J. Apostolakis
PH/SFT
1

The Geant4 toolkit and SFT
◦ Areas of SFT involvement
◦ Status for LHC production
 open issues and ongoing actions
◦ Future perspectives

The LCG simulation project
◦ GENSER - Event Generator
◦ Physics Validation

Manpower
2
The Geant4 Toolkit
Areas of SFT involvement
Geant4 Reviews
3

Designed for HEP

Open Architecture and full capabilities
◦ Especially LHC and future experiments
◦ nuclear physics, heavy ion experiments,
◦ also medical physics, space application.
◦ Flexible kernel, enabling ‘any type’ of application
◦ Powerful geometry modeler
◦ Physics models for EM, hadronic, weak interactions
 Choices with different accuracy, CPU performance needs, ..
◦ OO used to decouple implementations of physics models,
geometry shapes & navigation methods, ..

Developed by the Geant4 collaboration
◦ About 90 physicists & computer scientists from HEP
institutions & agencies, ESA and universities around the
world.
4


Team in IT (1994-2002), moved to PH/SFT (2003)
Constant focus of Simulation/G4 team on LHC and HEP
◦ Driven by LHC use/issues, physics and capability needs
 From RD44(1994-98) – with all LHC experiments as alpha users,
 through the G4 production ramp-up phase (1999-2003) to today.
◦ Initiative on Test Beam Comparisons (2000)
 Later part of Physics Validation in LCG Simulation Project
◦ New models for physics accuracy - at best CPU cost possible
 Introduce new Physics models, then push for optimisation of CPU
◦ Developing functionality for additional use cases
 E.g. biasing, scoring & parallel geometry for (cavern) background
◦ Liaison and support on all issues encountered by LHC experiments

Geant4 in production use by ATLAS, CMS, LHCb since 2004
◦ The engine of the detector simulation
◦ Enabled use of Geant4 in ROOT Virtual Monte Carlo (VMC)
 for use by ALICE, others
5

Coordination, release preparation (staff)

Physics Validation (mix of staff/associates/fellow)

Physics Model improvement (mainly by associates)

Geometry (mix)
◦ G4 Spokesperson/SB Chair, Release manager/QA
◦ Thin-target benchmark data
◦ Hadronic: String models, cascade, CHIPS
◦ EM Physics: Standard (HEP)
◦ Navigation; solids; tracking in field; persistency.
In the above areas support and maintenance are included - and
are an increasing fraction – as well as validation, testing and
development.



Testing: nightly integration tests, regression & release validation
Hosting web site and development tools
Contribute also to other areas, e.g.
◦ Performance improvement – collaborating with FNAL experts, LHC teams
◦ Training courses; multithreading prototype studies, ..
6



Organized by the Geant4 Oversight Board
Followed Geant4 Review 2007
◦ Previous reviews: 2001 full, 2002 delta
Addressed issues relevant to HEP users, medical &
space communities
◦ Reviewers from Atlas, CMS, Alice, medical, space, hadronic
MC code. [ 6 from HEP / 9 total ]

Topics covered included:

Report made 22 recommendations
◦ EM & Hadronic physics, validation, computing performance,
documentation.
◦ 2009 actions underway
◦ Next actions to be formulated at Collaboration meeting (Oct
2009)
7
Overall Status
Open Issues & Ongoing Actions
8

Geant4 latest releases in production in ATLAS, CMS and LHCb

Enables complex geometrical descriptions

Robustness in production
◦ All ongoing/starting productions with Geant4 9.2 (Dec 08) + patches
◦ Agreement to support 9.2 until Dec 2010 for LHC experiment use.
◦ From 10 microns to 100 meters (detector) and more
◦ Most productions see event failure rates below 10-5
 Occasional issues tracked down and addressed promptly

Combining ever better physics and low CPU is hard

EM processes targeted for LHC/HEP are ready
◦ Experiments choose ever-better models for physics reasons
◦ So performance (CPU & memory use) was/is/will be a key challenge
◦ Extensively validated at percent level
 Benchmark data and experiment test beams
◦ Some models are being improved, refined
◦ Large suite of benchmark (including simplified versions of LHC
calorimeter)
9

Physics accuracy goals:

Hadronic Models have limitations and applicable energy
range
◦ Describe known thin target data and test beam data
◦ Predictive power for unmeasured regions.
◦ Our physics lists mix different models
Today’s physics list QGSP_BERT has transitions between:
◦ High-energy : string models (QGS)
QGSP_BERT(_EMV) physics list is used in
◦ Intermediate: parameterised (LEP)
production
◦ low energy: cascade (BERTini)
• By LHC (and other HEP) experiments
All feed into de-excitation models (Preco)
• Also tried/used in other
applications
Studies with other physics lists ongoing.
10

Accurate Modeling is important tool

Many or most observables are modeled well (or adequately)

Some quantities have deficiencies

Following slides illustrate some results & issues
◦ For calibration, energy scale, jets, missing ET , t
◦ Energy response (E>20 GeV) and p/e,
◦ Shower profile for pion projectiles.
◦ Resolution (typically smaller than data),
◦ Shower profile for protons
◦ Energy response (8-20 GeV).
◦ Concentrating on open issues and investigations
◦ Comparisons with data are taken from presentations by experiments at
LCG Physics Validation meetings (recent, preliminary results)
 Result of test beam comparisons of ATLAS, CMS – in close contact
11
Pion longitudinal shower profile in stand-alone
ATLAS TileCal test-beam at 90o
Thanks to Atlas Tilecal
Data
MC within ~ ±10% up to 10 λ.
For Protons : -(20%-40%) at 10 λ.
12
ATLAS Tile
Problem of matching
models:

CMS & ATLAS reported
energy response shows
unphysical features
◦ Kink (9 GeV)
◦ Change of slope (25 GeV)


These are the
transition points
between models
Reproduced in simple
setups
◦ No detector effects
13
p- Fe : p0
Study single interaction: pon Fe
 Study fraction of energy
going into different particle
types
◦ for each hadronic model

Identified microscopic
reasons for transition issues
◦ Which differences of models

Identify potential anomalies
◦ Confirmed known model
limitations & find new ones
14
p- Fe: p

Search for thin-target data
◦ To distinguish between models
◦ To enable tuning / improvement

Find models that need
replacing
◦ LEP disagrees with benchmark
data

Assess potential alternatives
◦ QGS/Chips smoother than
QGS/P
◦ FTF spans from 3 – 20+ GeV

Prepare Revised physics lists
◦ Seek feedback from LHC
experiments

Improve physics models
15
Future perspectives
16

LHC experiments needs during first years
◦ Expanded Support in most areas – especially physics
 Identify source (& fix) discrepancies with data
◦ Improvement of hadronic and EM modeling
◦ High Stability & Robustness for large MC productions
◦ Performance improvement (CPU, memory)

LHC upgrade, ILC detectors, other
◦ Refined hadron shower modeling
 Better lateral profile (ILC), ..
◦ Part of EUDET project since start (2006); CALICE contacts
 Comparison with CALICE data probe different aspects of
hadronic showering – resulting future improvements in models
can benefit LHC
◦ We are following these key (physics) issues
 And will continue to do so – to the degree that manpower will
permit.
17

Support/Maintenance
◦ In our main areas (Geometry, Physics) and beyond;
◦ Performance improvement

Physics Validation & Improvement
◦ Hadronic Physics: Models & Combinations (lists)
 Identifying limitations and their causes (major role)
 Contributing to improving models

Geant4 Architecture Review (start: 2010)
◦ Participate in planned major overhaul
 Address design issues, drive to improve implementation

Integration & interaction with other SFT
tools/projects
◦ Latest: migration to SPI nightly testing system (2008/9);
◦ Multi-threading and multi-core: Geant4 as proto-project.
18

Depth of expertise in high-energy nuclear physics
◦ Experts with 10s years of experience few
 Several working for other projects (e.g. MCNP),
 Limitations of associates contracts.
◦ Simulation codes / models
 Several projects focused on model improvement on ion-ion
(UrQMD) or cosmic rays (QGSJET)

Data for validation of diffraction, p0 production, ..
◦ Difficulty to access data of some old experiments
 Measurements of HELIOS (diffraction on nuclei),
 Published data apparently corrected with faulty MC;
 Search for data of p0 production (ANL?).

Increased load of maintenance
◦ Author/developer departures
◦ Reduced manpower
19

Developed relations with model authors, experts
◦ Liege cascade (INCL) team
 A. Boudard (Saclay) joined Geant4
 INCL ported to Geant4; developments ongoing
◦ D. Cano (CIEMAT): low energy neutrons
 New databases for neutrons E < 20 MeV

Recruited to Geant4, amongst others
◦ J.M. Quesada
Molina (Univ. Sevilla)
 Made full revision of pre-compound (the P in QGSP) &
evaporation

Interfaces to other models
◦ DMPJET II.5: first for ion-ion (space funding)
 Plan to extend it to hadron-ion
20

Regular contacts with LHC experiments using G4 in production
(ATLAS, CMS, LHCb)
◦ Many questions answered and issues addressed;
◦ Liaisons in Atlas (JA), CMS (GC): attend experiment simulation meetings, ..

Dialog with FLUKA team in context of Simulation project

Collaboration with IT

Physics Validation Meeting

Increasing channels with CALICE/ILC

Geant4 Open User ‘Technical Forum’ since 2003
◦ Focus on validation of models; physics issues.
◦ Extensive & fast testing of each Geant4 Release using WLCG (+OSG);
◦ Work with Openlab on Performance (sequential & multithreaded).
◦ Forum for Physics issues, validation including test beams
◦ EUDET(2005), CALICE meetings (Mar & Sep 2009), CERN group.
◦ Spreading information of experience by HEP experiments, other users
◦ Agrees priority on ‘big’ requirements, feeds into Geant4 workplan.
21

In close contact with many Geant4
contributing teams/people, e.g.
◦
◦
◦
◦
SLAC: hadronics (including neutrons), kernel, Vis
KEK/Japan: kernel, particle properties, GUI
FNAL: hadronics, performance improvement
IN2P3 & INFN: EM Standard(HEP+), EM Low-
Energy
◦ Northeastern Univ.: multi-core and multi-threading

These are in addition to formal channels:
◦ Steering Board
◦ Working group meetings
22
Brief Overview
Event Generator Service
23


Part of LCG Application Area that spans
simulation
Encompasses
◦ SFT involvement in Geant4 and its support for
experiments
◦ GENSER Event Generator Project
 Hosted T. Sjöstrand- Pythia 8 development.
◦ Physics Validation (Geant4, Fluka)
 cross-comparisons on thin-target data
 Joint effort on test beam comparisons
 Regular meeting including LHC (& other) experiments
◦ Garfield
◦ FLUKA – liaison with CERN Fluka team.

Led by Gabriele Cosmo
24

Provide:
• a repository of validated
MC event generators,
and
• related tools,
useful for the
experimental and
theoretical LHC
communities.
LCG Simulation Project

GENSER
◦ Structure stable and used by experiments (ATLAS, CMS, LHCb)
◦ 25 generators installed (most of them with several versions)

Platforms supported: SLC4 and SLC5, 32- and 64-bits

For some generators, also Windows and Mac OS X
◦ New versions are installed as they are announced and tested

regression testing with respect to a reference version)
◦ Status of the generators and testing available on the Web
◦ Project planning meetings twice per year


Regular monthly technical meetings
HepMC
◦ Standard interface used heavily by the LHC community
◦ Meetings every 6 months to decide new releases (latest: 2.05, June 09)

MCDB
◦ Used in CMS productions
LCG Simulation Project

Staff

Associates

Fellow(s)

Students
◦ JA: Geant4 spokesperson (Steering Board chair), geometry, hadronic
physics.
◦ Gabriele Cosmo: release manager, geometry coordinator, quality
assurance.
◦ Gunter Folger: hadronics, integration testing, software tool manager.
◦ Alberto Ribon (LD - Dec 09): SIM physics validation project, GENSER Event
Generator service, G4 hadronic physics, Grid release validation.
◦ Vladimir Ivantchenko: EM coord. (HEP focus) + Hadronics; CMS 2nd contact.
◦ Mikhail Kossov: Hadronic physics, validation & CHIPS model.
◦ Vladimir Uzhinskiy: Hadronic string modeling & validation
◦ Andrea Dotti (July 09-): Hadronic validation & physics (diffraction.)
◦ Gabriele Camellini: Geometry & FLUGG (Fluka with G4 geometry)
◦ Victor Diaz: System Integration Testing (incl. move to SPI nightly system)
◦ Mary Tsagri (doctoral/MC-PAD): Modeling of neutrons in Gas detectors
27

Profile 2005-2009
◦ Staff departures (not replaced)
 Retirement of Geant4 testing coordinator (Mar 2008)
 LD: Simulation/GENSER coordinator (Jan 2008)
◦ Sparser fellows
◦ Decline in number of associates

Upcoming changes (end-2009)
◦ Departure of A. Ribon (LD) - post was opened.
◦ End of several associates contracts.
28

Geant4 new releases in production in ATLAS, CMS
and LHCb
◦ Robustness, improved performance

Comparisons with benchmark data & LHC exper. test
beams
◦ Good agreement in many quantities, discrepancies in others
◦ Key issues are being investigated, in particular model
transitions


Ongoing improvement and preparation for support of
LHC experiments’ data-taking
Challenges anticipated
◦
◦
◦
◦
Enabling existing model authors to contribute closely
Enabling collaboration with additional physics experts
Finding key data for comparison
Address new needs of future experiments
29
30

Coordination, release preparation
◦ G4 Spokesperson/SB Chair (JA), Release manager (GC)

Hadronic Physics:
◦ String models: V. Uzhinskyi (PJAS: 2009-), GF
◦ Cascade: GF; CHIPS: M. Kossov (PDAS)
◦ Validation: AR/new staff; A. Dotti (fellow)

EM Physics:
◦ Coordinator: V. Ivantchenko (PJAS: 2005-2009)

Geometry
◦ Coordinator (GC), deputy (JA), G.Cam.(student) - had
fellow

Testing: integration tests, release validation
31

Between LHC and other HEP experiments
◦ Hadronic physics is the same and experiment needs
for good modeling is invariant
◦ Also high precision in EM physics is a challenge

Also with many other users
◦ Physics below 10 or 1 GeV is very relevant for
energy deposition, fluctuations, missing energy, ..
◦ Low-energy (<1 GeV) hadronic physics is common
 model improvement helps most/all users
◦ Need for CPU-performant models in common
energy domain is widespread (HEP, medical phys.,
..)
32

Issue of early short hadronic showers(2004):

Addressed: added physics model, improvements
QGSP
◦ Introduced cascade (BERTini) – and improved it
◦ Added Quasi-Elastic channel - to QGS model (2007)

Agreement is better with QGSP_BERT
◦ Pion shower profile within 10% at 10 lambda
◦ Proton profile 20-40% low at 10 lambda
 Looking to improvements in diffraction to address this

Develop(ing) Fritiof (FTF) model as alternative
◦ Better results for protons than QGSP_BERT
33
Proton longitudinal shower profile in stand-alone
ATLAS TileCal test-beam at 90o
MC: -20% to -40% at 10 λ.
34
Pion energy response in
ATLAS barrel combined test-beam
G4.9
Bertini cascade model
increases the response.
QGSP_BERT shows the
best overall performance
for the linearity (within 4%).
2% above 10 GeV
4% below 10 GeV
35

Reports from CMS and ATLAS
◦ Non-monotonic energy response 8-12 GeV
◦ Kinks in response in transitions of physics models

Actions
◦
◦
◦
◦
Confirmed in simple setups
Investigations of properties of physics models
Identified issues with particular models
Feedback to improve modeling
 comparing with thin-target data
36
Pion energy response in
ATLAS stand-alone test-beams
HEC
G4.9
Tile
Bertini cascade increases response for Tile and HEC by 4-5%.
In Tile and HEC response ~2% too high with QGSP_BERT.
37
Pions and protons energy response
in CMS combined test-beam
4.9.2.b01
ECAL+HCAL
ECAL(mip)+HCAL
Agreement between data and simulation on energy
response is within systematic uncertainty
38
p- Fe : p-

Study fraction of energy
produced in reaction
Leading Partic
◦ for individual model level
◦ Plot the sum of the energies
of the secondaries of one
particle type (e.g. p0 or
protons)
 < S Epi-zero > / Ebeam


p0 production responsible
for most early energy
deposition
Transitions in QGSP_BERT
◦ BERT to LEP between 9.5-9.9
GeV
◦ LEP to QGS/Preco btwn 12-25
GeV
Pi- beam on Fe (one
interaction)
39
p- Fe : p+
Some key observations /
issue(s)

G4 Bertini cascade (BERT)
◦ Has large excess of energy in
protons and neutrons for E>3 GeV
 Revision is underway (@ SLAC)

QGS/Chips smoother than
QGS/P
◦ Appears plausable for E>10 GeV

LEP disagrees with other models

Revised physics lists prepared

Need data to compare (spectra)
◦ Looking to replace it (with FTF)
◦ First feedback from LHC
experiments
◦ Especially those relevant to p0
production (e.g. p+ + p-
40
Pion resolution in ATLAS stand-alone test-beams
HEC
Tile
Bertini cascade makes resolution better:
in Tile: better agreement with data (±10 %).
in HEC: MC resolution too good by -10%.
41
Pion resolution in ATLAS barrel combined test-beam
Bertini cascade makes
resolution better
MC predicts too good
resolution -5 ÷ -10%
42
Pions and protons energy resolution
in CMS combined test-beam
ECAL+HCAL
Resolution is too good in Monte Carlo
ECAL(mip)+HCAL
43
Pion lateral spread in stand-alone
ATLAS TileCal test-beam @90o
Bertini cascade makes
shower wider, which is in
better agreement with
data, but data are still a
bit wider.
44
Test Beam Comparisons: Short
Summary and Outlook
 The LHC experiments have carried out extensive
tests of the Geant4 physics models and validated
them with test beam data.
 ATLAS and CMS have chosen QGSP_BERT(_EMV)
as the default Physics List.
Fritiof-based Physics Lists, FTF_BIC and FTFP_BERT
show interesting features.
 There are some remaining issues in hadronic physics
1) Discontinuity in energy response at the model boundaries
2) Proton longitudinal shower profiles are shorter than data in
QGS-based Physics Lists (diffraction)
3) Lateral shower profiles are a bit narrower than data
(not an issue for LHC experiments, but for ILC it could be
important….)
45
1.
Identifying source of
physics deviations
◦
2.
2.
Transitions between
hadronic models
3.
◦
4.
Responsive to tuning, or
Intrinsic model limitations?
Trying to match mean,
RMS, of diverse observables
Extending Validation
◦
Key
◦
More benchmarks,
observables
Issues:
ongoing
New data
Shower shape (p good,
proton fair)
◦
Deficiencies of hadronic
models
◦
◦
3.
Typically many sources
could contribute
1.
4.
Investigating diffraction
Simplification in older
modeling
◦
◦
LEP Energy non-conservation
Cascade nucleon production.
◦
Energy Response (Atlas, CMS)
Steps or ‘Discontinuities’

In region 8-12 GeV
Comparisons
With FNAL & ITEP data 3-10
GeV
◦
With fine grained CALICE
data.
Recent
aspects (for each)
◦
46

Made 22 recommendations
◦ Covered 8 different areas:








EM physics
Hadronic physics
Computing Performance
Physics Validation
Release Validation
Documentation
User Support
Resources.
1.
2.
3.
4.
5.
6.
7.
Removal of less accurate EM models/options (overall simplification of
choices where possible)
Provide guidance to users for available EM models and options
Study assumptions and parameterisations of physics models in FLUKA
and improve corresponding hadronic models available in Geant4, where
needed
Continue in monitoring CPU performance, perform code reviews for
improvement
Development of multi-threading capability in Geant4
Conduct a code design assessment in view to also integrate threadsafety in code
Consolidation of the hadronic WG and low-energy WG web pages
1.
2.
3.
4.
5.
6.
7.
Provide pointers to physics benchmark results and tests and related
documentation
Adopt a range of metrics to characterize as broad a range of validation
results as possible
Acquire/redirect resources to improve documentation and keep up with
updates
Adopt periodic review of the documentation and improve documentation
design
Identify new tools to adopt for the software installation
Involvement of the user community in the log-term support of physics
models
Seek for additional support from all available sources; extend the
Collaboration to new Institutions

Focus areas of Internal GEANT4 Review:
◦ Key interfaces
◦ Design of Geant4 kernel

Goals (proposed)
◦
◦
◦
◦
◦

Streamline design for features added 1998-2009
Further improve performance, robustness
Enhance maintainability, test-ability
Adapt for multi-core / multi-threading
Evaluate other recent / new technologies.
Effort to be in parallel with
◦ Support for existing Geant4 version(s) in
production.
50