heavy ioninteractions. Need feedback from experiment!
33 / 33