Tutorials on PYTHIA and MadGraphsusy.phsx.ku.edu/~kckong/KWS2014/KWS2014.pdf · PYTHIA and HERWIG and produce semi-realistic reconstructed physics objects such as photons, electrons,

Tutorials on PYTHIA and MadGraph

K.C. Kong University of Kansas

Open KIAS 2014 Winter School on Collider Physics

January 19 - January 25, 2014

Fabio MaltoniFabio Maltoni TASI 2013, Boulder CO Fabio Maltoni

StatementsStatements TRUE FALSE IT DEPENDSI have

no clue

0 MC’s are black boxes, I don’t need to know the details as long as there are no bugs.

1 A MC generator produces “unweighted” events, i.e., events distributed as in Nature.

2 MC’s are based on a classical approximation (Markov Chain), QM effects are not included.

3The “Sudakov form factor” directly quantifies how likely it is for a parton to undergo branching.

4A calculation/code at NLO for a process provides NLO predictions for any IR safe observable.

5 Tree-level based MC’s are less accurate than those at NLO.

2

Test: How much do I know about MC’s?

Monday 10 June 2013


StatementsStatements TRUE FALSE IT DEPENDSI have

no clue

0 MC’s are black boxes, I don’t need to know the details as long as there are no bugs. ✓

1 A MC generator produces “unweighted” events, i.e., events distributed as in Nature. ✓

2 MC’s are based on a classical approximation (Markov Chain), QM effects are not included. ✓

3The “Sudakov form factor” directly quantifies how likely it is for a parton to undergo branching.

✓

4A calculation/code at NLO for a process provides NLO predictions for any IR safe observable.

✓

5 Tree-level based MC’s are less accurate than those at NLO. ✓

3


Monday 10 June 2013


Score Result Comment

≥5 Addict Always keep in mind that there are also other interesting activities in the field.

4 Excellent No problem in following these lectures.

3 Fair Check out carefully the missed topics.

≤2Room for

improvement Enroll in a MC crash course at your home

institution.

6 x no clue No clue

4


Monday 10 June 2013


Discoveries at hadron colliders

hard

shapepp→gg,gq,qq→jets+ET~~~~~~

Background shapes needed. Flexible MC for both signal and b a c k g r o u n d t u n e d a n d validated with data.

/

MichelangeloMangano®

5

“easy”

peakpp→H→4l

Background directly measured from data. TH needed only for p a r a m e t e r e x t r a c t i o n (Normalization, acceptance,...)

very hard

discriminantpp→H→W+W-

Background normalization and shapes known ver y wel l . I n t e r p l ay w i t h t he be s t theoretical predictions (via MC) and data.

Monday 10 June 2013


• Accurate and experimental friendly predictions for collider physics range from being very useful to strictly necessary.

• Confidence on possible excesses, evidences and eventually discoveries builds upon an intense (and often non-linear) process of description/prediction of data via MC’s.

• Both measurements and exclusions rely on accurate predictions.

9

Challenges for LHC physicists

Monday 10 June 2013


New generation (LHC) of MC tools

11

Experiment

Theory

LagrangianGauge invarianceQCDPartonsNLOResummation...

Detector simulationPions, Kaons, ...Reconstruction

B-tagging efficiencyBoosted decision tree

Neural network...

MC event generators

Monday 10 June 2013

Simulation of Collider events

MG/FR School, Beijing, May 22-26, 2013 Event Generation at Hadron Colliders Johan Alwall

Sherpa artist

39


1. High-Q Scattering2 2. Parton Shower

3. Hadronization 4. Underlying Event

☞ where new physics lies

☞ process dependent

☞ first principles description

☞ it can be systematically improved

40




☞ QCD -”known physics”☞ universal/ process independent☞ first principles description

41




☞ universal/ process independent

☞ model-based description

☞ low Q physics2

42




☞ energy and process dependent

☞ model-based description

☞ low Q physics2

43

MG/FR School, Beijing, May 22-26, 2013 Event Generation at Hadron Colliders Johan Alwall 44


Master formula

Zdx1dx2d�FS

Phase spaceintegral

fa(x1)fb(x2)

Parton densityfunctions

• Parton density (or distribution) functions:Process independent, determined by particle type

s = x1x2s• (s = collision energy of the collider)

• Difference between colliders given by parton luminocities

�ab!X(s, . . .)

Parton levelcross section

• Parton level cross section from matrix element

15


Monte Carlo Integration and Generation

σ =1

2s

!|M|2dΦ(n)

Calculations of cross section or decay widths involve integrations over high-dimension phase space of very peaked functions:

General and flexible method is needed

Dim[Φ(n)] ∼ 3n

20


1. pick x

3. pick 0<y<fmax f(x)

2. calculate f(x)

4. Compare:if f(x)>y accept event,

else reject it.

I= total tries

accepted= efficiency

Monte Carlo Event Generation

33


Improved by combining with importance sampling:

1. pick x distributed as p(x)

2. calculate f(x) and p(x)

3. pick 0<y<1

f(x)

4. Compare:if f(x)>y p(x) accept event,

else reject it.

much better efficiency!!!

Event generation

36


At the most basic level a Monte Carlo event generator is a program which produces particle physics events with the same probability as they occur in nature (virtual collider).

In practice it performs (a possibly large) number of (sometimes very difficult) integrals and then unweights to give the four momenta of the particles that interact with the detector (simulation).

Note that, at least among theorists, the definition of a “Monte Carlo program” also includes codes which don’t provide a fully exclusive information on the final state but only cross sections or distributions at the parton level, even when no unweighting can be performed (typically at NLO).

I will refer to these kind of codes as “MC integrators”.

41

MC Event generator: definition

Monday 10 June 2013

Event Generators

Event generators are softwares (tools) that generate simulated high-energy particle physics events

Event: a set of particle momenta

Many ToolsHadronic event generators

Pythia, Herwig, Sherpa, Isajet

simulate initial state composition and substructure, initial/final state shower, hadronization and further decay, as well as hard processes

Specialized event generators

MC@NLO, MCFM, Jimmy, Ariadne, AcerMC, Alpgen, TAUOLA

Multi-purpose parton level event generators

MG5, CalcHEP (Sherpa, Herwig, …)

Tools for LHC PhysicsA Repository For Beyond-the-Standard-Model Tools

http://www.ippp.dur.ac.uk/montecarlo/BSM

http://www.ippp.dur.ac.uk/montecarlo

MC4BSM workshop: May 18-23, 2014, Korea

Tools conference: 2008, 2010, 2012, 2014(?)

http://www.ippp.dur.ac.uk/montecarlo/BSM

Tools are useful but are only tools

Physics is more important.

Need to choose a RIGHT tool.

Should NOT trust tools 100%

made by humans

used by humans

Follow the instructions and need to know limitations

CalcHEPA package for evaluation of Feynman diagrams, integration over multi-particle phase space, and event generation

http://theory.sinp.msu.ru/~pukhov/calchep.html

Easy user interface and symbolic calculation is possible.

Linked to micrOmegas

LimitationsTree-level processes

Squared matrix element calculation

no spin information for outgoing particles

spin/polarization average amplitude

Limit on the number of external legs and the number of diagrams

Hard to evolve as an NLO calculator

Collider Physics using MG5

� �

��#�� $��

%��

Lagrangian

Model files

Parton Level Events

ISR / Parton Showering

Jet Reconstruction

Pretty Plots

Profit

FeynRules

MadGraph/MadEvent

Pythia

PGS / Delphes

� �

��#�� $��

%��

#��%��(��

�*'�)�#��*��

��

&��'��

��$��)��(��

#��

� �

��#�� $��

%��

#��%��(��

��

&��'��

��$��)��(��

� �

��#�� $��

%��

#��%��(��

�*'�)�#��*��

��

&��'��

��$��)��(��

#��

MadAnalysis / ExRootAnalysis

Johan Alwall - Simulation at the LHC 10

Simulation tools

Matrix element generators – for hard process

2

Diagrams for by MadGraph

MG5 2011 MG5

...

A bit of history

3

MG4

1994 Core MG4MadEvent

2002 MadEventMadOnia

2008 MadOniaMadWeight

2008 MadWeight

MadFKS

2009 MadFKS

MadDipolE

2007 MadDipole

aMC@NLO

MadLoop

2011 MadLoop

One code, to rule them all!

2013 MG5_aMC MadDM

MadGraph 5 Specs

4

• High-level language: Python

• Flexible and Modular => Developer friendly All-in-one distribution

• User-interface and automatic doc. => User friendly

• Involved algorithms => Performance increase

• Built-in testing suite => Reliability

• Complex data-structures allow for very general objectswhile keeping speed where needed.

Supported Models

6

Effective Theories N-Legs vertices, !N

color structures Sextets, !ijk, virtually all

Lorentz Structures All, thanks to Aloha

Spins supported 1, 1/2, (3/2), 2

Gauges Unitary, Feynman

Complex Mass Scheme Automatic Model ConversionAvailable for NLO too!

Model with loop info Import UFO Loop-models

Decay widths computation On-the-fly widths computation

Color CodeNew and in the public release! Planned / Ongoing progress

Done and will be made public for MG5 v2.0

Diagram Generationspeed benchmark

11

Very fast decay chains opening the way for new types of processes!

MadEvent5 now able to handle such large decay chains.

Event Generationspeed benchmark

20

No problem running pp>tt~jj on a laptop!

UIUC

FR/MG School on LHC Phenomenology, Sept 30-Oct 05 2012 MadGraph 5 Olivier Mattelaer Duke workshop, 2013-02-18 MadGraph Tutorial Olivier Mattelaer

BSM

Model Information

FeynRules Output

C. Degrande, C. Duhr, B. Fucks, D. Grellscheid, OM, T.Reiter

UFO

Basicaly No limitation

ALOHAAutomatic Creation of HELAS routines for ANY BSM theory

MadGraph building blockP. Aquino, W. Link, F. Maltoni, OM, T. Stelzer


Lagrangian

FeynArts

UFO

TeX Feynman Rules

Model-fileParticles, parameters, ...

FeynRules

MadGraph CalcHep Sherpa

Whizard GoSam Herwig

52

FeynRules

Monday 10 June 2013

MadGraph/MadEvent Structure

MadGraph5 and Going BeyondMadGraph (matrix element generator) is completely rewritten in python

New output includes C++ library for PYTHIA

Very fast / evolving to

automatic NLO QCD calculator (MadLoop)

automatic NLO corrections for BSM (MadGolem)

relic abundance calculator (MadDM)

OutlineIntroduction

PYTHIA

ttbar, slepton production, Wjj, LM6

exercise with ttbar events

PGS

MadGraph/MadEvent

ttbar

Project: a single Tprime production and decay to Higgs plus top quark

PGSPGS is a simulation of a generic high-energy physics collider detector with a tracking system, electromagnetic and hadronic calorimetry, and muon system. It is designed to take events generated with popular event generators like PYTHIA and HERWIG and produce semi-realistic reconstructed physics objects such as photons, electrons, muons, hadronically decaying taus, and hadronic jets (including b- and charm-tagging).

Ideally a high energy physics detector would tell us the four momenta of all outgoing particles in a hard collision:

Detector Effects and Simulation

τ+

τ-

dd_

what we want what we get

CMS

GEANT4

• the gold standard in high energy physics detector simulation software

• treats detector as “slabs” of particular material

• simulates in detail energy deposition from ionization, showering

• simulates all secondary interactions

• problem: takes (many) minutes of CPU per event!

PGS Philosophy

• interface to standard physics process generators (PYTHIA, HERWIG, ISAJET, ALPGEN, ...)

• perform very basic detector simulation with

‣ tracks

‣ calorimeter deposits

‣ muon ID

• reconstruct physics “objects”: γ, e, μ, τ, jet (b), MET from tracks/calorimeter

• parametrize where needed

Detector Simulation Goals

• detector acceptance

• detector efficiency

• detector resolution

• secondary interactions

- nuclear interactions

- brehmsstrahlung

- pair production

- multiple scattering

•multiple interactions (pileup)

•event reconstruction effects

PGS?

✔

✔

✔

✘

✘

✘

✔

✘

✔

HistoryMarch 1998: kickoff of Tevatron Run II SUSY-Higgs workshop

No Run II CDF/D0 simulations available then

Developed “SHW” simulation as average of CDF/D0

Published SHW Higgs report: hep-ph/0010338

Still a reliable resource for Tevatron Higgs physics

SHW -> PGS for Snowmass 2001

Used for VLHC, LHC, ILC, Tevatron comparison, especially by theorists

event generation

STDHEP common blocks

event simulation, object reconstruction

user analysis

user output

Flow of PGS

PGS Detector Simulation

• loop through all final-state HEPEVT particles

• if charged, make charged track (straight...)

• calorimeter deposits:

• gamma/electron: mostly electromagnetic

• hadron: mostly hadronic

• muon: minimum ionizing

•calorimeter is idealized, segmented in eta/phi

•resolutions are controllable parameters

PGS Resolutions

• tracking (B field, radius, sagitta)

✓ calculate sagitta, smear it, get pT

✓ includes possibility of charge confusion

• em calorimetry

ΔE/E = a + b/√E

• hadron calorimetry

ΔE/E = b/√E

PGS Jet Finding

• “top-down” (cone): find highest ET tower, then add to it nearby towers above some threshold, lying within a pre-set cone size (ΔR0); repeat until remaining highest ET tower is below some threshold

• “bottom-up” (kt jet): treat all towers (em+had) as “particles”; find all particle-particle distances min(kTi

2,kTj2)ΔRij

2/ΔR02 and

particle-”beam” distances kTi2 and if the

overall minimum is an ij, merge them; repeat until no merge-able pairs remain

PGS Electrons/Photons

• in real life electromagnetic showers are narrow; hadronic showers are wide

• in PGS, alas, there is no lateral spread

• we simply rely on the fact that the energy is deposited in the em section of the calorimeter

• start with clusters (kt jet alg.) and apply em fraction cuts, match with track

• apply calorimeter isolation cut (3x3 region)

PGS Electrons/Photons

• look at em fraction of cluster (single tower most likely)

• see if there is a track; no track ⇒ photon

• require sum of pT of other tracks in ΔR cone of 0.4 be less than 5 GeV

• require sum of energy in 3x3 collar region < 0.1 E

track

PGS muon efficiency

• efficiency about 97% out to |η| = 3 (depends totally on track efficiency)

muons, E > 20 GeV

TrackingOuter radius of tracker: 1 m

Magnetic field: 4 T

Sagitta resolution: 0.000005 m

Track finding efficiency: 0.98

Minimum track pt: 0.8 GeV

Tracking eta coverage: 2.4 (~10 degree)

MadGraph/MadEvent Structure

PGS Parameters

LHC ! parameter set name320 ! eta cells in calorimeter 200 ! phi cells in calorimeter0.0314159 ! eta width of calorimeter cells |eta| < 50.0314159 ! phi width of calorimeter cells0.0044 ! electromagnetic calorimeter resolution const0.024 ! electromagnetic calorimeter resolution * sqrt(E)0.8 ! hadronic calolrimeter resolution * sqrt(E)0.2 ! MET resolution0.01 ! calorimeter cell edge crack fractioncone ! jet finding algorithm (cone or ktjet)5.0 ! calorimeter trigger cluster finding seed threshold (GeV)1.0 ! calorimeter trigger cluster finding shoulder threshold (GeV)0.5 ! calorimeter kt cluster finder cone size (delta R)2.0 ! outer radius of tracker (m)4.0 ! magnetic field (T)0.000013 ! sagitta resolution (m)0.98 ! track finding efficiency1.00 ! minimum track pt (GeV/c)3.0 ! tracking eta coverage3.0 ! e/gamma eta coverage2.4 ! muon eta coverage2.0 ! tau eta coverage

User is free to change these...at his or

her own risk!

1

• calorimeters

PGS simulates calorimeter covering |η| < 4.1 with cell size ∆η ×∆φ = 0.1× 0.087266462 and 82 η cells and 72φ cells, and take the hadronic calorimeter energy resolution

σ =

!

1.2"

E/GeV

#

E (1)

and the electromagnetic calorimeter energy resolution

$ σ

E

%2

=

&

S√E

'2

+

&

N

E

'2

+ C2 (2)

where S = 0.0363 is the stochastic term, N = 0.124 the noise and C = 0.26 the constant term [1]. PGS ignorescalorimeter cell edge crack. Calorimeter cluster finding seed threshold is 3 GeV and calorimeter cluster findingshoulder threshold is 0.5 GeV.

• tracking

1.0 = outer radius of tracker (m)

4.0 = magnetic field (T)

0.000005 = sagitta resolution (m) ?

0.98 = track finding efficiency

0.8 = minimum track pt (GeV/c)

2.4 = tracking eta coverage

——————————————————

Take the input track 3-momentum, and return the smeared momentum, possibly with opposite charge. Theroutine does this by calculating the track sagitta from the track pt, and then smearing the sagitta by a fixedgaussian. The routine then converts back to pt, preserving the original direction in space. The new z momentumis taken from the original angle and the new track pt.

• muon

muon leaves little energy in the calorimeters, has a track, and travel all the way to the muon-detection systemoutside the calorimeters.

Muons with ET > 5 GeV and |ηµ| < 2.4 are reconstructed.

The global µ reconstruction efficiency is close to 1 up to |ηµ| < 2 and roughly 0.96 for 2 < |ηµ| < 2.4 [1]. PGStakes constant µ reconstruction efficiency, 0.985 for |ηµ| < 2.4

Muons are classified as isolated if the transverse energy, EisoT , from calorimeter energy sum within ∆R < 0.3

around a given tower excluding the seed is less 5 GeV, and EtrkT of additional tracks nearby in ∆R < 0.3

excluding any muon tracks is less than 5 GeV (check numbers). PGS uses minimum track pT = 0.8 GeV (samefor all tracks)

track finding efficiency = 0.98 (same for all tracks)

charge is determined from track....

charge misidentification?

• jets

Jets are defined as hadronic cluster with ET > 15 GeV within a cone with ∆R ="

∆η2 +∆φ2 = 0.5. PGSrequires |ηj | < 4.

jet misidentification?

• electron

ET > 10 GeV, |η| < 2.4, Ehcal

Eecal< 0.125, Eiso

T

ET< 0.1 within ∆R < 0.3 pisoT (∆R < 0.3)− pisoTmax(∆R < 0.15) < 5

GeV, 0.5 < Eecal

Etrk< 1.5

EisoT is the transverse energy from calorimeter energy sum within ∆R < 0.3 around a given tower excluding the

seed. ET is the transverse energy in calorimeter for a given tower. pisoT is the sum of all pT s around ∆R < 0.3for given η and φ. pisoTmax is the largest pT around ∆R < 0.15 for given η and φ.

2

FIG. 1. EisoT is the transverse energy Eiso

T from calorimeter energy sum within within ∆R < 0.3 around a given tower excludingthe seed and Etrk

T (Etrk) is the transverse energy (energy) of additional tracks nearby in ∆R < 0.3 excluding any muon tracks.Eiso

T , EtrkT and Etrk are calculated for the processes (a) pp → tt → W+W−bb → jjjjbb (b) pp → W+W−

→ µ+µ−νµ (c)pp → µµ∗

→ µ+µ−χ01χ

01

• photon

ET > 10 GeV, |η| < 2.4, Ehcal

Eecal< 0.125, Eiso

T

ET< 0.1 within ∆R < 0.3 number of tracks, niso < 1 within

∆R < 0.15, pisoT < 5 GeV within ∆R < 0.3, largest pT track, pisoTmax < 1 GeV within ∆R < 0.3,

• /ET

Missing ET is defined by summing (as a vector) the directed transverse energy deposited in all of the calorimetercells (treating each cell as a massless particle). This combines, ideally, the momenta of all photons, electrons,hadronically decaying taus, and jets. Adding to this the transverse momenta of any muons, whose energy ismeasured using the muon detection system. The magnitude of the resultant vector is the missing transverseenergy. (muon detection system works only out to |η| < 2.4, whereas the calorimeter extends to |η| < 4.1, somuons at large pseudorapidity can cause additional missed transverse momentum.) Muon leaves little energyin the calorimeters.

For /ET trigger, resolution is

σ

/ET= 0.2 (3)

• b-tagging

From a combined secondary vertex based B-tagging algorithm in CMS [1, 2], PGS parameterizes the probabilitieswith constants. Maximum η for tagging is 2.4.

b-tagging efficiency = 0.5 Non b-jet mistagging probability for c-jets = 0.06

Non b-jet mistagging probability for gluon-jets = 0.025

Non b-jet mistagging probability for uds-jets = 0.01

maximum eta for tagging = 2.4

60 cm z vertex fiducial cut ?

• taus

taus fake rate ?

[1] CMS Physics TDR, volume 1, CERN-LHCC-2006-001[2] C. Weiser, “A combined secondary vertex based B-tagging algorithm in CMS,” CERN-CMS-NOTE-2006-014

1

• calorimeters


σ =

!

1.2"

E/GeV

#

E (1)


$ σ

E

%2

=

&

S√E

'2

+

&

N

E

'2

+ C2 (2)


• tracking







——————————————————


• muon










• jets




• electron

ET > 10 GeV, |η| < 2.4, Ehcal

Eecal< 0.125, Eiso

T


GeV, 0.5 < Eecal

Etrk< 1.5



1

• calorimeters


σ =

!

1.2"

E/GeV

#

E (1)


$ σ

E

%2

=

&

S√E

'2

+

&

N

E

'2

+ C2 (2)


• tracking







——————————————————


• muon










• jets




• electron

ET > 10 GeV, |η| < 2.4, Ehcal

Eecal< 0.125, Eiso

T


GeV, 0.5 < Eecal

Etrk< 1.5



Example Olympics Output # typ eta phi pt jmas ntrk btag had/em dum1 dum2

0 1 3585

1 4 -1.312 3.143 104.54 21.59 19.0 0.0 1.22 0.0 0.0

2 4 -1.233 0.957 85.10 15.90 11.0 0.0 5.78 0.0 0.0

3 4 -2.939 1.139 38.38 26.74 20.0 0.0 63.11 0.0 0.0

4 4 3.226 5.123 37.37 34.33 8.0 0.0 1.10 0.0 0.0

5 4 -3.718 4.691 21.52 1.55 17.0 0.0 1.35 0.0 0.0

6 4 0.211 5.752 12.75 15.57 0.0 0.0 1.03 0.0 0.0

7 4 1.008 3.038 12.60 4.18 3.0 0.0 1.73 0.0 0.0

8 4 -2.106 4.275 7.93 2.75 19.0 0.0 3.32 0.0 0.0

9 6 0.000 6.008 15.64 0.00 0.0 0.0 0.00 0.0 0.0

0 2 3599

1 2 -1.317 3.638 3.36 0.11 -1.0 6.0 11.41 0.0 0.0

2 2 -1.388 1.845 12.23 0.11 1.0 10.0 0.10 0.0 0.0

3 4 -0.044 5.646 79.40 335.20 0.0 0.0 1.63 0.0 0.0

4 4 -0.341 1.677 56.31 32.28 8.0 0.0 5.10 0.0 0.0

5 4 -3.391 5.279 55.44 30.84 20.0 0.0 1.11 0.0 0.0

6 4 -1.242 3.464 36.02 34.93 9.0 0.0 2.23 0.0 0.0

7 4 3.875 2.981 23.08 25.33 12.0 0.0 1.78 0.0 0.0

8 4 -2.934 0.093 11.33 2.15 21.0 0.0 6.17 0.0 0.0

9 4 -1.584 4.694 11.12 2.39 18.0 0.0 5.91 0.0 0.0

10 4 -1.716 1.913 9.09 2.20 12.0 0.0 0.90 0.0 0.0

0 3 3585

1 4 0.523 0.059 225.21 48.39 19.0 0.0 3.19 0.0 0.0

2 4 1.336 3.220 228.44 3.75 10.0 0.0 10.04 0.0 0.0

3 4 2.918 0.007 62.64 123.09 13.0 0.0 1.53 0.0 0.0

4 4 2.888 3.307 39.08 6.84 13.0 0.0 0.51 0.0 0.0

5 4 -3.432 6.037 13.55 13.69 4.0 0.0 3.54 0.0 0.0

6 4 -1.444 2.410 11.78 4.33 4.0 0.0 1.06 0.0 0.0

7 4 2.065 1.650 11.82 2.55 14.0 0.0 3.07 0.0 0.0

8 4 2.221 2.814 8.24 2.63 14.0 0.0 2.87 0.0 0.0

9 4 -0.738 1.730 7.79 2.45 2.0 0.0 1.02 0.0 0.0

Fabio MaltoniFabio Maltoni TASI 2013, Boulder CO

Possible double counting

7

Parton shower

Mat

rix

elem

ents

...

...

...

...

Poss ible double count ing between partons from matrix elements and parton shower easily avoided by applying a cut in phase space

Monday 10 June 2013

Documents

Tutorials on PYTHIA and MadGraphsusy.phsx.ku.edu/~kckong/KWS2014/KWS2014.pdf · PYTHIA and HERWIG and produce semi-realistic reconstructed physics objects such as photons, electrons,