Paths to Geant4 evolution:
Refactoring, Reengineering and Physics
State-of-the-Art Simulation of
Photon Interactions with Matter
Maria Grazia Pia
M. Batič, M. Han, S. Hauf, G. Hoff, C. H. Kim, M. Kuster,
P. Saracco, H. Seo, G. Weidenspointner, A. Zoglauer
INFN Genova, Italy
J. Stefan Institute, Ljubljana, Slovenia
Hanyang University, Seoul, Korea
PUCRS, Porto Alegre, Brazil
EU XFEL GmbH, Hamburg, Germany
MPI Halbleiterlabor, Munich, Germany
Space Sciences Laboratory, UC Berkeley, USA
NSS 2012
30 October 2012, Disneyland, USA
Maria Grazia Pia, INFN Genova
“The code slowly sinks from engineering to hacking”.
“Refactoring is the process of changing a
software system in such a way that it does not
alter the external behavior of the code yet
improves its internal structure.”
“When you refactor you are improving the design of the code
after it has been written.”
Reengineering “seeks to transform a legacy
system into the system you would have built if
you had the luxury of hindsight and could have
known all the new requirements that you know
today.”
“A legacy is something valuable that you have inherited.”
“With rapid development tools and rapid turnover in personnel,
software
systems
can turn into legacies more quickly than you might imagine.”
Maria
Grazia Pia,
INFN Genova
Series of pilot projects
going on since 2008
Softwar
e
Physics
design
V&V
UQ
Focus on physics and
fundamental concepts
of particle transport
Highlights
No time to go into details
State-of-the-art, quantified
Archival
literature
Maria Grazia Pia, INFN Genova
Produce concrete
prototype
Quantify
current
future
Geant4 physics
capabilities
Identify
experimental
requirements
New physics
New performance
Code
Physic
s
V&V
Prune
Improve
Extend
physics
Maria Grazia Pia, INFN Genova
theory
exp.
MC
Assess the
state of the art
Refactor
(Re)design
software
Smells
If it stinks, change it.
Grandma Beck, discussing child-rearing philosophy
M. Fowler, K. Beck et al.,
Refactoring: Improving the Design
of Existing Code
Lazy Class
Duplicated Code
Speculative Generality
Long Method
Temporary Field
Large Class
Message Chains
Long Parameter List
Middle Man
Divergent Change
Inappropriate Intimacy
Shotgun Surgery
Alternative Classes with
Feature Envy
Different Interfaces
Data Clumps
Incomplete Library Class
Primitive Obsession
Data Class
Switch Statements
Refused Bequest
Parallel Inheritance Hierarchies
Maria Grazia Pia, INFN Genova
Electromagnetic smells
Coupling
stot and final state modeling
have been decoupled in hadronic
physics design since RD44
“model”
Total cross section
Whether a process occurs
Final state generation
How a process occurs
Dependencies
on other parts of the software
One needs a geometry
(and a full scale application)
to test (verify) a cross section
Difficult to test  no testing
often
Problem domain analysis
Improve domain decomposition
Maria Grazia Pia, INFN Genova
software design
TCrossSection
TGenerator
G4TRDPhotonProcess
Benefits
<<bind>>
Transparency
G4CrossSectionDataLib, G4GeneratorComptonDataLib
Ease of maintenance
<<typedef>>
Simplicity
of testing for V&V
G4ComptonDataLib
G4CrossSectio
<<bind>>
G4CrossSectionComptonStandard
Basic physics V&V can be
performed by means of
lightweight unit tests
<<bind>>
G4CrossSectionComptonPenelope,G4GeneratorComptonPenelope
Easier exploration of new models
Easier quantification of accuracy
<<typedef>>
G4ComptonPenelope
Bare-bone physical functionality
Decorations on top
Motion of atomic electrons
Doppler broadening
Binding effects
Maria Grazia Pia, INFN Genova
scattering
functions
Doppler
profiles
Vacancy creation
Atomic relaxation
Number one in the stink
M. Fowler,
Prune
parade is duplicated code Refactoring
physics
Objective quantification of smell
Efficiency w.r.t. experiment
Two Geant4 models, identical underlying
physics content (it used to be different)
stotal
5000
4000
600
“Livermore” Penelope
ds/dW
EPDL97
EPDL97
0.38±0.06
0.38±0.06
500
Counts
Counts
400
3000
2000
1000
s = 0.008 %
mean
7 10-5 %
300
200
100
s = 1%
mean
0.07%
0
0
−0.0004
−0.0002
0.0000
0.0002
sPenelope2008 - sEPDL
0.0004
sEPDL
Maria Grazia Pia, INFN Genova
−0.04
−0.02
0.00
0.02
dsPenelope2008 dsEPDL dsEPDL
(
)
dW
dW
dW
0.04
Code bloat
Burden on
• Software design
• Maintenance
• User support
Unnecessary complexity
Bremsstrahlung, evaporation, proton elastic scattering etc.
Trash
and redo
1.
2.
3.
4.
5.
6.
7.
Number one in the stink
parade is duplicated code
numbers
Bearden & Burr (1967)
Carlson
EADL
Sevier
ToI 1978 (Shirley)
ToI 1996 (Larkins)
Williams
Atomic{ binding energies
Carlson + Williams Carlson
( Shirley )
EADL
Source of epistemic uncertainties?
{
3246
IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 58, NO. 6, DECEMBER 2011
Evaluation of Atomic Electron Binding Energies
for Monte Carlo Particle Transport
23
pages
Maria Grazia Pia, Hee Seo, Matej Batic, Marcia Begalli, Chan Hyeong Kim, Lina Quintieri, and Paolo Saracco
p ionisation cross sections
difference w.r.t. experiment
electron emission, and precision simulation studies, such as microdosimetry, that involve the description of particle interactions with matter down to energies comparable with the scale
of atomic binding energies.
Simulation in these domains has been for an extended time
the object of specialized Monte Carlo codes; some general purpose Monte Carlo systems have devoted attention to these areas,
introducing functionality for the simulation of fluorescence,
PIXE (Particle Induced X-ray Emission) and microdosimetry.
In this context, emphasis has been placed on the development
and validation of the physics models implemented in the
simulation systems, while relatively limited effort has been
invested into verifying the adequacy of the atomic parameters
used by general purpose Monte Carlo codes with regard to the
requirements of new application domains.
This paper surveys atomic binding energies used by well
Index Terms—Geant4, ionization, Monte Carlo,
PIXE, simula-6
known Monte Carlo systems, including EGS [1], EGSnrc
tion, X-ray fluorescence.
[2], Geant4 [3], [4], ITS (Integrated Tiger Series) [5],
-8
MCNP/MCNPX [6], [7] and Penelope [8], and by some
specialized physics codes. These software systems use a vaI. INTRODUCTION
-10
riety of compilations of binding energies, which are derived
0
10
20
30
40
50
60
70
80
90
100
from experimental data or theoretical calculations; this paper
investigates
their
accuracy
and
their
effects
on
simulations.
HE simulation of particle interactions in matter involves
KL3 Eg w.r.t. experiment
Difference (eV)
other
10
Abstract—A survey of atomic binding energies used by general
purpose Monte Carlo systems is reported. Various
compilations of
8
these parameters have been evaluated; their accuracy is estimated
with respect to experimental data. Their effects on physical quan6 are highlighted:
tities relevant to Monte Carlo particle transport
X-ray fluorescence emission, electron and proton ionization cross
sections, and Doppler broadening in Compton
4 scattering. The effects due to different binding energies are quantified with respect
to experimental data. Among the examined compilations, EADL is
2
found in general a less suitable option to optimize
simulation accuracy; other compilations exhibit distinctive capabilities in specific applications, although in general their effects
on simulation
0
accuracy are rather similar. The results of the analysis provide
quantitative ground for the selection of binding energies to optimize the accuracy of Monte Carlo simulation-2in experimental use
cases. Recommendations on software design dealing with these parameters and on the improvement of data libraries
-4 for Monte Carlo
simulation are discussed.
Carlson Williams
EADL
Carlson
Maria Grazia Pia, INFN Genova
T
a number of atomic physics parameters, whose values affect physics models applied to particle transport and experimental observables calculated by the simulation. Despite the
mixed
Atomic number
II. COMPILATIONS OF ELECTRON BINDING ENERGIES
Algorithms
Popular belief
Physics model X is intrinsically slow
Baroque methods to combine it with “faster”
lower precision models and limit its use to
cases where one is willing to pay for higher
precision
This design introduces an additional computational
burden due to the effects of inheritance and the
combination algorithms themselves
Truth
Physics model X is intrinsically fast
But its computationally fast physics
functionality is spoiled by an inefficient
sampling algorithm
Maria Grazia Pia, INFN Genova
No code smell
► Spotted through
o in-depth code review
in the course of
software validation
►
Change the sampling algorithm!
The fastest algorithm
no algorithm at all
Shift modeling from algorithms to data
Maria Grazia Pia, INFN Genova
Merging models
Electron impact ionisation cross sections
Smoothing data
Guidance from
experimental data
(when available)
Data libraries
PIXE (sionisation: p)
BEB–DM (sionisation: e-)
Photon elastic
‒ RTAB (L. Kissel)
Example: LOESS local polynomial regression fitting
Beware: not optimized! Mathematical support in progress
Maria Grazia Pia, INFN Genova
experimental
SM
EPDL
10-8
10
-10
10
SM
EPDL
…no silver bullet
10-9
100
200
500
1000
2000
5000 10000
100
200
500
Energy (keV)
●
●
●●
●
●
●
●
1
●
●
●
●
-1
●
●
●
●
ds dW (b sr)
10
●
●●●●●
●
●●●●●
●●
●●●●
●●●
●
●●
●●●●
●●●
●●●●
●●●
●
●●●●
●●●
●
-3
●
●
●
●●
● ●
●
●
●
10
●
●
●●
●● ●● ●
-5
●●
10
-2
●
●
●
●
●●
●
●
●
Here our understanding of the
underlying physics phenomena
fails
10
●
-4
●
●
●
●●
● ● ●●●
●
●
● ● ●
● ●
●●
●●
●
●
10-6
10-7
●
10
Z=92
Smoothing
models (or data) that
exhibit significantly different
behaviour is not physically justified
●
●
●
ds dW (b sr)
10
5000
Energy (keV)
Z=82
●
●
1000 2000
-9
100
200
●
experimental
SM
EPDL
500
1000 2000
10
5000
Energy (keV)
Photon elastic scattering
cross section at 90°
Experimental data by Schumacher et al.,
Starr et al., Jackson et al., Moreh et al. etc.,
complete list in related publication
Maria Grazia Pia, INFN Genova
-8
100
experimental
SM
EPDL
Experimental data themselves must
be analyzed for inconsistencies and
200
500 1000 2000
possible5000systematics
Energy (keV)
How to solve this kind of problems?
Better theoretical calculations
New, dedicated experimental data
No easy solution
Refactoring data management
Today’s technology
‒ …keeping an eye on the new C++ Standard
Optimal container
Pruning data
Splitting files
Software design
Maria Grazia Pia, INFN Genova
Big refactoring
Well defined
responsibilities
and interactions
Maria Grazia Pia, INFN Genova
Geant4 Radioactive Decay
33Ba
photo-peaks
79.61 and 80.99 keV
Absolute validation
Experiment: Z. W. Bell (ORNL)
Statistical Decay
Validation
Po
218
84
3.1 m in
New
Figure algorithm
48: The upper part of the plot
j
i
ln( 2 )
ln( 2 )
_________
_________
2 6 .8 m in
3 .1 yr
0
0
ln( 2 )
ln( 2 )
_________
- _________
2 0 .0 m in
2 6 .8 m in
0
0
0
0
0
0
0
0
ln( 2 )
_________
1 .3 m in
0
0
0
ln( 2 )
ln( 2 )
_________
_________
0 .1 6 se c
2 0 .0 m in
0
0
ln( 2 )
_________
2 0 .0 m in
0
0
0
ln( 2 )
_________
0 .1 6 se c
0
0
0
0
-
Pb
214
82
2 6 .8 m in
Bi
Pb
210
82
2 2 . 3 yr
ln( 2 )
ln ( 2 )
_________
_________
2 2 .3 yr
1 .3 m in
0
ln ( 2 )
_________
2 2 .3 yr
Bi
210
83
5 days
Figure
Performance
Tl
210
81
1 .3 m in
Po
214
84
0 .1 6 se c
shows
the measured
Ba photo-peaks at
79.61 keV and 80.99 keV (dots). Also
shown are the background and peak
models fi tted to the data using HYPERMET. The background consist s of
a constant linear part (lower dashed
line) and a step-function modeling the
step of Compton component (upper
dashed line). The lower energy peak
is shown dotted, while the solid line is
the sum of the second peak and the
other component s. The low-energy exponential t ail is not dist inguishable for
these peaks. The lower part shows
the residuals in terms of σ uncertainties (solid area: 1σ , lines: 3σ ).
133
214
83
2 0 . 0 m in
Hg
206
80
8.1 m in
Figure 34: An illustration of how the Λ-matrix is fi lled using part of the 238 U decay chain as an example.
Geant4
refactored
for i > j and
Λ i i = − ki .
(35)
In Equation 34 bi j denotes the branching ratio from the jth to the i th component of the decay chain
n
with i = j+ 1 bi j = 1. An example is shown in Figure 34.
In computational practice Λ can be constructed by iterating through the decay chain and taking activation rates into account if necessary. Because Λ is independent of the actual nuclei numbers it must
only be constructed once for each set of nuclei and activations characterizing a given chain. Calculations
using different nuclei numbers can reuse these initial matrices as needed.
Using the above definitions the number of nuclei of each species in the decay chain in then given by
Maria Grazia Pia, INFN Genova
N( t ) = eΛt N(0)
More in
N28-6/7
(36)
new
algorithm
233 to the amplitude A
where α is a normalization factor
the photo-peak’s center; er f c(u) is the complementa
may vary with energy. To account for this the a poly
U decay
Paths to Geant4 evolution:
Refactoring, Reengineering and Physics
State-of-the-Art Simulation of
Photon Interactions with Matter
Maria Grazia Pia
M. Batič, M. Han, S. Hauf, G. Hoff, C. H. Kim, M. Kuster,
P. Saracco, H. Seo, G. Weidenspointner, A. Zoglauer
INFN Genova, Italy
J. Stefan Institute, Ljubljana, Slovenia
Hanyang University, Seoul, Korea
PUCRS, Porto Alegre, Brazil
EU XFEL GmbH, Hamburg, Germany
MPI Halbleiterlabor, Munich, Germany
Space Sciences Laboratory, UC Berkeley, USA
NSS 2012
30 October 2012, Disneyland, USA
Maria Grazia Pia, INFN Genova
Refactoring to improve
physics
Refactoring
M. Fowler, Refactoring
is the process of changing a software system in such a way that
it does not alter the external behavior of the code
yet improves its internal structure
Refactoring/reengineering legacy simulation code is an opportunity
•
•
•
•
To complement software testing with physics validation
To evaluate and compare the accuracy of various modeling options
To estimate the trade-off between accuracy and performance
To document the state-of-the-art for Monte Carlo transport
Electron-photon physics
Models and data libraries used by Geant4 and other
Monte Carlo codes
Maria Grazia Pia, INFN Genova
Photon elastic scattering
Physics
•
•
•
•
Rayleigh scattering
Thomson nuclear scattering
Delbrück scattering
Nuclear resonance scattering
Form factor approximation
“Livermore”
EPDL
“Penelope
2008”
“Penelope
2001”
“Standard”
EPDL
Baro et al.
Unknown
EPDL: non relativistic form factors
Thomson scattering in total cross
section only
• Non relativistic form factors
(Hubbell 1977) FLUKA, Geant4,MCNP, Penelope
• Relativistic form factors
nd order S-matrix
2
(Hubbell & Øverbø) EGS, ITS, XCOM
• Modified form factors
• Calculations by L. Kissel
(Schaupp et al.)
• Not yet exploited in Monte
• + anomalous scattering factors
Carlo codes
Maria Grazia Pia, INFN Genova
Z=13
Z=29
0.6
●
0.4
exp.
SM
EPDL
●
ds dW (b sr)
Heavy
materials
●
exp.
SM
EPDL
●
ds dW (b sr)
Forward
peaked
8
0.5
0.3
●
6
4
●
0.2
●
●
●
2
●●
●
0.1
●
●
●
●
●●
●
●
●
● ●
●●
●
●
●
●
●
●
0.0
5
10
20
●
50 100
Energy (keV)
500
●
●
●
0
●
2000
5
10
20
●
●
●
●
50 100
Energy (keV)
Z=50
●
500
2000
Z=82
50
150
●
●
●
●
40
●
●
●
ds dW (b sr)
●
exp.
SM
EPDL
●
30
100
20
exp.
SM
EPDL
●
●
ds dW (b sr)
●
●
●
●
●
50
●
●●
●
10
●
●
●
●
●
●●●
●● ●
0
5
Maria Grazia Pia, INFN Genova
10
20
●
●●
●●
●
●
●●
●●
●
●
●● ●
●●●
● ●●
●● ●
●
●
●
50 100
Energy (keV)
●
●
500
●
●
●
0
● ●
2000
5
10
20
●●
●
50 100
Energy (keV)
●
●
500
●
●
●● ●
2000
Comparison with experiment
> 4000 differential cross section measurements from the literature
5.41 keV ≤ E ≤ 39 MeV
0.5° ≤ θ ≤ 165°
69 target atoms, 3 ≤ Z ≤ 92
MFASF
300
RFASF
250
t-test
MF
RF
Counts
200
NF
150
Penelope2001
Penelope2008
100
EPDL
50
SM
smray
0
−15
−10
−5
0
5
10
15
−0.4
−0.2
0.0
0.2
0.4
Number of standard deviations
Confidence inter val
Statistical analysis of compatibility
Maria Grazia Pia, INFN Genova
+ categorical analysis
E=59.5 keV, Z=29
E=59.5 keV, Z=82
10.0
200
●
5.0
●●
●
●
●
●●
●
ds dW (b sr)
●
1.0
●
0.5
●
SM Rayleigh
●
●
20
●
●
●
●
● ●
10
●
●
5
●
●
●
●
●
●
●
●
●
●
0.2
● ●
2
●
●
0.1
0
100
10
50
●
experimental
EPDL
SM Rayleigh + NT
10
●
●●
●
10
●
●
●
●●
●
10
-3
●
●
● ●
●
●
●
●
50
100
Angle (degrees)
experimental
EPDL
SM Rayleigh + NT
●
●
●
● ●
●
●
●
●
-3
●
●
●
●
●
●
●
●
-4
●
●
●
● ●
●
●●
●
●
●
●
●
●
●
● ●
●●
●
●
●
●
●
●
●
10
0
150
●
10
●
●
●
10-2
●● ●
●
● ●
●
● ●
● ●
●
SM Rayleigh
●
●●
●●
●
●
●●
●
●
●
●
●
●●
●●
●
●
●
●●
●
●
-1
● ●
●
●
●
10
●
●
10
● ● ● ●
●●
100
●
●
●
●
●
1
SM Rayleigh
●
●
●
●
●
●
●●
50
●
●●
E=1.332 MeV, Z=82
●●
-2
0
ds dW (b sr)
ds dW (b sr)
150
●●
10
1
●
●
●
-1
● ●
E=661.6 keV, Z=82
●
●●●
Maria Grazia Pia, INFN Genova
●
●
●●
●
Angle (degrees)
●
1
● ●
●
●
●
●
Angle (degrees)
●
●●
●●
●●
●
●●
●
●
●
●
●
● ●
100
●
●
●
●●
●
●
●
●
●
Important to
account for it
especially at
higher energy
experimental
EPDL
SM Rayleigh + NT
50
SM Rayleigh
●
2.0
●
100
●
ds dW (b sr)
Nuclear
Thomson
scattering
experimental
EPDL
SM Rayleigh + NT
●
150
-5
0
50
100
Angle (degrees)
150
●
Model validation
State of the art
Identified through quantitative
validation w.r.t. experiment
Quantification
Compatibility with experiment: GoF test
(c2 test)
Comparison of models’ accuracy: categorical analysis
(Fisher’s exact test and c2 test)
Differential cross sections
Penelope Penelope
2001
2008
EPDL
Relativ. Non-Rel. Modified
FF
FF
FF
MFF
ASF
RFF
ASF
SM
NT
e
0.27
0.38
0.38
0.25
0.35
0.49
0.52
0.48
0.77
error
±0.05
±0.06
±0.06
±0.05
±0.06
±0.06
±0.06
±0.06
±0.05
e = fraction of test cases compatible with experiment
0.01 significance in the c2 test
S-matrix
Maria Grazia Pia, INFN Genova
Modified form factors
Z=50
Z=73
●
●
1
●
●
●●
●
10
●
●
●
10-2
●
●
10-3
●
10-4
●
10
●
●
●●
●
●●
●
-6
ds dW (b sr)
ds dW (b sr)
Other physics
to be
accounted for
at higher
energy
-1
●
●
10
-5
10
●
experimental
SM
EPDL
10-9
-10
100
200
500
1000
2000
5000 10000
100
200
500
Energy (keV)
●
●●
● ●
●
1
●
●
●
●
-1
●
●
●
●
ds dW (b sr)
10
●
●
●
●●●●●
●
●●●●●
●●
●●●●
●●●
●
●●●●
●
●●●
●●●
●●
●●
●●
●●●●
●
-3
●
●
●
●●
● ●
●
●
●
10
●●●
●
●
●●
●● ●● ●
-5
●●
●
10-9
100
10
200
1000 2000
Energy (keV)
●
●
●●
10
●
●
-4
●
●
●
●
● ● ●
● ●
●●
●●
●
●
-6
experimental
SM
EPDL
10-8
5000
●
●
●●
● ● ●●●
●
●
experimental
SM
EPDL
500
●
-2
●
10
10-7
Maria Grazia Pia, INFN Genova
5000
Z=92
●
ds dW (b sr)
10
1000 2000
Energy (keV)
Z=82
●
●
●
●
●
●● ●
●
●
10
●
●●
●
10-7
experimental
SM
EPDL
●
-8
●
●
● ●
100
200
500
1000 2000
Energy (keV)
5000
Z=47
●
2
1
300
●
2
1
400
500
600
4
3
●
●
2
●
1
●●
0
700
300
400
500
600
700
300
400
500
600
700
●
Energy (keV)
Energy (keV)
Z=53
Z=58
Z=64
10
4
3
2
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
6
4
●
Total cross section (b)
5
Total cross section (b)
Total cross section (b)
Energy (keV)
8
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
2
1
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
8
6
●
4
2
●
●
0
0
300
400
500
600
0
700
300
Energy (keV)
400
500
600
700
300
400
500
Energy (keV)
Energy (keV)
Z=73
Z=80
Z=69
600
700
15
10
8
●
6
4
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
10
●
Total cross section (b)
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
12
5
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
15
10
●
5
2
●
●
●
0
0
300
Maria Grazia Pia, INFN Genova
3
0
Total cross section (b)
Scarce
experimental
data
4
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
5●
●
0
Total cross section (b)
S-matrix
EPDL
MFASF
XCOM
Storm & Israel
Brennan & Cowan
Geant4 Standard
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
●
6







Z=50
6
Total cross section (b)
3
Total cross section (b)
SM
EPDL
MFASF
XCOM
Storm
Brennan
G4std
4
Total cross section (b)
Total
elastic
cross
section
Z=48
5
400
500
Energy (keV)
600
700
0
300
400
500
Energy (keV)
600
700
300
400
500
Energy (keV)
600
700
Photoelectric effect
Total cross section
Shell cross sections
Photoelectron angular distribution
Total cross section
EPDL
Kissel (RTAB)
Storm & Israel
Biggs & Lighthill
XCOM
Brennan & Cowan
McMaster
“Livermore”
EPDL
“Penelope 2008”
EPDL
“Penelope 2001”
EPDL
“Standard”
Biggs &
Lighthill
XCOM?
EPDL used down to 250 eV
Never validated, nor optimized
Maria Grazia Pia, INFN Genova
E > 1 keV
EPDL
Biggs
Brennan
Storm
XCOM
15
EPDL
15
10
Counts
Counts
Total cross
section
Difference
w.r.t.
experiment
E > 250 eV
5
0
10
5
-0.4
Penelope2008 - EPDL
-0.2
0.0
0.2
0
0.4
-0.4
-0.2
0.0
0.2
Relative difference
Relative difference
100 eV < E < 250 eV
E < 100 eV
0.4
300
40
EPDL
50
250
30
150
40
Counts
Counts
Counts
200
EPDL
20
30
100
20
50
10
10
0
-0.002
-0.001
0.000
0.001
Relative difference Penelope2008 - EPDL
0.002
0
-0.4
-0.2
0.0
0.2
0.4
0
-0.4
-0.2
0.0
0.2
0.4
Geant4
Relative difference
Relative difference
“Livermore” and “Penelope”:
Maria
Graziacross
Pia, INFN
Genova
Inadequate modeling or systematics in the measurements?
both
EPDL
sections
Photoelectric
effect
E = 661.6 keV
Total cross
section
10
8
exp.
p-valueEPDL
EPDL
RTAB
Biggs
Brennan
Storm
XCOM
6
4
30
20
10
0
0
●
1000
●
0.6
40
30
40
●
●
0.4
0.2
800
600
50
60
70
80
40
45
50
55
60
Atomic Number Z
Atomic Number Z
E = 123.6 keV
E = 59.54 keV
exp.
EPDL
RTAB
Biggs
Brennan
Storm
XCOM
exp.
p-valueEPDL
EPDL
RTAB
Biggs
Brennan
Storm
XCOM
3000
Total cross section (b)
exp.
EPDL
RTAB
Biggs
Brennan
Storm
XCOM
Total cross section (b)
Total cross section (b)
0.8
50
2
E = 2.62 MeV
E=2.62
MeV
●
exp.
EPDL
RTAB
Biggs
Brennan
Storm
XCOM
60
p-valueEPDL = 0.870
E > 50 keV
1.0
<0.001
Total cross section (b)
Total cross section (b)
12
E = 279.2 keV
2000
400
65
70
= 0.550
1000
●
●
200
0.0
50
55
60
65
70
75
Atomic Number Z
Maria Grazia Pia, INFN Genova
80
0
40
50
60
70
Atomic Number Z
80
0
20
30
40
50
Atomic Number Z
60
70
Z=2
Z = 10
0.4
0.3
0.2
0.1
0.10
0.15
0.20
0.25
3
1
0.10
0.15
0.20
Z = 18
Z = 36
0.5
0.11
0.12
E (keV)
0.25
3
1
0.09
0.10
0.11
0.12
E (keV)
Total cross section (Mb)
exp.
EPDL
RTAB
Brennan
30
20
10
0
0.2
0.4
0.6
0.8
E (keV)
Maria Grazia Pia, INFN Genova
1.0
Total cross
section
100 eV - 1 keV
2
Z = 54
40
0.30
exp.
EPDL
RTAB
Brennan
4
0
0.13
Photoelectric
effect
2
E (keV)
1.0
0.10
4
E (keV)
exp.
EPDL
RTAB
Brennan
0.09
exp.
EPDL
RTAB
Brennan
5
0
0.30
1.5
0.0
Total cross section (Mb)
0.5
0.0
Total cross section (Mb)
exp.
EPDL
RTAB
Brennan
0.6
Total cross section (Mb)
Total cross section (Mb)
6
0.13
Small experimental
sample
Noble gases
500 eV – 1 keV
p-valueEPDL = 0.593
250 – 500 eV
p-valueEPDL = 0.029
RTAB: similar to EPDL
Brennan-Cowan: less accurate
Z=1
2
Total cross section (Mb)
0.014
0.018
0.02
0.04
0.06
0.08
5
0
0.10
0.02
0.04
0.06
Z=8
Z = 10
Z = 18
0.06
0.08
600
400
200
0
0.10
exp.
EPDL
RTAB
Brennan
0.02
0.04
0.06
0.08
E (keV)
Z = 36
Z = 54
exp.
EPDL
RTAB
Brennan
50
40
30
20
10
0.06
E (keV)
0.08
0.10
80
0.10
exp.
EPDL
RTAB
Brennan
60
40
0.08
0.02 0.04 0.06 0.08 0.10
E (keV)
0.10
exp.
EPDL
RTAB
Brennan
40
30
20
10
0
0.02
0.04
0.06
0.08
0.10
E (keV)
Systematics in
experiment?
Inadequate
calculations?
20
0
50
Total cross section (Mb)
800
E (keV)
60
0.04
2
10
E (keV)
5
0.02
4
15
E (keV)
10
0.04
6
0
0.022
exp.
EPDL
RTAB
Brennan
0.02
8
exp.
EPDL
RTAB
Brennan
E (keV)
15
0
Maria Grazia Pia, INFN Genova
0.010
20
0
Measured
in gases
only
p-valueRTAB=0.977
exp.
EPDL
RTAB
Brennan
10
Total cross section (Mb)
4
Total cross section (Mb)
6
Total cross section (Mb)
10-100 eV
Total cross section (Mb)
Total cross
section
exp.
EPDL
RTAB
Brennan
8
0
Z=7
20
Total cross section (Mb)
Total cross section (Mb)
Photoelectric
effect
Z=2
Conclusions
A quantitatively validated, state-of-the-art
Monte Carlo code
Large effort
Supported by software design
a tremendous challenge
an inspiring motivation
Refactoring techniques and reengineering patterns contribute
to computational performance and facilitate validation
Quantitative validation of physics models
more about it in the last talk of this session
Physics insight is the key
Maria Grazia Pia, INFN Genova
…a big THANK YOU
to the
Maria CERN
Grazia Pia, Library!
INFN Genova