QCD and Monte Carlo Tools
Bryan Webber Cavendish Laboratory University of Cambridge
Bryan Webber 1 CERN-Fermilab HCP School 2017 Monte Carlo Event Generation
• Aim is to produce simulated (particle-level) datasets like those from real collider events ✤ i.e. lists of particle identities, momenta, ... ✤ simulate quantum effects by (pseudo)random numbers • Essential for: ✤ Designing new experiments and data analyses ✤ Correcting for detector and selection effects ✤ Testing the SM and measuring its parameters ✤ Estimating new signals and their backgrounds
Bryan Webber 2 CERN-Fermilab HCP School 2017 Monte Carlo Basics
Bryan Webber 3 CERN-Fermilab HCP School 2017 Monte Carlo Integration • Basis of all Monte Carlo methods: b 1 N I = f(x) dx (b a)f(x ) I ⇡ N i ⌘ N Za i=1 X weight wi
where x i are randomly (uniformly) distributed on [a,b].
Then I =limIN = E[w] , I = Var[w]/N • N !1 p where Var[w]=E[(w E[w])2]=E[w2] (E[w])2 b b 2 2 =(b a) f(x) dx f(x) dx V ⌘ Za Za ⇥ ⇤ ⇥ Central limit⇤ theorem: I = I V/N N ± P (< 1 ) = 68% p Bryan Webber 4 CERN-Fermilab HCP School 2017 Convergence • Monte Carlo integrals governed by Central Limit Theorem: error c.f. trapezium rule only if derivatives exist and are finite, e.g. Simpson’s rule
1 ⇡ I = 1 x2 dx = =0.785 0 4 Z p 2 ⇡2 pV = =0.223 r3 16 0.223 I =0.785 ± pN
Bryan Webber 5 CERN-Fermilab HCP School 2017 Importance Sampling • Convergence is improved by putting more points in regions where integrand is largest • Corresponds to a Jacobian transformation • Variance is reduced (weights “flattened”)
2/3 x3 ⇢ = x 3 0 2 1 4 ⇡ 2 ⇡2 Bryan Webber r 6 CERN-Fermilab HCP School 2017 Hit-and-Miss
• Accept points with probability = wi/wmax (provided all wi > 0) • Accepted points are distributed like real events
• MC efficiency eMC = E[w]/wmax improved by importance sampling
1 2 "MC =1/I =2/⇡ = 64% " = dx(1 x )/I =3/⇡ = 95% MC Z0 " (1 " ) 0.48 " (1 " ) 0.21 = MC MC = = MC MC = r N pN r N pN
Bryan Webber 7 CERN-Fermilab HCP School 2017 Multi-dimensional Integration
• Formalism extends trivially to many dimensions • Particle physics: very many dimensions, e.g. phase space = 3 dimensions per particles, LHC event ~ 250 hadrons. • Monte Carlo error remains • Trapezium rule • Simpson's rule
Bryan Webber 8 CERN-Fermilab HCP School 2017 Monte Carlo: Summary
Disadvantages of Monte Carlo: • Slow convergence in few dimensions. Advantages of Monte Carlo: • Fast convergence in many dimensions. • Arbitrarily complex integration regions (finite discontinuities not a problem). • Few points needed to get first estimate (“feasibility limit”). • Every additional point improves accuracy (“growth rate”). • Easy error estimate. • Hit-and-miss allows unweighted event generation, i.e. points distributed in phase space just like real events.
Bryan Webber 9 CERN-Fermilab HCP School 2017 Phase Space Generation
Phase space:
Two-body easy:
Bryan Webber 10 CERN-Fermilab HCP School 2017 Phase Space Generation
Other cases by recursive subdivision:
Or by ‘democratic’ algorithms: RAMBO, MAMBO Can be better, but matrix elements rarely flat.
Bryan Webber 11 CERN-Fermilab HCP School 2017 Particle Decays ` Simplest example W t ⌫ e.g. top quark decay: b p p p p t · ⇥ b ·
Breit-Wigner peak of W very strong - but can be removed by importance sampling: m2 M 2 m2 arctan W W (prove it!) W ⇥ M W W ⇥ Bryan Webber 12 CERN-Fermilab HCP School 2017 Associated Distributions
e.g. lepton momentum in top decays: Big advantage of Monte Carlo integration: Simply histogram any ` associated quantities. Almost any other technique requires new ⌫ integration for each observable. Can apply arbitrary cuts/ smearing.
Bryan Webber 13 CERN-Fermilab HCP School 2017 Hadron-Hadron Cross Sections
`+ pp p pq¯ = x2pp¯ Z _ p pq = x1pp p p¯
` 2 s = Ecm =4pppp¯ sˆ =4pqpq¯ = x1x2s
0 + Consider e.g. pp¯ Z ` ` • ! ! • Integrations over incoming parton momentum distributions: 1 1 (s)= dx1f(x1) dx2f(x2)ˆ (x1x2s) Z0 Z0 • Hard process cross section ˆ (ˆ s ) has strong peak, due to Z 0 resonance: needs importance sampling (like W in top decay)
Bryan Webber 14 CERN-Fermilab HCP School 2017 Drell-Yan production • Large number of measurements o Total and differential cross-sections in boson * (lepton) kinematics (y, pT, " ,…) o Cross-section ratios (W+/W-, W/Z, …) o Jet distributions • ATLAS/CMS and LHC-b complementary • Low & high x accessed by off-shell data • Large statistics, clean signature, excellent detector calibration à typical experimental systematics +~1% • ppLuminosity ` systematics ` cross (2-3%) andsection also other contributions! cancel in ratios
9 orders in cross-section • Understand proton structure: constrain PDFs • Probe higher-order pQCD
Full spin correlations calculations • Study non-perturbative efffects, soft gluon resummation, parton shower modelling • Test Monte Carlo tools M=psˆ [GeV/c2] 2 CMS-PAS-SMP-16-009 M=psˆ [GeV/c ] • Essential for precision physics at LHC G. Pásztor: Hard QCD @ LHC, LP2017 27 4⇡sˆ ` q ˆqq¯ Z0 `+` = ! ! 3M 2 (ˆs M 2 )2 + 2 M 2 Z Z Z Z + “Background” is qq¯ ⇤ ` ` • ! !
Bryan Webber 15 CERN-Fermilab HCP School 2017 Parton-Level Monte Carlo Calculations
Now we have everything we need to make parton-level cross section calculations and distributions Can be largely automated… • MADGRAPH • GRACE • COMPHEP • AMEGIC++ • ALPGEN But… • Fixed parton/jet multiplicity • No control of large higher-order corrections • Parton level Need hadron level event generators Bryan Webber 16 CERN-Fermilab HCP School 2017 Monte Carlo Event Generation
Bryan Webber 17 CERN-Fermilab HCP School 2017 Monte Carlo Event Generation
• Monte Carlo event generation: ✤ theoretical status and limitations • Recent improvements: ✤ perturbative and non-perturbative • Overview of results: ✤ W, Z, top, Higgs, BSM (+jets)
Bryan Webber 18 CERN-Fermilab HCP School 2017 4 4 Limits
0.0045 Quark-Quark Quark-Gluon 0.004 Gluon-Gluon
0.0035 CMS Simulation Preliminary
0.003 s = 8 TeV, WideJets |η| < 2.5 & |∆η | < 1.3 jj 0.0025
Normalized yield/GeV 0.002
0.0015
0.001
0.0005
0 1000 2000 3000 4000 5000 6000 Dijet Mass (GeV)
Figure 2: The reconstructed resonance mass spectrum generated with the PYTHIA MC simula- tion and Tune D6T for qq G qq, qg q qg, gg G gg for resonance masses of ⇥ ⇥ ⇥ ⇥ ⇥ ⇥ 1.0, 2.0, 3.0, 4.0, and 5.0 TeV. A high-mass dijet event
Figure 3: The event with the highest invariant mass: 3D view (left) and 2D view (right). The invariant massM of the= two 5.15 wide TeV jets is 5.15 TeV. • jj CMS PAS EXO-12-059
Bryan Webber 19 CERN-Fermilab HCP School 2017 LHC dijet
p
p Hard process
Bryan Webber 20 CERN-Fermilab HCP School 2017 LHC dijet
p
p Hard process Parton showers
Bryan Webber 21 CERN-Fermilab HCP School 2017 LHC dijet
p
p Hard process Parton showers Underlying event
Bryan Webber 22 CERN-Fermilab HCP School 2017 LHC dijet
p
p Hard process Parton showers Underlying event Confinement
Bryan Webber 23 CERN-Fermilab HCP School 2017 quark jet LHC dijet )
p beam ) jet beam p jet ) Hard process Parton showers Underlying event
Confinement ) Hadronization quark jet Bryan Webber 24 CERN-Fermilab HCP School 2017 Theoretical Status Exact fixed-order perturbation theory
p
p Hard process
Bryan Webber 25 CERN-Fermilab HCP School 2017 Theoretical Status Exact fixed-order perturbation theory Approximate all-order perturbation theory
p
p Hard process Parton showers
Bryan Webber 26 CERN-Fermilab HCP School 2017 quark jet Theoretical Status) Exact fixed-order perturbation theory Approximate all-order perturbation theory Semi-empirical local models only p beam ) jet beam p jet ) Hard process Parton showers Underlying event
Confinement ) Hadronization quark jet Bryan Webber 27 CERN-Fermilab HCP School 2017 QCD Factorization
1 2 2 2 2 2 pp X (Epp)= dx1 dx2 fi(x1,µ ) fj(x2,µ )ˆ ij X (x1x2Epp,µ ) ! ! Z0 ( ( momentum parton hard process fractions distributions cross section at scale µ2
• Jet formation and underlying event take place over a much longer time scale, with unit probability • Hence they cannot affect the cross section • Scale dependences of parton distributions and hard process cross section are perturbatively calculable, and cancel order by order
Bryan Webber 28 CERN-Fermilab HCP School 2017 Parton Shower Approximation
• Keep only most singular parts of QCD matrix elements: ↵S d⇠i d i Collinear d n+1 Pii(zi, i) dzi d n ⇠i =1 cos ✓i • ⇡ 2⇡ ⇠i 2⇡ i ↵ X p p d Soft d S ( T T ) i · j !d!d⇠ i d • n+1 ⇡ 2⇡ i · j p kp k i 2⇡ n i,j i j X · · ↵ ⇠ d! d = S ( T T ) ij d⇠ i d 2⇡ i · j ⇠ ⇠ ! i 2⇡ n i,j i j X ↵ d! d⇠ S ( T T )⇥(⇠ ⇠ ) i d ⇡ 2⇡ i · j ij i ! ⇠ n i,j i X j
✓ij >✓i ! =(1 z )E i i i ✓i ziEi Angular-ordered parton shower (or dipoles)
Bryan Webber 29 CERN-Fermilab HCP School 2017 Parton Shower Evolution
✓ ✓ Ei+1 hard 0 1 ✓3 z = ✓ ✓4 E process E0 E1 2E2 E3 E4 i ✓ 0 2 H 4 1+z A Pq q(z)= D ! 3 1 z R (E ,✓ ) O 1 1 N I Z (E2,✓2) A T (E ,✓ ) I 3 3 (E4,✓4) O N
E✓< Qmin
E0 Bryan Webber 30 CERN-Fermilab HCP School 2017 Parton Shower Evolution
✓ ✓ Ei+1 hard 0 1 ✓3 z = ✓ ✓4 E process E0 E1 2E2 E3 E4 i ✓ 0 2 H 4 1+z A Pq q(z)= D ! 3 1 z R (E E ,✓ ) (E1,✓1) O 0 1 1 N Pg g(z)= ! I 1+z4 +(1 z)4 ✓ Z (E2,✓2) 3 A z(1 z) T (E ,✓ ) I 3 3 (E4,✓4) 1 2 2 O Pg q = z +(1 z) N ! 2 ⇥ ⇤ E✓< Qmin
E E0 Bryan Webber 31 CERN-Fermilab HCP School 2017 Sudakov Factor
• Di(Q,Qmin) = probability for parton i to evolve from Q to Qmin without any resolvable emissions: ↵ E0 d! ✓0 d✓ (Q ,Q ) 1 C S 2 ⇥(!✓ Q )+... i 0 min ⇡ i ⇡ ! ✓ min Z Z ↵ Q 1 C S ln2 0 + ... ⇡ i ⇡ Q ✓ min ◆ ↵ Q exp C S ln2 0 ⇡ i ⇡ Q ✓ min ◆ • Cq = CF = 4/3, Cg = CA =3 • Then probability to evolve from Q1 to Q2 without resolvable emissions is Di(Q1,Qmin)/Di(Q2,Qmin)
• Given Q1, find Q2 by solving • Di(Q1,Qmin)/Di(Q2,Qmin) = Random #
Bryan Webber 32 CERN-Fermilab HCP School 2017 Hadronization Models
• In parton shower, relative transverse momenta evolve from a high scale Q towards lower values
• At a scale near LQCD~200 MeV, perturbation theory breaks Preconfinementdown and hadrons are formed
Planar approximation:• Before gluon that, at = scalescolour—anticolour ~ few x LQCD, there pair .is universal preconfinement of colour Follow colour structure of parton shower: colour-singlet pairs • Colour, flavour and momentum flows are only locally end up close inredistributed phase space by hadronization
Mass spectrumBryan Webber of colour-singlet pairs asymptotically33 CERN-Fermilab HCP School 2017 independent of energy, production mechanism, … Peaked at low mass
LHC Simulations 2 Bryan Webber Hadronization Models
• In parton shower, relative transverse momenta evolve from a high scale Q towards lower values
• At a scale near LQCD~200 MeV, perturbation theory breaks Preconfinementdown and hadrons are formed
Planar approximation:• Before gluon that, at = scalescolour—anticolour ~ few x LQCD, there pair. is universal preconfinement of colour Follow colour structure• Colour, of flavour parton and shower: momentum colour-singlet flows are only pairs locally end up close inredistributed phase space by hadronization
Mass spectrumBryan Webber of colour-singlet pairs asymptotically34 CERN-Fermilab HCP School 2017 independent of energy, production mechanism, … Peaked at low mass
LHC Simulations 2 Bryan Webber String Hadronization Model
• In parton shower, relative transverse momenta evolve from a high scale Q towards lower values
• At a scale near LQCD~200 MeV, perturbation theory breaks Preconfinementdown and hadrons are formed
Planar approximation:• Before gluon that, at = scalescolour—anticolour ~ few x LQCD, there pair. is universal preconfinement of colour Follow colour structure• Colour of flow parton dictates shower: how to colour-singlet connect hadronic pairs string end up close in(width phase ~ spacefew x LQCD) with shower
Mass spectrumBryan Webber of colour-singlet pairs asymptotically35 CERN-Fermilab HCP School 2017 independent of energy, production mechanism, … Peaked at low mass
LHC Simulations 2 Bryan Webber String Hadronization Model
• In parton shower, relative transverse momenta evolve from a high scale Q towards lower values
• At a scale near LQCD~200 MeV, perturbation theory breaks Preconfinementdown and hadrons are formed
Planar approximation:• Before gluon that, at = scalescolour—anticolour ~ few x LQCD, there pair. is universal preconfinement of colour Follow colour structure• Colour of flow parton dictates shower: how to colour-singlet connect hadronic pairs string end up close in(width phase ~ spacefew x LQCD) with shower
Mass spectrumBryan Webber of colour-singlet pairs asymptotically36 CERN-Fermilab HCP School 2017 independent of energy, production mechanism, … Peaked at low mass
LHC Simulations 2 Bryan Webber String Hadronization Model
• At short distances (large Q), QCD is like QED: colour field lines spread out (1/r potential) • At long distances, gluon self-attraction gives rise to colour string (linear potential, quark confinement) • Intense colour field induces quark-antiquark pair creation: hadronization
++ –
Bryan Webber 37 CERN-Fermilab HCP School 2017 Cluster Hadronization Model
• In parton shower, relative transverse momenta evolve from a high scale Q towards lower values
• At a scale near LQCD~200 MeV, perturbation theory Preconfinementbreaks down and hadrons are formed
Planar approximation:• Before gluon that, at = scalescolour—anticolour ~ few x LQCD, there pair. is universal preconfinement of colour Follow colour structure• Decay of preconfinedparton shower: clusters colour-singlet provides a direct pairs basis end up close infor phase hadronization space
Mass spectrumBryan Webber of colour-singlet pairs asymptotically38 CERN-Fermilab HCP School 2017 independent of energy, production mechanism, … Peaked at low mass
LHC Simulations 2 Bryan Webber Cluster Hadronization Model
• In parton shower, relative transverse momenta evolve from a high scale Q towards lower values
• At a scale near LQCD~200 MeV, perturbation theory Preconfinementbreaks down and hadrons are formed
Planar approximation:• Before gluon that, at = scalescolour—anticolour ~ few x LQCD, there pair. is universal preconfinement of colour Follow colour structure• Decay of preconfinedparton shower: clusters colour-singlet provides a direct pairs basis end up close infor phase hadronization space
Mass spectrumBryan Webber of colour-singlet pairs asymptotically39 CERN-Fermilab HCP School 2017 independent of energy, production mechanism, … Peaked at low mass
LHC Simulations 2 Bryan Webber Cluster Hadronization Model
• Mass distribution of preconfined clusters is universal • Phase-space decay model for most clusters • High-mass tail decays anisotropically (string-like)
Bryan Webber 40 CERN-Fermilab HCP School 2017 Hadronization Status
• No fundamental progress since 1980s ✤ Available non-perturbative methods (lattice, AdS/ QCD, ...) are not applicable • Less important in some respects in LHC era ✤ Jets, leptons and photons are observed objects, not hadrons • But still important for detector effects ✤ Jet response, heavy-flavour tagging, lepton and photon isolation, ...
Bryan Webber 41 CERN-Fermilab HCP School 2017 Multiparton Interaction Model (PYTHIA/JIMMY)
For small pt min and high energy inclusive parton—parton cross section is larger than total proton—proton cross section. Underlying! More than one parton—parton Event scatter per proton—proton(MPI)
NeedMultiple a model parton of spatial interactions distribution within in same proton collision • ! Perturbation theory gives n-scatter distributions LHC✤ SimulationsDepends 3 on density profile of proton Bryan Webber • Assume QCD 2-to-2 secondary collisions
✤ Need cutoff at low pT • Need to model colour flow ✤ Colour reconnections are necessary
Bryan Webber 42 CERN-Fermilab HCP School 2017 Underlying Event
〉 1.6 1.6
φ δ
Trans-min region ATLAS Trans-min region ATLAS η
1.4 [GeV] 1.4 δ -1 〉 -1
/ p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb T φ T ch δ 1.2 plead > 1 GeV 1.2 plead > 1 GeV η N
〈 T T δ / 1 T 1 p Σ
leading charged particle 0.8 〈 0.8
0.6 0.6
0.4 Data PYTHIA 8 Monash 0.4 Data PYTHIA 8 Monash PYTHIA 8 A14 Herwig7 PYTHIA 8 A14 Herwig7 0.2 PYTHIA 8 A2 Epos 0.2 PYTHIA 8 A2 Epos
towards 1 1 < 60 | | Model / Data 0.8 Model / Data 0.8
transverse (min) transverse (max) 5 10 15 20 25 30 5 10 15 20 25 30 plead [GeV] plead [GeV] 60 < < 120 60 < < 120 T T
| | 〉 1.6 | | 1.6
〉 2.4 3
φ φ δ
Trans-min region ATLAS δ Trans-min region ATLAS
η 2.2 Trans-max region ATLAS Trans-max region ATLAS 1.4 [GeV] 1.4 η δ 〉
-1 -1 [GeV]
δ away -1 〉 -1 / p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb
φ 2 2.5 T / pT > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb φ ch δ T T
lead ch lead δ
1.2 p> > 1 GeV 1.2 plead > 1 GeV lead 120 η
N
〈 T 1.8 pT > 1 GeV p > 1 GeV η N δ
〈 T T
| | δ / T 1.6 / 2 1 1 T p p Σ
1.4 Σ 〈
0.8 0.81.2 〈 1.5 0.6 0.61 0.8 1 0.4 0.4 Data PYTHIA 8 Monash 0.6 Data PYTHIA 8 Monash Data PYTHIA 8 Monash PYTHIA 8 A14 Herwig7 PYTHIA 8 A14 Herwig7 0.4 PYTHIA 8 A14 Herwig7 0.5 PYTHIA 8 A14 Herwig7 0.2 PYTHIA 8 A2 Epos 0.2 PYTHIA 8 A2 Epos 0.2 PYTHIA 8 A2 Epos PYTHIA 8 A2 Epos Figure 1: Definition of regions in the azimuthal angle with respect to the leading (highest-pT) charged particle, with arrows representing particles associated with the hard scattering process and the leading charged particle 1 1 highlighted in red. Conceptually, the presence of a hard-scatter particle on the right-hand side of the1 transverse 1 region, increasing its p , typically leads to thatModel / Data 0.8 side being identified as the “trans-max” and henceModel / Data the0.8 left-hand T Model / Data 0.8 Model / Data 0.8 side as the “trans-min”, with maximumATLAS sensitivity, JHEP to the 03(2017)157 UE. 5 10 15 20 25 30 5 10 15 20 25 30 P 5 10 15 20 25 30 5 10 15 20 25 30 lead lead p [GeV] plead [GeV] lead T pT [GeV] p [GeV] T T
〉 2.4 3
〉 1.6 φ Bryan Webber 43 CERN-Fermilab HCP School 2017 δ φ
2.2 ↵ Trans-max region ATLAS δ 0.8 Trans-max region ATLAS illustrated in Figure 1, the azimuthal angularη di erence with respect to the leading (highest-pT) chargedTrans-diff region ATLAS Trans-diff region ATLAS [GeV] δ η -1 〉 -1 [GeV]
1.4 δ 2 2.5 〉 / p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb-1 -1 φ T T / p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb p > 0.5 GeV, |η| < 2.5 s = 13 TeV, 1.6 nb ch