A REFLECTION ON SOFTWARE ENGINEERING IN HEP F.CARMINATI CHEP 2012, NEW YORK, MAY.
1 Alice Experience with Geant4 F.Carminati 1, I.González 2, I.Hrivnacova 3, A.Morsch 1 for the...
-
date post
21-Dec-2015 -
Category
Documents
-
view
215 -
download
2
Transcript of 1 Alice Experience with Geant4 F.Carminati 1, I.González 2, I.Hrivnacova 3, A.Morsch 1 for the...
1
Alice Experience with Geant4F.Carminati1, I.González2, I.Hrivnacova3, A.Morsch1
for the ALICE Collaboration
(1CERN, Geneva; 2IFCA, Cantabria; 3IPN, Orsay)
Isidro González
Instituto de Física de Cantabria
CHEP 2003
La Jolla, 24 March 2003
2
Outline
ALICE Experiment Virtual MC & Geant4 VMC Hadronic benchmarks
– ALICE interest– Proton thin-target benchmark– Neutron transmission benchmark
G4UIRoot Conclusions
3
The ALICE collaboration includes 1223
collaborators from 85 different institutes from
27 countries.
online systemmulti-level triggerfilter out backgroundreduce data volume
level 0 - special hardware8 kHz (160 GB/sec)
level 1 - embedded processors
level 2 - PCs
200 Hz (4 GB/sec)
30 Hz (2.5 GB/sec)
30 Hz
(1.25 GB/sec)
data recording &
offline analysis
Alice collaboration
Total weight 10,000tOverall diameter 16.00mOverall length 25mMagnetic Field 0.4Tesla
Alice Experiment
5
Virtual MC and Geant4
Virtual MC advantages
Provides an interface to Monte Carlo programs
No coupling between the user code and the concrete MC
– The same user application may be run with several MCs
2 MCs already implemented:– Geant3– Geant4
ALICE effort is now concentrated on including also Fluka
Geant4 VMC
Built as a new package external to Geant4
A big effort has been done in order to minimize the limitations
The geometry part is based on G3toG4
– From Geant4 4.0 there is support for reflections
– Limited support for “MANY” Overlapping volumes have to
be specified explicitly (via G4Gsbool function)
Detailed information in the presentation from I. Hrivnacova: The Virtual MonteCarlo or http://root.cern.ch/root/vmc/VirtualMC.html
6
Geant4 VMC and ALICE
ALICE background event
HIJING parameterization event generator
5000 primary particles (5.8 % of full background event)
Modular physics list according to the physics list in G4 example N04 (electromagnetic and hadronic physics)
Included 12 detectors and all structures
– ITS coarse geometry (due not resolved MANY)
The kinetic energy cuts equivalent to those in G3 were applied in G4 using a special process and user limits objects
Standard AliRoot magnetic field map
Results
Finished successfully– Protection against looping
particles Hits for 10 (from 12) detectors.
Missing: – ITS (coarse version does not
produce hits)– RICH (requires adding own
particles to the stack – not yet investigated)
Comparisons of hits x, z distribution
No detailed analysis yet 2 to 3 times slower than
Geant3– Still preliminary
7
Geant3 and Geant4 VMC in ALICEHits in the TPC
Geant3 Geant4
8
Geant3 and Geant4 VMC in ALICEHits in the TRD
Geant3 Geant4
9
Hadronic benchmarks: Reasons
Low momentum particle is of great concern for central ALICE and the forward muon spectrometer because:
– ALICE has a rather open geometry (no calorimetry to absorb particles)
– ALICE has a small magnetic field– Low momentum particles appear at the end of hadronic
showers
Residual background which limits the performance in central Pb-Pb collisions results from particles "leaking" through the front absorbers and beam-shield.
In the forward direction also the high-energy hadronic collisions are of importance.
10
Proton Thin-Target Benchmark
Experimental and simulation set-up Conservation laws Azimuthal distributions Comparisons with data: Double differential
cross sections Conclusions
Note:Revision of ALICE Note 2001-41 with Geant4.5.0 (patch 01)
11
Proton Thin TargetExperimental Set-Up
Beam energies: 113, 256, 597 & 800 MeV Neutron detectors at: 7.5º, 30º, 60º, 120º & 150º Detector angular width: 10º Materials: aluminium,
iron and lead Thin target only
one interaction Data information from
Los Alamos in: Nucl. Sci. Eng., Vol. 102, 110, 112 & 115
12
Proton Thin TargetSimulation Set-Up
Physics
Processes used:– Transportation– Proton Inelastic:
G4ProtonInelasticProcess
2 sets of models:– Parameterised (GHEISHA):
G4L(H)EProtonInelastic– Cascade and Precompound:
G4CascadeInterfaceG4PreCompoundModel
The Cascade code is new and “fresh” since 5.0
Geometry Very low cross sections:
Thin target is rarely “seen” CPU time expensive
One very large material block: One interaction always takes place Save CPU time
Stop every particle after the interaction: Store its cinematic properties
15
Conservation Laws
Systems in the reaction:1. Target nucleus
2. Incident proton
3. Emitted particles
4. Residual(s): unknown in the parameterised model
Conservation Laws:1. Energy (E)
2. Momentum (P)
3. Charge (Q)
4. Baryon Number (B)
16
Conservation Laws in the Parameterised Model
The residual(s) is unknown It must be calculated
– Assume only one fragment
Residual mass estimation: – Assume B-Q conservation:
We found negative values of Bres
and Qres
– Assume E-P conservation Eres and Pres are not correlated
unphysical values for Mres
Aluminum is the worst case
Energy Q<0 B<0 Nneu < 0
113 MeV 0.00 % 0.00 % 0.00 %
256 MeV 0.38 % 0.02 % 0.44 %
597 MeV 0.77 % 0.00 % 0.90 %
800 MeV 1.20 % 0.00 % 1.50 %
17
Conservation Laws in the Cascade & Precompound Models
There were some quantities not conserved in the initial tested versions (Precompound alone)
Charge and baryon number are now conserved Momentum is not conserved.
– But it was exactly conserved in previous versions (Precompound alone)
– Can be up to 30 MeV
– It is correlated with: The target mass number: the smaller A, the bigger non-conservation The incident proton energy: Non-conservation increases with proton energy
– For Lead it shows a strange bump
Energy is not conserved:– Precompound alone had a small non-conservation width of the order of a
few MeV
– Now the width is bigger and shows spikes.
18
Momentum non-conservation in the Cascade Model
Lig
ht n
ucle
usH
eavy
nuc
l eus
Proton energy
19
Energy non-conservation in theCascade Model
Precompoundalone
Cascade &Precompound
20
Azimuthal distributions
What, how, why?
Known bug in GEANT3
implementation of
GHEISHA
Expected to be flat
Separated for and
nucleons
Results
distributions are correct! However…
Parameterised model:– At 113 & 256 MeV: No is
produced– At 597 & 800 MeV:
Pions are produced in Aluminium and Iron
(Almost) no is produced for Lead
Cascade & Precompound models:
– Are now able to produce
x
y
z
p
23
Double differentials
Real comparison with data
We plot
Which model is better?…– With Precompound alone it was difficult to say
– Now Cascade & Precompound are much better than
the parameterised models
– Still we see big discrepancies for low incident proton
energies and light targets
ΩE dd
d2
24
Double Differentials
Parameterised
Precompound
25
Double Differentials
Parameterised
Precompound
27
Double Differential Ratio Al @ 256
Parameterised
Precompound
32
Conclusions Proton
We found several bugs in GEANT4 during proton inelastic scattering test development– Most of them are currently solved.
The parameterised model cannot satisfy ALICE physics requirements
The Precompound model combined with the new Cascade model:– Improves a lot the agreement with data for the double
differential cross sections!– Is able to produce pions in the reaction– But… introduces a new energy-momentum
non-conservation!
33
Neutron Transport Benchmark
Experimental and simulation set-up Simulation physics Flux distribution Conclusions
Note: Linux gcc 2.95 (supported compiler) was used
Note2: It has not been redone with the latest Geant4 version
34
Tiara Facility
36
Simulation set-up
Incident neutrons energy spectra. – Peak at 43 and 68 MeV
Test shield material and thickness:
– Iron (20 & 40 cm)– Concrete (25 & 50 cm)
x = 0, 20 & 40 cm
y
x
401 cm
ExperimentalSimulated
ExperimentalSimulated
37
Simulation Physics
Electromagnetic: for e± and Neutron decay Hadronic elastic and inelastic processes for neutron,
proton and alphas– Tabulated (G4) cross-sections for inelastic hadronic scattering– Precompound model is selected for inelastic hadronic
scattering
Neutron high precision (E < 20 MeV) code with extra processes:
– Fission– Capture
1 million events simulated for each case
38
Preliminary Results: 43 MeVTest Shield: Iron – Thickness: 20 cm
44
Preliminary Results: 43 MeVTest Shield: Concrete – Thickness: 50 cm
45
Preliminary Results: 68 MeVTest Shield: Concrete – Thickness: 50 cm
46
Conclusions Neutron
The MC peak, compared to the data, is narrower an higher
Low energy disagreement:– Attributed by H.P. to backscattering due to so simple
geometry– Needs more investigation
Though the simulation does not match the data:– Iron simulation shows better agreement than
Concrete– For concrete lower energies seem better
47
G4UIRoot
A GUI for Geant4:– Built with ROOT
…providing:– an easy way to explore G4
command tree– a quick inspection of
standard/error output
A C++ Interpreter (CINT)– That may allow run time access to
G4 classes– That certainly allows access to all
ROOT functionallity
More info in:http://home.cern.ch/iglez/alice/G4UIRoot
48
G4UIRoot Features
Full Geant4 command tree displayed in a “file system” like structure
– Availability clearly marked– Non available commands are
identified and cannot be selected. – The availability is correctly
updated with Geant4 status Normal Geant4 command
typing is also possible– Selecting a command in the tree
will automatically update the command line input widget and vice-versa
– Automatic command completion using the TAB key
– The navigation through the successful commands executed before may be done using the arrow keys
Full and short command help
External Geant4 macros and ROOT TBrowser accessible through the menu
Customisable main window title and pictures
Different windows for error and normal output with saving capabilities
History window with saving capabilities.
– History is always tracked.– Successful commands may be
recalled at any point hitting the up arrow at the command line.
Root interpreter (CINT) included
– It runs in the terminal.– Will give run-time access to
Geant4 if it is CINTified
49
Final conclusions
ALICE has done a big effort to use GEANT4 It is already integrated in AliRoot through the Virtual MC framework But the PPR production will be done with Geant3– The effort is now concentrated on bringing Fluka into the VMC.
Concerning the hadronic benchmarks: We see and important improvement in the quality of the models
But it seems there is still space for more– Some more work needs to be done in ALICE:
Test EGPLs and contribute with plots/experience Improve the results from the neutron transport benchmark
The ALICE effort has contributed: To spot bugs/deficiencies in Geant4 Most of them already
corrected! To develop new functionality (reflections, G3toG4) In providing an easy and clear way to compare Geant3 and Geant4
(and soon Fluka) in big applications via de VMC